American Journal of Public Health

May 1 2007, Volume 97, Issue 5 , pp. 780-957

LETTERS:

Gregory Todd Jones AGENT-BASED MODELING: USE WITH NECESSARY CAUTION Am J Public Health 2007 97: 780-781, 10.2105/AJPH.2006.109058. Igor Mezic, Paul J. Gruenewald, Dennis M. Gorman, and Jadranka Mezic MEZIC ET AL. RESPOND Am J Public Health 2007 97: 781-782, 10.2105/AJPH.2007.109710. ERRATUM:

ERRATA Am J Public Health 2007 97: 782, 10.2105/AJPH.2005.078121e. EDITOR'S CHOICE:

Farzana Kapadia Closing the Gaps Am J Public Health 2007 97: 783, 10.2105/AJPH.2007.112607. PUBLIC HEALTH THEN AND NOW:

Naomi Rogers Race and the Politics of Polio: Warm Springs, Tuskegee, and the March of Dimes Am J Public Health 2007 97: 784-795, 10.2105/AJPH.2006.095406.

FRAMING HEALTH MATTERS:

Jeremy Shiffman Generating Political Priority for Maternal Mortality Reduction in 5 Developing Countries Am J Public Health 2007 97: 796-803, 10.2105/AJPH.2006.095455. RESEARCH AND PRACTICE:

Jennifer Prah Ruger and Hak-Ju Kim Out-of-Pocket Healthcare Spending by the Poor and Chronically Ill in the Republic of Korea Am J Public Health 2007 97: 804-811, 10.2105/AJPH.2005.080184. Amanda Sacker, Richard D. Wiggins, Mel Bartley, and Peggy McDonough Self-Rated Health Trajectories in the United States and the United Kingdom: A Comparative Study Am J Public Health 2007 97: 812-818, 10.2105/AJPH.2006.092320. Richard T. Enander, Ronald N. Gagnon, R. Choudary Hanumara, Eugene Park, Thomas Armstrong, and David M. Gute Environmental Health Practice: Statistically Based Performance Measurement Am J Public Health 2007 97: 819-824, 10.2105/AJPH.2006.088021. Maria Argos, Faruque Parvez, Yu Chen, A.Z.M. Iftikhar Hussain, Hassina Momotaj, Geoffrey R. Howe, Joseph H. Graziano, and Habibul Ahsan Socioeconomic Status and Risk for Arsenic-Related Skin Lesions in Bangladesh Am J Public Health 2007 97: 825-831, 10.2105/AJPH.2005.078816. Robin Phinney, Sheldon Danziger, Harold A. Pollack, and Kristin Seefeldt Housing Instability Among Current and Former Welfare Recipients Am J Public Health 2007 97: 832-837, 10.2105/AJPH.2005.082677. Bruno Federico, Giuseppe Costa, and Anton E. Kunst Educational Inequalities in Initiation, Cessation, and Prevalence of Smoking Among 3 Italian Birth Cohorts Am J Public Health 2007 97: 838-845, 10.2105/AJPH.2005.067082. Joseph N.S. Eisenberg, James C. Scott, and Travis Porco Integrating Disease Control Strategies: Balancing Water Sanitation and Hygiene Interventions to Reduce Diarrheal Disease Burden Am J Public Health 2007 97: 846-852, 10.2105/AJPH.2006.086207. Zsigmond Kósa, György Széles, László Kardos, Karolina Kósa, Renáta Németh, Sándor Országh, Gabriella Fésüs, Martin McKee, Róza Ádány, and Zoltán Vokó A Comparative Health Survey of the Inhabitants of Roma Settlements in Hungary Am J Public Health 2007 97: 853-859, 10.2105/AJPH.2005.072173.

Marwan Khawaja and Rima R. Habib Husbands’ Involvement in Housework and Women’s Psychosocial Health: Findings From a Population-Based Study in Lebanon Am J Public Health 2007 97: 860-866, 10.2105/AJPH.2005.080374. Sônia Lansky, Elisabeth França, and Ichiro Kawachi Social Inequalities in Perinatal Mortality in Belo Horizonte, Brazil: The Role of Hospital Care Am J Public Health 2007 97: 867-873, 10.2105/AJPH.2005.075986. Mika Kivimäki, Debbie A. Lawlor, George Davey Smith, Anne Kouvonen, Marianna Virtanen, Marko Elovainio, and Jussi Vahtera Socioeconomic Position, Co-Occurrence of Behavior-Related Risk Factors, and Coronary Heart Disease: the Finnish Public Sector Study Am J Public Health 2007 97: 874-879, 10.2105/AJPH.2005.078691. Thomas W. Valente, Chich Ping Chou, and Mary Ann Pentz Community Coalitions as a System: Effects of Network Change on Adoption of Evidence-Based Substance Abuse Prevention Am J Public Health 2007 97: 880-886, 10.2105/AJPH.2005.063644. Coen H. van Gool, Gertrudis I.J.M. Kempen, Hans Bosma, Martin P.J. van Boxtel, Jelle Jolles, and Jacques T.M. van Eijk Associations Between Lifestyle and Depressed Mood: Longitudinal Results From the Maastricht Aging Study Am J Public Health 2007 97: 887-894, 10.2105/AJPH.2004.053199. Daniel Polsky, Sara J. Ross, Barbara L. Brush, and Julie Sochalski Trends in Characteristics and Country of Origin Among Foreign-Trained Nurses in the United States, 1990 and 2000 Am J Public Health 2007 97: 895-899, 10.2105/AJPH.2005.072330. Margaret A. Handley, Celeste Hall, Eric Sanford, Evie Diaz, Enrique Gonzalez-Mendez, Kaitie Drace, Robert Wilson, Mario Villalobos, and Mary Croughan Globalization, Binational Communities, and Imported Food Risks: Results of an Outbreak Investigation of Lead Poisoning in Monterey County, California Am J Public Health 2007 97: 900-906, 10.2105/AJPH.2005.074138. Kesha Baptiste-Roberts, Tiffany L. Gary, Gloria L.A. Beckles, Edward W. Gregg, Michelle Owens, Deborah Porterfield, and Michael M. Engelgau Family History of Diabetes, Awareness of Risk Factors, and Health Behaviors Among African Americans Am J Public Health 2007 97: 907-912, 10.2105/AJPH.2005.077032. Chantal Matkin Dolan, Helena Kraemer, Warren Browner, Kristine Ensrud, and Jennifer L. Kelsey Associations Between Body Composition, Anthropometry, and Mortality in Women Aged 65 Years and Older Am J Public Health 2007 97: 913-918, 10.2105/AJPH.2005.084178. Kathleen A. Cagney, Christopher R. Browning, and Danielle M. Wallace The Latino Paradox in Neighborhood Context: The Case of Asthma and Other Respiratory Conditions Am J Public Health 2007 97: 919-925, 10.2105/AJPH.2005.071472. Arleen F. Brown, Alfonso Ang, and Anne R. Pebley The Relationship Between Neighborhood Characteristics and Self-Rated Health for Adults With Chronic Conditions Am J Public Health 2007 97: 926-932, 10.2105/AJPH.2005.069443. Gilbert C. Gee, Jorge Delva, and David T. Takeuchi Relationships Between Self-Reported Unfair Treatment and Prescription Medication Use, Illicit Drug Use, and Alcohol Dependence Among Filipino Americans Am J Public Health 2007 97: 933-940, 10.2105/AJPH.2005.075739. Daniel J. Whitaker, Tadesse Haileyesus, Monica Swahn, and Linda S. Saltzman Differences in Frequency of Violence and Reported Injury Between Relationships With Reciprocal and Nonreciprocal Intimate Partner Violence Am J Public Health 2007 97: 941-947, 10.2105/AJPH.2005.079020. MARKETPLACE:

MARKETPLACE Am J Public Health 2007 97: 948-950. JOB OPPORTUNITIES:

JOB OPPORTUNITIES Am J Public Health 2007 97: 951-957.

 LETTERS

AGENT-BASED MODELING: highly stylized environment.5 Whether this lattice can be bordered or unbordered (i.e., USE WITH NECESSARY CAUTION teaches us about drinking behavior is an em- moving off the right side puts you on the left pirical question.4 Similarly, the “bar”3(p2058) is side) and conversion can be asynchronous or Agent-based modeling—a tool employing nothing more than a probabilistic sink that synchronous (all agents determine the new multiple interacting agents to reveal emer- guarantees to attract the “drinkers” who in no agent type before all agents convert together). gent properties of systems that are not prop- way “chose to spend a greater portion of their I replicated a family of models with NetLogo erties of the individual agents themselves1— time at this site”3(p2059) (emphasis added). 3.1.1 (Center for Connected Learning and has made significant in-roads into many Ecological validity and simple face validity Computer-Based Modeling, Evanston, Ill) to social science disciplines,2 including public are also important concerns. In one instance, determine the likely characteristics of Gorman health. I congratulate Dennis Gorman and Gorman et al. sought to identify ecological et al.’s models, with good results (Figure 1). his colleagues on their effort to leverage components of drinking behavior that they (The “bar” model runs 20000 periods in ap- these new tools to address specific public suggest might be thought of as a college pop- proximately 40 seconds on a 1.6 Ghz Pentium health concerns.3 However, given the novelty ulation that moves from a specific site with a processor; models are available from the au- of agent-based modeling methodology and given probability each day, operationalized thor upon request.) This replication also the resistance of those favoring traditional as a random walk.3(p2056) However, each of helped me to sort out the mathematical error analytic methods,4 great care in application is these students will have his or her own sink in Gorman et al.’s Figure 13(p2056) (The ordi- necessary to increase the likelihood of ac- in the form of an apartment or a dorm room nary proportion for conversion from suscepti- ceptance. In this spirit, I offer a few concerns. that he or she should return to (at least on bles to drinkers, Most important, it is imperative to be clear most days). Gorman et al.’s random-walk di() di() about what agent-based models have to offer. method quickly produces probabilistic = , si()++ ri () di () ti() Gorman et al. claim that their “models dem- homelessness. onstrate that the basic dynamics underlying Finally, precision is required in specifying shown on the state diagram, rather than dou- social influences on drinking behavior are models and reporting results. For example, a ble this proportion as specified in the body shaped by contacts between drinkers and focused by characteristics of drinking environments.”3(p2055) The offered models demonstrate no such thing. At most, the models provide insight into the spatial dynamics of agents following programmed rules that respond stochastically to other agents in a

Letters to the editor referring to a recent Journal article are encouraged up to 3 months after the article’s appearance. By submitting a letter to the editor, the author gives permission for its publication in the Journal. Letters should not duplicate material being published or submitted elsewhere. The editors reserve the right to edit and abridge letters and to publish responses. Text is limited to 400 words and 10 references. Submit online at www.ajph.org for immediate Web posting, or at submit.ajph.org for later print publication. Online responses are automatically considered for print publication. Queries should be addressed to 3 the Editor-in-Chief, Mary E. Northridge, PhD, FIGURE 1—Replication of Gorman et al.’s primary model: P=.3; γ=.3; ρ=.3 on an MPH, at [email protected]. unbordered lattice with synchronous conversions.

780 | Letters American Journal of Public Health | May 2007, Vol 97, No. 5  LETTERS 

Note. Drinkers probability of moving toward bar=0.5 and probability of moving away from bar=0.1. FIGURE 2—Replication of Gorman et al.’s3 “bar” model: P=.3; γ=.3; ρ=.3.

text of the diagram is what I believe was and Conflict Resolution, Atlanta, and the Computa- intended.) tional Laboratory for Complex Adaptive Systems, Atlanta. With this replicated model, I was also Requests for reprints should be sent to Gregory Todd able to investigate concerns about the au- Jones, PhD, MBA, MPA, JD, Georgia State University thors’ claim that with a “bar” on the lattice, College of Law, 140 Decatur Street, Atlanta, GA 30303 (e-mail: [email protected]). susceptibles leveled “off at a constant nonzero doi:10.2105/AJPH.2006.109058 value.”3(p2059) Analytically, this could not be the case. Indeed, the clustering of drinkers Acknowledgments makes the conversion of a susceptible on a G.T. Jones’s research was funded, in part, by the Wil- random walk a low-probability event, but the liam and Flora Hewlett Foundation. susceptibles do eventually go to zero (taking a mean of 10822 periods over 50 runs), a References characteristic missed because of incorrect as- 1. Flake GW. The Computational Beauty of Nature. Cambridge, Mass: MIT Press, 1998. sumptions of scale (Figure 2). For agent-based modeling to meet its 2. Bankes SC. Agent-based modeling: a revolution? Proc Natl Acad Sci U S A. 2002;99(suppl 3): promise, practitioners should resist the temp- 7199–7200. tation to overstate the implications of model 3. Gorman DM, Mezic J, Mezic I, Gruenewald PJ. outcomes, carefully design agent-based mod- Agent-based modeling of drinking behavior: els with an eye toward ecological validity, a preliminary model and potential applications to and provide precise specifications and results theory and practice. Am J Public Health. 2006;96: 2055–2060. based on families of models subjected to re- 4. de Marchi S. Computational and Mathematical peated tests. Modeling in the Social Sciences. New York, NY: Cam- bridge University Press; 2005. Gregory Todd Jones, PhD, MBA, MPA, JD 5. Bonabeau E. Agent-based modeling: methods and techniques for simulating human systems. Proc Natl Acad Sci U S A. 2002;99(suppl 3):7280–7287. About the Author 6. Lempert R. Agent-based modeling as organiza- The author is with the College of Law, Georgia State tional and public policy simulators. Proc Natl Acad Sci University, Atlanta, Ga, the Consortium on Negotiation U S A. 2002;99(suppl 3):7195–7196.

May 2007, Vol 97, No. 5 | American Journal of Public Health Letters | 781  LETTERS 

MEZIC ET AL. RESPOND

We would like to thank Jones for replicating our modeling efforts and identifying the typo- graphical error (not mathematical error) in the equation presented in the text of Figure 1 (where “=” is replaced by “+”).1 We are glad to see that his replication of the basic model and agent rules lead to the same conclusions drawn in our article. We also are pleased to have the opportunity to clarify some important points about the relevance of this modeling exercise to the very real problems communities experience with alcohol availability (whether through bars or other sources) and alcohol-related problems. The public health importance of these nascent modeling efforts should be underscored. We fully acknowledged that agent-based models represent “a simplified version of the real-world processes of interest,” and we accept that such simulations should be evaluated in the context of real data. Indeed, we were very careful to describe both the details and the limitations of our models and not overstate their potential implications. Further- more, our work since the article’s publication that uses data from several US cities demonstrates that contact with realistic situations can be established.2 The models we presented demonstrate the feasibility of modeling relationships between sources of alcohol and related problems with simple systems of agent-based rules. Our further work demonstrates that these simple rules, applied in 2 dimensions and reflecting the geographic distributions of outlets in community areas, may also predict the incidence of alcohol-related problems across community areas.3 Thus, we stand by our original statement that the “models demonstrate that the basic dynamics underlying social influences on drinking behavior are shaped by contacts between drinkers and focused by characteristics of drinking environments.”2(p2055) Contrary to the observation of Jones, our models do not produce “probabilistic homelessness.” Indeed, the timescale of a day is represented by 1 step. Thus, “visits to a bar” means only that an agent has been there on a particular day. The agent’s presence on that day implies that it might also be in the vicinity of (or visit) the same bar the next day, and so

May 2007, Vol 97, No. 5 | American Journal of Public Health Letters | 781  LETTERS 

we place agents in the neighborhood in the 2. Gorman DM, Mezic J, Mezic I, Gruenewald PJ. Annual Meeting of the National Institute on Alcoholism next time step. Of course the agents “return Agent-based modeling of drinking behavior: and Alcohol Abuse National Advisory Council; June 8, a preliminary model and potential applications to 2006; Bethesda, Md. home” every day, but it is not necessary for us theory and practice. Am J Pub Health. 2006;96: 4. Deuflhard P, Huisinga W, Fischer A, Schütte C. to model this process because it is not our in- 2055–2060. Identification of almost invariant aggregates in retention to capture effects of drinking at home. 3. Gruenewald PJ, Remer L, Mezic I, et al. Ecosys- versible nearly uncoupled Markov chains. Linear Alge- Finally, as Jones correctly points out, any fi- tem models of alcohol problems. Paper presented at: bra Appl. 2000;315:39–59. nite-state Markov chain with transition probabilities defined as in our models will ultimately have susceptibles decay to zero, with or without the presence of bars. However, on the timescales we were interested in (several years), the population levels off to a plateau. This plateau effect is indicative of the fact that in an infinite system, the susceptibles would not necessarily decay to zero; the effect represents a genuine metastable state in which the system is maintained for a long time. These types of systems and phenomena are well known and ubiquitous in dynamical systems theory.4 Thus, although Jones’s remark is tech- nically correct, it is misleading with regard to larger populations in which metastable states that maintain drinking problems may last far beyond the time scales of interest to public health (the time scale of his simulation, which has a mean of 10822 periods, is equivalent in the model to about 30 years).

Igor Mezic, PhD Paul J. Gruenewald, PhD Dennis M. Gorman, PhD Jadranka Mezic, MS

About the Authors Igor Mezic is with the Department of Mechanical and En- vironmental Engineering, University of California, Santa Barbara. Paul J. Gruenewald is with the Prevention Re- search Center, Berkeley, Calif. Dennis M. Gorman is with the Department of Epidemiology and Biostatistics, School of Rural Public Health, Texas A&M Health Science Center, College Station. Jadranka Mezic is with Aimdyn, Inc, Santa Barbara. Requests for reprints should be sent to Dennis M. Gorman, Department of Epidemiology & Biostatistics, School of Rural Public Health, TAMU 1266, College Station, TX 77843-1266. doi:10.2105/AJPH.2007.109710

Acknowledgement The modeling project described herein is funded by the National Institute on Alcohol Abuse and Alcoholism (contract ADM N01AA410012).

References 1. Jones GT. Agent-based modeling: use with necessary caution. Am J Pub Health. 2007;97:780–781.

782 | Letters American Journal of Public Health | May 2007, Vol 97, No. 5  LETTERS 

ERRATA In: Ferreira MU, da Silva-Nunes M, Bertolino CN, Malafronte RS, Muniz PT, Cardoso MA. Anemia and iron deficiency in school children, adolescents, and adults: a commu- nit-based study in rural Amazonia. Am J Public Health. 2007;97:237–239. doi:10. 2105/AJPH.2005.078121. A total was improperly reported. On page 237, the first sentence in the abstract is: “We investigated the prevalence and risk factors of anemia and iron deficiency in 389 rural Amazonians aged 5–90 years in Acre, Brazil.” A finding was improperly reported. On page 237, the second sentence in the Discus- sion section is: “However, because less than 20% of anemia in the population was attributable to iron deficiency, widespread iron supplementation alone is likely to have only a limited impact on the overall prevalence of anemia among people aged 5 years or older.” doi:10.2105/AJPH.2005.078121e

782 | Letters American Journal of Public Health | May 2007, Vol 97, No. 5 AMERICAN JOURNAL OF PUBLIC HEALTH

EDITOR-IN-CHIEF Mary E. Northridge, PhD, MPH DEPUTY EDITOR Farzana Kapadia, PhD  EDITOR’S CHOICE FEATURE EDITOR Gabriel N. Stover, MPA ASSOCIATE EDITORS Mary T. Bassett, MD, MPH Leslie Beitsch, MD, JD Felipe González Castro, PhD, MSW Michael R. Greenberg, PhD Sofia Gruskin, JD, MIA Closing the Gaps Said Ibrahim, MD, MPH Stewart J. Landers, JD, MCP Stella M.Yu, ScD, MPH I’m proud to join the editorial team of the perpetuate health disparities, such as the ASSOCIATE EDITOR FOR STATISTICS AND EVALUATION American Journal of Public Health in the position availability of trained health care professionals Roger Vaughan, DrPH, MS of deputy editor as we strive to put forth impor- at the national level and adequate health ser- INTERNATIONAL ASSOCIATE EDITORS tant research that elucidates the pathways that vice facilities at the community level. There is Kenneth Rochel de Camargo Jr, MD, PhD fuel health inequities. In addition, we aim to also a pervasive undercurrent through most of (Rio de Janeiro, Brazil) Daniel Tarantola, MD (Sydney,Australia) promote programs and policies as valuable tools the included papers of social inequalities that DEPARTMENT EDITORS John Colmers, MPH toward influencing global and national agendas influence the health and well-being of entire Government, Politics, and Law to reduce health inequalities. It is with these populations as well as individuals. Elizabeth Fee, PhD, and Theodore M. Brown, PhD goals at the forefront of our mission that this The Framing Health Matters contribution by Images of Health month’s issue is dedicated to the theme, “Health Shiffman deserves special mention as it seeks to Public Health Then and Now Voices From the Past Inequalities: Locally, Nationally, and Globally.” answer the question, “What factors influence Bernard M. Dickens, PhD, LLD, FRSC Providing rigorous scientific research on political leaders in developing countries to pay Health Policy and Ethics Forum health inequalities across complex settings attention to particular health causes?” Case Kenneth R. McLeroy, PhD, and and interwoven levels is challenging yet criti- studies in Guatemala, Honduras, India, Indone- Deborah Holtzman, PhD, MSW Framing Health Matters cal. Examples of disparities research from a sia, and Nigeria were then conducted. While EDITORIAL BOARD Neil Hann, MPH, CHES (2007), Chair variety of global, domestic, local, and personal each country had “unique political and socio- Hector Balcazar, PhD (2008) perspectives are evident in this issue. Thus, by economic circumstances,” Shiffman identified a Bobbie Berkowitz, PhD, RN (2008) examining the interrelatedness of these per- number of transnational and domestic factors Bonnie Duran, DrPhH (2008) spectives, not only can we gain a more com- that, when working in concert, placed greater Vanessa Northington Gamble, MD, PhD (2009) Alice M. Horowitz, PhD, MA (2009) prehensive understanding of how health dis- political value on reducing maternal mortality. Cassandra L. Joubert, ScD (2009) parities occur but we can also examine how Shiffman also cautioned that the “appearance Michael D. Kogan, PhD (2007) to best implement appropriate solutions. of an issue on a national policy agenda is only Marsha D. Lillie-Blanton, DrPH (2007) A decade ago I traveled to Cape Town, one among multiple factors that stands behind Christie Zunker Malpede, MA (2007) South Africa, and observed firsthand the con- policy effectiveness. . . . ” Notwithstanding the Allan Steckler, DrPH (2009) ditions that engender disparities. The challenges and complexities, this framework Henrie M.Treadwell, PhD (2008) apartheid government had been dismantled demonstrates how a better understanding of Terrie F.Wetle, PhD (2007) Lynne S.Wilcox, MD, MPH (2009) as recently as 1994, and the deplorable gaps the reasons for health disparities at multiple Siu G.Wong, OD, MPH (2007) in health status by race/ethnicity, socioeco- levels can bring a problem to national attention STAFF Georges C. Benjamin, MD, FACP nomic position, and region of the country and provoke action. Executive Director/Publisher were profound. The then newly elected gov- Together, the research published in this Ellen T. Meyer, Director of Publications ernment placed equality for all citizens at the issue honors one of the fundamental premises Nancy Johnson, MA, Executive Editor forefront of its political agenda. Nonetheless, of efforts to reduce health disparities. That is, Brian Selzer, Journal Production Manager Alexandra T. Stupple, Assistant Production Editor the social and economic hierarchies that the by working with individuals in affected com- Sarah Smith, Production Coordinator munities burdened by adverse social and eco- preapartheid legislation had created, and sus- Ashell Alston, Director of Advertising tained, continued to limit access to the most nomic conditions, we acknowledge their expe- Maya Ribault, Editorial Assistant basic necessities of life for the most disadvan- riences and wisdom in advancing their own Jennifer Strass, Graphic Designer taged segments of society. In the case of health so we can further close the gap be- Dana Jones, Reviews Coordinator South Africa, it was not a lack of political will tween the haves and have nots. Vivian Tinsley, Subscriptions Coordinator but a confluence of historical, social, and eco- FREELANCE STAFF Janis Foster, Greg Edmondson, Michele Quirk, Gretchen Becker,Alisa Guerzon, nomic factors that limited the ability of its citi- Farzana Kapadia, PhD Alison Moore, John Alexander, Beth Staehle, zens to attain health equity and responsibly Deputy Editor, AJPH Trish Weisman, Noelle Boughanmi, Copyeditors care for themselves and their own families. Alison Moore, Chris Filiatreau, Chrysa Cullather, Alexe van Beuren, Eileen Wolfberg, Proofreaders With this experience in mind, I viewed the doi:10.2105/AJPH.2007.112607 Vanessa Sifford, Michele Pryor, Graphic Designers current compilation of papers in this issue Aleisha Kropf, Image Consultant with attention to the various pathways that

May 2007, Vol 97, No. 5 | American Journal of Public Health Editor’s Choice | 783 PUBLIC HEALTH THEN AND NOW

Race and the Politics of Polio Warm Springs, Tuskegee, and the March of Dimes

| Naomi Rogers, PhD

The Tuskegee Institute opened a polio center in 1941, funded by the March of Dimes. training was intended to compen- O’Connor hinted at the political The center’s founding was the result of a new visibility of Black polio survivors and sate for the many hospitals in Al- calculations that had led to the the growing political embarrassment around the policy of the Georgia Warm Springs abama and across the region that opening of the center. During the polio rehabilitation center, which Franklin Roosevelt had founded in the 1920s be- refused to treat Black patients or 193 0s the systematic neglect of accept Black providers on their Black polio victims had become fore he became president and which had maintained a Whites-only policy of admis- staff. Orthopedic care at the publicly visible and politically sion. This policy, reflecting the ubiquitous norm of race-segregated health facilities Tuskegee Center, O’Connor embarrassing. Most conspicu- of the era, was also sustained by a persuasive scientific argument about polio itself: boasted, would help “hundreds of ously, the polio rehabilitation cen- that Blacks were not susceptible to the disease. children who might otherwise be ter in Warm Springs, Ga, which After a decade of civil rights activism, this notion of polio as a White disease was doomed to a lifetime of Roosevelt, himself a polio sur- challenged, and Black health professionals, emboldened by a new integrationist epi- crippling.”2 Yet from the outset, vivor, had founded, accepted demiology, demanded that in polio, as in American medicine at large, health care should the center, with only 36 beds and only White patients. This policy, be provided regardless of race, color, or creed. (Am J Public Health. 2007;97:784–795. limited outpatient facilities, could reflecting the ubiquitous norm of doi:10.2105/AJPH.2006.095406) help only a fraction of the patients race-segregated health facilities, who sought its care. Implicit in was sustained by a persuasive sci- O’Connor’s remarks, then, was an entific argument about polio it- IN MAY 1939, BASIL O’CONNOR, acknowledgment of what physi- self. Blacks, medical experts in- director of the National Founda- cian W. Montague Cobb would sisted, were not susceptible to this tion for Infantile Paralysis, an- excoriate a few years later as the disease, and therefore research nounced that the foundation enduring “Negro medical and treatment efforts that focused would provide $161350, its ghetto.”3 on Black patients were neither largest single grant, to build and The center opened with much medically necessary nor fiscally staff a polio center at the fanfare in January 1941, marked justified. Tuskegee Institute. The Tuskegee by a ceremony broadcast nation- It was true that few Black polio Infantile Paralysis Center, O’Con- ally on the radio. In a special victims were reported during the nor claimed, would “provide the message read at the event by 192 0s and 1930s, even with the most modern treatment for col- O’Connor, President Franklin growing number of epidemics in ored infantile paralysis victims” Roosevelt extolled the Tuskegee the Northeast and Midwest. And and would “train Negro doctors Center as “a perfect setting for a in the South, where the majority and surgeons for orthopedic hospital unit to care for infantile of Black Americans lived until work,” as well as orthopedic paralysis victims” based in “a hos- after World War II, there were nurses and physical therapists.1 pital completely staffed and di- few outbreaks of polio at all. Designing the center as a site for rected by competent Negro doc- Nonetheless, leaders of the Black both health care and professional tors.”4 Neither Roosevelt nor medical profession argued that

784 | Public Health Then and Now | Peer Reviewed | Rogers American Journal of Public Health | May 2007, Vol 97, No. 5 PUBLIC HEALTH THEN AND NOW

too many Black polio cases were to care for Black polio survivors vaccine trials and the subsequent missed as the result of medical and to train Black health profes- national vaccination programs.7 racism and neglect: families had sionals. During the late 1940s, in Thus, it was a calculated con- limited access to doctors and hos- response to civil rights activism vergence of politics, civil rights ac- pitals, and inadequately trained and Cold War race politics, the tivism, and philanthropy that first Black health professionals were foundation gradually began to use made minority polio survivors vis- unable to diagnose polio’s am- its funding to try to integrate ible, then affirmed that they de- biguous early symptoms. “I firmly training programs and health fa- served care from specialists (albeit believe,” Black orthopedist John cilities. Officials featured Black in separate institutions), and fi- Watson Chenault asserted, that boys and girls as March of Dimes nally, as the notion of polio as a the statistics used to argue for a poster children and boasted that White disease became disrep- lower incidence of disease among the foundation’s efforts were con- utable, transformed Blacks into Blacks “are due to the notoriously tributing to a breakdown in med- appropriate subjects of research poor treatment facilities available ical segregation. With this change and recipients of medical science for Negroes and[,] as much as I in philanthropy policy and the advances. This new politics of hate to admit it, the failure of so growing visibility of Black polio polio made Black and White bod- many of our men to recognize the cases, the science of polio also ies analogous and forced a new disease.”5 shifted and the theory of polio’s epidemiology. Clinical visibility Only after a decade of pressure racial susceptibility faded. Invigo- from Black activists was the no- rated by this integrationist epide- tion of polio as a White disease miology, civil rights activists de- effectively challenged. Funding manded that in polio, as in from the National Foundation for American medicine at large, Orthopedic care at the Tuskegee Center, Infantile Paralysis (popularly health care should be provided O’Connor boasted, would help ‘hundreds of called the March of Dimes) at first “regardless of race, color or supported separatist health facili- creed.”6 Black children were children who might otherwise be doomed to a ties such as the Tuskegee Center made part of the 1954 Salk “lifetime of crippling.’ 2 Yet from the outset, the center, with only 36 beds and limited outpatient facilities, could help only a fraction of the patients who sought its care.”

The Infantile Paralysis Center at the Tuskegee Institute, c. 1945. Source. March of Dimes Archives, White Plains, NY.

May 2007, Vol 97, No. 5 | American Journal of Public Health Rogers | Peer Reviewed | Public Health Then and Now | 785 PUBLIC HEALTH THEN AND NOW

informed social awareness of neg- The low numbers of paralyzed Even at the time, there was a lect, which in turn empowered Black children—in a 1924 De- powerful piece of evidence that larger claims for social justice and troit epidemic, for example, there could have argued against the medical equity. were only 5 in 300 cases10 —did race theory of differential sus- not surprise most medical ex- ceptibility: White Southerners, POLIO, A WHITE DISEASE? perts. More susceptible to some the other missing group in early diseases and impervious to oth- 20th-century polio epidemics. In the early decades of the ers, the constitutions of “primi- Without widespread electrifica- 20th century, when polio epi- tive” races were contrasted with tion or water filtration systems, demics first appeared in North the complex and delicate bodies the South had such a low level America, medical professionals of the “civilized” peoples of of sanitation (which led to mild and the lay public shared the con- Northern European heritage. The infant infection and widespread viction that Blacks were rarely in- most sophisticated expositor of adult immunity) that the region fected, whereas with Whites, this argument was Rockefeller saw no major polio epidemics polio was common and tran- Hospital clinician George Draper. until the late 1940s.14 Yet polio scended class lines. Epidemiolo- A polio expert who had been clearly did affect Black commu- gists initially saw wealthy White studying the disease since the nities. During the 1916 epi- polio patients as anomalies, best early 1910s and had directed demic, for example, health explained by an unlucky associa- Roosevelt’s own therapy in the officials in Maryland reported Rita Reed from Blue Island, Ill, the tion with infected members of the 192 0s, Draper believed that eu- more Black cases than White first African American March of urban poor.8 However, by the late genic factors were at the heart of cases but were unable to ex- Dimes poster child, 1947. 192 0s, polio was reconceived as polio’s confusing epidemiology. plain why.15 Source. March of Dimes Archives, “everyone’s disease.”9 In his 1917 text on polio, ex- Facing a segregated health sys- White Plains, NY. panded in 1935, Draper pointed tem and Black nurses and doc- to the “well-grown, plump” tors without specialized training White, native-born children with in polio, anxious families sought widely spaced upper front teeth out any expert who claimed and “delicate” teenagers who some scientific insight into the filled hospital beds and doctors’ disease. During the 1920s, offices during polio outbreaks.11 George Washington Carver, the The argument that a child’s sus- eminent agricultural chemist at ceptibility could be the result of a the Tuskegee Institute, began to constitutional makeup rather use a special peanut oil he had than an acquired immunity fit developed to massage a few well within broader professional young men whose muscles had and popular assumptions about been debilitated by polio. In the biological differences between early 1930s, he told reporters Blacks and Whites.12 When about his work, adding, “it has White physicians studied Blacks been given out that I have found as the “syphilis-soaked race,” for a cure. I have not, but it looks example, racialized genetics and hopeful.” Hundreds of people social psychology combined. wrote to him, and families More vulnerable to syphilis, less brought their paralyzed children susceptible to polio—both pieces to Tuskegee to ask his advice.16 of evidence were enlisted to es- In 1934, when a small group of tablish the pathological and alien Black Democrats decided to or- nature of the Black body.13 ganize a ball as part of Roo- Today, we would argue that sevelt’s polio fundraising, they poverty, poor hygiene and nutri- proposed naming the event after tion, and unequal access to Carver, “the Tuskegee scientist, health care could explain most whose peanut oil products . . . race disparities in health and have aided treatment for infantile disease in American history. paralysis.”17

786 | Public Health Then and Now | Peer Reviewed | Rogers American Journal of Public Health | May 2007, Vol 97, No. 5 PUBLIC HEALTH THEN AND NOW

WARM SPRINGS— Springs patient praised in 1932 as According to a whispering cam- A SEGREGATED POLIO a “most diverting person . . . tall, paign, polio had left him addicted REFUGE gaunt . . . with her sun hat and to drugs, so erratic that he re- bony hands,” expert in “pouring quired a straightjacket, and was By the time Roosevelt was olive oil on skins that have not incontinent, sexually impotent, elected to the White House in known the sun, adjusting bathing and helplessly crippled.23 1932, he had made polio—a dis- suits, fetching towels, wheeling One key ingredient in Roo- ease he claimed to have con- chairs . . . [and urging patients] sevelt’s New Deal strategy was to quered—his personal charity. Dur- ‘’Cose yo’ can learn to swim. . . . remake the Democratic Party by ing the 1920s, he bought Warm Fust thing yo’ knows yo’ll be going attracting new voters, especially Springs, then a run-down thermal right across the pool by yo’self. Blacks.24 During the 1936 presi- springs resort, and turned it into a Yas, ma’am.’”21 dential campaign, the Republican polio rehabilitation center, for- Roosevelt encouraged the Party sought prominent Blacks to mally incorporating it as the Geor- trustees of the Warm Springs combat claims that the Democra- gia Warm Springs Foundation, a Foundation to use him as a patron tic Party was the party of civil nonprofit company eligible for tax- for polio fundraising, and annual rights. Warm Springs policies be- free gifts.18 Birthday Ball campaigns began in came part of this debate, and dis- Warm Springs gradually be- 1934 (held annually on January tinctions between Roosevelt’s came a refuge for an elite group 30, Roosevelt’s birthday). At first New Deal programs and the re- of the disabled. Merging medical the funds were intended to create sort were frequently blurred. In a and social rehabilitation, the cen- a permanent endowment for radio speech supporting Governor ter gave polio survivors the op- Warm Springs. But gradually the Alfred Landon for president, Mar- portunity to recover physically Birthday Ball organizers redirected ion Moore Day, the daughter of and mentally. But from the begin- the money to the local communi- Frederick Randolph Moore, a for- ning, this refuge was for White ties that had raised it. The signifi- mer minister to Liberia, offered as survivors only. Typical of the po- cance of this philanthropic policy evidence of the Democrats’ neg- litical calculations by Roosevelt’s shift away from Warm Springs was lect the $100000 Blacks had New Deal administration, which not widely appreciated by the contributed to the Warm Springs fought the Depression with un- American public, however, and Foundation during the Birthday precedented federal funding while Warm Springs continued to be Ball celebrations, while “no Negro accepting racist policies that ac- viewed as a national polio center.22 children suffering from infantile commodated the Southern status paralysis can be admitted quo, Roosevelt was careful not to RACE, POLITICS, AND there.”25 Reverend J.S. Bookens challenge what he saw as Geor- POLIO of the African Methodist Episco- gia’s “local customs” and kept the pal Zion Church in Mobile, Ala, Warm Springs patients, adminis- Segregation at Warm Springs tried to have his paralyzed 9-year- trators, and medical staff White.19 became a source of political dis- old son admitted to Warm Springs Of course, Blacks lived and quiet when Roosevelt first ran for and was told “Negroes [are] never worked at Warm Springs as maids, reelection in 1936. The close admitted to that institution.” This waiters, body servants (assistants connection between the president case was widely discussed in the used to lift disabled patients), laun- and Warm Springs, a place he Black press and spurred Walter dresses, gardeners, and janitors. By continued to use as a personal re- White, secretary of the National the mid-1930s, 43 of 93 employ- treat, led those campaigning Association for the Advancement ees were Black. The center’s against him to scrutinize its ad- of Colored People, to remind White employees lived in the main mission policies as part of the Eleanor Roosevelt that segrega- hotel and the surrounding house- fierce campaign. There were ru- tion at Warm Springs was the rea- keeping cottages; Black employees mors that Warm Springs was a son his association refused to lived in a special “colored dormi- moneymaking scam for the De- sponsor Birthday Ball fund rais- tory” and in the servants’ room in mocratic Party and for Roosevelt ing.26 In a letter published in the the hotel basement.20 The sub- personally. The president was Chicago Tribune, one angry writer servience and hard work expected also said to have deceived the cited Warm Springs to show how of these employees was exempli- American people about the ef- little New Deal officials cared fied in Sarah, whom a Warm fects of polio on his own body. about racial justice:

May 2007, Vol 97, No. 5 | American Journal of Public Health Rogers | Peer Reviewed | Public Health Then and Now | 787 PUBLIC HEALTH THEN AND NOW

There is a place in Georgia duplicating existing orthopedic portrait of Lincoln to the wall.” named Warm Springs where the hospitals. We feel our national Prominent intellectuals formed President has endowed, or par- program is helping more colored tially maintains, a sanitarium for people than the putting up of what was known as the Black the treatment of infantile paraly- buildings and constructing of Cabinet, an informal but influen- sis. I have no doubt but what the pools which could service but a tial group of New Deal advisors, humblest, most ragged, and illit- small number.28 erate little white child in the land and Black doctors, nurses, and so- would be admitted there for As polio fundraising was recast cial workers gained positions as treatment, but the most cultured, as a national campaign, Warm administrators of some New Deal refined, and well clothed Negro child would be denied admit- Springs’ admission policy became programs, although usually with tance simply because it was a even more potent as a symbol of little say about the eligibility and 27 Negro. medical racism. “Since Negroes rights of their clients.31 are contributing . . . and since Eleanor Roosevelt and, less en- Declining to answer such they are sufferers of infantile thusiastically, her husband began charges publicly, “because of the paralysis,” a National Urban to urge the Warm Springs trustees intense heat of this year’s presi- League official told Eleanor Roo- to turn one of the center’s residential campaign,” Charles Irwin, sevelt, the Birthday Ball campaign dential cottages into a house for chief surgeon at Warm Springs, must include “their needs in the patients of color and to build prepared a 6-page response that general program of curing and them a small pool for hydrother- could be sent out in reply to preventing this dreaded malady,” apy. Just at this time the first for- private inquiries, explaining why a change that “would be heartily mal epidemiological analysis of welcomed by ten million other- polio and race appeared. Written Typical of the political calculations by Roosevelt’s wise socially disinherited Ameri- by Paul Harmon, a Chicago public New Deal administration, which fought the can citizens.”29 Warm Springs seg- health official, and published in Depression with unprecedented federal funding regation had become an awkward the prestigious Journal of Infectious reminder of the Roosevelt admin- Diseases, it provided a strong ar- “while accepting racist policies that accommodated istration’s reluctance to enforce gument for maintaining the status the Southern status quo, Roosevelt was careful not social justice and civil rights. quo at Warm Springs. to challenge what he saw as Georgia’s ‘local cus- “There has never been an ade- DEFENDING MEDICAL quate survey of the incidence of toms’ and kept the Warm Springs patients, SEGREGATION this disease among racial groups,” administrators, and medical staff White. Harmon announced, and he made Because it could be neatly but- the question of race central by an- tressed by statistics, the suscepti- alyzing the numbers of Black and Warm Springs did not “accept col- bility argument provided an expe- White and “Oriental” polio ored patients.” Irwin frankly as- dient way to counter the cases.32 Statistics from the 1916 sumed the need ”for racial segre- groundswell of protests against New York epidemic, showing a gation at most medical facilities. racism in polio care. Still, it was a Black morbidity rate of 241 per Warm Springs, he said, was “not a fragile defense, suggesting both 100 000 of population compared general orthopedic hospital. It that Black cases did not exist and with 383 for Whites, suggested treats and studies nothing but In- that building separate health serv- “that the disease is relatively infre- fantile Paralysis. It maintains no ices for them was too expensive. quent in the colored race.”33 Har- wards, separate clinics or segre- Such arguments were familiar mon also analyzed outbreaks in gated rooms. Aid and pay patients parts of what Cobb would de- the early 1930s in North Carolina, share the same facilities. We can- scribe as a policy of “secondhand Alabama, and Mississippi, which not take colored people for this hospitals” defended by White pro- showed a 2 to 4 times higher inci- reason.”28 ponents as “not new, but . . . bet- dence in the “white race,” and in Irwin’s second line of defense ter than anything [a Negro] . . . the 1930 San Francisco epidemic, was about money: has or can get now.”30 only 10 of 234 cases were Asian The Trustees deem it more wise In 1936, Roosevelt was re- Americans, supporting the suscep- to spend public monies received elected with an extraordinary tibility theory.34 for enlarging the national fight so 76% of Black voters who chose, Harmon did include some dis- that a greater number may be helped back home than to spend as one Black editor in Pittsburgh turbing counterevidence. Records such monies in huge buildings famously phrased it, to “turn the showed a higher incidence of

788 | Public Health Then and Now | Peer Reviewed | Rogers American Journal of Public Health | May 2007, Vol 97, No. 5 PUBLIC HEALTH THEN AND NOW

Black cases than White cases in already treated the local Black believed that only troublemakers Maryland in 1916 and in community. like “professional colored promot- Chicago in 1930 and a higher fa- Trustee A. Graeme Mitchell, a ers” were raising “this colored tality rate than Whites in New pediatrician who directed Cincin- question,” he admitted that “we York’s 1931 epidemic.35 All of his nati’s Children’s Hospital, was have a psychological problem statistics demonstrated significant convinced “that there is no which we probably should en- numbers of Black polio survivors, chance of having a cottage for Ne- deavor to meet to whatever ex- but he felt that the epidemiologi- groes at Warm Springs.” It was tent we are not meeting it now.”41 cal records were unreliable “for obvious, Mitchell explained, that In a similarly awkward de- the difficulty that the negro pre- Black and White patients could fense that acknowledged the ex- sents to statisticians is real.” With not swim in the same pool or istence of Black polio patients, tortured reasoning, Harmon sug- eat in the same dining room.39 business manager Henry N. gested that higher mortality rates Warm Springs Foundation Direc- Hooper and chief surgeon Irwin in New York might be the result tor O’Connor, Roosevelt’s long- insisted that the outpatient clinic of the “non-reporting of mild standing friend and adviser, based at Warm Springs was already re- cases among negros [sic],” which his reluctance on the susceptibility sponsive to the needs of local would mean there were even argument, because “it has always Black health professionals and more unrecognized Black cases. been my understanding that the their “colored orthopedic pa- “It is possible that opportunity colored race was not very suscep- tients.” “Where surgery is indi- for contact infection” rather than tible to this disease.” O’Connor cated,” Irwin noted, these pa- a “genuine racial immunity . . . drew the trustees’ attention to the tients were sent to “one of the is the determining factor in the special epidemiological analysis Atlanta hospitals for colored peo- data reported,” Harmon con- he had asked former Warm ple.”42 Hooper argued that the cluded, but his doubtful tone al- Springs physician Leroy Hubbard limited “susceptibility of the col- lowed readers to remain skepti- to complete, which concluded ored people” meant that only cal of such a conclusion and left that “the attack rate in the colored White patients could provide White guests and black waiters at differential racial susceptibility race is somewhat lower than in useful “material for education the Warm Springs dining room, c. uncontested as the best overall the white, indicating that there and for study by our orthopedic 1950. 36 43 explanation. may be a [sic] slightly less suscep- and physical therapy staffs.” In Source. March of Dimes Archives, The politics of race and polio tibility.”40 Although O’Connor any case, Hooper added, as “our White Plains, NY. were exacerbated in early 1937 when an article on George Wash- ington Carver in Reader’s Digest featured a photograph of Carver seated beside piles of letters from the parents of paralyzed children.37 Within weeks, in private letters, conversations, and memoranda, the White trustees of the Warm Springs Foundation and the Birthday Ball committee began to try to work out a solution.38 All the trustees disliked the idea of integrating Warm Springs, but many felt uncomfortable having to say so. Their discussions covered 3 issues: (1) the financial cost of installing separate facilities at Warm Springs, (2) the political cost of keeping Warm Springs White, and (3) the public relations risk of demon- strating that Warm Springs staff

May 2007, Vol 97, No. 5 | American Journal of Public Health Rogers | Peer Reviewed | Public Health Then and Now | 789 PUBLIC HEALTH THEN AND NOW

work to civilization compares if history is true, with that of Pas- teur, the eminent French chemist.”48 More sharply, a Chicago Defender article headed “We Donated, But They Left Us Out” claimed that the Warm Springs board of trustees, which “comprises some of the aristocrats of the South . . . took particular care not to include a Race institution in its research program” and had defended this policy by arguing that “the Negro should solve his problem with reference to this plague through local medical practitioners, because statistics show that it is most prevalent among White people.”49 Epidemiological statistics that were believed to prove polio’s differential racial susceptibility were being used to jus- Warm Springs movie theater interior, facilities do not lend themselves businessmen, and bringing med- tify medical segregation and to with white picket fence separating to the comfortable housing and ical racism into the public eye. deny Blacks legitimacy as both pa- White patients and staff from Black treatment of resident colored In August 1937, the announce- tients and scientists. employees, c. 1950. cases . . . we do not feel that we ment of the Birthday Ball’s expen- Source. March of Dimes Archives, White Plains, NY. could make such patients com- ditures for that year, with no men- THE TUSKEGEE SOLUTION fortable both physically and psy- tion of Tuskegee or any other chologically.”44 Black institution, outraged Black In September 1937, Roosevelt The trustees rejected the idea community leaders who had orga- announced the formation of the of a special cottage at Warm nized fundraisers as a way of both National Foundation for Infantile Springs for “colored victims” but fighting polio and showing their Paralysis to “direct and coordinate agreed that the solution should be newfound support for the Democ- the fight against this disease in all based on the familiar model of in- ratic Party. Letters to Roosevelt, its phases.” The new foundation stitutional separatism. Warm the Warm Springs trustees, and to (soon known popularly by its Springs officials were instructed to Paul de Kruif, secretary of the campaign slogan, the March of investigate “the possibility of our Birthday Ball’s Research Commit- Dimes) would be directed by establishing relations with an insti- tee and the nation’s most famous O’Connor, and the president tution already equipped for the science journalist, were reprinted would not “hold any official posi- care of colored people.”45 Perhaps and debated in the Black press.47 tion in it.” With Warm Springs at a new polio unit based in an De Kruif defended his committee’s trustees no longer in charge of na- established Black hospital that neglect of Carver’s polio work tional polio fundraising, the Geor- could train nurses, physical thera- with the unconvincing argument gia center could become just one pists, doctors, and brace makers, that Birthday Ball funds were of a number of regional centers “considerably more would be ac- available for prevention, not treat- caring for polio patients and train- complished, the results would be ment. “In these crucial hours,” ing specialists.50 obtained at a lower cost and criti- Chicago businessman James Hale A few months before this an- cism would both be avoided and Porter replied, “a scientific remedy nouncement, the Tuskegee Insti- unjustified.”46 A separatist solu- from a student of this disease is an tute’s Andrew Memorial Hospital tion would enable Warm Springs angel of mercy to suffering mil- had opened a small unit for dis- to avoid breaching the comfort lions,” adding, with a pointed ref- abled children and hired John level of current patients and staff, erence to de Kruif’s best-selling Chenault, one of the nation’s few angering local politicians and Microbe Hunters, “Dr. Carver’s Black orthopedic surgeons, as its

790 | Public Health Then and Now | Peer Reviewed | Rogers American Journal of Public Health | May 2007, Vol 97, No. 5 PUBLIC HEALTH THEN AND NOW

director. The 12-bed unit was a and the traditions of an unin- Cold War prestige abroad meant calculated step in a campaign by formed past.”56 that moderate civil rights mea- institute president Frederick Dou- The center quickly became a sures could potentially “make glass Patterson and Midian Oth- symbol of Black progress. Its na- credible the government’s ello Bousfield, director of the tional prominence enabled the argument about race and democ- Rosenwald Fund’s Division of staff to confront the theory of racy.”61 In 1944 when the March Negro Health. Buttressed by sup- polio’s racial susceptibility and to of Dimes funded a nonsegregated port from Reverend J.S. Bookens, make visible the neglect of dis- polio hospital during a particu- Mary Bethune, and other Black abled patients of color. In the larly fierce polio epidemic in activists, and building on public Journal of the National Medical As- Hickory, NC, a Black Baltimore, interest in Carver’s polio work, sociation, Chenault directly ad- Md, newspaper pointedly noted Patterson and Bousfield were qui- dressed the susceptibility argu- that if disease is color blind, etly urging O’Connor to consider ment with his own analysis of “there is cause to wonder why Tuskegee as the site for a polio evidence from Alabama and they can’t do the same thing the center.51 Georgia. Harmon’s work, he ar- year round.”62 National founda- In April 1939, Roosevelt made gued, had capably summarized tion executives began to argue his first official visit to the previous epidemiological knowl- that their funding was contribut- Tuskegee Institute. He praised the edge, but new evidence showed ing to a gradual breakdown in dis- beauty of the campus, visited the that Black patients’ unequal ac- crimination, including “notable re- Veterans Hospital and shook cess to care and their physicians’ versals of admission policies of hands with patients in wheel- inadequate professional training medical schools, associations, and chairs, and talked briefly with had skewed the statistical reports, hospitals.”63 A turning point came Carver.52 O’Connor came in May leaving “many features” of polio’s in September 1945 when Bynum to speak at the institute’s com- epidemiology “still debatable.”57 announced at the annual meeting mencement ceremonies, where he Meanwhile at the National of National Medical Association’s announced that the National Foundation for Infantile Paralysis, board of trustees that Black polio Foundation for Infantile Paralysis O’Connor hired university admin- patients were now being treated would fund the Infantile Paralysis istrator Charles Hudson Bynum to at Warm Springs.64 Center.53 The 3-story redbrick direct the foundation’s new inter- On the popular front, March of building, designed by Tuskegee ar- racial activities division.58 In Dimes organizers began to en- chitect Louis E. Fry, officially Negro Digest, Bynum reminded courage local chapters to inte- opened in 1941.54 “Like many readers that the National Founda- grate, arguing that “accurate statis- other institutions under the spon- tion provided funding “on the tics” now showed that “the sorship of the National Founda- same basis that polio strikes— disease attacked all races.” A few tion,” O’Connor proclaimed at the regardless of race, color or creed” Black boys and girls were chosen dedication, this new center would and that there was “no evidence for local, regional, and national make “a valuable contribution to of any racial susceptibility to the campaigns.65 These slow changes the solution of the problem,” and disease.”59 In the New York Times, in public relations and profes- headed by the “brilliant young Bynum claimed that March of sional training did not resolve specialist” John Chenault, it was Dimes funding enabled hospitals questions of access, however. “destined to become a great ortho- nationwide to treat Black polio Polio survivor Wilma Rudolph, fu- pedic center for the Negro people patients and Black physicians, ture Olympic track star who won of the United States.”55 But while nurses, and physical therapists to 3 gold medals in 1960, later re- arguing that polio was “notori- train at not only Tuskegee but called traveling 50 miles in a seg- ously no respecter of persons,” also Black hospitals in Nashville, regated bus with her mother from O’Connor simultaneously blamed Tenn; Durham, NC; and Chicago, her home in Clarksville, Tenn, to the plight of “twisted, distorted Ill. Training courses for polio be treated at Meharry’s polio and disabled children and adults” emergency volunteers were now clinic in Nashville.66 on the ignorance of Black parents also open to Black women.60 who did not “know when, and Integration occurred more fit- A NEW VISIBILITY how, and where” to secure “expert fully, but by the late 1940s, the medical care” and who had not fear that race discrimination at With the opening of the “abandon[ed] the superstitions home could damage America’s Tuskegee Infantile Paralysis

May 2007, Vol 97, No. 5 | American Journal of Public Health Rogers | Peer Reviewed | Public Health Then and Now | 791 PUBLIC HEALTH THEN AND NOW

physician speaking for the March Black scientists and technicians at Tuskegee’s Carver Research of Dimes called a Shreveport, La, Foundation produced the HeLa cells used for evaluating the epidemic unusual because “usu- 72 ally polio strikes blonde, blue- vaccine. After the results were announced, Bynum, eyed persons at a far greater “ 70 known in the Black press as ‘Mr Polio,’ boasted that rate.” Leading scientists, how- the results were a ‘triumph of racial cooperation.’73 ever, found the differential susceptibility theory less convincing. In 1946 Harry Weaver, newly ap- pointed as director of research at ” the National Foundation, discovered this when he wrote to virolo- gist Thomas Francis asking about research on race and polio: “I have been under the impression that most people believed that there was less poliomyelitis among Negroes than Whites.” No, Francis told Weaver, “the incidence by race . . . was essentially the same”; poor statistical collection had previously skewed the statistics and “there has been a tendency in the past not to seek out colored cases as well as White.”71 When Francis later directed the massive clinical trial of the Salk polio vaccine funded by March of Dimes, Black children were made part of the research, and Black medical leaders, in- March of Dimes official Charles H. Center and the prominent support opportunity to make places for cluding Chenault and Matthew Bynum accepting a check from Mrs of the nation’s largest disease phi- our best brains, our most capable Walker, president of the National J.A. Jackson, secretary of the Grand Chapter of the Order of the Eastern lanthropy, Black leaders had a hands, our most dynamic person- Medical Association, were invited Star of Virginia, December 3, 1955. platform to talk in general about alities, whether they be Negro or to attend the historic announce- Source. Afro-American Newspaper race and medicine, health care ac- white,” for “as a nation, we cannot ment ceremony at the University Archives and Research Center, cess, and the training of profes- continue to squander the abilities of Michigan. During the 1954 Baltimore, Md. sionals. March of Dimes money of our people without lessening vaccine trial, Black activists shored up Tuskegee’s financial our capacity for world leader- praised the unusual integration of troubles, and by 1948 O’Connor ship.”68 Highly visible to the audi- Black and White professionals as had become president of the insti- ence was the impressive transfor- “white nurses assisted Negro tute’s board of trustees. In No- mation that March of Dimes physicians in administering the vember 1950, to commemorate support had made in the renova- vaccine” and the way that “public the center’s 10th anniversary, tion and expansion of institute health officials, many of whom 300 people came to hear buildings, including a new nursing had never taken notice of Negro speeches by O’Connor, Chenault, school, nurses’ residence, and out- children in the community[,] su- Bynum, and Mrs Bettye Steele patient department.69 pervised the tests in person.” Turner, the head of Tuskegee’s George Draper’s arguments Black scientists and technicians at March of Dimes chapter.67 With about susceptibility, though, did Tuskegee’s Carver Research a flourish of Cold War rhetoric not vanish. In a 1951 story re- Foundation produced the HeLa O’Connor declared that “we must ported as “Polio Strikes Negroes cells used for evaluating the vac- continue to broaden the field of 1st in Louisiana,” a White cine.72 After the results were

792 | Public Health Then and Now | Peer Reviewed | Rogers American Journal of Public Health | May 2007, Vol 97, No. 5 PUBLIC HEALTH THEN AND NOW

announced, Bynum, known in the were no longer content to work care and professional training at Black press as “Mr Polio,” boasted with White business leaders to Tuskegee, funded by the March that the results were a “triumph support separatist hospitals of Dimes, ran in tandem with the of racial cooperation.”73 through what medical historian US Public Health Service’s This new science of polio, how- Preston Reynolds has called “well- Tuskegee syphilis experiment. ever, did not counter pervasive established patterns of civility.”76 But polio lacked the virulent customs of medical racism. Thus, The new civil rights movement racist connotations associated in May 1954, the month that the pervaded medicine, as physicians, with a sexually transmitted dis- Supreme Court announced its nurses, and other activists began ease. More than this, the senti- Brown v. Board of Education deci- to work for “the death of Jim mentalized appeal of disabled sion, Black children in Mont- Crow.”77 During a 1957 March of children, rather than adults fac- gomery, Ala, had to wait to re- Dimes campaign, gospel singer ing the so-called penalties of ceive their Salk shots on the front Mahalia Jackson refused to per- promiscuity, helped to make lawns of local White public form in a segregated hall and re- polio a more palatable issue for schools. They were forbidden to minded local organizers of the civil rights activism. use the restrooms inside.74 March of Dimes’ national policy, The fight to desegregate Warm Warm Springs remained segre- “which is dedicated to all people, Springs made visible the narra- gated for many years. By the end regardless of race.” She was not, tives of Black patients and em- of the 1940s it had set up a few she explained, “urging my people boldened claims for compassion “emergency” beds for local Black to turn their backs on the drive and equity.80 Although medical patients, but there were no Black against polio. I know what sick- practices remained infused with physicians, nurses, therapists, or ness is. I think race hatred is a racism, the turmoil allowed the administrators, and the Warm sickness too.”78 Amid the slow de- appalling inequities in access and Springs movie theater had an in- segregation of hospitals across the quality of care among disabled door picket fence indicating South, the Tuskegee Center lost minorities and the idea of polio as where Black employees could sit, its regional distinctiveness, and in a White disease to come under separate from the White patients 1975 it closed its doors.79 critical scrutiny. Under the leaky and staff, in the worst seats, the umbrella of Roosevelt’s pluralist two front rows.75 By the mid- CONCLUSION New Deal, although conspicu- 1960s, the widespread use of the ously without its proactive sup- Salk and Sabin polio vaccines led The struggle to desegregate port, Blacks were able to alter the to fewer polio cases and a less Warm Springs and to vanquish official record and assert the right critical need for rehabilitative racism in polio care and special- to an equal standard of health care. The Warm Springs center ist training was a difficult battle care. As the fight against Jim became primarily a tourist attrac- fought at a time when polio, Crow medicine grew more force- tion commemorating the legacy of race, and politics converged on ful in the 1950s, Blacks—now Franklin Roosevelt. In the 1980s, the national stage. During the recognized as susceptible to the however, with the emergence of 1910s and 1920s defenders of polio virus and as deserving recip- postpolio syndrome, the Warm medical segregation had argued ients of the advances of medical Springs center’s history of rehabil- that Blacks were simply not sus- science—waited eagerly for their itative expertise gave it new im- ceptible to the disease, but epi- children to receive the Salk vac- portance, and today the Roosevelt demiological statistics nonethe- cine. But all too often it was Warm Springs Institute for Reha- less revealed numbers of Black offered to children waiting on the bilitation includes a postpolio polio cases whose very exis- lawn outside the local White clinic and other programs for tence, activists pointed out, had school door. ■ brain injury rehabilitation, or- been kept invisible by medical thotics, and prosthetics. racism and neglect. By the Even before the national de- 193 0s, polio became a civil About the Author cline in polio cases, the Tuskegee rights issue that Roosevelt’s De- Naomi Rogers is with the Section of the Center began to represent an out- mocratic advisers and the direc- History of Medicine and the Women’s, dated symbol of medical accom- tors of the Warm Springs center Gender, and Sexuality Studies Program, Yale University, New Haven, Conn. modationism. By the mid-1950s could not ignore. The building of Requests for reprints should be sent to many Black health professionals a special center for rehabilitative Naomi Rogers, PhD, Section of the History

May 2007, Vol 97, No. 5 | American Journal of Public Health Rogers | Peer Reviewed | Public Health Then and Now | 793 PUBLIC HEALTH THEN AND NOW

of Medicine, Yale School of Medicine, PO with questions such as “could there be Edmund J. Sass with George Gottfried 22. See David L. Sills, The Volunteers: Box 208015, New Haven, CT 06520- something about race, something in your and Anthony Sorem, eds., Polio’s Legacy: Means and End in a National Organiza- 8015 (e-mail: [email protected]). genes, that makes you less—or more— An Oral History (Lanham, MD: Univer- tion (Glencoe, Ill: Free Press, 1957), likely to get polio?” (48) and “apparently sity Press of America, 1996). 42–43; Paul, History, 305–307; Lipp- the poliovirus has a peculiar affection for man, Squire of Warm Springs, 203–204. Acknowledgments 15. Harmon, “Incidence,” 333. Germanic fold and southern Europeans” 23. See George Wolfskill and John A. Thanks to David Arguss, Ted Brown, 16. “May See Paralysis Treatment,” New (49); Richard L. Bruno, The Polio Paradox: Hudson, All But the People: Franklin D. Elizabeth Fee, Meg Hyre, Beth Linker, York Times, October 31, 1933; “Peanut Uncovering the Hidden History of Polio to Roosevelt and His Critics, 1933–1939 Steven Mawdsley, David Rose, Emilie Oil Helps in Paralysis Cure,” Washington Understand and Treat “Post-Polio Syn- (London: Macmillan, 1969), 4–16; and Townes, John Harley Warner, and Daniel Post, December 31, 1933; “Mineral Oil drome” and Chronic Fatigue (New York: Lippman, Squire of Warm Springs, Wilson for their comments and sugges- Helps Cure Ailing Boys,” Los Angeles Warner Books, 2002). 64–67, 187–198. tions; to Marilyn Benaderet at the Afro- Times, January 1, 1934; “Cripples Beg 10.Paul H. Harmon, “The Racial Inci- American Newspapers Archives for her Cure of Negro Scientist,” Washington 24.Kevin J. McMahon, Reconsidering dence of Poliomyelitis in the United assistance in tracking down photographs; Post, January 7, 1934. On Carver’s polio Roosevelt on Race: How the Presidency States with Special Reference to the and to David Rose at the March of Dimes work, see Linda O. McMurry, George Paved the Way to Brown (Chicago: Uni- Negro,” Journal of Infectious Diseases 58 Archives for all his research help. Washington Carver: Scientist and Symbol versity of Chicago Press, 2004). (1936): 331. For an important analysis (New York: Oxford University Press, of race and the history of epidemiology 25. “Affront to Negro Laid to New 19 81), 242–55; and Rackham Holt, Human Participant Protection see Harry M. Marks, “Epidemiologists Deal,” New York Times, November 1, George Washington Carver: An American No human subject research was involved Explain Pellagra: Gender, Race and Polit- 193 6. Biography (Garden City, NY: Doubleday, in this study. ical Economy in the Work of Edgar 26. “Colored Help Warm Springs; Can’t Doran and Co, 1943), 299. For Carver’s Sydenstricker,” Journal of the History of Get Aid,” Chicago Tribune, October 15, description of the 2020 letters he had Medicine and Allied Sciences 58 (2003): 193 6; “Application of Race Lad to Warm References received during 1934, see Carver to Mrs 34–55. Springs Signal for Buck Passing,” Chicago Hardwick, December 16, 1934, 1. “Paralysis Center Set Up for Ne- Defender, October 24, 1936. See also 11.Draper quoted this description from reprinted in Gary R. Kremer, George groes,” New York Times, May 22, 1939. “Warm Springs Bars Negro Boy,” New his 1917 text Acute Poliomyelitis in his Washington Carver: In His Own Words York Sun, October 14, 1936; and Walter 2. Basil O’Connor, “Education in In- later, more popular book Infantile Paraly- (Columbia, MO: University of Missouri White to Eleanor Roosevelt, October 20, fantile Paralysis: An Address” (15 Janu- sis (New York: Appleton-Century, 1935), Press, 1987), 146. On Roosevelt’s tactful 1936, both quoted in Chappell and ary 1941), quoted in Edith P. Chappell 59; see also Paul, History, 161–66; responses to Carver, see David M. Oshin- Hume, “A Black Oasis,” 34–37. and John F. Hume, “A Black Oasis: Gould, Summer Plague, 70–73. sky, Polio: An American Story (New York: Tuskegee’s Fight Against Infantile Paraly- Oxford University Press, 2005), 66. 27. George W. Holbert, “Reply to Mr. 12. On theories of acquired immunity sis, 1941–1975” (Tuskegee University, Ickes,” Chicago Tribune, July 15, 1936. 19 87, unpublished), copy in March of during the 1910s, see Paul, History, 17. McMurry, Scientist and Symbol, 28.C.E. Irwin, “[Report on] Georgia Dimes Archives, White Plains, NY, 194. 131–33. On research projects (probably 253–54; “Negro Citizens to Give Ball inspired by George Draper’s work) for Roosevelt,” Washington Post, January Warm Springs Foundation, Warm 3. W. Montague Cobb, Medical Care funded by Rockefeller Foundation during 14 , 19 3 4 . Springs, Georgia,” October 20, 1936, and the Plight of the Negro (New York: the 1916 epidemic that included detect- President’s Secretary’s File, Subject: 18. “Georgia” was probably added be- National Association for the Advance- ing “physical characteristics,” see Paul, Warm Springs: 1936, Box 170, Franklin cause there were other well-known ment of Colored People, 1947), 6. History, 152. On the history of race and D. Roosevelt Presidential Library, Hyde warm springs linked to health resorts. On 4. “Statement at the Dedication of the disability, see Douglas C. Baynton, “Dis- Park, NY. Roosevelt and Warm Springs, see Hugh Infantile Paralysis Unit,” Atlanta Daily ability and the Justification of Inequality Gregory Gallagher, FDR’s Splendid De- 29. Jesse O. Thomas to Mrs Roosevelt Herald, February 6, 1940. in American History,” in Paul K. Long- ception (New York: Dodd, Mead, 1985); [abstr], January 25, 1938, President’s more and Lauri Umansky, eds., New Dis- 5. John W. Chenault, “Infantile Paraly- Dawn W. Houck and Amos Kiewe, Personal File 76, Warm Springs, Georgia, ability History: American Perspectives sis (Acute Anterior Poliomyelitis),” Jour- FDR’s Body Politics: The Rhetoric of Dis- 193 8, Franklin D. Roosevelt Presidential (New York: New York University Press, nal of the National Medical Association ability (College Station, TX: Texas A&M Library, Hyde Park, NY. 2001), 33–57. 33 (1941): 220–26, quote 221. Press, 2003); Theo Lippman Jr., The 30. Cobb, Medical Care and the Plight of 6. Charles H. Bynum, “Dimes Against 13. See, for example, Dorothy Roberts, Squire of Warm Springs: F.D.R. in Geor- the Negro, 20; see also “Voices From the Death,” Negro Digest 5 (1947), 82. Killing the Black Body: Race, Reproduction, gia, 1924–1945 (Chicago: Playboy Past: Medical Progress and African and the Meaning of Liberty (New York: Press, 1977); and Turnley Walker, Roo- Americans,” American Journal of Public 7. Jane S. Smith, Patenting the Sun: Pantheon, 1997); James H. Jones, Bad sevelt and the Warm Springs Story (New Health 92 (2002): 191–94. Polio and the Salk Vaccine (New York: Blood: The Tuskegee Syphilis Experiment York: A. A. Wyn, 1953). 31. Nancy J. Weiss, Farewell to the Party William Morrow, 1990), 273. (London: Free Press, 1981); Vanessa 19. See Ira Katznelson, When Affirma- of Lincoln : Black Politics in the Age of 8. See Naomi Rogers, Dirt and Dis- Northington Gamble, Making a Place for tive Action Was White: An Untold History FDR (Princeton, NJ: Princeton University ease: Polio Before FDR (New Brunswick, Ourselves: The Black Hospital Movement, of Racial Inequality in Twentieth-Century Press, 1983); McMahon, Reconsidering NJ: Rutgers University Press, 1992); John 1920–1945 (New York: Oxford Univer- America (New York: Norton, 2005), 42; Roosevelt on Race; Darlene Clark Hine, R. Paul, A History of Poliomyelitis (New sity Press, 1995); and Susan M. Reverby, Wilson, Living with Polio, 75–76. “Black Professionals and Race Conscious- Haven, CT: Yale University Press, 1971); “More Than a Metaphor: An Overview ness: Origins of the Civil Rights Move- To ny Gould, A Summer Plague: Polio and of the Scholarship of the Study,” in 20. Arthur Carpenter, “Special Report ment, 1890–1950,” Journal of American Its Survivors (New Haven, CT: Yale Uni- Tuskegee’s Truths: Rethinking the Tuskegee to Trustees: Georgia Warm Springs History 89 (2003): 1279–94. versity Press, 1995); Daniel J. Wilson, Study, ed. Susan M. Reverby (Chapel Hill, Foundation,” March 1, 1933, President’s Living with Polio: The Epidemic and Its NC: University of North Carolina Press, Secretary’s File, Subject: Warm Springs: 32. Harmon, “The Racial Incidence of Survivors (Chicago: University of Chicago 2000), 1–11. 1933, Box 170, Franklin D. Roosevelt Poliomyelitis,” 331. Presidential Library, Hyde Park, NY. Press, 2005); and Amy L. Fairchild, 14 . On White patients’ experiences of 33. Ibid. “The Polio Narratives: Dialogues with polio care in the 1930s, 1940s, and 21.Reinette Lovewell Donnelly, “Play- 34. Ibid., 332–36. FDR,” Bulletin of the History of Medicine 1950s, see Kathryn Black, In the Shadow ing Polio at Warm Springs,” Polio Chroni- 35. Ibid., 333. 75 (2001): 488–534. of Polio: A Personal and Social History cle 1 (1932): n.p. On Roosevelt’s per- 9. But see one contemporary argument (Reading, MA: Addison-Wesley, 1996); sonal efforts to give “Aunt Sarah” an 36. Ibid., 336. After the 1936 polio epi- against this “central dogma” of polio epi- Thomas M. Daniel and Frederick C. Rob- “old-age security” pension, see Walker, demic in Chicago, the governor of Illinois demiology, supporting “the power of bins, eds., Polio (Rochester, NY: Univer- Roosevelt and the Warm Springs Story, set up a new division for handicapped genes in determining susceptibility” (51) sity of Rochester Press, 1997); and 260–261. children with Harmon as its first director.

794 | Public Health Then and Now | Peer Reviewed | Rogers American Journal of Public Health | May 2007, Vol 97, No. 5 PUBLIC HEALTH THEN AND NOW

37. McMurry, Scientist and Symbol, Byrd and Linda A. Clayton, An American Treats Negro Polio Victims,” Chicago De- Archives; and see Elaine M. Strauss, In 253; James Saxon Childers, “A Boy Who Health Dilemma: Race, Medicine, and fender, September 8, 1945. My Heart I’m Still Dancing (New Was Traded for a Horse,” American Health Care in the United States, Rochelle, NY: self published, 1979), 65. Robert Darr, “The Deep South Magazine (1932), reprinted in Reader’s 1900–2000 (New York: Routledge, 90–91, cited in Fairchild “The Polio Speaks,” Louisiana Weekly, January 18, Digest 30 (1937): 5–9; and see “Ex- 2002); Hine, “Black Professionals and Narratives,” 531, fn. 211. 19 47; “Polio Grants Made to Train Ne- Slave Aids Paralytics,” New York Times, Race Consciousness,” 1279–94; and E. groes,” New York Times, September 26, 76.P. Preston Reynolds, “Hospitals and July 20, 1937. H. Beardsley, “Making Separate, Equal: 19 47; “Marching Against Polio,” Atlanta Civil Rights, 1945–1963: The Case of Black Physicians and the Problems of 38. Basil O’Connor to My Dear Mr Daily World, January 14, 1948. The Simkins v Moses H. Cone Memorial Hospi- Medical Segregation in the Pre-World President, March 3, 1937, President’s 19 47 poster girl “Negro child Rita Reed” tal,” Annals of Internal Medicine 12 6 War II South,” Bulletin of the History of Secretary’s File, Subject: Warm Springs: was a 5-year-old polio patient from Blue (1997): 898–906. Medicine 57(1983): 382–396. 1937, Box 171, Franklin D. Roosevelt Island, IL; Joe Willie Brown was a poster 77. Hine, “Black Professionals and Race Presidential Library, Hyde Park, NY. boy for the 1948 campaign. 52. “Roosevelt Thrilled by Tuskegee Consciousness,” 1294; and see Gamble, 39. “Memorandum of replies re ques- Choir”; “President Meets Dr. Carver, 66.Wilma Rudolph, Wilma (New York, Making a Place for Ourselves, and P. Pre- tion of a cottage for negroes at Warm Tuskegee Wizard,” Pittsburgh Courier, NY: Signet Books, 1977), cited in Wil- ston Reynolds, “Professional and Hospital Springs,” [enclosed with] O’Connor to April 8, 1939. son, Living with Polio, 146. Discrimination and the US Court of Ap- My Dear Mr President, March 3, 1937. peals Fourth Circuit 1956–1967,” Amer- 53. Chappell and Hume, “A Black 67.“Tuskegee to Hold Polio Confer- ican Journal of Public Health 94 (2004): 40. “Memorandum: Poliomyelitis, in re- Oasis,” 42. ence,” New York Times, November 26, 710–720. lation to white and colored populations 54. “Statement at the Dedication of the 1950. in U.S.,” Leroy W. Hubbard to Basil O’- 78.“Race Bias Stops Mahalia Jackson,” Infantile Paralysis Unit”; “Infantile Paraly- 68. “Negro Leaders’ Contribution in Connor, March 9, 1937, [enclosed in] Chicago Defender, January 19, 1957. sis to Be Treated Here,” Atlanta Daily Field of Health Praised by Basil O’Con- Basil O’Connor to Dear Mr President, Herald, February 6, 1940. nor, President of National Foundation for 79. Chappell and Hume, “A Black March 10, 1937, President’s Secretary’s 55. O’Connor, “Education in Infantile Infantile Paralysis,” Globe and Indepen- Oasis.” File, Subject: Warm Springs: 1937, Box Paralysis,” 196–97; see also O’Connor’s dent (Nashville), December 1, 1950; “O’- 171, Franklin D. Roosevelt Presidential 80.On the dynamics of civil rights and address quoted in “New Polio Center to Connor Praises Contribution of Negro Library, Hyde Park, NY. clinical visibility, see Keith Wailoo, Dying Leaders to Health,” Montgomery (Al- Aid Negro Paralysis Victims,” Macon in the City of the Blues: Sickle Cell Anemia 41. O’Connor to My Dear Mr President, abama) Advertiser, November 28, 1950. (Georgia) Telegraph, June 25, 1939. and the Politics of Race and Health March 3, 1937. 56. O’Connor “Education in Infantile 69. Kimberly Ferren Carter, “Trumpets (Chapel Hill, NC: North Carolina Univer- 42. “Memorandum of replies.” Paralysis,” 194, 195; see also “‘Do Ye of Attack: Collaborative Efforts between sity Press, 2001). 43. Henry Hooper to Dear Mr O’Con- Also Unto Them,’” Black Dispatch (Okla- Nursing and Philanthropies to Care for nor, May 8, 1937, President’s Personal homa City), January 18, 1941. the Child Crippled with Polio 1930 to 1959,” Public Health Nursing 18 (2 001): File 76, Warm Springs, Georgia, Franklin 57. Chenault, “Infantile Paralysis,” 221. 254–61. D. Roosevelt Presidential Library, Hyde See also Chenault’s speech at the ground Park, NY. breaking ceremony for the Tuskegee 70.“Polio Strikes Negroes 1st in 44. “Memorandum of replies.” Center in January 1940, on “recent sur- Louisiana,” Washington Post, August 21, veys” which showed “no appreciable 1951. 45. O’Connor to My Dear Mr President, variation between the white and Negro May 11, 1937. 71. Harry Weaver to Thomas Francis, races”; “Infantile Paralysis Center December 6, 1946; Francis to Weaver, 46. Hooper to O’Connor, May 8, 1937. Launched at Tuskegee,” Birmingham (Al- December 10, 1946; quoted in Oshin- abama) News, January 12, 1940. 47. James Hale Porter to Franklin De- sky, Polio, 66–67. lano Roosevelt, August 3, 1937 [letter 58. “Negro Polio Victims Gets Founda- 72. “Salk Vaccine Effectiveness Is Trib- reprinted] in “What The People Say,” tion Aid,” New York Times, December 21, ute to Cooperation,” (Baltimore) Afro- Chicago Defender, August 14, 1937. 19 45. American, April 23, 1955; R.W. Brown 48. “Urges Nation Use Dr. Carver in In- 59. Charles H. Bynum, “Dimes Against and J. H. Henderson, “The Mass Produc- fantile Paralysis Fight,” Chicago Defender, Death,” Negro Digest 5 (1947), 82. tion and Distribution of HeLa cells at the September 25, 1937. Tuskegee Institute, 1953–55,” Journal of 60. “Negro Polio Victims Aided in All the History of Medicine and Allied Sciences 49.“We Donated, But They Left Us States,” New York Times, December 18, 38 (1983): 413–431. Out,” Chicago Defender, August 7, 1937. 1946; “Negro Groups Aided by Polio 73. Enoc P. Waters, “Salk Vaccine Is 50. Franklin D. Roosevelt, “Foreword,” Foundation,” New York Times, January 9, Triumph of Racial Cooperation,” Chicago Annual Report of Georgia Warm Springs 1949. Defender, April 23, 1955. Foundation 1940, Disability History Mu- 61. Mary L. Dudziak, Cold War Civil seum, http://disabilitymuseum.org/ Rights: Race and the Image of American 74. Bea Wright in McCutcheon Tran- lib/docs/2168.htm (accessed January 31, Democracy (Princeton, NJ: Princeton Uni- script, quoted in Smith, Patenting the Sun 2007). versity Press, 2000), 14. (New York: William Morrow, 1990), 273. 51.Walcott, “Tuskegee Institute Infan- 62. “$329.03 Is Raised in Atlanta Polio tile Paralysis Center,” 14; “Dr J. S. Drive,” Atlanta Daily World, February 7, 75. Charles H. Bynum, “Upholding a Brookens, AME Editor, Dies On Train,” 19 41; “Disease Knows No Color Line,” Pledge,” Opportunity 24 (1946): 22. Chicago Defender, September 22, 1951; (Baltimore) Afro-American, January 1, After outrage at their poor treatment, “Roosevelt to See Noted Dr. Carver on 1949. these patients were removed to a gen- Tuskegee Visit,” Birmingham Age-Herald, eral hospital in Atlanta; “Don’t Say We 63. Catherine Worthingham, “Profes- March 30, 1939. For the broader pic- Didn’t Tell You,” Chicago Defender, April sional Education and Poliomyelitis,” Jour- ture on Black medical leaders and phil- 5, 1947; “Polio Grants Made to Train nal of the National Medical Association anthropic funding, see P. Preston Negroes,” New York Times, September 43 (1951): 24. Reynolds, “Dr Louis T. Wright and the 26, 1947; Gould, Summer Plague, NAACP: Pioneers in Hospital Racial Inte- 64. “Says Warm Springs Treats Negro 18 9–191; Photo Album, Box 3, photos gration,” American Journal of Public Polio Victims Now, Atlanta Daily Herald, 909 and G418, Georgia Warm Springs Health 90 (2000): 883–87; W. Michael September 8, 1945; “Warm Springs Foundation Record, March of Dimes

May 2007, Vol 97, No. 5 | American Journal of Public Health Rogers | Peer Reviewed | Public Health Then and Now | 795  FRAMING HEALTH MATTERS 

Generating Political Priority for Maternal Mortality Reduction in 5 Developing Countries

| Jeremy Shiffman, PhD

attention from safe motherhood researchers: I conducted case studies on the level of political priority given to maternal mortality reduction in 5 countries: Guatemala, Honduras, India, Indonesia, and Nigeria. Guatemala, Honduras, India, Indonesia, and Among the factors that shaped political priority were international agency ef- Nigeria (Table 1). The MDGs call for a de- forts to establish a global norm about the unacceptability of maternal death; crease in the world’s maternal mortality ratio those agencies’ provision of financial and technical resources; the degree of co- by 75% from 1990 levels by the year 2015. hesion among national safe motherhood policy communities; the presence of With an estimated 585000 maternal deaths national political champions to promote the cause; the deployment of credible in the year 1990,10 and little evidence of de- evidence to show policymakers a problem existed; the generation of clear policy cline since then,11 much change is needed over alternatives to demonstrate the problem was surmountable; and the organization the next decade if the maternal health goal is of attention-generating events to create national visibility for the issue. to be achieved. Between 2003 and 2006, I The experiences of these 5 countries offer guidance on how political priority prepared individual studies for each country can be generated for other health causes in developing countries. (Am J Public on agenda setting for this cause.12–16 I bring to- Health. 2007;97:796–803. doi:10.2105/AJPH.2006.095455) gether results from these 5 studies to draw out implications for health priority generation in There is a strong emphasis on health in the What are the barriers to political attention other resource-poor countries. Millennium Development Goals (MDGs; pov- for health? How does a lack of political sup- erty alleviation objectives agreed to by United port affect the achievement of health goals? METHODS Nations [UN] member countries).1 Goals 4, 5, How can such support be generated? Political and 6 are concerned with child mortality, scientists have referred to these issues as chal- I used a process-tracing methodology in maternal mortality, HIV/AIDS, and malaria.2 lenges in agenda setting and generating politi- each of the 5 studies, a qualitative case study In 2003, a High Level Forum—a venue for cal priority,6,7 ensuring that political leaders research strategy commonly employed in po- dialogue among senior policymakers from consider an issue to be worthy of sustained at- litical science.17 Process tracing uses multiple governments, aid agencies, foundations, and tention and will back up that attention with the sources of information to minimize bias, es- other organizations—formed to find ways to provision of financial, human, and technical re- tablish common patterns of causality, and re- accelerate the slow progress toward the real- sources commensurate with the severity of the veal social and political processes—the major ization of the MDGs regarding health.3–5 problem. We know priority is present when: goal of this research.17 Process tracing in par- Many factors undoubtedly stand behind (1) national political leaders publicly and pri- ticular and case study methodologies more this slow progress, including insufficient donor vately express sustained concern for the issue; generally have received increasing attention resources, lack of consensus on intervention (2) the government, through an authoritative in political science inquiry in recent years be- strategies, and weak health systems. Another decisionmaking process, enacts policies that cause of their unique capacity to consider po- potential contributor, one that has attracted offer widely embraced strategies to address the litical and social phenomena in their real-life little research attention, may be the difficulty problem; and (3) the government allocates and context, with particular attention to historical in generating national political support for releases public budgets commensurate with influences.18 More commonly used methodol- particular health goals. Even if national poli- the problem’s gravity. Agenda setting is the ogies in public health and medical research, cymakers recognize the existence of health first stage of the public policy process during including randomized controlled experiments, problems, have sufficient donor resources, which some issues are given attention by poli- structured surveys, and statistical analysis of and are cognizant of MDGs, there is no guar- cymakers and others receive minimal attention health service utilization, do not normally antee they will prioritize these issues or take or are neglected completely. Scholars have have these advantages.17 ,18 action. Policymakers in developing countries identified systematic features in the agenda-set- In the study for each country, I investi- are burdened with thousands of issues and ting process that shape the likelihood that any gated the same 2 questions: to what extent is have limited resources to deal with them, as given issue will receive policy attention.6,8,9 maternal mortality reduction on the national well as conflicting political imperatives. Goals Drawing on this political science scholarship, policy agenda, and what factors have facili- targeting improved health must compete for I examined the state of political priority for tated or obstructed political priority for the policy attention and resources in these diffi- maternal mortality reduction in 5 developing cause? Drawing from existing public policy cult political circumstances. countries that have attracted considerable research, the studies were exploratory in

796 | Framing Health Matters | Peer Reviewed | Shiffman American Journal of Public Health | May 2007, Vol 97, No. 5  FRAMING HEALTH MATTERS 

nature, investigating these 2 questions to identify which factors may be at work in each country. A limitation of the case study approach is the difficulty of controlling for confounding influences on the outcome of interest. As such, inferences in this study must be understood as propositions that require further research, ideally in comparative context. Each country was selected because of the significant attention it has attracted in safe motherhood scholarship and because of its potential to reveal underlying dynamics of agenda setting for maternal mortality reduction. Honduras is 1 of only a handful of developing countries to have experienced a documented significant decline in maternal mortality since the advent of the Global Safe Motherhood Initiative in 1987.19 (This has been confirmed by 2 reliable reproductive age mortality surveys, the gold standard in maternal mortality measurement that documents every maternal death over the course of a year.20,21) Its Central American neighbor, As part of a Unicef-funded Safe Motherhood program, Arik Dhelbeny, 18 years old and 6 months pregnant, attends a prenatal check-up in Rumbek Hospital, Southern Sudan. Photograph by Georgina Guatemala, provides an interesting contrast— Cranston. Available at http://www.Galbe.com. despite its greater wealth than Honduras and its receipt of significant donor resources for safe motherhood, it has a higher maternal consultation of national safe motherhood In addition to the interviews, multiple doc- mortality ratio and there is no firm evidence meeting reports; review of government, donor uments were carefully read and cross- of maternal mortality change over the same and nongovernmental organization (NGO) checked to develop a history of safe mother- time period.14 Indonesia has been the focus of documents; and by asking interviewees whom hood initiatives in each country, to evaluate extensive safe motherhood research because they considered to be most centrally involved the state of political priority for maternal of its long-standing problem with maternal in safe motherhood. mortality reduction, and to facilitate analysis mortality and a unique initiative begun in Across the 5 countries, I carried out a total of the factors that shaped the level of priority. 1989 to place a midwife in each of its more of 124 interviews, of which 75 were con- Documents included demographic and health than 60000 villages to address this problem. ducted jointly with research collaborators. In- and other surveys; government policy docu- India and Nigeria rank first and second glob- terviewees included former ministers and sec- ments, health reports, and technical guide- ally in numbers of maternal deaths annually, retaries of health, maternal and child health lines on obstetric care; documents from bilat- together contributing approximately one third division heads, officials in other government eral and multilateral donors; national of the international total.11 If maternal mortal- ministries and agencies, parliamentarians, bi- government development plans; reports from ity does not decline significantly in these 2 lateral donors, multilateral agency representa- foundations and NGOs; and published re- countries, it is unlikely the MDG concerning tives, NGO officials, and academics. Most in- search on safe motherhood and maternal maternal health can be achieved. terviews lasted between 1 and 2 hours. The mortality. These documents were gathered I used 5 kinds of data collection methods interviews were not transcribed, but I took from libraries in government, NGO, and for each country’s study: interviews with offi- detailed notes during each. Although there donor offices; direct solicitation from intervie- cials involved in safe motherhood policy, ob- were some common questions asked of most wees; research libraries in the United States; servation of implementation sites, government interviewees, including their assessment of and Web-based searches. In addition, villages, reports and documents, donor agency reports, the state of political priority for the cause, I local health centers, and hospitals where safe and published research on safe motherhood. I did not employ a uniform survey instrument, motherhood activities were being imple- identified key individuals involved in safe because each interviewee had unique knowl- mented were observed. motherhood policymaking and implementa- edge about safe motherhood in his or her Once all the material had been collected, tion, and I conducted between 20 and 30 country. Instead, I asked each interviewee interview notes and documents were re- semistructured interviews in each country. open-ended questions in an exploratory way viewed and facts were checked across multi- These key individuals were identified through to elicit that unique knowledge. ple sources to develop a history of safe

May 2007, Vol 97, No. 5 | American Journal of Public Health Shiffman | Peer Reviewed | Framing Health Matters | 797  FRAMING HEALTH MATTERS 

TABLE 1—Safe Motherhood and Economic Indicators, by Country: Mid-1990s to Early 2000s world. In Indonesia, priority for safe motherhood rose from near obscurity to national 21, 52,53 54,55 56 57,58 39,59,60 Indicator Honduras Guatemala Indonesia India Nigeria prominence over the period 1988 to 1997 Maternal mortality ratioa 108 153 307 540 704 and received the direct attention of President Percentage of women delivering 62 42 40 35 33 Suharto, although after his fall from political in health institutions power it declined. Nominally on the policy Percentage of women delivering 56 41 66 42 35 agenda in India since independence, maternal with medical attendants mortality reduction took a backseat to other Most common biomedical Hemorrhaging, Hemorrhaging, NA Hemorrhaging, Hemorrhaging, health causes for several decades, but in causes of death hypertension, infection, anemia, infection, 2005, it rose to receive meaningful national infection hypertension infection unsafe political attention. Until 2000, the cause re- abortion ceived no significant attention in Guatemala GDP per capita (purchasing power 2800 5200 3700 3400 1000 and Nigeria, and it remains a neglected issue. parity in 2005) in US$b Nine factors, each identified in previous research on agenda setting, shaped the degree Note. GDP=gross domestic product; NA=not available. to which maternal mortality reduction aNumber of maternal deaths per 100000 live births.The Honduran and Guatemalan maternal mortality ratios are highly reliable because they are population figures derived from reproductive-age mortality surveys—the gold standard in maternal emerged on the national policy agendas of mortality measurement—that investigate every death over the course of the year and therefore do not require confidence these 5 countries (Table 2). These can be intervals.The Indonesian, Indian, and Nigerian maternal mortality ratios are estimates from representative surveys. divided into 3 categories: transnational influ- bGDP per capita is 2005 estimate as reported in the Central Intelligence Agency World Factbook.51 ences, domestic advocacy, and the national political environment. motherhood policy attention in each country. RESULTS Interview notes were compared with one Transnational Influences another and with written documents to ver- The 5 countries varied considerably in the It was a group of international advocates ify and extract information on major devel- degree to which the cause of maternal mortal- and organizations that first put safe mother- opments in the history of attention to safe ity reduction had received political priority (in- hood on the global agenda. They used several motherhood in each country and to facilitate dicated by the 3 criteria in the introductory mechanisms to influence national political the development of propositions with regard section). In Honduras, political priority was systems to embrace the cause. to which factors may have shaped this atten- very high; in Indonesia, high; in India, moder- Norm promotion. International relations tion. This comparison and cross-checking of ate (with a recent rise); and in Guatemala and scholars, particularly those who emphasize information from interviews was crucial be- Nigeria, low.12–16 In Honduras, safe mother- the role of ideational factors in politics, argue cause respondents often did not remember hood became one of the country’s foremost that countries form their policy preferences accurately when particular developments oc- health priorities, and between 1990 and not simply through national political curred. Individuals involved in safe mother- 19 97, the country experienced a 40% decline processes, but also through participation in hood in each country also reviewed their in its maternal mortality ratio,20,21 one of the the international political arena, which helps country’s case study reports to check for fac- most significant reductions in such a short shape societal norms and therefore policy tual accuracy. time span ever documented in the developing preferences.7,22 For instance, officials from

TABLE 2—Factors Influencing the Degree to Which Maternal Mortality Reduction Appeared on National Policy Agendas: Guatemala, Honduras, India, Indonesia, and Nigeria, Early 1990s to Mid-2000s

Factor Category Description

Norm promotion Transnational influence Efforts by international agencies to establish a global norm concerning the unacceptability of maternal death Resource provision Transnational influence The offer of financial and technical resources by international agencies to address maternal mortality Policy community cohesion Domestic advocacy The degree to which national safe motherhood promoters coalesced as a political force pushing the government to act Political entrepreneurship Domestic advocacy The presence of respected and capable national political champions willing to promote the cause Credible indicators Domestic advocacy The availability and strategic deployment of evidence to demonstrate the presence of a maternal mortality problem Focusing events Domestic advocacy The organization of forums to generate national attention for the cause Clear policy alternatives Domestic advocacy The availability of clear policy alternatives to demonstrate to political leaders that the problem is surmountable Political transitions National political environment Political changes, such as democratization, that positively or adversely affect prospects for safe motherhood promotion Competing health priorities National political environment Priority for other health causes that divert policymaker attention away from maternal mortality reduction

798 | Framing Health Matters | Peer Reviewed | Shiffman American Journal of Public Health | May 2007, Vol 97, No. 5  FRAMING HEALTH MATTERS 

international organizations involved in health Compulsion does not seem to have been Policy communities are networks of actors from seek to persuade national officials to prioritize at work with respect to safe motherhood. different types of organizations—government particular health causes such as HIV/AIDS The enticement of resources, however, did agencies, legislatures, NGOs, and others—com- prevention.23 Also, national health officials shape the behavior of these 5 countries and mitted to common causes. Among the factors influence one another at international confer- provided material backing for the norm- that shape their degree of influence are their ences and other forums. promotion efforts of international actors. In levels of moral authority, knowledge, and co- The international shaping of norms was an Honduras, the US Agency for International herence.33 influential force in the decisions of countries to Development (USAID) provided US $57.3 Safe motherhood policy communities first take up the cause of maternal mortality re- million to the health sector between 1988 to formed in each of these 5 countries, consist- duction. A 1985 Lancet article titled “Maternal 2000, a large portion directed toward ma- ing of Ministries of Health doctors, parliamen- mortality—a neglected tragedy. Where is the M ternal mortality reduction.27 The UN Fund tarians, obstetrician-gynecologists, health-fo- in MCH?” played a key role in bringing the for Population Activities, Pan American cused nationals employed by donor agencies, issue of maternal death to the attention of in- Health Organization, and the World Bank and other individuals and groups. All held ternational health officials.24 An international also provided safe motherhood financing moral authority by virtue of their commit- meeting in 1987 in Nairobi, Kenya, also was and technical assistance in the early ment to a humanitarian cause: reduction of influential and launched the Global Safe Moth- 19 9 0s.13 A primary financier for safe moth- maternal death levels. Because the group was erhood Initiative that aimed to lower global erhood in the early 1990s—the USAID’s largely composed of medical experts, all also maternal deaths by at least half by the year MotherCare program—made Guatemala one held knowledge-based authority, and policy- 2000. Thereafter, the Inter-Agency Group for of its focal countries.14 In Indonesia, the makers deferred to them on technical issues Safe Motherhood formed to focus global atten- World Bank offered support for a safe moth- about safe motherhood. The communities dif- tion on the issue, bringing together multiple in- erhood program through a US $104 million fered, however, in degree of coherence. Some ternational agencies.25 A series of UN-spon- loan to support a national population pro- coalesced into tight networks, transforming sored international conferences throughout the gram covering the period 1991 to 1996.28 their moral and knowledge-based authority 1990s reaffirmed the global commitment to In the early 1990s, at least 6 other donor into political influence and pushing their gov- reduce maternal mortality by 50%. Most re- agencies also began safe motherhood proj- ernments to act.12 ,13 Others struggled to cently, maternal mortality reduction received a ects.12 In Nigeria, the Department for Inter- come together and therefore had limited place in the MDGs, and a partnership for ma- national Development is funding a 7-year agenda-setting influence.14–15 ternal, newborn, and child health formed, link- project with a main concern for safe mother- In Honduras, for instance, a highly effective ing formerly separate initiatives. hood.29 USAID, World Health Organization, working group formed in 1990 that became These international initiatives created con- UNICEF, the UN Fund for Population Activi- the unofficial center for national safe mother- cern among many national health officials ties, the World Bank, and the MacArthur hood efforts.13 Meeting regularly over several about the problem of maternal mortality. Offi- and Packard Foundations are also funding years, the group included members of the cials from Guatemala, Nigeria, Indonesia, and safe motherhood activities.15 In India, the Ministry of Health’s division of maternal and India attended the Nairobi conference, and 19 92 Child Survival and Safe Motherhood child health, Pan American Health Organiza- officials from all 5 countries participated in program received US $214.5 million from tion, USAID, the UN Fund for Population Ac- regional conferences on safe motherhood as the World Bank and US $67.8 million from tivities, UNICEF, and other donors and agen- follow-up to this meeting. By the early 1990s, UNICEF,30 and donors have offered exten- cies. The group produced a national plan of all had launched new national safe mother- sive financing since 1997 for both phases of action for maternal mortality reduction for hood activities.12–16 the government’s Reproductive and Child the period 1991 to 1995. Members also trav- Resource provision. International relations Health Program, whose aims include mater- eled to each of the country’s regions, facilitat- scholars have identified several other forms nal mortality reduction.31 ing the development of local action plans and of transnational influence on the policy prefer- mobilizing regional health bureaucracies in ences of countries.26 One mechanism is com- Domestic Advocacy service of safe motherhood. In Guatemala pulsion, such as the leverage the International Transnational actors brought the issue of and Nigeria, by contrast, cohesive networks Monetary Fund wields when it threatens to maternal mortality to the global agenda but have yet to form, and linkages among safe deny loans to countries that face severe finan- could not institutionalize the cause in na- motherhood promoters remain informal.14 ,15 cial crises if they do not adopt structural ad- tional political systems on their own. National Political entrepreneurship. Public policy schol- justment programs. Another mechanism is re- adoption and sustainability required domestic ars have found that individual national political source provision: the enticement of financial advocacy. entrepreneurs also shape agenda setting.6,34 and technical assistance from the International Policy community cohesion. Political scientists These political entrepeneurs are politically in- Monetary Fund and other organizations to have argued that the structure and organization fluential and particularly capable individuals governments if they agree to adopt particular of policy communities shape how successful willing to exert effort to advance a cause. Not priorities and policies. they will be in influencing national priorities.6,32 just any person can play such a role, however.

May 2007, Vol 97, No. 5 | American Journal of Public Health Shiffman | Peer Reviewed | Framing Health Matters | 799  FRAMING HEALTH MATTERS 

Research has shown that effective political en- policymakers is the presence of a clear indica- the policy priorities of states and local govern- trepreneurs possess certain distinct features: tor to highlight the issue, such as a maternal ments, the absence of subnational data has they are knowledgeable about the issue, they mortality ratio to indicate maternal death lev- contributed to a situation in which most state are persistent, they have excellent coalition- els.6,35 These make a difference because they governors and local government heads are un- building skills, they articulate vision amid com- have the uniquely powerful effect of giving vis- aware of problems in their own areas and plexity, they have a credibility that facilitates ibility to that which has remained hidden, serv- avoid acting on maternal mortality.15 the generation of resources, they generate ing not just monitoring purposes, the way they Focusing events. Focusing events—large- commitment by appealing to important social are traditionally understood, but also as cata- scale happenings such as crises, conferences, values, they are aware of the critical challenges lysts that may provoke political elites to act. and discoveries that attract notice from wide in their environments, they infuse colleagues Where no such indicators are available, policy- audiences—also have agenda-setting power.9 and subordinates with a sense of mission, and makers may ignore the issue either because They function much like indicators, bringing they are strong in rhetorical skills.12 they are unaware of the existence of a prob- visibility to hidden issues. Policy communities were more effective lem or are unconvinced in the absence of evi- Focusing events helped put safe mother- where political entrepreneurs for safe mother- dence that any problem exists. hood on the global agenda and shaped promo- hood emerged to lead them. Indonesia’s safe In Guatemala and Honduras, maternal mor- tion of the issue in all 5 countries. The interna- motherhood policy community had tality studies sparked national efforts to ad- tional Nairobi conference was the first example particularly effective leadership.12 In 1995, dress the issue, and in Indonesia, such a study of such an event. In 1988, soon after this con- the Assistant Minister of Women’s Roles, contributed to rekindling an existing initiative. ference, Indonesia’s first national seminar on considered to be among the most effective In Guatemala, a midlevel official in the Min- safe motherhood was held, with President leaders in the Indonesian bureaucracy in istry of Health responded to a Central Ameri- Suharto delivering the keynote address.40 In the social development sector, came up can regional call for action on safe mother- 1990, after the Nairobi conference, Nigerian with the idea of a national campaign to raise hood by organizing a national mortality attendees organized a national safe mother- attention to the plight of pregnant women. He survey for women of reproductive age.36 Com- hood conference, convened by the Society for single-handedly convinced President Suharto pleted in 1991, the study found a maternal Obstetrics and Gynecology of Nigeria. A for- to take a direct role in the campaign, gener- mortality ratio of 248 deaths per 100000 mal Central American launch of the Global ated additional budgetary appropriations for live births, far higher than health officials ex- Safe Motherhood Initiative was held in maternal mortality reduction, and mobilized pected, prompting the Minister of Health to Guatemala City in 1992 and attended by Hon- provincial and local governments to address declare maternal mortality reduction a priority duran and Guatemalan delegates.41 In 2000, the issue. Ministry of Health doctors also issue.36 A former Honduran Ministry of the White Ribbon Alliance of India, a group of played key entrepreneurial roles, sparking a Health official working for the Pan American organizations that promoted safe motherhood flurry of activities inside the Ministry of Health Organization led the organization of in the country, organized a march, led by a Health in follow-up to the Nairobi conference. his country’s first mortality survey study for parliamentarian who was also a movie star, to In Guatemala, India, and Nigeria, a number women of reproductive age in 1990. The re- the Taj Mahal—a monument built to commem- of capable individuals in government and civil search revealed a maternal mortality ratio of orate the death of a sultan’s wife in childbirth. society had promoted the safe motherhood 182, nearly 4 times the previously accepted The event generated national media coverage cause. In Guatemala, the former vice-president figure.20 The official and his colleagues ac- of the country’s high levels of maternal death. of parliament collaborated with the UN Fund tively publicized the study’s results. By the Most recently, New Delhi hosted the 2005 for Population Activities to secure the passage end of 1990 a new health minister had com- World Health Day, whose theme was maternal of a reproductive health bill. In India, the for- mented in the national media on the study, and child health. Prime Minister Manmohan mer Secretary of Health and Family Welfare noting that the country had a serious problem Singh met with leaders of several United Na- actively promotes the issue as head of with maternal mortality.37 In Indonesia, the tions agencies and spoke publicly about mater- an NGO. In Nigeria, the House Chairwoman decision of the Assistant Minister of Women’s nal mortality specifically. for Women Affairs and Youth Development is Roles to launch a campaign was a direct result Clear policy alternatives. Agenda-setting re- leading an effort for a bill on maternal mortal- of his alarm over a high maternal mortality searchers have found that policymakers are ity reduction. However, in none of these 3 ratio of 390 reported in the 1994 Indonesian more likely to act on an issue if they are pre- countries has any individual emerged as a rec- Demographic and Health Survey.12 , 3 8 sented with clear proposals that convince ognized leader of a safe motherhood policy The absence of credible evidence has con- them that a problem is surmountable.6,32 If community, nor has anyone played the politi- tributed to inertia in Nigeria. Although reliable policy communities have not generated clear cal mobilization role that the Assistant Minis- data exist to confirm a national maternal mor- and widely accepted proposals, policymakers ter of Women’s Roles did in Indonesia. tality problem,39 there are no disaggregated are unlikely to pay attention to their con- Credible indicators. Agenda-setting scholars data for Nigeria’s 36 states or 774 local gov- cerns, because political elites prefer to allo- have demonstrated that among the factors that ernments. In a federal political system where cate resources toward problems they believe shape whether an issue rises to the attention of the national government has little control over can be effectively addressed.

800 | Framing Health Matters | Peer Reviewed | Shiffman American Journal of Public Health | May 2007, Vol 97, No. 5  FRAMING HEALTH MATTERS 

The rise of the safe motherhood cause on cially in India and Nigeria) that give national more visible causes that gain them votes and to the Indonesian policy agenda was facili- governments little control over subnational pol- political capital, such as road construction. tated by just such a proposal. A village mid- itics, and endemic corruption. Two factors, Competing health priorities. Most health wife program, begun in 1989, reflected a however, were particularly critical: political sectors in developing countries are strapped concern that women in rural Indonesia had transitions and competing health priorities. for resources, and health causes must com- poor access to medical care during their Political transitions. Political scientists have pete against one another for scarce funding. pregnancies. The Ministry of Health managed found that major political transitions and re- Donors contribute to this competition by pro- to place a midwife in most of Indonesia’s forms such as democratization and public sec- viding limited funding for health, and by 68000 villages to ensure that pregnant tor decentralization alter public priorities by promoting health causes that recipient gov- women could get both prenatal and delivery giving new actors agenda-setting power, and ernments may not consider to be priorities.47 assistance.12 This program drew the attention by changing the processes by which public In India, until recently, maternal mortality of President Suharto, who at one point inter- policies are made and implemented.44,45 The reduction took a backseat to population convened directly to hasten its implementation.12 same reform may have the opposite effect on trol, child mortality reduction, and polio erad- It is not yet clear whether this program has the prioritization of any given issue, depend- ication.16 In 1946, prior to independence, a reduced maternal mortality; however, policy- ing on the context. commission established a plan for a national makers perceived that it would and were In Nigeria, democratization facilitated policy primary health care system, creating a cadre therefore willing to devote resources to the attention for safe motherhood.15 In 1999, the of maternal and child health workers. From program, giving the village midwife program country experienced a transition to a semi- the beginning, these workers emphasized the an agenda-setting effect. democratic political system after decades of health of children rather than mothers. Mater- Globally, the safe motherhood movement military-authoritarian rule, creating the political nal health work was further diluted in 1966 lost momentum in the 1990s as a result of in- space for social issues, such as maternal mortal- when the government established a family ternal disagreements on intervention strategies ity reduction, to appear on the national agenda. planning program with its own administrative (although recently these disagreements are Under a democratic political system, the gov- structure and introduced population control being transcended). One of the more con- ernment has faced increased pressure to be ac- targets, creating incentives for maternal and tentious debates was whether scarce resources countable to its constituents, with direct impact child health workers to focus primarily on should be concentrated on ensuring the pres- on safe motherhood. One manifestation has contraception promotion.48 In the 1980s, ence of skilled attendants at all births or on been the creation of a national poverty allevia- these workers took on additional roles, pro- making emergency obstetric care available for tion program, entitled National Economic Em- moting immunization and polio eradication. It women who experience complications at de- powerment and Development Strategy, which is only in India’s most recent national health livery.25,42 Also, international safe motherhood has become an overarching national framework programs that maternal mortality reduction promoters were criticized for lacking evidence for social change and which explicitly lists ma- objectives have gained prominence alongside to back up their proposed interventions and ternal mortality reduction as an objective. An- population control and child mortality goals. for their use of unclear terminology surround- other indicator is a public budget line devoted Nigeria has faced a similar problem with ing interventions.43 Another problem has to safe motherhood, the first time such an ap- respect to HIV/AIDS. This cause has become been a difficulty in measuring maternal mor- propriation has been made. a funding priority for donors, particularly the tality and connecting process indicators to In Indonesia, by contrast, democratization United States through the president’s Emer- health outcomes. These difficulties may have and a subsequent reform—public sector gency Plan for AIDS Relief program, making resulted in reduced leverage to convince na- decentralization—may have hurt safe mother- Nigeria a focal point in the fight against the tional policymakers to prioritize the cause. hood.46 Indonesian political entrepreneurs disease. Donors even convinced the govern- had convinced Suharto to prioritize safe ment to establish a special commission National Political Environment motherhood, and the president and these en- explicitly devoted to control of the disease.49 The quality of political advocacy by interna- trepreneurs used the authoritarian political in- Attention to maternal mortality and other re- tional and national safe motherhood promoters frastructure to push subnational governments productive health causes has suffered influenced the degree to which the issue re- to prioritize the issue. In 1998, Suharto fell as NGOs pursue AIDS money and local gov- ceived policy attention in these 5 countries. from power and the country democratized, ernments receive signals from the political The political and social environments in which and in 1999 political and financial power was center to prioritize HIV/AIDS over other these advocates worked and over which they decentralized to district governments. As a re- problems that are just as serious. had little control also shaped policy attention. sult, the capacity of the central government— Many such factors were influential, including including the Ministry of Health—to command CONCLUSION cultural barriers that place low value on district governments to implement its priori- women’s lives, the ethnic composition of soci- ties weakened substantially. Few district Findings from the case studies indicate that eties, civil strife, weak national health infra- heads now view maternal mortality reduction the level of political priority for maternal mor- structures, federalist political structures (espe- as a priority, preferring to devote resources to tality reduction varies considerably across the

May 2007, Vol 97, No. 5 | American Journal of Public Health Shiffman | Peer Reviewed | Framing Health Matters | 801  FRAMING HEALTH MATTERS 

5 countries and that 9 factors may stand be- public health scholarship on the developing Third, the case studies demonstrate that in- hind this variance and help explain how the world50 and, like agenda setting, requires con- ternational donor prioritization and resources, issue emerged on the global health agenda in siderably more research. and effective medical and technical interven- the first place. Transnational actors first put These limitations notwithstanding, 3 impli- tions, although critical, are far from sufficient maternal mortality reduction on the global cations emerge from this study for achieving for achieving health objectives. Attaining the agenda, promoting a norm that maternal death national health objectives. First, the case stud- goal is as much a national political challenge was unacceptable and generating the interest ies support findings from prior public policy as it is a medical or technical challenge. Pol- of national health officials with financial and research about systematic features to the icy communities in settings with significant technical resources to address the problem. agenda-setting process that increase the likeli- health problems need to develop careful po- National advocates then achieved varying de- hood that national advocates will be effective litical strategies to ensure that their national grees of success in promoting the cause. They in moving political elites to action. Specifi- leaders give these issues the attention and were most successful when they formed cohe- cally, national health advocates are more resources they deserve. sive policy communities, were led by respected likely to be effective if they: national political entrepreneurs, deployed cred- 1. Coalesce into unified policy communities, ible indicators to show a serious problem ex- About the Author translating their potential moral and isted, organized focusing events such as na- Jeremy Shiffman is with the Maxwell School of Citizenship knowledge-based authority into political and Public Affairs of Syracuse University, Syracuse, NY, tional forums to promote visibility for the power and pressing national political offi- and the Center for Global Development, Washington, DC. cause, and developed clear policy alternatives Requests for reprints should be sent to Jeremy Shiffman, cials to act. to demonstrate to national leaders that the 306 Eggers Hall, The Maxwell School of Syracuse Univer- 2. Bring into their communities respected sity, Syracuse, NY 13244-1020 (e-mail jrshiffm@maxwell. problem was surmountable. Many factors in and well-connected national political entre- syr.edu). their political environments shaped the effec- preneurs with track records in placing pub- tiveness of their efforts, but 2 were key: major lic health issues on national agendas. Contribution political reforms, including democratic transi- The author conceptualized, carried out, and wrote the 3. Develop credible measures that mark the tions and public sector decentralization that al- results of the comparative study. severity of this problem, and make political tered the policymaking process, and the degree leaders aware of these measures so they of resource competition with other priorities, Human Participant Protection cannot plausibly deny that a problem exists. No approval was required. such as population control and HIV/AIDS. 4. Organize large-scale focusing events such The replicated case study methodology im- as national forums to generate widespread Acknowledgments poses limits on inferring causality and gener- attention to the issue. The author is grateful to the MacArthur Foundation, the alizing results. In-depth exploration of these Bill and Melinda Gates Institute for Population and Repro- 5. Present leaders with clear policy alterna- countries facilitated the development and ex- ductive Health of the Johns Hopkins Bloomberg School of tives proven to be effective, so that policy- Public Health, Syracuse University, and the US Govern- amination of propositions that concern the makers come to believe the problem can ment’s National Security Education Program for providing generation of political priority. In the absence funding for the 5 case studies. The author thanks the Cen- be surmounted and know what they are of additional comparative inquiry, one cannot ter for Global Development for providing a visiting fellow expected to do. position during which study results were written up. be certain that the factors identified were the The author also thanks the research collaborators primary forces at work nor of their causal Second, the case studies point to the dis- on the country case studies: Ana Lucía Garcés del weight. Also, each of these 5 countries has tinctiveness of national circumstances and the Valle, Friday Okonofua, Ana Patricia Salazar, Cynthia Stanton, and Rajani Ved. In addition, the author would unique political and socioeconomic circum- need to consider these in political strategy de- like to express his appreciation to the many individuals stances, so one must be cautious in generaliz- velopment. Safe motherhood emerged— who agreed to be interviewed for this study, without ing to other settings. Another limitation of or failed to emerge—on national agendas be- whom it would have been impossible to evaluate the safe motherhood experiences of each country. this study is that it focuses on the agenda- cause of country-specific combinations of fac- setting stage in the public policy process but tors. Although there were regularities to the References not on implementation. The appearance of an process including the centrality of credible ev- 1. United Nations. UN Millennium Development issue on a national policy agenda is only 1 of idence and the power of political entrepre- Goals. Available at: http://www.un.org/millenniumgoals. multiple factors that stand behind policy ef- neurship, generating priority for the cause Accessed May 22, 2006. fectiveness and is hardly enough to ensure was not formulaic, and some factors were at 2. World Health Organization. Health and the Mil- lennium Development Goals. Available at: http://www. that the political system will carry out plans work in certain countries but not others. Suc- who.int/mdg. Accessed May 22, 2006. or that these plans will be successful in reduc- cessful policy communities understood the 3. World Health Organization, World Bank. High- ing maternal mortality. Implementation, like distinct characteristics of their political envi- Level Forum on the Health MDGs [overview]. Avail- agenda setting, is a politically infused process, ronments and used an intuitive understanding able at: http://www.hlfhealthmdgs.org. Accessed May 22, 2006. and implementation bottlenecks may emerge of agenda-setting mechanisms to develop po- 4. Summary of Discussions and Action Points: Third at all levels of the system. The subject of im- litical strategies appropriate to the national High-Level Forum on the Health MDGs. Paris, France: plementation has received little attention in context. High-Level Forum on the Health MDGs; 2005.

802 | Framing Health Matters | Peer Reviewed | Shiffman American Journal of Public Health | May 2007, Vol 97, No. 5  FRAMING HEALTH MATTERS 

5. World Health Organization, World Bank. High- 26. Stone D. Learning lessons and transferring policy 45. Cheema GS, Rondinelli DA, eds. Decentralization Level Forum on the Health MDGs [January 2004 across time, space and disciplines. Polit. 1999;19:51–59. and Development: Policy Implementation in Developing meeting]. Available at: http://www.hlfhealthmdgs.org/ 27. Honduras Project Paper: Health Sector II. Washington, Countries. Beverly Hills, Calif: Sage; 1983. January2004Mtg.asp. Accessed May 22, 2006. DC: US Agency for International Development; 1988. 46. Utomo B, Shiffman J, Marchal B, Coates A, De 6. Kingdon JW. Agendas, Alternatives and Public Poli- 28. Staff Appraisal Report Indonesia: Fifth Population Brouwere V. Sustaining Priority for Safe Motherhood in cies. Boston, Mass, and Toronto, Ontario: Little, Brown Project (Family Planning and Safe Motherhood). Wash- Indonesia Under Political Transition [working paper]. and Company; 1984. ington, DC: The World Bank; 1991. Aberdeen, Scotland: Initiative for Maternal Mortality Programme Assessment; 2005. 7. Keck ME, Sikkink K. Activists Beyond Borders: 29. Duby F. Resource Materials for Better Health: Advocacy Networks in International Politics. Ithaca, NY: Round Table Report Safe Motherhood, Report on Confer- 47. Banerji D. Alma-Ata showed the route to effective Cornell; 1998. ence in Cross Rivers State. Abuja, Nigeria: Partnership resource allocations for health. Bull World Health 8. Baumgartner FR, Jones BD. Agendas and Instabil- for Transforming Health Systems; 2004. Organ. 2004;82:707–708. ity in American Politics. Chicago, Ill, and London, En- 30. Implementation Completion Report: Child Survival 48. Mavalankar D. State of maternal health. In: gland: University of Chicago Press; 1993. and Safe Motherhood Project. Delhi, India: The World Dadhich JP, Vinod P, eds. State of India’s Newborns Re- 9. Birkland TA. After Disaster: Agenda Setting, Public Bank; 1997. port. New Delhi, India: National Neonatology Forum, Policy, and Focusing Events. Washington, DC: George- Save the Children/US; 2004:27–42. 31.Government of India. Reproductive and Child Health town University Press; 1997. Programme. New Delhi, India: Department of Family 49. Nudging the Giant: The Story of the POLICY Project 10.World Health Organization, United Nations Chil- Welfare, Ministry of Health and Family Welfare; 2005. Nigeria 1999–2004. Abuja, Nigeria: US Agency for dren’s Fund. Revised 1990 Estimates of Maternal Mor- 32. Sabatier P. The advocacy coalition framework: International Development Nigeria, the POLICY Proj- tality: A New Approach by WHO and UNICEF. Geneva, revision and relevance for Europe. J Eur Public Policy. ect; 2005. Switzerland: World Health Organization; 1996. 1998;5:98–130. 50. Task Force on Health Systems Research. Informed 11. United Nations Children’s Fund, United Nations 33. Haas PM. Introduction: epistemic communities choices for attaining the Millennium Development Fund for Population Activities, World Health Organiza- and international policy coordination. Int Organ. 19 92; Goals: towards an international cooperative agenda for tion. Maternal Mortality in 2000: Estimates Developed 46:1–35. health-systems research. Lancet. 2004;364:997–1003. by WHO, UNICEF, and UNFPA. Geneva, Switzerland: 51. Central Intelligence Agency. World Factbook. World Health Organization; 2004. 34. Doig JW, Hargrove EC, eds. Leadership and Inno- vation: A Biographical Perspective on Entrepreneurs in Available at: http://www.cia.gov/cia/publications/ 12. Shiffman J. Generating political will for safe moth- Government. Baltimore, Md, and London, England: factbook/geos. Accessed April 23, 2006. erhood in Indonesia. Soc Sci Med. 2003;56:1197–1207. Johns Hopkins University Press; 1987. 52. Honduras: Encuesta Nacional de Epidemiología y 13. Shiffman J, Stanton C, Salazar AP. The emergence 35. Walker JL. Performance gaps, policy research, and Salud Familiar 2001. Tegucigalpa, Honduras: Hon- of political priority for safe motherhood in Honduras. political entrepreneurs: toward a theory of agenda set- duran Ministry of Public Health, Ashonplafa, US Health Policy Plan. 2004;19:380–390. ting. Policy Stud J. 1974;3:112–116. Agency for International Development, Centers for 14 . Shiffman J, Garcés Del Valle AL. Political history Disease Control and Prevention, Management Sciences 36. Medina H. Study of Maternal Mortality in and disparities in safe motherhood between Guatemala for Health; 2001. Guatemala [In Spanish]. Guatemala City. Guatemala: and Honduras. Popul Dev Rev. 2006;32:53–80. Ministerio de Salud Pública y Asistencia Social; 1989. 53. State of World Population 2005. New York, NY: 15. Shiffman J, Okonofua F. The state of political United Nations Fund for Population Activities; 2005. 37. Salud intentara reducer embarazos de Hondureñas. priority for safe motherhood in Nigeria. Br J Obstet Available at: http://www.unfpa.org/swp/2005/pdf/ [Health ministry seeks to reduce pregnancies among Gynaecol. 2007;114:127–133. en_swp05.pdf. Accessed October 20, 2005. Hondurans.] La Tribuna [Tegucigalpa, Honduras]. Janu- 16. Shiffman J, Ved R. The state of political priority for ary 31, 1991. 54. Línea Basal de Mortalidad Materna Para el Año safe motherhood in India. Br J Obstet Gynaecol. In press. 38. Central Bureau of Statistics, State Ministry of 2000. [Baseline study on maternal mortality for the 17. Yin R. Case Study Research: Design and Methods. Population/National Family Planning Coordinating year 2000.] Guatemala City, Guatemala: Ministerio de 2nd ed. Thousand Oaks, Calif: Sage; 1994. Board, Ministry of Health, Macro International Inc. Salud Pública y Asistencia Social; 2003. 18.Brady HE, Collier D, eds. Rethinking Social In- Indonesia Demographic and Health Survey 1994. Calver- 55. Guatemala, Encuesta Nacional de Salud Materno quiry: Diverse Tools, Shared Standards. Oxford, England: ton, Md: Central Bureau of Statistics, Macro Interna- Infantil 2002. [Guatemala national maternal and child Rowman and Littlefield; 2004. tional Inc; 1995. health survey 2002.] Guatemala City, Guatemala: 19. Danel I. Maternal Mortality Reduction, Honduras, 39. Multiple Indicator Cluster Survey 1999. Lagos, Ministerio de Salud Pública y Asistencia Social, Insti- 1990–1997: A Case Study. Washington, DC: World Nigeria: Federal Office of Statistics, United Nations tuto Nacional de Estadística, Centers for Disease Con- Bank; 1998. Children’s Fund; 2000. trol and Prevention; 2003. 20. Castellanos M, Ochoa JC, David V. Investigation of 40. Shah U, Sudomo S. Assessment of Maternal Health 56. Indonesia Demographic and Health Survey 2002- the Mortality of Women of Reproductive Age With an Situation and Health Services: Safe Motherhood. Jakarta, 2003. Calverton, Md: Badan Pusat Statistik-Statistics Emphasis on Maternal Mortality [in Spanish]. Washing- Indonesia: Ministry of Health, United Nations Develop- Indonesia, ORC Macro; 2003. ton, DC: Pan American Health Organization; 1990. ment Program, World Health Organization; 1991. 57. National Family Health Survey (NFHS-2) 1998-99. 21. Meléndez JH, Ochoa JC, Villanueva Y. Investigation 41. Guatemala Safe Motherhood Declaration. Bombay, India: International Institute for Population on Maternal Mortality and Women of Reproductive Age in Guatemala City, Guatemala: Asociación Pro-Bienestar Sciences; 2000. Honduras: Final Report Corresponding to the Year 1997 de la Familia de Guatemala, Family Care International, 58. Registrar General of India. Sample Registration [in Spanish]. Tegucigalpa, Honduras: Pan American Inter-American Parliamentary Group on Population Bulletin 33:1. New Delhi, India: Office of the Registrar Health Organization/World Health Organization; 1999. and Development; 1992. General; 1999. 22. Finnemore M. National Interests in International 42. Maine D, Rosenfield A. Commentary—The safe 59. Nigeria Demographic and Health Survey 2003. Society. Ithaca, NY: Cornell University Press; 1996. motherhood initiative: why has it stalled? Am J Public Calverton, Md: National Population Commission 23. Ogden J, Walt G, Lush L. The politics of ‘brand- Health. 1999;89:480–482. [Nigeria], ORC Macro; 2004. ing’ in policy transfer: the case of DOTS for tuberculo- 43. Hussein J, Clapham S. Message in a bottle: sinking 60. Child Survival, Protection and Development in Nige- sis control. Soc Sci Med. 2003;57:179–188. in a sea of safe motherhood concepts. Health Policy. ria: Key Social Statistics. Abuja, Nigeria: National Plan- 24.Rosenfield A, Maine D. Maternal mortality—a 2005;73:294–302. ning Commission, United Nations Children’s Fund; neglected tragedy. Where is the M in MCH? Lancet. 44. Linz JJ, Stepan AC. Problems of Democratic Transi- 1998. Cited in Fatusi A. Maternal Mortality Situation 19 85;2:83–85. tion and Consolidation: Southern Europe, South America and Determinants in Nigeria. A Review Commissioned by 25. Tinker A, Koblinsky MA. Making Motherhood Safe. and Post-communist Europe. Baltimore, Md: Johns Hop- Federal Ministry of Health. Abuja, Nigeria: Federal Min- Washington, DC: The World Bank; 1993. kins University Press; 1996. istry of Health; 2004.

May 2007, Vol 97, No. 5 | American Journal of Public Health Shiffman | Peer Reviewed | Framing Health Matters | 803  RESEARCH AND PRACTICE 

Out-of-Pocket Healthcare Spending by the Poor and Chronically Ill in the Republic of Korea

| Jennifer Prah Ruger, PhD, and Hak-Ju Kim, PhD

The equity and efficiency of health care sys- Objectives. We estimated out-of-pocket health care spending and out-of-pocket tems is an important policy issue as evidenced spending burden ratio employing household equivalent income in the Republic by the 2000 World Health Report, which of Korea. We examined variations in out-of-pocket spending, estimated out-of- ranked nations according to health care sys- pocket spending burden ratio employing household equivalent income, and iden- tem performance. According to the report, the tified factors associated with out-of-pocket spending. Republic of Korea (hereafter Korea) ranked Methods. We used the 1998 Korean National Health and Nutrition Survey, a 53rd on fairness and 58th overall (out of nationally representative survey of 39060 individuals. Our analyses examined 18 0); the United States ranked 54th on fair- out-of-pocket spending, out-of-pocket spending burden ratio, and health care ness and 37th overall.1 However, despite use by socioeconomic status, insurance type, health care facility type, and chronic these rankings, knowledge about the equity of condition after we controlled for sociodemographic variables. Results. The lowest income quintile spent 12.5% of their total income out-of- health financing in Korea is limited. pocket on medical expenditures, which was 6 times that of the highest income In 1989, after about a decade of compre- quintile (2%). Among those with 3 or more chronic conditions, low-income Ko- hensive national health reforms, Korea reans had the highest out-of-pocket spending burden ratio (20%), which was 5 achieved universal health insurance coverage times the spending burden among high-income Koreans (4%). In multivariate at a low cost to the government by offering analyses, the number of chronic conditions, insurance type, health care use, and limited benefits, charging high copayments health care facility type were associated with out-of-pocket spending. and coinsurance rates, imposing low fees on Conclusions. Out-of-pocket spending in Korea is regressive, because lower-income providers, and restricting fee growth to the groups pay disproportionately more of their income compared with higher-income level of general inflation. The National Health groups. Low-income individuals with multiple chronic conditions are particularly vul- Institue (NHI) excludes some services, includ- nerable. (Am J Public Health. 2007;97:804–811. doi:10.2105/AJPH.2005.080184) ing expensive diagnostic tests such as ultra- sonography and magnetic resonance imaging. did not associate their findings with socioeco- Among low-income and chronically ill NHI involves significant cost sharing and re- nomic status, insurance type, or chronic ill- groups, it is important to know the extent to quires most Koreans to pay for portions of in- ness. Yang studied patient medical bills from 3 which the burden of out-of-pocket spending is patient and outpatient care through coinsur- general hospitals.7 Out-of-pocket spending was mitigated by Medical Aid, Korea’s assistance ance and copayments,2 which are payments higher than that was stipulated by law be- program for the poor. According to official for services above the health insurance pre- cause of items that were not covered by NHI: statistics from 1998, only 3% of the popula- mium. The cost-sharing schedules set by the in 1996, patient out-of-pocket spending paid tion was eligible for Medical Aid even though Korean Ministry of Health and Welfare apply for 51% to 67% of inpatient services and 12% of Koreans’ income fell at or below the to all services and medical facilities but vary 63% to 94% of outpatient services. Both poverty line.14 Low-income individuals who by type of service (per visit) and facility.2 studies were consistent with earlier work.8 did not have Medical Aid would be expected Moreover, there are insufficient caps on cost A few studies have examined patient- to have a higher burden of out-of-pocket sharing, which ranged from 35% to 45% of reported data and found cost sharing ranged spending compared with those who had Med- the total cost per visit in 2005.3 This combi- from 34% to 45% for inpatient services and ical Aid; however, some of these individuals nation of limited benefits and high cost shar- 64% to 67% for outpatient services.9–13 likely fell into the “other” insurance category ing has created gaps in coverage that burden However, these data were not derived from in the 1998 Korean National Health and Nu- the poor and chronically ill.4,5 nationally representative surveys, and income trition Survey (KHNS). Previous studies have estimated the magni- and health variables were unavailable. They Among individuals with incomes in the tude of out-of-pocket spending for high-cost also did not explore variations in out-of- lowest-income quintiles, it is also important items and hospital care. Kim et al. showed that pocket spending by insurance and health care to know how the burden of out-of-pocket patients’ cost sharing on average accounted facility type or by individual characteristics, spending differs between those who use pub- for 52% of total hospital revenues (40% for particularly chronic illness. Thus, previous es- lic versus private medical facilities, because inpatient, 67% for outpatient)6; however, they timates have provided an incomplete picture public facilities are expected to charge lower did not identify characteristics of persons who of the impact of health care costs on low- unregulated prices for benefits not covered showed high out-of-pocket spending, and they income chronically ill populations. by NHI, and they are not allowed to demand

804 | Research and Practice | Peer Reviewed | Ruger and Kim American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

“informal treatment charges.”15 Compared represent national estimates. We calculated covered).18 Individuals must meet income and with private facilities, public health centers standard errors with Stata version 7.0 (Stata asset criteria to be eligible for Medical Aid; vaccinate children at roughly one third the Corp, College Station, Tex) survey modules therefore, people who fall below the poverty price, treat ambulatory patients at half to one that could accommodate the complex survey line for income but have assets (e.g., a truck third the cost, and hospitalize patients with design. used for work) may be excluded. The “other” chronic diseases at about one fifth the cost.15 The 1998 KHNS samples were collected category is a catchall for those who are poor We studied aspects of equity in health fi- from the general population (excluding island but not eligible for Medical Aid, many of nancing in Korea with the nationally represen- residents), which included 13523 households whom fail to pay premiums and are at risk of tative 1998 KHNS.16 Specifically, we esti- and 39060 household members. The aver- losing their coverage. mated out-of-pocket spending in Korea by age response rate was 91% and ranged from The KHNS provided high-quality data on income group, level of chronic illness, and we 87% (Nutrition Survey) to 95% (Health Be- household income, so we further categorized also examined variations in spending by insur- havior Survey). individuals into 5 quintiles ranked by their ance type, occupation, and health care use monthly household income. The lowest and facility type, and we identified factors as- Definitions and Measurement quintile (0%–20%) is the 20% of the popu- sociated with out-of-pocket spending. We also The KHNS definition of a chronic condi- lation with the lowest household income in estimated out-of-pocket spending burden ratio tion was taken from the 1995 US National 19 98. The highest quintile (80%–100%) is employing household equivalent income. Be- Health Interview Survey and included 14 dis- the 20% of the population with the highest cause of the recent strong interest in universal ease diagnostic-code variables. The clinical household income. To determine household health insurance coverage in the United States classification system defined a chronic condi- equivalent income, which reflects total in- and in middle- and low-income countries that tion as one that lasted or was expected to last come and the number of adults and children are pursuing health financing and insurance 3 or more months. Because individuals could in the household, we used the Organisation reform, lessons learned from Korea may pro- have as many as 14 chronic conditions, the for Economic Co-operation and Develop- vide insights for policymakers worldwide. number of conditions was aggregated for ment Equivalence Scale, which equals

each respondent. 1+0.7(Na –1) + 0.5Nc, where Na = number

METHODS The KHNS collected information on 7 of adults and Nc = number of children; a types of health care facilities as defined by weight of 1 was attached to the household Data Source and Sample Korean medical law: general professional hos- head. In recent years, this measure has been The 1998 KHNS was conducted from No- pital, general hospital, hospital, clinic, Oriental used for international comparisons of pov- vember 1 through December 31, 1998. It clinic, public health center, and pharmacy.17 erty and income inequality19 but has had was sponsored by the Korean Ministry of Different facilities provide different services; limited use in health care studies. Health and Welfare and is the most compre- for example, Oriental medical facilities pro- KHNS survey respondents were asked hensive nationally representative data set on vide Oriental treatments such as herbal medi- about their spending for outpatient services health care use and expenditures. The KHNS cine and acupuncture, which are mostly not during the past 2 weeks and during the past had 4 components: the Health Interview Sur- covered by NHI except for inexpensive ser- year for inpatient services. We estimated indi- vey, the Health Examination Survey, the vices, such as acupuncture.17 The KHNS also viduals’ annual out-of-pocket spending with Health Behavior Survey, and the Nutrition collected data on the number of outpatient the following equation: annual out-of-pocket Survey. The Health Interview Survey and the visits and inpatient hospitalizations. spending=inpatient services expense during Health Behavior Survey estimated the na- The KHNS categorized the type of NHI the past year+outpatient services expense tional prevalence of selected diseases and risk into 5 groups: government employee and during the past 2 weeks×2.2×12; this factors with data from household interviews teacher insurance (offered by a single health method has been used in other studies.20,21 about morbidity, limitation of activity, health insurer; approximately 9% of the population Thus, reported out-of-pocket spending is total care use, and health behaviors. The Health covered), private employee insurance (offered expenditures that inpatients and outpatients Examination Survey estimated national popu- by multiple private insurers; approximately paid directly to hospitals and other facilities. lation reference distributions of selected 34% of the population covered), self-employed It includes coinsurance, copayments, and all health parameters, and the Nutrition Survey regional insurance (offered by multiple private cash payments for services, pharmaceuticals, collected information on dietary practices and insurers; approximately 53% of the popula- supplies, and items not covered by NHI. It nutrition intake. tion covered), Medical Aid (part of the Korean does not include health insurance premiums, The KHNS used a stratified multistage public assistance system that offers free insur- deductibles, or expenses such as medical probability sampling design (including sample ance to eligible poor individuals; benefits are home care or items not related to health. stratification and clustering). As a result, we the same as NHI; approximately 2% of the We also computed the out-of-pocket spend- needed a sampling weight for unbiased na- population covered), and other (those who do ing burden ratio to assess equity among sub- tional estimates, and all of our estimates have not receive any type of NHI or Medical Aid groups of the population. Out-of-pocket been weighted (per KHNS specifications) to coverage; approximately 1% of the population spending burden ratio is the ratio of average

May 2007, Vol 97, No. 5 | American Journal of Public Health Ruger and Kim | Peer Reviewed | Research and Practice | 805  RESEARCH AND PRACTICE 

out-of-pocket spending to household equiva- TABLE 1—Respondent Characteristics by Number of Chronic Conditions: Korean National lent income and is estimated as the mean of Health and Nutrition Survey (KNHS), 1998 the ratios between the 2 numbers for each individual (as opposed to the ratio of the mean Number of Chronic Conditions, % of out-of-pocket spending to the mean of in- Characteristics Total Population None 1 2 ≥3 come). Income was defined as the individual’s Total population 100 33.7 30.4 18.0 17.9 share of household income (household equiv- Age, y alent income). We did not cap individual out- 0–19 18.9 52.4 34.6 9.7 3.3 of-pocket spending at 100% of income prima- 20–44 49.4 38.0 31.1 18.5 12.4 rily because preliminary analysis suggested 45–65 18.4 19.8 28.2 21.1 31.0 that such spending may exceed 100%. 66–79 12.1 9.0 24.6 24.0 42.5 ≥80 1.2 13.3 34.5 22.7 29.5 Statistical Analysis Gender We used bivariate and multivariate analy- Male 48.2 36.4 32.1 17.6 13.8 ses to examine variations in out-of-pocket Female 51.8 30.9 29.0 18.4 21.8 spending according to independent variables, Insurance type including socioeconomic factors, health care Government employees and teachers 9.2 28.4 33.2 19.8 18.6 service use (e.g., outpatient visits, inpatient Private employees 34.4 35.9 30.9 17.3 15.9 hospitalizations), health care facility type, in- Self-employed 53.2 33.5 29.9 18.3 18.3 surance type, and number of chronic condi- Medical Aid 2.1 20.8 22.9 16.2 40.2 tions. We used linear multivariate regression Othera 1.1 33.7 37.2 13.1 16.0 to assess multivariate associations between in- Occupationb dependent variables and out-of-pocket spend- Professional and management 5.6 38.2 33.5 18.5 9.8 ing. For multivariate analyses, we modeled White collar 7.8 41.0 31.8 18.2 8.9 the natural logarithm of household income Sales 13.5 31.4 30.7 21.6 16.3 and out-of-pocket spending, because the re- Farming and fishery 7.3 13.0 24.1 22.8 40.2 spective distributions of these variables were Labor 15.2 29.1 29.3 21.0 20.7 skewed and required log transformation. Al- Military 0.2 44.3 37.7 18.0 0.0 though different specifications were used, a Student 12.3 55.7 31.1 9.4 3.8 double-logarithmic regression model (for Other 38.2 30.6 31.2 17.3 20.9 income and out-of-pocket spending) that in- Household income quintile cluded age, chronic conditions, insurance 1 (0–20%) 20 22.2 26.2 19.5 32.2 type, health care service use, and health care 22036.5 28.3 17.9 17.3 facility type best fit the data. We ran 2 regres- 32034.1 30.3 19.7 15.9 sion models with and without control for 42036.3 33.4 16.7 13.7 health care service use to measure the direct 5 (80–100%) 20 37 33.4 16.7 12.9 effects of insurance type, health status, and in- a come. We used the following software to con- Other comprises those who do not receive any type of NHI or Medical Aid coverage; they are approximately 1% of the population. bOccupation categories were those used by the KNHS. duct analyses: Excel 2000 (Microsoft Corp, Redmond, Wash), SAS version 8.2 (SAS Insti- tute Inc, Cary, NC), and Stata version 7.0. men had none. The prevalence of 3 or more chronic conditions (32%), whereas the high- RESULTS chronic conditions was significantly higher est quintile had the lowest prevalence (13%). (40%) among those with Medical Aid than This pattern continued for the second-lowest Respondent Characteristics and Number among those with other types of insurance. quintile, which had the second-highest preva- of Chronic Conditions The prevalence of 3 or more chronic condi- lence of 3 or more chronic conditions (17%). Table 1 shows respondents’ age, gender, in- tions was similar among those with govern- surance type, occupation, household income ment employee and teachers insurance (19%) Out-of-Pocket Spending and quintile, and number of chronic conditions. and self-employed insurance (18%), but it was Out-of-Pocket Spending Burden Ratio The prevalence of chronic conditions in- slightly lower among those with private em- In 1998, the mean annual out-of-pocket creased with age, except after age 80 years. A ployee insurance (16%). spending per person was 215700 won higher percentage of women had 3 or more The lowest income quintile also had a dis- (Table 2), or roughly US $179. On average, chronic conditions, and a higher percentage of proportionately high prevalence of 3 or more out-of-pocket spending by low-income

806 | Research and Practice | Peer Reviewed | Ruger and Kim American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 2—Mean Annual Out-of-Pocket Spending (OPS) and Percentage of Out-of-Pocket chronic condition and ranged from 1% for Spending Burden Ratio (OPBR) by Socioeconomic Characteristics and Number of Chronic no chronic conditions to 11% for 3 or more Conditions: Korean National Health and Nutrition Survey (KNHS), 1998 conditions. Low-income Koreans with chronic condi- Number of Chronic Conditions tions had an even higher spending burden Total Population None 1 2 ≥3 that increased with the number of conditions. Characteristic OPS OPBR, % OPS OPBR, % OPS OPBR, % OPS OPBR, % OPS OPBR, % The second-lowest income quintile spent an average of 225000 won (6% of total in- Total population 215700 4.8 68000 1.1 231400 4.5 261500 5.6 419300 11.4 come) on health care. As shown in Table 2, Age, y the lowest income quintile with 3 or more 0–19 92900 2.3 36600 0.8 162500 4.0 102600 2.3 226700 7.8 chronic conditions spent an average of 20–44 214200 4.6 85900 1.7 237500 4.5 273700 7.0 460100 10.5 362800 won (20% of total income) annually 45–65 274100 6.5 75700 1.6 240800 6.2 322100 7.4 398600 9.4 for health care. Moreover, the second-lowest 66–79 321500 8.2 29200 0.7 300000 9.7 257000 6.5 432200 10.0 income quintile with 3 or more chronic con- ≥80 253700 3.5 5500 0.1 490900 3.1 109100 4.0 199800 5.2 ditions spent 514300 won on average on Gender health care, and although their spending as a Male 188000 4.5 42000 0.8 234200 5.6 219200 5.8 425600 10.1 percentage of total income (13%) was less Female 241500 5.4 96600 2.0 228500 4.8 299300 7.1 415600 9.6 than that of the lowest income quintile, it was Insurance type 3 times the percentage of the highest quintile Government employees 302600 6.0 69500 0.9 316400 4.9 269300 7.4 670000 14.4 (4%). The Gini coefficient for out-of-pocket and teachers spending on health care in Korea was 0.7, Private employees 201200 4.1 55100 1.0 213500 4.3 268900 6.0 433100 8.3 whereas the coefficient for income was 0.3, Self-employed 216900 5.5 78400 1.7 234700 6.0 262200 6.7 395100 10.3 which suggests regressive financing (data not Medical Aid 125600 3.8 33400 1.2 126600 2.7 159500 4.9 159000 5.2 shown). Othera 53900 1.8 30700 1.2 48200 1.2 46700 2.3 121700 4.1 The level and burden of out-of-pocket Occupationb spending varied by type of health insurance Professional and 250400 7.0 86900 1.5 424000 12.2 313600 10.4 174900 4.5 (Table 2). Mean out-of-pocket spending and management out-of-pocket spending burden ratio were White-collar 206000 4.0 64000 1.2 249200 5.0 209300 4.3 699700 12.7 lowest among the “other” group (poor Kore- Sales 212100 5.0 59300 1.4 204900 5.2 359000 9.4 325000 5.5 ans who did not receive Medical Aid; 53900 Farming and fishery 221500 9.2 36700 1.3 175900 9.1 217600 7.6 310700 12.8 won, or 2% of annual income) and the Med- Labor 192300 3.8 52200 1.2 180300 2.9 176700 4.0 421900 8.5 ical Aid group (125600 won, or 4% of an- Military 155500 2.6 10500 0.1 373900 6.2 55400 2.3 0 0.0 nual income). It was highest among those Student 121700 2.9 78500 1.4 185100 5.0 112400 3.4 258300 7.2 with government employee or teacher insur- Other 252800 5.2 71700 1.5 247900 4.5 300900 6.4 485400 10.7 ance (302600 won, or 6% of annual in- Household income quintile come) and private employee insurance 1 (0–20%) 228500 12.5 56000 1.6 218400 11.7 216400 11.8 362800 19.5 (201200 won, or 4% of annual income). 2 225000 5.6 101600 1.9 164000 4.1 293800 7.7 514300 13.4 Those with regional self-employed insurance 3 167000 3.3 63100 1.1 139200 3.0 208800 3.9 390600 7.8 spent 216900 won, or 6% of their annual in- 4 223000 3.1 49000 0.6 324500 4.6 178500 2.5 491400 7.0 come. Medical Aid appears to have helped 5 (80–100%) 232800 2.0 65600 0.6 274700 2.3 413900 3.5 369200 3.5 offset some of the costs incurred by the poor Note. Values were weighted in accordance with Korean National Health and Nutrition Survey specifications. OPS in 1000 won. and chronically ill Koreans. Nonetheless, out- a Other comprises those who do not receive any type of NHI or Medical Aid coverage; they are approximately 1% of the population. of-pocket spending burden ratio of 4% to 5% bOccupation categories were those used by the KNHS. suggest persistent gaps in coverage.

Use Rates and Out-of-Pocket Spending Koreans accounted for a significant percent- the highest quintile (2% of total income). by Health Care Facility Type age of income. Although Koreans spent an Mean out-of-pocket spending increased with Although low-income individuals had a estimated 215700 won (5% of total income) additional chronic conditions from 68 000 higher out-of-pocket spending burden ratio, on average on health care, Koreans in the won for no chronic condition to 419300 their use rates were similar to those in other lowest income quintile spent 228 500 won won for 3 or more conditions (Table 2). categories for nearly all types of health care (13% of total income). This out-of-pocket The out-of-pocket spending burden ratio facilities, except public health centers, which spending burden ratio was 6 times that of rose at an increasing rate for each additional were used by 3% of the lowest income

May 2007, Vol 97, No. 5 | American Journal of Public Health Ruger and Kim | Peer Reviewed | Research and Practice | 807  RESEARCH AND PRACTICE 

TABLE 3—Health Services Use and Out-of-Pocket Spending Burden Ratio (OPBR), by Income Quintile and Number of Chronic Conditions and Type of Health Care Facility: Korean National Health and Nutrition Survey, 1998

General Professional Number of Hospital General Hospital Hospital Clinic Oriental Clinic Public HealthCenter Pharmacy Chronic Conditions Use Rate, % OPBR Use Rate, % OPBR Use Rate, % OPBR Use Rate, % OPBR Use Rate, % OPBR Use Rate, % OPBR Use Rate, % OPBR

Quintile 1 (0–20%) 0 0.0 65.2 0.2 24.9 0.2 23.3 1.0 20.0 0.0 8.6 0.1 6.7 1.2 6.0 1 0.2 119.4 0.5 31.4 0.7 38.5 1.3 40.8 0.2 10.4 0.3 3.2 1.7 22.6 2 0.3 14.1 0.5 58.0 0.7 48.0 1.3 31.6 0.1 53.9 0.6 16.8 1.3 27.4 ≥3 0.5 81.0 0.8 48.6 1.4 79.4 2.9 35.2 0.2 200.1a 1.6 12.5 2.7 37.7 Total 1.1 68.8 2.0 46.4 2.9 56.8 6.4 34.6 0.5 104.7a 2.6 11.7 6.7 28.1 Quintile 2 0 0.1 10.7 0.4 15.6 0.4 8.1 2.0 10.0 0.0 370.3a 0.1 0.4 2.1 4.8 1 0.4 23.6 0.4 19.3 0.6 12.0 1.6 9.9 0.2 10.9 0.2 1.5 2.2 5.8 2 0.2 49.6 0.3 9.4 0.5 11.5 1.1 18.5 0.2 62.3 0.2 19.8 1.4 14.3 ≥3 0.4 23.3 0.4 29.8 0.8 23.7 1.5 18.4 0.1 107.7a 0.3 8.2 1.7 14.3 Total 1.1 28.7 1.5 20.5 2.4 13.3 6.2 14.6 0.6 83.1 0.8 7.6 7.4 9.2 Quintile 3 0 0.1 13.1 0.5 5.7 0.5 13.0 2.6 5.0 0.0 11.6 0.1 2.8 2.2 2.1 1 0.3 22.8 0.4 31.7 0.7 9.6 1.9 9.9 0.4 2.5 0.2 2.1 2.0 3.2 2 0.2 45.2 0.3 15.8 0.5 14.5 1.2 12.0 0.2 3.0 0.2 2.1 1.2 7.9 ≥3 0.2 37.0 0.4 37.3 0.6 16.0 1.2 12.3 0.2 64.3 0.3 6.6 1.5 9.2 Total 0.7 27.8 1.7 28.0 2.2 13.3 6.9 10.0 0.8 35.3 0.7 4.1 7.0 5.4 Quintile 4 0 0.2 7.5 0.4 5.3 0.4 10.7 2.6 4.1 0.1 13.3 0.1 1.4 2.3 2.1 1 0.5 39.1 0.4 15.7 0.6 9.2 1.8 11.4 0.4 66.6 0.1 2.8 2.3 6.4 2 0.3 31.6 0.2 35.8 0.4 14.6 0.9 5.9 0.2 12.3 0.1 2.2 1.3 3.5 ≥3 0.2 27.0 0.2 15.0 0.6 7.2 1.0 7.5 0.1 51.3 0.1 3.7 1.2 9.7 Total 1.2 31.1 1.3 16.2 1.9 10.5 6.4 7.6 0.7 49.3 0.3 2.8 7.1 5.3 Quintile 5 (80–100%) 0 0.2 3.3 0.5 6.5 0.5 7.9 2.6 3.8 0.1 40.4 0.1 0.6 2.3 1.4 1 0.2 10.0 0.5 5.4 0.7 13.5 2.0 7.6 0.4 24.1 0.1 10.9 2.0 3.3 2 0.2 16.9 0.3 13.0 0.5 15.2 1.1 4.6 0.3 16.9 0.0 2.4 1.2 4.9 ≥3 0.2 13.2 0.2 10.6 0.4 10.2 0.9 6.4 0.1 4.1 0.1 4.3 1.1 3.9 Total 0.9 11.8 1.5 7.8 2.1 11.4 6.6 5.9 0.9 16.4 0.4 4.2 6.4 3.1

Note. Health care facilities are listed by type as defined by Korean medical law.Values are weighted in accordance with Korean National Health and Nutrition Survey specifications. 0.0% means no person in the income group used the specific health service. Use rate represents the percentage of people who used services, including inpatient and outpatient services. aRepresents spending on outpatient services during the previous 2 weeks; in oriental clinics, people visit to purchase tonics for general, rather than urgent treatments, so type of purchase may affect this percentage.

quintile but were used by only 0.4% of the clinics (104.7% of income), followed by gen- spending, therefore, was spread unevenly highest quintile (Table 3). Among those with eral professional hospitals (69%). It was low- among income groups. Oriental clinics had 3 or more chronic conditions, the lowest- est at public health centers (12%; Table 3). the highest out-of-pocket spending burden income individuals were about 16 times Out-of-pocket spending among the lowest ratio among the lowest-income Koreans with more likely to use public health centers com- income quintile for oriental clinic and hospi- 3 or more chronic conditions (200% of in- pared with the highest quintile (Table 3). tal facilities was approximately 6 times as come) and represented the largest difference Among the lowest income quintile, the burdensome (in terms of the out-of-pocket in the out-of-pocket spending burden ratio out-of-pocket spending burden ratio for spending burden ratio) compared with the between low-income and high-income health care services was highest at Oriental highest quintile. The burden of out-of-pocket Koreans.

808 | Research and Practice | Peer Reviewed | Ruger and Kim American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 4—Multivariate Analysis of Out-of-Pocket Spending: Korean National Health and Oriental clinics had higher out-of-pocket Nutrition Survey, 1998 spending compared with patients who used public health centers. In Model 2, the number a Model 1 Model 2 of chronic illnesses was more strongly associ- Standardized Coefficient Standardized Coefficient ated with out-of-pocket spending than was in- Predictor (95% CI) t P (95% CI) tP surance type. Demographics Women 0.011 (–0.011, 0.068) 1.4 .161 –0.001 (–0.046, 0.044) –0.06 .954 DISCUSSION Age 0.064 (0.003, 0.005) 6.46 <.001 0.045 (0.001, 0.004) 4.05 <.001 Health service use Korea achieved universal coverage at low Visit days 0.376 (0.208, 0.226) 47.08 <.001 ...... cost by offering limited benefits, by requir- Hospitalization days 0.265 (0.027, 0.031) 33.08 <.001 ...... ing high copayments and coinsurance rates, Health status imposing low fees on providers, and by re- No. of chronic conditions 0.105 (0.097, 0.140) 10.87 <.001 0.154 (0.149, 0.197) 14.19 <.001 stricting fee growth to the level of general 18 , 2 2 Household income inflation. Consequently, the health care Household income (log) 0.071 (0.088, 0.145) 8.06 <.001 0.052 (0.054, 0.118) 5.28 <.001 system leaves many citizens relatively un- Insurance type protected. Financial barriers to access of Government employees 0.125 (0.360, 0.699) 6.13 <.001 0.077 (0.134, 0.515) 3.34 .001 care can become insurmountable among Private employers 0.185 (0.333, 0.655) 6.01 <.001 0.111 (0.116, 0.477) 3.21 .001 those with low incomes, particularly among Self-employed 0.177 (0.288, 0.605) 5.53 <.001 0.090 (0.049, 0.405) 2.50 .013 patients with multiple chronic condi- 4,5,23 Other 0.025 (0.107, 0.705) 2.66 .008 0.002 (–0.304, 0.368) 0.19 .852 tions. High cost sharing, high fees for Medical Aid (reference) uninsured services, and the widespread Health care service type practice of informal treatment charges can Hospital and clinic 0.149 (0.355, 0.500) 11.59 <.001 0.178 (0.428, 0.590) 12.28 <.001 constrain the ability to afford necessary Oriental clinic 0.037 (0.202, 0.528) 4.39 <.001 0.043 (0.242, 0.608) 4.54 <.001 care. According to the National Health In- Pharmacy 0.084 (0.152, 0.284) 6.47 <.001 0.088 (0.153, 0.302) 6.00 <.001 surance Corporation, the out-of-pocket Public health center spending for outpatient services accounts (reference) for 65% of total expenditures, and about R 2 0.248 0.049 one half of the patient’s share (34%) is non- Adjusted R 2 0.247 0.048 insurance charges—18% for legal and 16% 24 F 303.69 <.001 56.15 <.001 for illegal informal treatment charges. Medical Aid covers roughly 2% to 3% of Note. Models show analysis of individuals’ estimated annual out-of-pocket spending for health care services. the population. Medical Aid recipients and aModel 2 shows the impact of health and income variables without including control for health service use. those in the lowest income quintiles have a significantly higher burden of illness on average compared with other groups. Because the An out-of-pocket spending burden ratio Model 1 explained 24.7% (P<.001) and Korean NHI uses cost sharing and a restricted greater than 100% does not necessarily mean Model 2 explained 5% (P<.001) of the ad- benefits package to reduce health care spend- an individual spent more than their income justed variance in the dependent variable ing and control costs, it is important to exam- for a medical service. Because of the financial (P<.0001). Regression coefficients of income ine the impact of these policies on vulnerable burden, actual use rate is low. For instance, and insurance type in Model 2 were signifi- populations. the use rate for Oriental clinics, where most cantly lower compared with Model 1. Among Our findings show that, on average, both services are not covered, appears to be less the independent variables in Model 1, outpa- the annual out-of-pocket spending and the than 1%. tient visits were most strongly associated with out-of pocket burden spending ratios increase out-of-pocket spending. One additional visit with the number of chronic conditions. This Multivariate Results day for outpatient services increased out-of- association persisted even after we controlled Table 4 shows standardized coefficients pocket spending by 37%. Inpatient hospital- for a number of factors. Moreover, the aver- and partial multivariate coefficients of deter- ization days also had highly significant effects age out-of pocket spending burden ratio was 6 mination for the multivariate linear regression on the dependent variable, even after we ad- times higher among Koreans in the lowest in- models. We examined possible colinearity is- justed for age, health, and economic condi- come quintile (13%) compared with those in sues by calculating variance inflation factors tions. One additional hospitalization day in- the highest quintile. Low-income individuals for each variable (results of 1 and 1.4 con- creased out-of-pocket spending by 26%. with 3 or more chronic conditions had the firmed colinearity were not a problem). Patients who used hospitals, pharmacies, and highest out-of pocket spending burden ratio

May 2007, Vol 97, No. 5 | American Journal of Public Health Ruger and Kim | Peer Reviewed | Research and Practice | 809  RESEARCH AND PRACTICE 

(20%) compared with their high-income coun- spending costs, which might be considerable adjustments in fee schedules and premiums terparts with 3 or more chronic conditions among patients with multiple chronic illnesses. for system sustainability.31 Empowering low- (4%). Studying the effect of health insurance income and chronically ill individuals to be- on those with 3 or more chronic conditions Conclusions come involved in health system design and showed that those with Medical Aid had lower Our results compare favorably with studies operation is a critical step.32 out-of pocket spending burden ratios com- of out-of-pocket spending in the United We found that low-income individuals with pared with all other insurance groups, with States,27,28 which also found increased out-of- multiple chronic conditions were especially the exception of the “other” category. Never- pocket spending associated with the number vulnerable to cost sharing and coverage restric- theless, Medical Aid recipients experienced an of chronic conditions.29 Moreover, in 1997 tions because they need and use more ser- out-of pocket spending burden ratio of up to low-income older Americans who were not vices, including those individuals with limited 5% on average, which was more than that of enrolled in Medicaid and traditional Medi- coverage. Further research is necessary for bet- highest-income quintile. Employing household care spent 30% of their income out-of-pocket ter understanding the association between out- equivalent income, which incorporates size for health care compared with 23% among of-pocket spending, insurance type, and access and composition of households, enables more those who were enrolled in a Medicare to care among those with significant health accurate estimates of this burden. health maintenance organization.27 Com- problems. It would be especially useful to Thus, our estimates show that although pared with other Organisation for Economic assess the impact over time of health status Medical Aid provides financial protection for a Co-operation and Development countries, on health care service use and out-of-pocket small percentage of low-income NHI benefici- Korea, the United States, and Mexico have spending, particularly the degree to which low- aries, most low-income Koreans experience a the highest private share (percentage) of total income and chronically ill Koreans are forgo- significant burden of medical expenses. As a health care expenditures.30 ing health care or are becoming indebted be- result, many low-income and chronically ill Out-of-pocket payments in Korea, which cause of costs. These findings offer lessons for Koreans have high out-of-pocket spending on represent almost half of overall health financ- the United States and middle-income and low- health care. Analysis of health care service use ing, are regressive for at least 3 reasons.31 income countries that are pursuing health fi- patterns suggests that, despite significant eco- First, payments are high and are unrelated to nancing and insurance reform, particularly in nomic burdens, low-income and chronically ill the ability to pay, particularly for uncovered the Middle East (e.g., Morocco) and in Latin Koreans may sustain use rates by incurring services. Recent efforts to address this issue America (e.g., Mexico). significant debt to obtain care, because uncov- were implemented in July 2004. Since then, ered costs exceed annual income. Our first individuals’ payments on covered services that model found that health care service use vari- cost more than 3000000 won within a 6- About the Authors ables, insurance type, and the number of month period can be reimbursed by the Na- Jennifer Prah Ruger is with the Department of Epidemiol- chronic conditions explained a great deal of tional Health Insurance Corporation. These ogy and Public Health, Yale School of Medicine, New Haven, Conn, the Yale Graduate School of Arts and Sci- the variability in health expenditures, whereas measures require evaluation. Second, cost ences, New Haven, and the Yale Law School, New Haven. the number of chronic conditions was a more sharing on NHI-covered services is waived Hak-Ju Kim is with the Department of Social Welfare, important variable in our second model. only for Medical Aid first-grade beneficiaries, Gyeongsang National University, Jinju, Kyungnam, Repub- lic of Korea. who are a small percentage of the population. Requests for reprints should be sent to Jennifer Prah Ruger, Limitations Third, cost-sharing rates have no annual cap, PhD, Yale University, 60 College St, PO Box 208034, Our study has some limitations. First, it although they are cut by 50% for high-cost New Haven, CT 06520-8034 (e-mail: jennifer.ruger@ yale.edu). 28 might be biased because of failures to respond claims. Social insurance contributions—the This article was accepted April 1, 2006. to the survey and missing values, although second largest financing source—are related to previous studies of the KHNS have not de- income, although contributions are propor- Contributors 25,26 tected bias. Second, household income tional to income only up to a ceiling (25 mil- Both authors originated the study, contributed to the was self-reported and prone to error, although lion won per month). Our analysis shows that study design, and wrote the article. H.J. Kim managed the data and completed the statistical analyses, with it was confirmed by other household mem- many low-income individuals may not receive input from J.P. Ruger. H.J. Kim had full access to the bers. Third, personal interviews were con- Medical Aid protection. As a result, many peo- data. ducted at the end of 1998, just after the Asian ple with low incomes pay a substantial share economic crisis; household income may have of their incomes out-of-pocket for health care. Acknowledgments been significantly lower than normal and may Thus, strategies for reform should include J. P. Ruger is funded in part by a Career Devel- thus have increased out-of pocket spending (1) setting caps for low-income individuals’ opment Award from the US National Institutes of Health (grant K01DA01635801). H. J. Kim burden ratios. Fourth, the out-of-pocket spend- financial burden, (2) determining premiums was funded in part by the Korea National Re- ing in our study did not reflect the 9% in- and copayments on the basis of income levels, search Foundation (grant KRF–2002–003- crease in medical fees allowed by the govern- (3) expanding benefits, (4) removing informal B00142). We thank Keeho Lee of the Korea Gallop Organiza- ment beginning in mid-2000. Fifth, we did treatment charges, (5) changing NHI’s cost- tion and Sungjin Ahn and Yunsung Chung of the De- not consider nonmedical out-of-pocket sharing structure, and (6) making upward partment of Statistics at Gyeongsang National University

810 | Research and Practice | Peer Reviewed | Ruger and Kim American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

for data clarification and technical assistance. We thank Health and Welfare Statistics. Seoul, Korea: Ministry of Susan Gatchel, Kimberly Hannon, and Linda Sage for Health and Welfare; 2002. administrative, research, and editing assistance. 18. Won SJ. Theory of Social Security. Seoul, Korea: Yangsuhwon Press; 2002. Human Participant Protection 19. Atkinson AB, Rainwater L, Smeeding T. Income No protocol approval was needed for this study. Distribution in OECD Countries: Evidence From the Lux- embourg Income Study. Social Policy Studies No. 18. References Paris, France: Organisation for Economic Co-operation Elder Abuse: and Development; 1995. 1. World Health Organization (WHO). The World A Public Health Health Report 2000 Health Systems: Improving Perfor- 20. Coughlin TA, Liu K, McBride TD. Severely dis- mance. Geneva, Switzerland: WHO; 2000. Available at: abled elderly persons with financially catastrophic Perspective http://www.who.int/whr2001/2001/archives/2000/ healthcare expenses: sources and determinants. Geron- en/index.htm. Accessed March 11, 2004. tologist. 19 92;31:391–403. 2. Ministry of Health and Welfare. 2002 White Book 21. Stum MS, Bauer JW, Delaney PJ. Disabled elders’ on Social Welfare. Seoul, Korea: Ministry of Health and out-of-pocket home care expenses: examining financial Welfare; 2002. burden. J Consumer Affairs. 1998;32:82–105. 3. Shin YJ, Yoo WS, Ha HY, Chung SH. Patient’s cost 22. Chung HJ. Out-of-pocket spending by types of dis- sharing for in-patient services in Hospitals. Hanyang J ease. Health Insurance Forum. 2002;1:73–97. Med. 2000;20:79–88. 23. Ruger JP, Jamison DT, Bloom D. Health and the 4. Xu K, Evans DB, Kawabata K, Zeramdini R, economy. In: Merson MH, Black RE, Mills AJ, eds. In- Klavus J, Murray CJL. Household catastrophic health ternational Public Health. Gaithersburg, Md: Aspen Pub- expenditure: a multicountry analysis. Lancet. 2003; lishers; 2001:617–660. 362:111–117. 24. Shin YJ. Expanding benefit coverage in National 5. Ruger JP. Catastrophic health expenditures. Lancet. Health Insurance. Health Welfare Policy Forum. 2003;8: 2003;362:996–997. 34–45. 6. Kim CY, Lee JS, Kang GW, Kim YI. Magnitude of 25. Kim MH, Kim SH. A study on factors causing the helps foster and create a patient’s cost sharing for hospital services in the Na- burden of medical expenses to the elderly with chronic tional Health Insurance in Korea. Korean J Health Policy disease. Korean J Soc Welfare. 2002;48:150–178. national dialogue that Admin. 1999;9:1–14. focuses on systems and 26. Jung SH, Kim HJ. An analysis of health care ex- 7. Yang BM. Coinsurance rates of medical services. penditures for the elderly households. J Welfare Aged. methods concerning Med Sci. 1996;3:80–87. 2001;12:129–151. identification of the problem 8. Kee HS, Moon OR. An analysis of the magnitude 27.Gross DJ, Alexcih L, Gibson MJ, Corea J, Caplan C, and prevention. of self-payment under the health insurance exclusion Brangan N. Out-of-pocket health spending by poor and Edited by Randal W. Summers clauses. J Inst Health Environ Sci. 19 93;3:18–32. near-poor elderly Medicare beneficiaries. Health Serv Res. 1999;34:241–254. and Allan M. Hoffman 9. Noh IC, Kim SC, Lee CS, Han HK. A Study on Cost-Sharing in the National Health Insurance. Seoul, 28. Goldman DP, Zissimopoulos JM. High out-of- ISBN: 0-87553-050-8 Korea: Korean Institute of Health and Affairs; 1989. pocket health care spending by the elderly. Health Aff (Millwood). 2003;22:194–202. 10. Lee BS. A Study on Co-Payments of the National Member: $ 20.25 Health Insurance in Korea. Seoul, Korea: Suhkang Uni- 29.Hwang W, Weller W, Ireys H, Anderson G. Out- versity Press; 1994. of-pocket medical spending for care of chronic condi- Non-Member: $ 28.95 tions. Health Aff (Millwood). 2001;20:267–278. 11. Choi BH, Noh IC, Shin JK, Lee SY. Cost-Sharing and Benefit Structure of Health Insurance. Seoul, Korea: 30.Organisation for Economic Co-operation and De- Korean Institute of Health and Affairs; 1997. velopment (OECD). OECD Health Data 2003: a Com- parative Analysis of 30 Countries. Paris, France: OECD; 12. Shin JK. Benefit coverage of the National Health www.aphabookstore.org 2003. Insurance. Health Welfare Policy Forum. 19 97;9:17–27. 1-866-320-2742 toll free 31.Organisation for Economic Co-operation and De- 13. Lee JK. Out-of-pocket spending for uninsured velopment (OECD). OECD Reviews of Health Care Sys- 1-866-361-2742 fax medical services. Health Insurance Forum. 2002;1: tems: Korea. Paris, France: OECD; 2003. 51–72. 32. Gwatkin DR, Buiya A, Victora CG. Making health 14 . Ministry of Health and Welfare. Major Administra- systems more equitable. Lancet. 2004;364: tive Statistics 1998. Seoul, Korea: Ministry of Health 1273–1280. and Welfare; 1999. 15. Choi BH. Fall Conference of Korean Social Security Association: Crisis and Task in Securing the Public Health System: Reform Alternatives of Korean Social Insurance— Proceedings. Seoul, Korea: Korean Social Security Asso- ciation; 2002:7–59. 16.Korean Institutes of Health and Affairs, Ministry of Health and Welfare. 1998 National Health and Nu- trition Survey: Overlook. Seoul, Korea: Ministry of Health and Welfare; 1999. 17. Ministry of Health and Welfare. Yearbook of

May 2007, Vol 97, No. 5 | American Journal of Public Health Ruger and Kim | Peer Reviewed | Research and Practice | 811  RESEARCH AND PRACTICE 

Self-Rated Health Trajectories in the United States and the United Kingdom: A Comparative Study

| Amanda Sacker, PhD, Richard D. Wiggins, PhD, Mel Bartley, PhD, and Peggy McDonough, PhD

The cause of poorer health of the US popula- Objectives. We reviewed literature on comparative social policy and life tion compared with that of other developed course research and compared associations between health and socioeco- nations has been much debated in recent nomic circumstances during an 11-year period in the United States and the 1–3 years. Some suggest that differences in United Kingdom. population health stem from restricted access Methods. We obtained data from the US Panel Study of Income Dynamics and to resources at the individual level and public the British Household Panel Survey (1990–2002). We used latent transition anal- underinvestment in the human, physical, and ysis to examine change in self-rated health from one discrete state to another; social fabric of society within countries, in- these health trajectories were then associated with socioeconomic measures at cluding health and welfare policies.4 The the beginning and at the end of the study period. more generous, comprehensive, and universal Results. We identified good and poor latent health states, which remained relatively stable over time. When change occurred, decline rather than improve- state programs of social democratic welfare ment was more likely. UK populations were in better health compared with US governments have already been compared populations and were more likely to improve over time. Labor market participa- with the more financially limited and less action was more strongly associated with good health in the United Kingdom than 5–7 cessible programs in the United States. Yet, in the United States. despite these insights, at least 2 significant is- Conclusions. National policies and practices may be keeping more US work- sues remain relatively unexplored in compar- ers than UK workers who are in poor health employed, but British policies may ative research on socioeconomic inequalities give UK workers the chance to return to better health and to the labor force. (Am in health. First, most comparative work on J Public Health. 2007;97:812–818. doi:10.2105/AJPH.2006.092320) health differences has focused on aggregate measures of inequality.8–11 However, if we are to better understand how policies contribute country and differences in these patterns be- Kingdom ranked squarely alongside other Eu- to, maintain, and reduce social inequalities in tween countries. This approach also allowed ropean countries in lifting those at risk out of health, we need between-country compar- us to make stronger statements about the so- relative income poverty via tax and benefits isons of health and inequality at the individ- cial causes and consequences of different systems.24,25 During the 1990s, however, ual level. health patterns. there was convergence between US and UK A second issue is that most comparative We compared health trajectories and their welfare reforms.15 Although we did not test research relies on cross-sectional data12–14 associations with socioeconomic variables in specific hypotheses associated with this devel- despite widespread acknowledgment that so- the United States and the United Kingdom opment, the reform period provides the con- cioeconomic conditions and health have a during the 1990s. The United Kingdom is an text within which we interpreted population complex time-dependent relationship15 and interesting comparator because, like the health patterns in the 2 countries and some analysis of this relationship requires longitudi- United States, it is considered to be a liberal possible causes for these patterns. nal repeated-measures data. For example, re- welfare state,21,22 although some of its policies When comparing health trajectories, we cent research on individual health change or are more closely shared with European social asked 2 questions: what are patterns of indi- trajectories shows that health patterns are democratic welfare states. Recent UK welfare vidual health change in the United States and more variable than previously thought. On reforms resemble the means-testing and the United Kingdom, and how are these pat- average, physical health and function may de- welfare-to-work programs that now dominate terns associated with antecedent and subse- cline with age,16 but there is considerable in- the US social assistance agenda, but the provi- quent socioeconomic circumstances? We dividual variation in this overall pattern.17–20 sion of universal health care and child bene- used data from the US Panel Study of In- This suggests that the population health dis- fits in the United Kingdom are just 2 exam- come Dynamics (PSID) and the British advantage in the United States at one point in ples of important differences in agendas.23 Household Panel Survey (BHPS) to investi- time may tell us very little about national dif- Furthermore, although poverty rates in the gate individual health patterns with a latent ferences in health across individuals’ life United Kingdom and the United States were transition model.26 This approach built upon courses. Because health has stable and dy- higher throughout the 1990s compared with earlier work in which we modeled individual namic components, we investigated patterns the Organisation for Economic Co-operation growth curves as a continuous function of of population health over time within a given and Development average, the United self-rated health over time.27,28 However,

812 | Research and Practice | Peer Reviewed | Sacker et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

health trajectories may be better represented TABLE 1—Respondents’ Self-Rated Health Reported at the Initial Survey Interview and as movement between discrete stages that in- Odds Ratios for Reporting Poor Health 1 Year Later, by Socioeconomic Characteristics: volve not only stable periods or unidirec- PSID, United States, 1990, and BHPS, United Kingdom, 1991 tional change but also intermittent deterioration or improvement. We asked whether and 1990 PSID (N=4042) 1991 BHPS (N=4116) a a how health changes during an 8-year period, Socioeconomic Characteristics % (SE) OR (95% CI) % (SE) OR (95% CI) and whether the reciprocal association be- Self-rated health (PSID rating/BHPS rating) tween health trajectories and socioeconomic Excellent/excellent 30.23 (1.11) 33.46 (0.77) circumstances over time can inform us about Very good/good 37.16 (0.97) 45.97 (0.76) the processes that underlie cross-sectional na- Good/fair 25.97 (1.04) 14.75 (0.56) tional differences in health. Fair/poor 5.66 (0.51) 4.48 (0.29) Poor/very poor 0.99 (0.18) 1.34 (0.18) METHODS Employment status Employed 92.62 (0.60) 1.00 (reference) 80.57 (0.69) 1.00 (reference) Data Unemployed 2.87 (0.30) 1.16 (0.69, 1.96) 5.08 (0.38) 1.49 (0.96, 2.32) The data for our study were from the Economically inactiveb 4.51 (0.44) 1.31 (0.85, 2.03) 14.35 (0.56) 1.73 (1.25, 2.41) 19 90 to 2001 waves of the PSID and the Low incomec 20.00 (1.19) 1.51 (1.21, 1.89) 20.00 (0.86) 1.75 (1.34, 2.30) 19 91 to 2002 waves of the BHPS, which Routine or semiroutine occupationd 28.05 (1.44) 1.73 (1.38, 2.17) 29.09 (0.95) 1.84 (1.44, 2.34) are ongoing studies of representative samples of adults and children living in families Note. PSID=Panel Study of Income Dynamics; BHPS=British Household Panel Survey, OR=odds ratio, CI=confidence interval. in the United States and the United King- aFor reporting poor health 1 year later. dom, respectively.29,30 The PSID began with bEconomically inactive includes those persons who are early retirees, permanently or temporarily disabled, involved in family a national sample of nearly 5000 house- care or keeping house, students, involved in workfare or government training schemes, in prison, or involved in other nonwork activities. holds in 1968. Individuals were interviewed cLow income was defined as those persons in the bottom 20% of adjusted household income.Adjusted incomes were derived annually until 1997 and biannually there- by dividing household income by the square root of household size. d after. The PSID sample (n = 4042) com- Routine and semiroutine occupations are regulated by short-term labor contracts, exchange wages for labor, are highly unsupervised, and have little or no need for employee discretion. prised household heads and their partners who responded to the self-reported health question, were aged 25 to 55 years in 1991, and who had complete covariate data in when the wording of the question was identi- (1992 to 2000) to analyze health transitions 19 9 0. The BHPS was initiated in 1991 and cal to that of the PSID. during an 8-year period. is an annual survey of approximately 5500 Table 1 shows respondents’ descriptions Socioeconomic variables were measured in private households composed of approxi- of their health at the beginning of the obser- 19 90 and 2000 for the PSID and in 1991 mately 9000 individuals aged 16 years and vation period. Without a “very good” cate- and 2001 for the BHPS. Respondents were older. The BHPS sample (n = 4116) com- gory, more of the UK respondents (33.5%) classified as employed, unemployed, or eco- prised household members aged 25 to 55 reported their health as excellent compared nomically inactive. The economically inactive years in 1992 who had self-reported health with US respondents (30.2%), but fewer US included those who were not employed be- data that year and complete covariate data respondents rated their health as less than cause they were early retirees, permanently in 1991. good (6.7%) compared with UK respondents or temporarily disabled, involved in family (20.6%). Differences in the marginal distri- care or keeping house, students, involved in Measures butions of health measures that were incon- workfare or government training schemes, in In the PSID, respondents were asked, sistent with reports of poorer health in the prison, or involved in other nonwork activi- “Would you say your health in general is ex- US together with differences in response cat- ties (such as unpaid charity work). Low in- cellent, very good, good, fair, poor?” This egory labels suggest that the 2 variables come was defined as being in the bottom question was asked each year from 1990 to were not equivalent measures in their raw 20% of adjusted household income. Ad- 2001, except in 1998 and 2000. In the states. This evidence underpinned the ra- justed incomes were derived by dividing BHPS, respondents were asked, “Please think tionale for examining latent health rather household income by the square root of back over the last 12 months about how your than the observed responses for this compar- household size.31 We used routine or semi- health has been. Compared with people of ative study. To account for the different routine occupation as a dummy variable. Em- your own age, would you say that your health health measure in the 1999 BHPS and the ployees in routine and semiroutine occupa- has on the whole been excellent, good, fair, skipped interview years of the PSID, we tions are regulated by short-term labor poor, very poor?” This question was asked used self-rated health for alternate years of contracts, exchanging wages for labor in annually from 1991 to 2002, except in 1999 the PSID (1991 to 1999) and the BHPS highly supervised conditions with little or no

May 2007, Vol 97, No. 5 | American Journal of Public Health Sacker et al. | Peer Reviewed | Research and Practice | 813  RESEARCH AND PRACTICE 

need for employee discretion. For the UK Third, the stability of and change in the un- probability weights accounted for differential data, this variable was defined by the Office observed latent health states must be charted. sampling in the 2 surveys, and standard errors of National Statistics classification of an indi- Initial health state probabilities at time t0 were adjusted for the clustered sample design. vidual’s current or most recent occupation.32 were regressed on the set of background vari- For the US data, it was identified by a 3-digit ables, similar to step 1. In the LTA, change RESULTS occupation code from the Census of Popula- was modeled as the probability of a transition tion Alphabetical Index of Industries and Occu- from one health state to another at time tn , a Identifying and Validating the Number of pation.33 Self-reports of specific health prob- probability that depended on the health state Underlying or Latent Classes lems were asked of respondents in both at tn–1 . This LTA model is known also as a The first step identified 2 latent health surveys (in 2001 for the PSID and 2002 for hidden first-order Markov model; the se- states underlying the observed responses to the BHPS), which enabled us to validate our quence of transitions or movement between the self-rated health question: a good health latent class analysis of self-rated health. The health states over time is an individual’s tra- state and a poor health state. There were following conditions were recorded in both jectory.35 The transition probabilities at time considerable cross-national similarities in the surveys: cardiovascular disease, diabetes, tn did not depend on health state at tn–2 and distribution of response categories for the 2 cancer, lung disease, and self-reported emo- were the same for all tn . A model that al- health states (data not shown). Those who tional problems. lowed the transition probabilities to be influ- were in underlying good health also had a enced by the background variables produced very high probability (P >0.94) of endorsing Analysis unstable estimates. the top 2 categories of the 5-point self-rated We analyzed health trajectories with latent Fourth, transition patterns must be summa- health questionnaire item. Those who were transition analysis (LTA), which examines rized. The LTA estimates the probability of in poor health were more variable in their movement into and out of latent health an individual being in n health states on 5 responses, and most reported the middle classes over time, and we considered the occasions, which provides n5 possible pat- category. socioeconomic circumstances associated with terns for describing health transitions; we Table 1 shows the odds for poor health at this movement. summarized these patterns into a smaller baseline as predicted by the socioeconomic LTA requires a number of discrete steps. number of discrete trajectories. The probabil- variables 1 year earlier, after control for demo- First, an optimal number of latent classes, or ity of an individual having a particular health graphic variables. In the United States, low in- unobserved categories of health status, must trajectory is the sum of the probabilities of come and a routine or semiroutine occupation be identified such that, in any year, respon- belonging to the transition patterns that make in 1990 increased the odds for being in a poor dents within the same latent class are homo- up that trajectory. health state rather than a good health state in geneous with respect to their observed re- Fifth, socioeconomic differences in transi- 19 91. Employment status in 1990 did not sponses to self-rated health, and respondents tion patterns must be estimated. To examine have an independent effect on health state. In in different latent classes have dissimilar re- the reciprocal influences of health and socio- the United Kingdom, those who were economi- sponses. The latent class model takes mea- economic circumstances over time, we esti- cally inactive in 1991 were more likely to be surement error into account by allowing la- mated trajectory-specific prevalence rates of in a poor health state than a good health state tent states to diverge from what is imposed by the socioeconomic variables twice: 1 year be- in 1992. The same was true for having low in- the common practice of dichotomizing ob- fore the assessment of the health trajectories come or a routine or semiroutine occupation. served responses into good and poor health. (1990 in the PSID and 1991 in the BHPS), The health status of those who were in a In addition to employment status, low in- and 2 years after the end of the health trajec- good health state or a poor health state was come, and occupation, the latent health states tories (2001 in the PSID and 2002 in the confirmed during the validation step (data not were regressed on work-limiting illness, age in BHPS). The precision of the estimates uses shown). Despite some reporting differences years, gender, race/ethnicity (White or non- individual probabilities of having each health between the United States and the United White), education (completed by age 16 years trajectory established in the previous step. Kingdom, few individuals who were in a good and completed after age 16 years), and quali- (More details about the LTA are available as latent health state in either country reported fications (no undergraduate degree or equiva- a supplement to the online version of this any health problems. Less than 10% of those lent or undergraduate degree level or higher), article.) who were in a good latent health state re- all of which were measured in 1990 (PSID) All analyses were conducted with Mplus v4 ported any health problem in either survey, or 1991 (BHPS). software,36 which applies robust maximum whereas with the less healthy, more than Second, the discriminant validity of the likelihood estimation with the assumption that 74% reported 1 or more health problems. latent states must be verified. We used data missing data are missing at random.37 This from 2001 (PSID) and 2002 (BHPS) in a made full use of data from individuals who Charting Stability and Change in Latent confirmatory analysis to estimate prevalence did not respond to all questions, who dropped Health States rates of the selected health problems in each out of the survey, or were not interviewed The LTA model estimated that respon- latent health class.34 in 1 or more waves of the survey. Inverse dents in the UK survey were healthier both

814 | Research and Practice | Peer Reviewed | Sacker et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

at baseline (75% in a good health state in health state 2 years later was estimated moved into a poor health state (P = 0.052; 19 92) and at the end of the observation pe- to be 0.047 (95% confidence interval 95% CI = 0.040, 0.065), but there was a riod (63% in a good health state in 2000). In [CI] = 0.032, 0.062), whereas the probability significantly greater probability of recovery 19 91, 55% of US respondents were in a of moving from a poor health state to a good in the United Kingdom (P = 0.028; 95% good health state, and by 1999, only 48% health state was 0.010 (95% CI = 0.003, CI=0.013, 0.044). Thus, the cross-sectional were. In the United States, the probability of 0.017). Transition probabilities were quite finding of better health in the United King- moving from a good health state to a poor similar among the UK respondents who dom was parallel with more movement from poor health to good health among UK respondents. Nevertheless, the chances of recovery for both populations were still slim. TABLE 2—Estimated Class Counts for Latent Class Patterns During the 5 Biennial Surveys: PSID, United States, 1991–2001, and BHPS, United Kingdom, 1992–2002 Summarizing Transition Patterns The LTA model predicted the proportion Trajectory Groupa Latent Class Patternb PSID, no. (proportion) BHPS, no. (proportion) of individuals who experienced each of the Stable good 1 1 1 1 1 1847 (0.46) 2480 (0.60) 32 possible health transitions during the 8- Declining 1 1 1 1 2 92 (0.02) 137 (0.03) year period (Table 2). We grouped the transi- 1 1 1 2 2 95 (0.02) 141 (0.03) tion patterns into 5 trajectories: stable good 1 1 2 2 2 99 (0.02) 144 (0.04) (5 waves in a healthy state), declining 1 2 2 2 2 103 (0.03) 148 (0.04) (healthy at start to less healthy at end), im- Improved 2 1 1 1 1 16 (0.00) 25 (0.01) proved (less healthy at start to healthy at 2 2 1 1 1 16 (0.00) 26 (0.01) end), stable poor (5 occasions in a less 2 2 2 1 1 17 (0.00) 26 (0.01) healthy state), and intermittent change (all 2 2 2 2 1 18 (0.00) 27 (0.01) other combinations). The largest groups were Stable poor 2 2 2 2 2 1728 (0.43) 927 (0.23) those that remained in good health during Intermittent change 1 1 1 2 1 1 (0.00) 4 (0.00) the whole period. Next came a large minority 1 1 2 1 1 1 (0.00) 4 (0.00) in both populations who stayed in poor 1 1 2 1 2 0 (0.00) 0 (0.00) health. Considerably fewer individuals had 1 1 2 2 1 1 (0.00) 4 (0.00) declining health trajectories, and less than 1 2 1 1 1 1 (0.00) 4 (0.00) 3% were in the improved trajectory in either 1 2 1 1 2 0 (0.00) 0 (0.00) survey. The remaining participants (< 1%) 1 2 1 2 1 0 (0.00) 0 (0.00) had indeterminate trajectories and are not 1 2 1 2 2 0 (0.00) 0 (0.00) discussed further. 1 2 2 1 1 1 (0.00) 4 (0.00) Estimating Socioeconomic Differences 1 2 2 1 2 0 (0.00) 0 (0.00) in Transition Patterns 1 2 2 2 1 1 (0.00) 4 (0.00) Table 3 shows the association between 2 1 1 1 2 1 (0.00) 1 (0.00) the socioeconomic variables and health be- 2 1 1 2 1 0 (0.00) 0 (0.00) fore and after measurement and identifica- 2 1 1 2 2 1 (0.00) 1 (0.00) tion of the health trajectories. We examined 2 1 2 1 1 0 (0.00) 0 (0.00) health trajectories within each country and 2 1 2 1 2 0 (0.00) 0 (0.00) found that those individuals who were in sta- 2 1 2 2 1 0 (0.00) 0 (0.00) ble poor health were the worst off socioeco- 2 1 2 2 2 1 (0.00) 1 (0.00) nomically, and the opposite was the case for 2 2 1 1 2 1 (0.00) 1 (0.00) those who were in stable good health. At 2 2 1 2 1 0 (0.00) 0 (0.00) baseline, the socioeconomic characteristics 2 2 1 2 2 1 (0.00) 1 (0.00) of individuals who were in the declining 2 2 2 1 2 1 (0.00) 1 (0.00) health trajectory more closely resembled Total 4042 (100.00) 4116 (100.00) those of the stable good health group. Note. PSID=Panel Study of Income Dynamics; BHPS=British Household Panel Survey. Eleven years later, these 2 groups became aTransition patterns were grouped into 5 trajectories: stable good (5 waves in a healthy state), declining (healthy at start to more differentiated as the proportions of less healthy at end), improved (less healthy at start to healthy at end), stable poor (5 waves in a less healthy state), and nonemployed and low-income individuals in intermittent change (all other combinations). bPattern for self-rated health every 2 years from 1991 to 1999 (PSID) and 1992 to 2000 (BHPS); 1=good health latent the declining health group grew compared class, and 2=poor health latent class. with those in the stable good health group. The socioeconomic profile of those who

May 2007, Vol 97, No. 5 | American Journal of Public Health Sacker et al. | Peer Reviewed | Research and Practice | 815  RESEARCH AND PRACTICE 

TABLE 3—Socioeconomic Characteristics and Estimated Percentage of Latent Health socioeconomic position. It is important to Trajectory Members Before and After Study Initiation: PSID, United States, 1991–2001, note the higher levels of nonemployment in and BHPS, United Kingdom, 1992–2002 the United Kingdom compared with the United States, regardless of health or occa- Stable Good Health, Declining Health, Improved Health, Stable Poor Health, sion, although the magnitude of this differ- Characteristic Estimated % (95% CI) Estimated % (95% CI) Estimated % (95% CI) Estimated % (95% CI) ence declined considerably over time be- PSID cause labor force participation increased Percentage responding 45.8% 9.6% 1.6% 42.8% among all health groups in the UK popula- 1990 tion, with the exception of those in the de- Employed 95.2 (94.1, 96.3) 95.1 (93.8, 96.4) 92.7 (90.5, 95.0) 89.3 (87.7, 90.9) clining health group. Unemployed 1.8 (1.2, 2.4) 1.8 (1.1, 2.5) 2.3 (1.5, 3.1) 4.2 (3.4, 5.1) As already noted, there were many similar- Economically inactivea 3.0 (2.1, 3.8) 3.1 (1.9, 4.3) 5.0 (3.0, 6.9) 6.5 (5.1, 7.8) ities in the socioeconomic profiles of the Low incomeb 12.7 (10.7, 14.7) 14.9 (12.3, 17.5) 24.5 (20.1, 28.9) 28.7 (25.4, 32.1) health trajectory groups between the 2 coun- Routine or semiroutine 18.5 (15.9, 21.0) 23.8 (19.7, 27.9) 33.8 (28.8, 38.8) 39.0 (35.5, 42.5) tries both at baseline and at the end of the occupationc study, but differences also were evident. 2001 Notably, improved health in the United King- Employed 92.4 (90.9, 93.9) 85.0 (82.3, 87.7) 90.4 (86.2, 94.7) 85.3 (83.6, 87.0) dom was associated with significant increases Unemployed 1.0 (0.5, 1.6) 2.4 (0.9, 3.8) 1.1 (0.7, 1.6) 2.9 (2.1, 3.8) in economic activity, but this was not the case Economically inactivea 6.5 (5.2, 7.8) 12.6 (10.1, 15.2) 8.4 (4.2, 12.6) 11.8 (10.3, 13.3) in the United States. However, improved Low incomeb 11.7 (10.1, 13.3) 20.1 (16.5, 23.6) 19.3 (13.2, 25.5) 29.8 (26.3, 33.3) health was associated with increases in Routine or semiroutine 14.6 (12.4, 16.8) 18.8 (14.9, 22.8) 20.5 (14.3, 26.6) 27.5 (24.8, 30.3) income in the United States, a change over occupationc time that was not observed for the stable BHPS poor health group. Thus, we found signifi- Percentage responding 60.3% 13.8% 2.5% 22.6% cantly higher rates of economic inactivity 1991 among those who were in poor health in the Employed 85.6 (84.4, 86.9) 84.4 (82.3, 86.5) 68.0 (63.0, 73.1) 66.1 (63.0, 69.2) United Kingdom compared with in the United Unemployed 4.0 (3.3, 4.7) 5.1 (3.8, 6.5) 7.6 (4.5, 10.8) 7.6 (6.0, 9.2) States, but this was combined with a signifi- Economically inactivea 10.4 (9.3, 11.4) 10.5 (8.9, 12.1) 24.3 (19.8, 28.9) 26.3 (23.6, 29.1) cant return to the UK labor market among Low incomeb 15.3 (13.8, 16.9) 17.4 (15.2, 19.5) 30.1 (25.3, 34.9) 32.9 (29.3, 36.4) those whose health improved over time. Routine or semiroutine 23.1 (21.1, 25.1) 27.5 (24.9, 30.0) 40.5 (35.9, 45.1) 44.8 (41.7, 47.9) occupationc DISCUSSION 2002 Employed 90.4 (89.3, 91.6) 82.5 (80.2, 84.8) 84.4 (80.9, 87.8) 71.2 (68.5, 73.9) Crossnational Comparisons Unemployed 1.2 (0.8, 1.6) 2.3 (1.3, 3.3) 1.5 (0.5, 2.4) 2.4 (1.5, 3.2) Our study is the first to compare the health Economically inactivea 8.3 (7.2, 9.4) 15.2 (12.9, 17.5) 14.2 (10.6, 17.7) 26.4 (23.8, 29.0) trajectories of individuals in the United States Low incomeb 14.4 (12.7, 16.1) 21.6 (18.5, 24.6) 29.7 (23.8, 35.5) 33.6 (30.2, 36.9) with those of individuals in the United King- Routine or semiroutine 22.0 (19.9, 24.1) 26.9 (23.4, 30.3) 32.0 (25.8, 38.1) 34.1 (30.4, 37.9) dom and to examine the role of socioeco- occupationc nomic circumstances in these patterns. As such, we have made 3 contributions to com- Note. PSID=Panel Study of Income Dynamics, BHPS=British Household Panel Study; CI=confidence interval. aEconomically inactive includes those that are early retirees, permanently or temporarily disabled, involved in family care or parative research on health and social in- keeping house, students, involved in workfare or government training schemes, in prison, or involved in other nonwork equalities. First, we suggest that, although activities. there may be an overall health advantage in bLow income was defined as those in the bottom 20% of adjusted household income.Adjusted incomes were derived by dividing household income by the square root of household size. the United Kingdom, the distribution of latent cRoutine and semiroutine occupations are regulated by short-term labor contracts, exchange wages for labor, are highly health states and change in health states over unsupervised, and have little or no need for employee discretion. time were quite similar in the 2 countries. Four main trajectories placed the vast majority of the 2 populations in stable health states were in the improved health group in both more favorable. Despite this progress, the so- (good or poor) during the 8-year period, and countries was initially similar to that of the cioeconomic conditions of the improved the minority was in the declining or improved stable poor health group; however, with health group never caught up with those of groups. The health of the US population was time, a gap evolved between them as the the stable good health group. no more likely to deteriorate than the health employment, income, and occupational sta- We next compared between-country of the UK population, but the latter popula- tus of the improved health group became associations between health trajectories and tion had a higher likelihood for improvement.

816 | Research and Practice | Peer Reviewed | Sacker et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Health and Socioeconomic the study period.38 When faced with poor classes with understandable measurement Circumstances employment prospects, applying for disability heterogeneity. The analysis also accounted for Second, we have shown processional asso- benefits may have become more attractive the uncertainty in allocating individuals to la- ciations between health and socioeconomic to unemployed individuals who had health tent classes and to trajectory groups. Because circumstances. In fact, when time is consid- problems, especially those who had few com- of this, comparisons of the latent class distri- ered, the associations look quite different petitive skills.39,40 Another factor is access to butions with transition probabilities between from the cross-sectional findings. The latter health care. The universal health insurance the United States and the United Kingdom suggested that employment status exerted no system in the United Kingdom contrasts were possible.41 effect on subsequent health state in the sharply with the largely private system in the United States (Table 1); however, nonemploy- United States, which ties insurance benefits Conclusions ment was clearly associated with health tra- (when they exist) to employment. Because Our study of health trajectories paints a jectories (Table 3). Another observation re- of this constraint, individuals in the United more complex picture of comparative health garding process is that we found evidence for States may be forced to continue working to in both the United States and the United King- both social causation mechanisms (socioeco- ensure access to medical care, regardless of dom than has been previously shown—i.e., the nomic disadvantage causes poor health) and their health. health status of the UK population was not health selection mechanisms (i.e., that social Although these explanations are plausible only better but also more likely to improve mobility affects health). Social causation for cross-sectional national differences in em- over time. This finding suggests that it is im- would predict more nonemployment, lower ployment status and health, they do not nec- portant for future comparative health research income, and routine or semiroutine occupa- essarily account for differences in change: to consider the associations between individ- tional status among the declining health during the study period, nonemployment fell ual and structural factors and health change group compared with the stable good health in the United Kingdom (except among the and health state. In this regard, we found clear group. This was found among BHPS respon- declining health group), and it rose in the evidence of the socioeconomic consequences dents and, to a lesser extent, among PSID re- United States. Moreover, improved health was of health change. However, our comparative spondents. Those who had improved health associated with improved chances for employ- examination raised intriguing questions about also had a small advantage in initial occupa- ment in the United Kingdom but not in the what processes and structures are keeping tional class and unemployment levels com- United States. Although the US pattern is con- more US than UK individuals employed while pared with those who had stable poor health sistent with the aging of this population co- at the same time returning proportionally in the PSID, but this was not found in the hort and their associated withdrawal from the more UK individuals with improved health to BHPS. Health selection was shown in the labor force, the results for the United King- the labor force. No matter what explanations PSID by a widening gap in the socioeco- dom are more puzzling. They may be associ- are ultimately determined through continuing nomic conditions of those in the declining ated with the sharper drop in unemployment comparative research, the evidence challenges health group between 1990 and 2001 com- in the United Kingdom (from 7.5% to 5.8%) researchers to answer several questions. What pared with those in the stable good health compared with the United States (from 5.5% are the individual-level processes that sustain group, and there was less unemployment and to 4.8%) during the study period.38 More sat- persistent states of good or poor health? What low income in 2001 among those in the im- isfying explanations may emerge from further do these processes have in common? To what proved health group compared with those in research; nevertheless, our findings suggest extent can governments adapt employment the stable poor health group. Final economic that individual factors and structural factors opportunities and access to health care to fa- activity rates among the declining health interact in complex ways. cilitate ready engagement with the labor mar- group and the improved health group suggest ket as health improves? The answers to these health selection processes had little impact on Measurement Equivalent questions are fundamental for reducing social socioeconomic inequalities in the health of Finally, a third contribution of our research inequalities in health and thus improving the BHPSrespondents in 2002. is that it offers a practical way for dealing population health of all nations. Among all between-country differences in with a common problem in comparative re- process, the socioeconomic characteristic that search—the lack of equivalence in measure- stood out was employment status. Nonem- ment. LTA, when undertaken with appropri- About the Authors Amanda Sacker and Mel Bartley are with the Depart- ployment was higher in the United Kingdom ate checks for validation, appears to be a ment of Epidemiology and Public Health, University Col- compared with the United States throughout useful means for managing this. We found lege London, England. Richard D. Wiggins is with the the study period, an observation that may that a 2-class model fitted the data for both Department of Sociology, City University, London. Peggy McDonough is with the Department of Public Health Sci- partly be understood by macroeconomic con- countries. The pattern of conditional response ences, University of Toronto, Toronto, Ontario. ditions and social policies. For example, the probabilities for each class was similar, but Requests for reprints should be sent to Amanda Sacker, unemployment rate was, on average, 2.4% not completely invariant, across both coun- PhD, Dept of Epidemiology and Public Health, University College London, 1–19 Torrington Pl, London, WC1E higher in the United Kingdom compared with tries. Thus, the nonequivalent raw measures 6BT, UK (e-mail: [email protected]). the United States on an annual basis during were translated into homogenous latent This article was accepted May 21, 2006.

May 2007, Vol 97, No. 5 | American Journal of Public Health Sacker et al. | Peer Reviewed | Research and Practice | 817  RESEARCH AND PRACTICE 

Contributors data from seven post-communist countries. Soc Sci Med. British Household Panel Survey User Manual Volume A: A. Sacker originated the study, conducted the data 2000;51:1343–1350. Introduction, Technical Report and Appendices. Col- analyses, and drafted the article. P. McDonough drafted chester, England: University of Essex; 2003. 13.Pampel FC. Inequality, diffusion, and the status the article and conducted literature reviews. M. Bartley gradient in smoking. Soc Probl. 2002;49:35–57. 31. Kunst A, Bos V, Mackenbach J, the EU Working and R.D. Wiggins assisted with writing the article and Group on Socioeconomic Inequalities in Health. Moni- interpreting the findings. All authors were actively in- 14 . Sanmartin C, Ross NA, Tremblay S, Wolfson M, toring Socioeconomic Inequalities in Health in the Euro- volved in the editing process. Dunn JR, Lynch J. Labour market income inequality pean Union: Guidelines and Illustrations. Rotterdam, the and mortality in North American metropolitan areas. Netherlands: Department of Public Health, Erasmus J Epidemiol Community Health. 2003;57:792–797. Acknowledgments University; 2001. 15. House J. Understanding social factors and inequal- 32. Office for National Statistics. The National Statis- This study was funded in part by the Canadian Insti- ities in health: 20th century progress and 21st century tics Socio-economic Classification User Manual. Bas- tutes of Health Research (grant PPR–79227). A. Sacker prospects. Health Soc Behavior. 2002;43:125–142. also was supported by the Medical Research Council ingstoke, England: Palgrave Macmillan; 2005. 16. Kahng SK, Dunkle RE, Jackson JS. The relation- (grant G0100222) and the Economic and Social Re- 33. US Dept of Commerce, Bureau of the Census. ship between the trajectory of body mass index and search Council (grant L326253061). M. Bartley was 1990 Census of the Population: Alphabetical Index of health trajectory among older adults—multilevel model- supported by the Economic and Social Research Coun- Industries and Occupations. Publication CPH-R-3 ed. ing analyses. Res Aging. 2004;26:31–61. cil (grant RES000230588). Washington, DC: US Government Printing Office; Data from the British Household Panel Survey were 17. Aldwin CM, Spiro A, Levenson MR, Bosse R. 19 92. supplied by the Economic and Social Research Council Longitudinal findings from the Normative Aging Study. 34.Croudace TJ, Jarvelin M-R, Wadsworth MEJ, Jones data archive. 1. Does mental-health change with age. Psychol Aging PB. Developmental typology of trajectories to nighttime Note. Those who carried out the original collection 1989;4:295–306. and analysis of the data bear no responsibility for its bladder control: epidemiologic application of longitudi- further analysis and interpretation. 18. Aldwin CM, Spiro A, Levenson MR, Cupertino AP. nal latent class analysis. Am J Epidemiol. 2003;157: Longitudinal findings from the normative aging study: 834–842. III. Personality, individual health trajectories, and mor- 35. Raudenbush SW. Comparing personal trajectories tality. Psychol Aging. 2001;16:450–465. Human Participants Protection and drawing causal inferences from longitudinal data. No protocol approval was required for this study. 19. McDonough P, Berglund P. Histories of poverty Annu Rev Psychol. 2001;52:501–525. and self-rated health trajectories. J Health Soc Behav. 36. Muthén LK, Muthén BO. Mplus Statistical Analysis 2003;44:200–216. References with Latent Variables. User’s Guide. 4th ed. Los Angeles, 1. Jenkins C, Runyan D. What’s killing Americans in 20. Clipp E, Pavalko E, Elder G. Trajectories of health: Calif: Muthén & Muthén; 1998–2006. in concept and empirical pattern. Behav, Health Aging. the prime of life? Int J Health Serv. 2005;35:291–311. 37. Little TD, Rubin DB. Statistical Analysis With 19 92;2:159–179. 2. Kunitz S, Pesis-Katz I. Mortality of White Ameri- Missing Data. 2nd ed. New York, NY: John Wiley and cans, African Americans, and Canadians: the causes 21. Esping-Andersen G. The Three Worlds of Welfare Sons; 2002. Capitalism. Cambridge, England: Polity Press; 1990. and consequences for health of welfare state institu- 38. Bureau of Labor Statistics. Comparative Civilian tions and policies. Milbank Q. 2005;83:5–39. 22. Esping-Andersen G. Social Foundations of Post- Labor Force Statistics, 10 Countries, 1960–2004. 3. Banks J, Marmot M, Oldfield Z, Smith J. Disease industrial Societies. Oxford, England: University Press; Washington, DC: US Dept of Labor; 2005. 1999. and disadvantage in the United States and in England. 39. Autor DH, Duggan MG. The rise in disability rolls JAMA. 2006;295:2037–2045. 23. Walker R, Wiseman M, eds. The Welfare We and the decline in unemployment. Q J Econ. 2003;118: 4. Lynch J, Smith GD, Harper S, et al. Is income in- Want? The British Challenge for American Reform. 157–205. Bristol, UK: The Policy Press; 2003. equality a determinant of population health? Part 1. 40.Walker R, Howard M. The Making of a Welfare A systematic review. Milbank Q. 2004;82:5–99. 24. Förster M, d’Ercole MM. Income Distribution and Class? Benefit Receipt in Britain. Bristol, England: The Poverty in OECD Countries in the Second Half of the 5. Leisering L, Leibried S. Time and Poverty in West- Policy Press; 2000. 1990s: OECD Social, Employment and Migration ern Welfare States. New York, NY: Cambridge Univer- 41. McCutcheon AL, Hagenaars JA. A comparative Working Papers; 2005 18 Feb 2005. Report No: 22 sity Press; 1999. social research with multi-sample latent class models. DELSA/ELSA/WD/SEM(2005)1. 6. Smeeding T, Rainwater L, Burtless G. US poverty In: Rost J, Langeheine R, eds. Applications of Latent 25. Moller S, Bradley D, Huber E, Nielsen F, Stephens in cross-national context. In: Danziger SH, Haveman RH, Trait and Latent Class Models in the Social Sciences. JD. Determinants of relative poverty in advanced capi- eds. Understanding Poverty. New York, NY: Russell Sage New York, NY: Waxmann; 1997:266–277. talist democracies. Am Socio Rev. 2003;68:22–51. Foundation; 2001:162–189. 26. Collins LM, Hyatt SL, Graham JW. Latent transi- 7. Walker R. Social Security and Welfare: Concepts tion analysis as a way of testing models of stage- and Comparisons. New York, NY: Open University sequential change in longitudinal data. In: Little T, Press; 2005. Schnabel K, Baumert J, eds. Modeling Longitudinal and 8. Gravelle H, Wildman J, Sutton M. Income, income Multilevel Data. Mahwah, NJ: Lawrence Erlbaum Asso- inequality and health: what can we learn from aggre- ciates; 2000;147–161. gate data? Soc Sci Med. 2002;54:577–589. 27. McDonough P, Sacker A, Wiggins RD. Time on 9. Judge K. Income distribution and life expectancy: my side? Life course trajectories of poverty and health. a critical appraisal. BMJ. 19 95;311:1282–1287. Soc Sci Med. 2005;61:1795–1808. 10. Legrand J. Inequalities in health—some interna- 28. Sacker A, Clarke P, Wiggins RD, Bartley M. Social tional comparisons. Euro Econ Rev. 19 87;31:182–191. dynamics of health inequalities: a growth curve analysis of aging and self assessed health in the British 11. Mellor J, Milyo J. Reexamining the evidence of an household panel survey 1991–2001. J Epidemiol Com- ecological association between income inequality and munity Health. 2005;59:495–501. health. J Health Polit Policy Law. 2001;26:487–522. 29. Hill M. The Panel Study of Income Dynamics: A 12. Bobak M, Pikhart H, Rose R, Hertzman C, User’s Guide. Newbury Park, Calif: Sage; 1992. Marmot M. Socioeconomic factors, material inequalities, and perceived control in self-rated health: cross-sectional 30.Taylor MF, Brice J, Buck N, Prentice-Lane E, eds.

818 | Research and Practice | Peer Reviewed | Sacker et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Environmental Health Practice: Statistically Based Performance Measurement

| Richard T. Enander, PhD, Ronald N. Gagnon, MBA, R. Choudary Hanumara, PhD, Eugene Park, PhD, Thomas Armstrong, BS, and David M. Gute, PhD, MPH

Interest in the use of statistical measures to Objectives. State environmental and health protection agencies have tradi- evaluate the effectiveness of environmental tionally relied on a facility-by-facility inspection-enforcement paradigm to achieve and human health protection programs has compliance with government regulations. We evaluated the effectiveness of a increased sharply in recent years. In 3 con- new approach that uses a self-certification random sampling design. secutive grant rounds (ending with the Methods. Comprehensive environmental and occupational health data from a 2006 solicitation), for example, the US En- 3-year statewide industry self-certification initiative were collected from repre- vironmental Protection Agency’s National sentative automotive refinishing facilities located in Rhode Island. Statistical com- Center for Environmental Innovation parisons between baseline and postintervention data facilitated a quantitative awarded competitive grants to 14 states for evaluation of statewide performance. the development of pilot and large-scale per- Results. The analysis of field data collected from 82 randomly selected automotive refinishing facilities showed statistically significant improvements (P<.05, formance measurement–based initiatives Fisher exact test) in 4 major performance categories: occupational health and modeled after the Massachusetts Environ- 1,2 safety, air pollution control, hazardous waste management, and wastewater dis- mental Results Program (ERP). This indus- charge. Statistical significance was also shown when a modified Bonferroni ad- try self-certification approach is viewed as a justment for multiple comparisons was performed. simple and efficient way to control chemical Conclusions. Our findings suggest that the new self-certification approach to hazards and environmental releases. The environmental and worker protection is effective and can be used as an adjunct program accommodates both multimedia to further enhance state and federal enforcement programs. (Am J Public Health. pollution (air, water, and hazardous waste, 2007;97:819–824. doi:10.2105/AJPH.2006.088021) for example) and occupational health issues and reaches a larger segment of the regu- automotive refinishing industry). By using state agencies to several US industry sectors, lated community than traditional enforce- analytic techniques that follow accepted sta- including confined animal feeding operations, ment programs. The US Environmental tistical theory and epidemiological practice, lithographic and screen printing, dry cleaning, Protection Agency describes the model in we show how state environmental health photo processing, automotive refinishing and this way: protection agencies can enhance accountabil- repair, and gasoline service stations. State ity and inform programmatic decisionmaking agency experience suggests that the model is The Environmental Results Program combines compliance assistance, self-audit/certification, through the collection and analysis of quanti- broadly applicable and may be used to ad- and statistically based inspections and perform- tative data on industry performance. dress other US and international public ance measurement in order to strengthen or In practice, the model proceeds in a se- health issues such as the control of infectious replace an existing regulatory structure. . . . [R]egulators educate facilities about their envi- quential manner: (1) baseline inspections diseases (e.g., health department food protec- ronmental impacts and obligations, as well as document industry conditions before the tion programs), biological waste treatment, voluntary best practices they can use to allevi- program launch, (2) industry sector-wide pro- and low-level radioactive waste storage ate potential impacts. Facilities are then required to self-evaluate and certify compliance. gram implementation (i.e., self-certification operations. By conducting “before and after” inspections and government-sponsored technical and and applying statistical analysis, regulators can compliance assistance), and (3) postinterven- AUTOMOTIVE REFINISHING leverage limited inspection and enforcement resources to verify compliance, measure per- tion facility audits coupled with performance formance, and institute lasting improvements.3 measurement over a defined time interval. An estimated 48730 automotive refinish- Compliance performance continues to be ing facilities, employing 197965 technicians, The objectives of our study were to tracked and data are analyzed longitudinally engage in collision repair nationwide.4 Most (1) evaluate the statistically based perform- for trends. A key aspect of this approach is of these shops have an annual sales volume ance measurement component of the model, that, with proper program design and rigor- of less than US$1 million and operate with and (2) assess the overall effectiveness of ous data evaluation, only a sample set of varying degrees of technological sophistica- the approach in a small-business industry field inspections are needed to obtain statisti- tion; they can generally be characterized as sector that is characterized by a range of cally valid results for industry performance. forming a small, nonunion independent busi- human health and environmental risks (the This approach is currently being applied by ness sector.4

May 2007, Vol 97, No. 5 | American Journal of Public Health Enander et al. | Peer Reviewed | Research and Practice | 819  RESEARCH AND PRACTICE 

α Occupational and environmental health re- plan data are then followed prospectively for power=0.80 [Z0.80 =0.842], =.05 search conducted in the automotive refinish- performance trends. [Z0.95 =1.645]), the total sample size required ing sector has shown a significant potential To quantitatively assess whether an im- was calculated as for workplace exposures to and environmen- provement in industry-wide performance oc-   2 0...842 0 24++ 0 2275 1 . 645 2 ×× 0 . 525 0 . 475 tal releases of various toxicants, including iso- curred, data were collected from a statistically (4) n = 2    04. − 0065.  cyanate aerosols, welding fumes and gases, predetermined number of randomly selected metal-bearing particulates, silica and nuisance facilities, which had been inspected at base- or approximately 97 (97/2 or 49 baseline dusts, and chemical vapors.5,6 The National line and postintervention, and were compared plus 49 postintervention audits). α Institute for Occupational Safety and Health and subjected to statistical analysis. Choosing different values for P1, P2, , and as well as the International Agency for Re- power resulted in different sample size esti- search on Cancer reported on the potential Determining Sample Size, Power, mates. In order to detect a smaller effect size for adverse health effects (including cancer, Significance, and Proportions (difference between proportions P1 and P2), asthma, kidney disease, and central nervous The following equation8 was used to de- larger samples were needed. The actual sam- system effects) among painters who are ex- termine the total number of baseline and ple sizes we used were n1 =40 for baseline 7 posed to toxicants on the job. postintervention field audits needed to com- audits and n2 =42 for postintervention audits. In 2003, after more than 2 years of indus- pare binomial proportions for assessing over- Pre- and postintervention audits were con- try research and stakeholder meetings to con- all performance: ducted on independent cohorts that were ran- sider the hazards inherent in automotive refin- 2 domly drawn from the entire universe of 367  ΖΖ− +−+− −−βαPPPP()11() 21 PP() ishing as well as the economic and technical n = 2  11 12 21  licensed refinishing facilities. (1)  ∆  limits of individual facilities, the Rhode Island   Department of Environmental Management Data Collection where P1 = baseline compliance rate, (RIDEM)—in partnership with the Rhode Is- Field data were collected during pre- and P2 = postimplementation compliance rate, land Department of Health and the University ∆ postintervention facility audits. Research Ran- =|P1 –P2|, Z1–␣ = significance level test of Rhode Island—launched a voluntary state- domizer (Wesleyan University, Middletown, statistic, Z1–␤=power test statistic, and wide certification program to improve regula- Conn), a computer-based random number P =(P1 + P2)/2. tory compliance with both environmental and In Equation 1, sample size is determined on generator, was used to select shops for inclu- occupational health standards. This compre- the basis of an assumed difference between sion in the study. After 40 baseline audits hensive certification program was designed to 2proportions: the compliance rates before were completed, the auto body initiative was address worker training requirements, haz- launched at an industry-wide workshop that (P1) and after (P2) program implementation: ardous waste management, air quality, health was followed by a statewide mailing of self- Pa= ssumed baseline compliance rate and safety, and wastewater discharge among (2) 1 certification checklists and guidance manuals. no.. shops in compliance 367 licensed refinishing facilities located in = During baseline audits, technical assistance total number of shops Rhode Island. Demographically, the average was offered regarding regulatory interpreta- Pa= ssumed compliance rate postimplementatioon Rhode Island automotive refinishing shop em- (3) 2 tion, compliance methods, and engineering no. shops in compliance ployed 5 people, processed 7 vehicles per = and pollution prevention recommendations. week, and was composed of mostly White total number of shops A 6-month interval was then allowed for facil- men who had a high-school education.7 The objective of the calculation was to ob- ities to conduct self-evaluations and submit tain 2 samples (1 pre- and 1 postintervention) completed checklist and corrective action plan METHODS of sufficient size to allow the detection of a data to RIDEM. During this interval, project difference (i.e., an improvement) in compli- partners provided more than 25 on-site Self-certification initiatives typically collect ance rates if one truly existed. To calculate worker health and safety audits (conducted by binary, count, and descriptive data that are total sample size for baseline and postinter- the Rhode Island Department of Health), 10 submitted in the form of questionnaire re- vention inspections, compliance rate propor- university-led pollution prevention assess- sponses and corrective action plans, and col- tions (P1 and P2), significance level, and ments, numerous telephone consultations, and lected during independent baseline and post- power had to be prespecified. We chose the multiple statewide workshops that reached intervention facility audits conducted by state conventional significant (α) level of 5% and more than 200 individuals. Incentives to par- agency staff. Corrective action plans are writ- power of 80%. On the basis of previous field ticipate in the program included reduced inten summaries submitted to a state agency experience, we estimated the industry’s mean spection priority for certified shops, the ability that indicate the amount of time needed (but baseline performance (P1)—relative to all envi- to correct violations without gravity-based not greater than a specified limit) to come ronmental as well as health and safety re- penalties, advanced preparation for environ- into compliance with any single regulatory quirements—to be 40% and assumed a post- mental and worker health and safety enforce- requirement. Industry participation rate, self- intervention improvement of 25%. With ment inspections (e.g., complaint based), no- certification response, and corrective action these choices (where P1 =0.40, P2 =0.65, cost technical and compliance assistance, and

820 | Research and Practice | Peer Reviewed | Enander et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 1—Methylene Chloride (MeCl) TABLE 2—Automotive Refinishing Facility Baseline and Postintervention Compliance 2x 2 Table Performance Comparisons for Each Performance Indicator

Observed Baseline Postintervention Statistical Comparison Intervention MeCl+ MeCl– Data Set Sample Size Proportion Sample Size Proportion Percentage changea b c n1 p1 n2 p2 (95% CI ) P Postintervention a b r1

Preintervention c d r2 Hazardous waste management

s1 s2 N Authorized agents 40 0.28 43 0.44 16 (. . .) .088 Container inspections 16 0.06 9 0.22 16 .287 Note. P value for the test=Probability of observed data set+Probabilities of extreme data sets. Probability of Container labeling 39 0.21 44 0.39 18 (. . .) .059

observed data set=r1!r2!s1!s2!/N!a!b!c!d!. Contingency plans 16 0.06 9 0.22 16 .287 Probabilities of extreme data sets computed by EPA identification number 32 0.88 44 0.86 –2 (. . .) . . . rearrangement of observed table values reducing “a” by 1 with fixed marginals. Personnel training records 16 0.06 9 0.22 16 .287 Secondary containment 16 0.63 9 0.56 –7 (. . .) . . . Manifest tracking 39 0.56 44 0.89 33 (15, 51) .001* Air pollution control the reduced environmental and worker liabil- Compliant surface coatings 40 1.00 40 1.00 0 (. . .) . . . ity associated with facility improvements. Dust control 39 0.33 40 0.48 15 (. . .) .146 Enclosed spray gun cleaner 40 0.83 41 0.88 5 (. . .) .360 Hypothesis Testing HVLP spray equipment 40 1.00 40 1.00 0 (. . .) . . . To test whether performance had improved Methylene chloride usage 40 0.67 41 0.95 28 (12, 44) .001* during the first-round interval of the program, Solvent rag storage 36 0.81 41 0.88 7 (. . .) .287 the Fisher exact probability test (1-tailed) was Ventilated sanding equipment 39 0.31 40 0.30 –1 (. . .) . . . used because observed cell frequencies in the Wastewater discharge 2×2 table (Table 1) were relatively small. For Discharge signage 39 0.00 42 0.48 48 (33, 63) <.001* example, to determine whether the ERP ini- Unpermitted floor drains 40 0.67 39 0.69 2 (. . .) .531 tiative was successful at reducing methylene Washwater runoff 38 0.37 42 0.74 37 (17, 57) .001* chloride usage among auto body shops, the Worker health and safety following hypotheses were tested using the Employee medical examination 33 0.33 35 0.46 13 (. . .) .214 9 Fisher exact method, where Ho =no differ- Hazard communication program 32 0.28 37 0.46 18 (. . .) .101 ence in the proportion of shops using methyl- Lockout/tagout program 33 0.06 34 0.56 50 (31, 69) <.001* ene chloride at baseline and postintervention Personal protective equipment 33 0.09 38 0.63 54 (36, 72) <.001* and Ha =reduction in the usage of methylene Safety/health poster 36 0.42 40 0.83 41 (21, 61) <.001* chloride postintervention. Respiratory protection 30 0.33 38 0.61 28 (5, 51) .023 For the methylene chloride usage compli- Note.CI=confidence interval; EPA=Environmental Protection Agency; HVLP=high volume low pressure. ance performance indicator, shops were ad- a Calculated as 100(p2 – p1). vised to stop using methylene chloride-based b α × 95% CIs calculated for indicators showing statistical significance at =.05; 95% CIs calculated as (p2 –p1)±1.96 square paint strippers. All other hypotheses tested root [p1(1.00–p1)/n1 + p2(1.00–p2)/n2]. c were similar: H = no difference in the pro- P values were calculated with the Fisher exact test online, available at http://home.clara.net/sisa/fisher.htm; P values o calculated only for performance indicators showing improvement (1-tailed test). portion of shops in compliance with a spe- *P≤.001; Simes-modified Bonferroni adjustment for multiple comparisons. cific indicator (Table 2) at baseline and postintervention, and Ha = improvement in compliance postintervention. Written guidance and technical assistance were provided Test for Multiple Comparisons smallest P value was compared against .05/ to specifically address each area where com- In addition to checking each P value against (k–1), and so on until the largest P value was pliance improvement was being sought. the nominal level of significance of .05 to de- compared against .05. This method is less The Fisher exact test was used to calculate termine which indicators showed improve- conservative than the usual Bonferroni adjust- P values for each performance indicator. Al- ment, a modified Bonferroni adjustment11 was ment and has higher power. though a test for multiple comparisons was used to avoid inflation of the overall type I also performed, discussions of performance error rate. The approach was to order the P RESULTS improvements are generally on the basis of values from the smallest to the largest. If the significance tests performed with the type I number of indicators was k, then the smallest Of 367 licensed refinishing facilities, 171 error fixed at .05 for each indicator.10 P value was compared against .05/k, the next individual shops voluntarily certified to their

May 2007, Vol 97, No. 5 | American Journal of Public Health Enander et al. | Peer Reviewed | Research and Practice | 821  RESEARCH AND PRACTICE 

performance and an additional 15 filed non- federal Occupational Safety and Health Ad- The total number of auto body shops that applicability statements (valid Rhode Island ministration (OSHA) requirements did not were inspected, and to which the specified Department of Business Regulation auto re- apply to 1-person owner-operated shops, performance indicator was found to apply, is finishing licenses were held, but the work was which resulted in n1 <40 for all 6 “worker given in Table 2. Compliance rate proportions subcontracted out). Among these 171 facili- health and safety” indicators. resulting from all postintervention facility in- ties, 74 (43%) filed 271 individual corrective Table 2 shows the proportion of shops that spections are presented in column p2 and action plans along with their completed self- were in compliance with each indicator at ranged from 0.22 (compliance with required certification checklists. baseline. Performance estimates ranged from weekly hazardous waste “Container inspec- Using Equation 1 and the set of assump- poor (0.00 or 0 out of 39 shops with signage tions” and the maintenance of appropriate tions specified in the Methods section, we posted over facility sinks prohibiting the dis- “Personnel training records”) to 1.00 (the use estimated the total number (n) of shops charge of chemicals) to excellent (1.00 or 40 of “Compliant coatings” and “HVLP spray needed in each sample to be 49. In order to out of 40 shops used compliant coatings and equipment”). During the time between base- estimate the industry’s baseline performance, high-volume low-pressure spray guns). The line and postintervention audits, staff from 51 auto body shops were randomly selected overall mean was 43%. the Rhode Island Department of Health con- from a population of 367 licensees. Of the A combined total of 42 randomly selected ducted 72 voluntary health and safety audits. original 51 shops targeted for on-site visits, post-ERP implementation facility audits of The audits detected 487 serious hazards, 73 40 baseline audits were ultimately conducted both certified (18) and noncertified (24) other-than-serious hazards, 25 regulatory by RIDEM nonregulatory staff (3 shops re- shops were conducted. RIDEM nonregulatory hazards, and 1 imminent danger hazard, all fused entry, 1 was out of business, and 7 did technical assistance staff conducted 18 audits of which were corrected (the Department of not perform refinishing on site). from a universe of 171 certified facilities (ac- Health maintains a policy that all OSHA vio- Baseline audit results are presented in cess to 2 additional shops could not be lations discovered during nonregulatory site Table 2. These data show that although 40 gained); enforcement personnel performed assessments must be corrected within a given baseline audits were conducted, selected per- 24 regulatory inspections drawn from a pool time frame). formance indicators often applied to less than of 196 noncertified shops. Although technical A total of 74 auto body shops (43%, 74 out 40 shops. For example, certain hazardous assistance staff gained access to only 18 of of 171 ERP participants) filed 271 individual waste management requirements did not the 20 randomly selected certified facilities, corrective action plans along with completed apply to shops that only engaged in satellite data for all 20 shops were included in the self-certification forms to RIDEM; the range accumulation (i.e., storing <55 gallons of haz- Fisher exact test calculations because correc- was 1–37 plans per shop. Of these, 234 from ardous waste near the point of generation). tive action statements acknowledging regula- 69 shops were found to be valid. Figure 1 This exclusion applied to performance indica- tory deficiencies were submitted to RIDEM shows the submissions to RIDEM by program- tors “container inspections,” “contingency and were assumed to be accurate indicators matic area: air, water, hazardous waste, and plans,” “personnel training records,” and “sec- of nonperformance for statistical analysis occupational health; subcategories for each of ondary containment” where n1 =16. Similarly, purposes. the 4 major divisions are also shown.

FIGURE 1—Number of corrective action plans received from 69 certified shops, organized by major category.

822 | Research and Practice | Peer Reviewed | Enander et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Performance Results and Statistically employees who are subject to the federal res- After the traditional facility-by-facility inspec- Significant Improvements piratory protection standard (e.g., professional tion approach, on-site audits were largely Percentage differences between postinter- spray painters) undergo an evaluation by a single media in nature, that is, individual facil- vention and baseline proportions (p2–p1) physician or other licensed health care profes- ity audits considered only air, water, haz- are shown in Table 2. These differences sional and receive clearance to wear a respi- ardous waste, or occupational health issues at ranged from a net decrease in performance rator. Though the 13% improvement found the time of inspection, but not all 4 simulta- of –7% (secondary spill containment for during field inspections was not statistically neously. The probability that an individual hazardous waste) to a 54% improvement in significant (P=.21; Table 2), 20 of the 69 shop would receive a comprehensive, multi- the use of personal protective equipment shops that submitted corrective action plans media inspection in any given year was virtu- and a mean improvement of 22% for all (i.e., self-reported noncompliance) docu- ally zero. By comparison, the certification pro- nonnegative indicators. mented that they were deficient relative to gram achieved a 49% participation rate and To determine which performance measures this regulatory requirement and that compli- comprehensive self-assessments addressed all achieved statistical significance, the Fisher ance would be achieved within 2 months. major air, water, hazardous waste, pollution exact test was applied to all indicators that Overall, the number of corrective action plans prevention, and occupational health and showed a measurable improvement: P values submitted by the ERP-participating shops pro- safety standards within 6 months. In addition, for these 19 indicators are shown in Table 1. vided supporting evidence that facility owners education and outreach materials (compliance Confidence intervals were also calculated for made a real effort to improve compliance workbook, self-audit checklist, and fact sheet 8 of the 19 indicators that showed signifi- with environmental and health and safety mailings) were sent to 100% of all licensed cance at α<.05. The Simes-modified Bonfer- regulatory requirements. facilities. roni adjustment for multiple comparisons re- On the basis of the success of the auto- duced the number of statistically significant DISCUSSION motive refinishing industry initiative, Rhode indicators from 8 to 7. Island expanded the self-certification ap- In general, our results showed at least 1 Rhode Island’s experience confirms that self- proach to exterior lead-based paint removal statistically significant improvement in each of certification programs can produce measurable contractors, auto salvage yards, and under- the 4 major health and environmental cate- and statistically significant improvements in ground storage tank sectors. To enhance im- gories. Although a number of statistically sig- environmental performance, as originally re- plementation, Rhode Island also began to au- nificant improvements were found, the level ported by the state of Massachusetts.2 The sta- tomate the field inspector data collection and of confidence we placed in any single per- tistical methods used to determine sample size statistical analysis components of these pro- formance indicator varied. For example, a sta- and evaluate field inspection data highlighted grams, starting with the underground storage tistically significant indicator that had a large the importance of the prespecified input pa- tank sector. Automation included converting effect size where compliance was also directly rameters used in Equation 1; especially, the field inspection checklists into an electronic observable (e.g., wastewater discharge signage assumed mean level of baseline performance form and using tablet personal computers posted directly over a facility sink [48%, (P1) and expected improvement or target effect with up-loading capability for digital photos. P<.01] or displaying the OSHA-required Job size in compliance rate proportions. Statistically Automation and expansion of the certification Safety and Health Protection poster in an significant improvements in environmental per- program makes it possible for the state to area where all employees could see it [41%, formance, for example, were found only for achieve broader compliance with fewer on- P<.01]) was interpreted to carry more weight those indicators that showed a difference be- site inspections and improved efficiency. than a statistically significant improvement for tween pre- and postintervention compliance Although a voluntary approach worked a multicomponent performance indicator for rate proportions of 28% or more (Table 2), as well for the Rhode Island automotive refin- ∆ which individual program elements may not assumed ( =|P1 – P2 |=0.25; Equation 1) at ishing industry, this article reports on a pro- have been thoroughly verified by RIDEM in- the start of the program. gram that was implemented with a great spectors (e.g., the OSHA requirement for a Before adopting the certification approach, amount of interaction with all stakeholders, comprehensive workplace respiratory protec- we estimated that the 4 RIDEM divisions active program partners (including both the tion program [28%, P<.03]). In addition, no most responsible for regulatory compliance in regulatory and academic sectors), all operat- change in facility performance regarding the the auto body sector (Office of Air Resources, ing within a geographically compact state, usage of high-volume low-pressure spray guns Office of Waste Management, Office of Com- which benefits from an ability to implement and compliant coatings provided a good indi- pliance and Inspection, and Office of Water comprehensive programs with willing mem- cation that facilities were consistently comply- Resources) collectively inspected fewer than bers of the regulated community. The success ing with RIDEM requirements. 5% of all licensed facilities in any given year of voluntary initiatives may be limited in re- In other cases, even though statistical sig- and most inspections were complaint driven. gions where these cofactors are not present. nificance was not achieved, real improve- Inspection coverage by state and federal A number of specific factors contributed to ments were found to occur as a result of the agencies responsible for occupational health the initial success of the program, including program. For example, OSHA requires that and safety was thought to be about the same. increased awareness within the industry of a

May 2007, Vol 97, No. 5 | American Journal of Public Health Enander et al. | Peer Reviewed | Research and Practice | 823  RESEARCH AND PRACTICE 

coordinated enforcement effort in which non- Contributors Injury Prevention for participants would be subject to random regu- R.T. Enander led the writing, data analysis, and oversaw project implementation with R.N. Gagnon who Children and Adolescents latory inspections and extensive research and originated and co-designed the project. R.C. Hanumara Research, Practice, and outreach activities conducted previous and guided and oversaw all statistical analyses. E. Park and Advocacy subsequent to program kickoff. T. Armstrong conducted field investigations, data collection, data management, and were instrumental in the In conclusion, the automotive refinishing project launch and implementation. D.M. Gute assisted case study demonstrates that voluntary com- with conceptualizing the original field study and con- pliance with human health and environmental tributed to the writing. standards can be quantitatively assessed by auditing a relatively small number of facilities Acknowledgments The authors would like to thank Steven DeGabriele both pre- and postintervention. Taken to- and Tara Velazquez of the Massachusetts Department gether, analysis of the random field-audit data, of Environmental Protection, who helped launch self-certification checklist submissions, and in- Rhode Island’s first self-certification initiative. We also acknowledge R. Scott Bowles, Gregory Ondich, George formation provided in corrective action plans Frantz, and Beth Termini of the US Environmental Pro- indicated that the program succeeded in tection Agency, for their support and assistance, as well improving compliance performance. As con- as Michael Crow of ERP Consulting for his helpful comments. structed, the self-certification approach is an it- erative process by which modifications in Human Participant Protection guidance materials and checklists are made No protocol approval was needed for this study. in response to regulatory changes, periodic The chapters in this book take a detailed look at industry self-assessments are conducted as re- References child and adolescent injuries, incorporating research, practice, and advocacy with recommen- quired, and individual improvements in facil- 1. US Environmental Protection Agency. Innovative dations for future work. Each chapter provides ity performance and higher industry-wide par- Environmental Permitting, States Implementing ERP. Available at: http://www.epa.gov/ooaujeag/permits/ an account of a child/adolescent's particular in- ticipation rates are continually sought. As with erp/states.htm. Accessed February 3, 2006. jury; the injury's pertinent risk factors; up-to- any traditional or nontraditional state-based 2. Massachusetts Department of Environmental Pro- date research findings and how research has environmental health initiative, however, the tection. Environmental Results Program. Available at: led to successful action and practice; the impor- sustainability and long-term success of indus- http://mass.gov/dep/service/about11.htm. Accessed November 15, 2005. tance of the role of advocacy; and future re- try self-certification programs will depend 3. US Environmental Protection Agency. Innovative search, practice, and advocacy efforts. upon adequate funding, agency stewardship, Environmental Permitting. Environmental Results Pro- gram. Available at: http://www.epa.gov/ooaujeag/ ISBN 0-87553-068-0 • softcover • 2006 and organizational support. Finally, the results $27.30 APHA Members • $39.00 Nonmembers permits/erp/what.htm. Accessed July 21, 2005. of this study should not be interpreted to 4. I-Car Education Foundation. Snapshot of the Colli- mean that voluntary self-certification programs sion Industry: Executive Summary of a 2004 Survey. American Public Health are an acceptable alternative to state and fed- Available at: http://www.i-car.com/pdf/education_ Association eral regulation, but rather that voluntary pro- foundation/snapshot.pdf. Accessed August 2, 2005. 800 I Street, NW, Washington, DC 20001 5. Enander RT, Gute DM, Cohen HJ, Brown LC, grams can be a useful adjunct that results in www.apha.org Desmaris AMC, Missaghian R. Chemical characteriza- significant performance improvements when tion of sanding dust and methylene chloride usage in TO ORDER:web www.aphabookstore.org coupled with field inspections and strong automotive refinishing: implications for occupational email [email protected] enforcement. and environmental health. Am Ind Hyg Assoc J. 2002; 63:741–749. fax 888.361.APHA 6. Enander RT, Cohen HJ, Gute DM, Brown LC, phone 888.320.APHA M-F 8am-5pm EST Desmaris AMC, Missaghian R. Lead and methylene About the Authors chloride exposures among automotive repair techni- Richard T. Enander, Ronald N. Gagnon, and Thomas cians. J Occ Env Hyg. 2004;1:119–125. Armstrong, are with the Rhode Island Department of Envi- 7. Enander RT, Gute DM, Missaghian R. Survey of ronmental Management, Office of Technical and Customer risk reduction and pollution prevention practices in the Assistance, Providence, RI. R. Choudary Hanumara is with Rhode Island automotive refinishing industry. Am Ind the Department of Computer Science and Statistics, Uni- Hyg Assoc J. 19 9 8;59:478–489. versity of Rhode Island, Providence. Eugene Park is with 8. Woolson RF, Clark WR. Statistical Methods for the the Department of Chemical Engineering, University of Analysis of Biomedical Data. 2nd ed. New York, NY: Rhode Island, Providence. David M. Gute is with the De- Wiley; 2002. partment of Civil and Environmental Engineering, Tufts 9. Swinscow TDV, Campbell MJ. Statistics at Square University, Medford, Mass. One. 10th ed. London: BMJ Books; 2002. Requests for reprints should be sent to Richard T. Enander, PhD, Rhode Island Department of Environmen- 10. Ott RL, Longnecker M. Statistical Methods and tal Management, Office of Technical and Customer Assis- Data Analysis. 5th ed. Duxbury; 2001. tance, 235 Promenade St, Providence, RI 02908 (e-mail: 11. Simes RJ. An improved Bonferroni procedure for [email protected]). multiple tests of significance. Biometrika. 19 8 6;73: This article was accepted August 11, 2006. 751–754.

824 | Research and Practice | Peer Reviewed | Enander et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Socioeconomic Status and Risk for Arsenic-Related Skin Lesions in Bangladesh

| Maria Argos, MPH, Faruque Parvez, MPH, Yu Chen, PhD, A.Z.M. Iftikhar Hussain, MD, MPH, Hassina Momotaj, MD, MPH, Geoffrey R. Howe, PhD, Joseph H. Graziano, PhD, and Habibul Ahsan, MD, MMedSc

Beginning in the 1970s, hand-pumped tube Objectives. Arsenic contamination of groundwater is a severe public health wells were installed in Bangladesh to provide crisis in Bangladesh, where the population is exposed to arsenic in drinking water a source of pathogen-free groundwater for through tube wells used for groundwater collection. In this study, we explored the consumption. Since then, the number of tube association between socioeconomic status and arsenic toxicity. wells within the country has exponentially in- Methods. We used baseline data from 11438 men and women who were re- creased, and these wells are the population’s cruited into the Health Effects of Arsenic Longitudinal Study (HEALS), a prospec- primary source of drinking water. Unfortu- tive cohort study on the health effects of arsenic exposure in Bangladesh. We con- nately, the groundwater in Bangladesh has ducted analyses with logistic regression and generalized estimating equations. been naturally contaminated with high levels Results. We found a strong dose–response association with all measures of of arsenic, a phenomenon discovered only arsenic exposure and skin lesions. We also found that the effect of arsenic was modified by land ownership on a multiplicative scale, with an increased risk after decades of exposure and an epidemic of among non–land owners associated with well water arsenic (P=.04) and urinary arsenical skin lesions.1,2 total arsenic concentrations (P=.03). Many of the human health effects of arsenic Conclusions. Our study provides insight into potentially modifiable host char- have been proven, and arsenic has been classi- acteristics and identifies factors that may effectively target susceptible population 3 fied as a human carcinogen. Epidemiologic subgroups for appropriate interventions. (Am J Public Health. 2007;97:825–831. evidence has shown a significant association doi:10.2105/AJPH.2005.078816) between the consumption of arsenic through drinking water and cancers of the skin, lung, bladder, liver, and kidney.4–8 Premalignant previous studies of arsenic exposure have ex- concentration as part of a precohort survey. A skin lesions are an early manifestation and plicitly examined effect modification by SES complete enumeration of the study area pop- hallmark of arsenic toxicity and may indicate in arsenic toxicity. We sought to determine ulation (n=65876) was conducted as a part increased future risk for arsenic-related whether selected SES indicator variables of this survey. The preliminary well survey cancer.9 However, there is evidence of marked modify the association between various mea- findings and details of the study methods interindividual variability in arsenic-induced sures of arsenic exposure and premalignant used have been published elsewhere.14 From premalignant skin lesions and cancer.9 It has skin lesions among 11438 men and women the enumeration, a list of eligible married been suggested that differential susceptibility in Araihazar, Bangladesh. couples was generated, and regular users of to arsenic-induced skin lesions may be attrib- tube wells were selected for recruitment into uted to host characteristics10 ,11 ; therefore, be- METHODS the study. Individuals were eligible for partici- cause of the magnitude of the arsenic problem pation if they were married, lived in the bari in Bangladesh, a thorough examination of po- In 2000, the Health Effects of Arsenic Lon- (a cluster of household dwellings occupied by tentially modifiable host characteristics is war- gitudinal Study (HEALS)—an ongoing popula- members of the extended family) for at least ranted. tion-based study of both the short-term and 3 years, and were aged 18 to 75 years. The Socioeconomic status (SES) has been iden- long-term health effects of arsenic exposure— rationale for these selection criteria has been tified as an important determinant of health was launched in Araihazar, Bangladesh. The published elsewhere.13 across a broad range of health issues.12 SES is selection of cohort participants, study design, The field team recruited couples in the a multidimensional construct that includes ed- and methods have been described in detail order they were recorded on the precohort ucation, occupation, income, wealth, and resi- elsewhere and are briefly summarized here.13 survey list. Once an individual was located dence. However, the measure of this con- (i.e., matched by name and age and by hus- struct in less-developed countries creates a Participants band’s name for female respondents), eligibil- challenge because adequate data may not be The cohort participants are residents of ity was verified. Eligible individuals and their available. Therefore, the examination of sev- Araihazar, Bangladesh—a well-defined spouses were recruited into the study. If the eral measures of SES is advantageous be- 25-square-kilometer rural area east of the spouse was not present during the initial visit, cause it increases the likelihood that some di- capital city of Dhaka. All tube wells in the the recruiter returned to the bari on another mension of the SES construct is captured. No area (n=5966) were first tested for arsenic day to recruit the spouse. The recruiter

May 2007, Vol 97, No. 5 | American Journal of Public Health Argos et al. | Peer Reviewed | Research and Practice | 825  RESEARCH AND PRACTICE 

continued to enroll eligible individuals and keratosis was characterized by the general ownership, television ownership, and food their spouses until a maximum of 3 couples thickening of the skin on palms and soles.16 frequency for all participants; weekly house- from each tube well was obtained; this restric- Among the 714 skin lesion cases, 421 hold cooking oil consumption was asked only tion was adopted to reduce the correlation (59.0%) had melanosis only; 194 (27.2%) of female respondents. The amount reported among individuals in the cohort. Between Oc- had melanosis and keratosis; 44 (6.2%) had by the female participant was adjusted for tober 2000 and May 2002, 11746 men and leucomelanosis and keratosis; 29 (4.1%) total household size and also was used for women were recruited; this included 4801 had leucomelanosis only; 14 (2.0%) had her spouse. The food index was created by married couples and 2144 participants whose melanosis, leucomelanosis, and keratosis; and adding the total number of days per week spouses were not enrolled in the cohort. 12 (1.7%) had melanosis and leucomelanosis. that each participant reported eating fish, beef There was a 97.5% participation rate among We dichotomized skin lesion status into pres- or lamb, poultry, eggs, or dried beans. These those who were approached for recruitment. ence or absence of any skin lesions. food items were selected as indicators of Comparison with the precohort survey household wealth on the basis of their corre- Exposure Assessment showed that our cohort population was a rep- lation with the other selected SES indicators. Three measures of individual-level arsenic resentative sample of the study population in Household cooking oil consumption and the exposure were assessed: well water arsenic terms of arsenic and geographic distributions food index were selected as indicators of SES concentration, creatinine-adjusted urinary total (data not shown). independent of their potential contribution to arsenic concentration, and cumulative arsenic nutritional status. All continuous SES vari- exposure (CAE). Well water was analyzed for Questionnaire ables were dichotomized at their median for arsenic concentration with graphite furnace A structured questionnaire composed in our analyses. The correlation among these atomic absorption spectrometry that has a de- Bengali was administered to participants by SES indicator variables was weak (Pearson tection limit of 5 µg/L.17 Water samples that trained interviewers. The questionnaire as- correlation coefficients 0.22–0.35), which had less than 5 µg/L were subsequently rean- sessed sociodemographic characteristics, cur- suggests that they each captured different alyzed with inductively coupled plasma-mass rent and past tube well use, typical water components of SES variability. spectrometry that has a detection limit of 0.1 consumption patterns, food consumption fre- µg/L.18 Urinary total arsenic concentration quency of 39 items common to the popula- Statistical Analysis was analyzed with graphite furnace atomic ab- tion, occupational exposures, and smoking Of the 11746 individuals enrolled at base- sorption spectrometry in accordance with the habits. The semiquantitative food consump- line, 308 individuals (<3%) were eliminated method used by Nixon et al.19 Urinary creati- tion frequency questionnaire was developed from our analysis. Reasons for exclusion in- nine was measured with a colorimetric Sigma for this target population and validated.15 The cluded having declined a clinical examination Diagnostics Kit (Sigma Diagnostics, St Louis, interviewers were blind to the tube well water (n=210), having skin lesions unrelated to ar- Mo), and urinary total arsenic concentration arsenic concentrations for the study region. senic (n=96), and having not reported age was subsequently expressed as micrograms (n=2). Thus, the sample eligible for our cross- per gram creatinine. CAE was calculated by Clinical Examination sectional analysis was 11438. Additionally, multiplying well water arsenic concentration During the baseline interview, each partici- we excluded from all regression analyses times the estimated amount of water con- pant was examined by a trained physician for those participants who were missing data on sumed per year times the number of years the the presence of premalignant skin lesions in at least 1 SES indicator or other covariates tube well had been used for drinking.13 A con- accordance with a structured protocol.13 The (n=353). Individuals who were missing ar- stant arsenic concentration in each tube well physicians were specially trained for the de- senic exposure data were excluded only from was assumed for calculation of the CAE for tection and diagnosis of these skin lesions. the analysis of the missing exposure measure— the duration of use by each participant. This Premalignant skin lesions were categorized as 323 individuals for urinary total arsenic con- assumption was previously established for melanosis, leucomelanosis, and keratosis.16 centration data and 381 for CAE data. Well tube wells in our study area.14 Furthermore, The presence, location, and extent of each water data were not missing for any partici- individuals who reported either use of another skin lesion type were documented. Physicians pant. Our logistic regression analyses used the tube well before their current well or use of were blind to the participants’ well water ar- generalized estimating equation method and a secondary tube well in addition to their cur- senic concentration, and no histopathologic included 11085 individuals (681 with skin le- rent well had those additional sources of expo- confirmation of these lesions was conducted. sions and 10404 without lesions) for well sure included in their CAE calculations. Prevalent cases of premalignant skin le- water arsenic concentration, 10771 individu- sions (n=714) were ascertained during the SES Variables als (666 with skin lesions and 10105 without baseline survey. Melanosis was characterized Education, land ownership, television own- lesions) for urinary total arsenic concentration, by the hyperpigmentation of the skin over ership, weekly household cooking oil con- and 10719 individuals (668 with skin lesions wide body surface areas, leucomelanosis was sumption, and a food index were considered and 10051 without lesions) for CAE. characterized by the hypopigmentation of to be indicators of SES.20 The baseline ques- We used linear regression to evaluate the the skin over wide body surface areas, and tionnaire measured years of education, land association between SES indicators after we

826 | Research and Practice | Peer Reviewed | Argos et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

adjusted for gender, age, and body mass index RESULTS cases and 984.1 mg among nonlesion con- (kg/m2). This was done separately for both trols. Results from the linear regression mod- well water arsenic concentration and creati- The baseline clinical evaluation identified els showed television ownership to be the nine-adjusted urinary total arsenic concentra- 714 premalignant skin lesion cases. The aver- only SES indicator consistently associated tion. For descriptive purposes, we used 2- age baseline well water arsenic concentration with arsenic exposure, with individuals who sided χ2 tests to compare subjects by skin was 165.8 µg/L among arsenical skin lesion did not own a television having higher well lesion status with respect to gender, age, and cases and 97.2 µg/L among nonlesion con- water and urinary total arsenic concentrations SES indicators. We used logistic regression to trols, the average baseline creatinine-adjusted (data not shown) than television owners. estimate prevalence odds ratios (ORs) and urinary total arsenic concentration was 426.6 Table 1 shows the distribution of cohort 95% confidence intervals (CIs) for the associa- µg/g creatinine among arsenical skin lesion participants by skin lesion status with respect tion between these characteristics and skin le- cases and 271.5 µg/g creatinine among non- to gender, age, and SES indicators. Men and sion status after we adjusted for gender, age, lesion controls, and the average baseline CAE individuals aged 35 years and older had the and well water arsenic concentration. Partici- was 2213.2 mg among arsenical skin lesion highest prevalence of skin lesions. We also pants were then categorized into quintiles of well water arsenic, urinary total arsenic con- TABLE 1—Baseline Skin Lesion Status Among 11438 Adults in the HEALS Cohort: centration, and CAE separately in accordance Araihazar, Bangladesh, 2000–2002 with the distribution among the total cohort eligible for analysis (n=11438). The quintile Skin Lesion Status Age-Adjusted Skin a containing the lowest level for each measure Sample Present (n=714), Absent (n=10724), Lesion Prevalence, % POR b a was the reference category. Characteristic No. (%) No. (%) P Men Women (95% CI ) We defined premalignant skin lesions—the Gender 0.001 outcome of interest in this analysis—as the Women 130 (18.2) 6432 (60.0) NA 2.27 1.0 (NA) presence of any type of arsenic-related Men 584 (81.8) 4292 (40.0) 9.96 NA 4.8 (3.9, 5.9) skin lesion (melanosis, leucomelanosis, or Age, y 0.001 keratosis). Prevalence ORs and their 95% CIs 18–35 141 (19.7) 5418 (50.5) 3.68 0.74 1.0 (NA) were estimated with logistic regression mod- 36–75 573 (80.3) 5306 (49.5) 9.34 2.34 2.7 (2.2, 3.3) els. Because several individuals drank water Education, y 0.001 from the same tube well, we used the gener- >2 290 (40.6) 5391 (50.3) 7.96 1.62 1.0 (NA) 21 alized estimating equation method to calcu- ≤2 424 (59.4) 5327 (49.7) 12.40 2.73 1.6 (1.3, 1.8) late effect estimates and their CIs, which ac- Missing 6 (0.1) counted for the correlated exposure data. Owns a television 0.001 Although different measures of relative risk Yes 175 (24.5) 3780 (35.2) 6.79 1.64 1.0 (NA) can be estimated from data of cross-sectional No 537 (75.2) 6944 (64.8) 11.60 2.61 1.7 (1.4, 2.0) studies, we deemed prevalence ORs an ap- Missing 2 (0.3) 22 propriate measure of effect for this study. Owns land 0.091 Prevalence odds ratios were adjusted for all Yes 331 (46.4) 5321 (49.6) 7.88 2.19 1.0 (NA) SES indicator variables, gender, age (continu- No 382 (53.5) 5387 (50.2) 11.82 2.30 1.3 (1.1, 1.5) ous), body mass index (continuous), smoking Missing 1 (0.1) 16 (0.2) status (current or past cigarette smoker vs Cooking oil usec (mL/week) 0.001 never smoked), and occupation (indicators for >125 258 (36.1) 5172 (48.2) 8.22 1.69 1.0 (NA) laborers, unemployed, and homemakers vs ≤125 430 (60.2) 5320 (49.6) 12.09 2.79 1.5 (1.3, 1.8) business). Missing 26 (3.6) 232 (2.2) Multiplicative interaction was assessed by Food indexd (days) 0.001 evaluating the Wald statistic for the cross- >6 312 (43.7) 5533 (51.6) 8.06 1.71 1.0 (NA) product interaction term from the regression ≤6 402 (56.3) 5191 (48.4) 12.36 2.77 1.6 (1.3, 1.8) model and was interpreted as the statistical probability for interaction. A statistical proba- Note. HEALS=Health Effects of Arsenic Longitudinal Study, POR=prevalence odds ratio, CI=confidence interval; NA=not applicable. bility of less than 0.05 for the interaction aAdjusted for gender, age, and well water arsenic concentration. term suggested that the slope of the 2 bTwo-sided χ2 test; missing values were excluded from the statistical comparison. c dose–response trends was statistically differ- Adjusted for household size. dThe food index was created by adding the total number of days per week that each participant reported eating fish, beef or ent when stratified by the effect modifier. All lamb, poultry, eggs, or dried beans.These food items were indicators of household wealth on the basis of their correlation analyses were performed with SAS version with the other selected socioeconomic indicators. 9.0 (SAS Institute Inc, Cary, NC).

May 2007, Vol 97, No. 5 | American Journal of Public Health Argos et al. | Peer Reviewed | Research and Practice | 827  RESEARCH AND PRACTICE 

TABLE 2—Prevalence Odds Ratios (95% Confidence Intervals) for Premalignant Skin effect of arsenic on skin lesion risk by SES. Lesions Associated With Well Water Arsenic Concentration by Socioeconomic Status Because of the relative homogeneity of our Indicators Among 11438 Adults in the HEALS Cohort: Araihazar, Bangladesh, 2000–2002 study population, these findings are rather striking. We used several indicators of SES a Well Water Arsenic Concentration, µg/L that are intercorrelated; however, each vari- Sample Characteristic <7 7–38 39–90 91–177 >177 Interaction P able explains a different component of SES Education, y 0.51 variability in this population. We found that ≤2 1.0 2.01 (1.14, 3.53) 3.62 (2.11, 6.21) 3.77 (2.20, 6.45) 5.72 (3.39, 9.64) land ownership appears to be the best indica- >2 1.0 1.82 (1.04, 3.19) 2.22 (1.29, 3.82) 2.82 (1.67, 4.77) 4.42 (2.65, 7.36) tor of SES for this rural population with re- Owns a television 0.95 gard to the health effects of arsenic. No 1.0 2.07 (1.26, 3.40) 2.83 (1.75, 4.59) 3.36 (2.10, 5.39) 5.27 (3.32, 8.37) The food index was evaluated for bias that Yes 1.0 1.53 (0.78, 3.02) 3.10 (1.67, 5.75) 3.15 (1.69, 5.87) 4.56 (2.44, 8.52) may have arisen from the potential misclassi- Owns land 0.04 fication of vegetarian participants as having No 1.0 1.78 (0.99, 3.21) 3.17 (1.80, 5.57) 4.47 (2.59, 7.74) 6.04 (3.53, 10.34) low SES. Only a small proportion of our Yes 1.0 2.08 (1.23, 3.51) 2.72 (1.65, 4.49) 2.39 (1.43, 3.99) 4.30 (2.64, 7.01) study cohort appeared to be strictly vegetar- Cooking oil,b mL/week 0.93 ian (n=709), which was defined as no con- ≤125 1.0 2.07 (1.18, 3.65) 3.98 (2.32, 6.84) 3.66 (2.13, 6.29) 5.55 (3.29, 9.37) sumption of meat, fish, or eggs. In an analysis >125 1.0 1.75 (1.00, 3.05) 1.84 (1.06, 3.17) 2.91 (1.74, 4.86) 4.57 (2.76, 7.57) that excluded individuals who were identified Food indexc,days 0.99 as strict vegetarians, we evaluated effect mod- ≤6 1.0 1.98 (1.12, 3.51) 2.90 (1.66, 5.07) 3.18 (1.83, 5.51) 5.15 (3.04, 8.72) ification of the associations between arsenic >6 1.0 1.82 (1.06, 3.15) 2.88 (1.72, 4.83) 3.36 (2.02, 5.59) 4.85 (2.95, 7.96) exposure measures and arsenical skin lesions by the food index, and we found that the Note. HEALS =Health Effects of Arsenic Longitudinal Study. Prevalence odds ratios were estimated with the generalized effect estimates and overall risk trends ap- estimating question and were adjusted for gender, age, body mass index, smoking status, occupation, and all other variables presented in the table. peared to be very similar to those in the over- aQuintiles of exposure; quintile 1, <7 µg/L, was the reference group. all study population (data not shown). Thus, b Adjusted for household size. we believe no significant bias of the food c The food index was created by adding the total number of days per week that each participant reported eating fish, beef or lamb, poultry, eggs, or dried beans.These food items were indicators of household wealth on the basis of their correlation index was introduced with the inclusion of with the other selected socioeconomic indicators. vegetarian participants in our analyses. Our dose–response findings are consistent with previous studies that examined the asso- saw a significant difference in skin lesion concentration by land ownership, with a ciation between arsenic exposure and prema- prevalence by SES indicator strata, with the higher risk among non–land owners lignant skin lesions.23,24 There is an underly- highest skin lesion prevalence consistently (P =.04). Table 3 shows the adjusted preva- ing biological rationale for also exploring among the categories indicating lower SES. lence ORs and 95% CIs for the associations differential susceptibility to arsenic-induced The association between SES and premalig- between urinary total arsenic concentration skin lesions. Molecular epidemiologic evi- nant skin lesions was examined with logistic and premalignant skin lesions stratified by dence suggests that there is marked variabil- regression, and we saw significant associations SES indicators. We found significant effect ity in arsenic metabolism among exposed in- between all SES indicators and skin lesions modification by land ownership (P = .03) for dividuals.25 Factors that include gender, age, after we adjusted for gender, age, and well the effects of urinary total arsenic concentra- smoking status, nutrition, and genetic poly- water arsenic concentration. An increased tions. Table 4 shows similar dose–response morphisms may influence arsenic metabo- risk for skin lesions was consistently found trends for the associations between CAE and lism.26 Because of the extent of arsenic expo- among the lower-SES categories. premalignant skin lesions; however, there sure within the Bangladeshi population and A strong dose–response association was were no interactions with SES indicators. the public health crisis that has resulted from found between all measures of arsenic expo- this exposure, our study focused on modifi- sure and premalignant skin lesions in our DISCUSSION able characteristics and can be used to readily cross-sectional analysis. Table 2 shows ad- identify subpopulations at greater risk for the justed prevalence ORs and their 95% CIs for In this cross-sectional analysis of baseline health effects of arsenic exposure. the associations between well water arsenic data from a large cohort study, we observed Recent studies by Hadi and Parveen27 and exposure and premalignant skin lesions strati- borderline significant effect modification of Sikder et al.28 found SES to be a significant fied by SES indicators. We observed signifi- the association between premalignant skin le- risk factor for prevalent skin lesions, although cant effect modification on a multiplicative sions and arsenic exposure by land ownership. no adjustment for the level of arsenic expo- scale for the association between premalig- Our population-based study was the first, to sure was made in either analysis. In our nant skin lesions and well water arsenic our knowledge, to show this heterogeneity of study, we had a sufficiently large sample for

828 | Research and Practice | Peer Reviewed | Argos et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 3—Prevalence Odds Ratios (95% Confidence Intervals) for Premalignant Skin CAE. This arsenic exposure measure is a Lesions Associated With Creatinine-Adjusted Urinary Total Arsenic Concentration by function of both dose and duration. Our find- Socioeconomic Status Indicators Among 11438 Adults in the HEALS Cohort: Araihazar, ings suggest that interaction between SES and Bangladesh, 2000–2002 arsenic exposure may be associated with a mechanism for which dose of exposure is im- a Creatinine-Adjusted Urinary Total Arsenic Concentration, µg/g Cr portant. The CAE measure is problematic in Sample Characteristic ≤35 36–66 67–114 115–204 >204 Interaction P this regard because of the fact that an individ- Education, y 0.93 ual with a high-dose exposure of short dura- ≤2 1.0 1.75 (1.09, 2.79) 2.04 (1.29, 3.21) 2.48 (1.56, 3.93) 5.16 (3.31, 8.03) tion would have the same cumulative expo- >2 1.0 1.75 (1.07, 2.86) 2.42 (1.52, 3.87) 3.55 (2.19, 5.75) 5.02 (3.10, 8.13) sure as an individual with a low-dose Owns a television 0.09 exposure of long duration. Thus, evaluating No 1.0 1.55 (1.04, 2.30) 1.96 (1.32, 2.91) 2.41 (1.62, 3.58) 4.31 (2.94, 6.32) effect modification with CAE may attenuate Yes 1.0 2.21 (1.19, 4.09) 2.70 (1.53, 4.78) 4.32 (2.39, 7.82) 7.66 (4.22, 13.90) any modifying effect. Owns land 0.03 Although the etiology of the interaction be- No 1.0 1.38 (0.85, 2.23) 2.13 (1.35, 3.36) 2.58 (1.60, 4.15) 5.92 (3.79, 9.26) tween SES and arsenic is presently not under- Yes 1.0 2.14 (1.33, 3.43) 2.27 (1.44, 3.58) 3.28 (2.06, 5.22) 4.13 (2.59, 6.60) stood and needs to be explored further, it is Cooking oil use,b an important factor for a public health inter- mL/week 0.63 vention in this population. Our study popula- ≤125 1.0 1.63 (1.00, 2.64) 2.44 (1.55, 3.86) 2.68 (1.67, 4.30) 5.40 (3.43, 8.50) tion is an agrarian society. Land-based agri- >125 1.0 1.91 (1.19, 3.06) 1.86 (1.16, 3.00) 3.29 (2.05, 5.28) 4.70 (2.94, 7.50) culture provides livelihood for the majority of Food index,c days 0.18 the rural population, and land is a meaningful ≤6 1.0 1.53 (0.93, 2.51) 2.08 (1.29, 3.35) 2.42 (1.48, 3.94) 4.27 (2.70, 6.76) measure of SES. By evaluating land owner- >6 1.0 2.02 (1.26, 3.22) 2.38 (1.50, 3.77) 3.58 (2.25, 5.69) 6.48 (4.09, 10.25) ship, we have identified a higher-risk sub- population among arsenic-exposed individu- Note. HEALS = Health Effects of Arsenic Longitudinal Study. Prevalence odds ratios were estimated with the als. Future population-based interventions generalized estimating equation and were adjusted for gender, age, body mass index, smoking status, occupation, and all variables presented in the table. that are focused on the prevention of arsenic- aQuintiles of exposure; quintile 1, ≤35 µg/g Cr, was the reference group. induced lesions should be designed with the b Adjusted for household size. specific needs and characteristics of this c The food index was created by adding the total number of days per week that each participant reported eating fish, beef or lamb, poultry, eggs, or dried beans.These food items were indicators of household wealth on the basis of their correlation higher-risk population in mind. with the other selected socioeconomic indicators. The assessment of arsenical skin lesions in this study enhanced our knowledge about individuals who are susceptible to arsenic. Skin evaluating effect modification by SES, and we structured protocol. Furthermore, both inter- lesions are much more common than skin found a significant interaction between ar- viewers and physicians were blind to the well cancers among individuals who are exposed senic exposure and land ownership with re- water arsenic concentration of participants. It to arsenic from drinking water.29 Unlike skin spect to premalignant skin lesion risk. is unlikely that laboratory analyses of well cancer, which has a latency of decades, skin The use of prevalent cases of premalignant water and urinary total arsenic concentrations lesions appear within a few years of arsenic skin lesions is unlikely to have biased our would yield systematically different exposure exposure. Epidemiologic evidence suggests findings. Our assumption is based on the fact measures by SES. Therefore, we believe it is that skin lesions may be associated with in- that mortality from arsenic-induced lesions is improbable that an ascertainment bias by SES creased risk for arsenic-induced cancers.30,31 not immediate; therefore, we did not expect could have distorted the findings of this study. Therefore, skin lesions may be considered an survivor bias to be present in this analysis. Epidemiologic research of other exposures, intermediate endpoint for arsenic-induced We also did not expect prevalent cases to be diseases, and populations has frequently cancers. Knowledge or characteristics of sus- sufficiently different from incident cases of shown health disparity and variations in dis- ceptible individuals who may also be at a skin lesions with regard to the modifying ef- ease prevalence by SES. Therefore, our find- higher risk for arsenic-induced cancers is criti- fect of SES on the arsenic-induced skin lesion ing of increased risk for arsenic-induced skin cal for intervention and prevention of future association. lesions among lower-SES individuals was not disease burden in this population. Modification of the association between ar- an unexpected one. The consistency of this Future studies of susceptibility to arsenic- senic exposure and skin lesions by SES may finding across arsenic exposure measures induced disease should explore the mechanistic be because of differential ascertainment of ex- (well water arsenic and urinary total arsenic pathways by which SES modifies the associa- posure status or skin lesions by SES. Skin le- concentrations) by land ownership also is not tion between arsenic exposure and premalig- sions were assessed through a physical exami- surprising. However, contrary to our previous nant skin lesions. Potential mechanisms include nation by trained physicians who followed a belief, effect modification was not seen with nutrition, health behaviors and awareness, and

May 2007, Vol 97, No. 5 | American Journal of Public Health Argos et al. | Peer Reviewed | Research and Practice | 829  RESEARCH AND PRACTICE 

TABLE 4—Prevalence Odds Ratios (95% Confidence Intervals) for Premalignant Skin with analyzing the data and interpreting the analyses. A.Z.M.I. Hussain and H. Momotaj assisted with plan- Lesions Associated With Cumulative Arsenic Exposure by Socioeconomic Status Indicators ning and conducting the study. G.R. Howe and Among 11438 Adults in the HEALS Cohort: Araihazar, Bangladesh, 2000–2002 J.H. Graziano assisted with planning the study, interpreting the analyses, and reviewing the article. H. Cumulative Arsenic Exposure, mga Ahsan assisted with planning of the study, interpreting Sample Characteristic <62 62–224 225–583 584–1490 >1490 Interaction P the analyses, and writing and reviewing the article.

Education, y 0.54 Acknowledgments ≤2 1.0 1.89 (1.14, 3.15) 3.08 (1.85, 5.12) 3.90 (2.36, 6.44) 5.41 (3.30, 8.88) This research was supported by the US National Insti- >2 1.0 1.55 (0.88, 2.72) 1.92 (1.10, 3.36) 3.09 (1.81, 5.26) 5.29 (3.18, 8.79) tute of Health (grants P42-ES–10349, P30-ES–09089, R01-CA–102484, and R01-CA–107431). Owns a television 0.79 We thank the HEALS staff and the study partici- No 1.0 1.86 (1.16, 2.98) 2.64 (1.65, 4.23) 3.63 (2.30, 5.72) 5.49 (3.51, 8.58) pants for their important contributions. Yes 1.0 1.42 (0.73, 2.76) 2.27 (1.22, 4.24) 3.29 (1.79, 6.04) 5.14 (2.81, 9.39) Owns land 0.22 Human Participant Protection The study protocol was approved by the institutional No 1.0 1.63 (0.95, 2.79) 3.11 (1.84, 5.26) 3.80 (2.27, 6.36) 6.27 (3.79, 10.36) review board of Columbia University and the Yes 1.0 1.87 (1.11, 3.15) 2.06 (1.22, 3.49) 3.33 (2.01, 5.51) 4.59 (2.81, 7.50) Bangladesh Medical Research Council. Informed con- Cooking oil use,b 0.38 sent was obtained from all participants. mL/week ≤125 1.0 1.50 (0.89, 2.53) 2.66 (1.61, 4.39) 3.80 (2.32, 6.20) 5.48 (3.38, 8.88) References 1. Chowdhury UK, Biswas BK, Chowdhury TR, et al. >125 1.0 2.11 (1.25, 3.56) 2.33 (1.34, 4.02) 3.08 (1.82, 5.22) 5.20 (3.15, 8.57) Groundwater arsenic contamination in Bangladesh and c Food index, days 0.91 West Bengal, India. Environ Health Perspect. 2000;108: ≤6 1.0 1.54 (0.89, 2.66) 2.60 (1.52, 4.45) 3.50 (2.07, 5.92) 5.08 (3.04, 8.50) 393–397. >6 1.0 1.99 (1.18, 3.35) 2.46 (1.46, 4.14) 3.51 (2.11, 5.85) 5.78 (3.54, 9.42) 2. Tondel M, Rahman M, Magnuson A, Chowdhury IA, Faruquee MH, Ahmad SA. The relationship of ar- Note. HEALS = Health Effects of Arsenic Longitudinal Study. Prevalence odds ratios were estimated with the senic levels in drinking water and the prevalence rate generalized estimating equation and were adjusted for gender, age, body mass index, smoking status, occupation, and of skin lesions in Bangladesh. Environ Health Perspect. all variables presented in the table. 1999;107:727–729. a Quintiles of exposure; quintile 1, <62 µg, was the reference group. 3. International Agency for Research on Cancer b Adjusted for household size. (IARC). Monographs on the Evaluation of the Carcino- c The food index was created by adding the total number of days per week that each participant reported eating fish, beef or genic Risk to Humans: Arsenic and Arsenic Compounds lamb, poultry, eggs, or dried beans.These food items were indicators of household wealth on the basis of their correlation (Group 1). Lyon, France: IARC; 1987. with the other selected socioeconomic indicators. 4. Smith AH, Goycolea M, Haque R, Biggs ML. Marked increase in bladder and lung cancer mortality in a region of Northern Chile due to arsenic in drink- access to and use of medical care. For example, About the Authors ing water. Am J Epidemiol. 1998;147:660–669. it has been suggested that nutritional status Maria Argos is with the Department of Epidemiology, 5. Chen CJ, Chen CW, Wu MM, Kuo TL. Cancer Mailman School of Public Health, Columbia University, potential in liver, lung, bladder and kidney due to in- may affect one’s ability to methylate arsenic, New York, NY. Faruque Parvez is with the Department of gested inorganic arsenic in drinking water. Br J Cancer. which may be a factor in the differential sus- Environmental Health Sciences, Mailman School of Public 19 92;66:888–892. ceptibility to arsenic-induced disease.25 Future Health, Columbia University, New York. At the time of the 6. Hopenhayn-Rich C, Biggs ML, Fuchs A, et al. study, Yu Chen and Geoffrey R. Howe were with the De- studies also should examine other host charac- Bladder cancer mortality associated with arsenic in partment of Epidemiology, Mailman School of Public drinking water in Argentina. Epidemiology. 1996;7: teristics associated with differential susceptibil- Health, Columbia University, New York. A.Z.M. Iftikhar 117–124. ity to arsenic-induced disease. Hussain and Hassina Momotaj were with the National In- stitute of Preventive and Social Medicine, Dhaka, 7. Hopenhayn-Rich C, Biggs ML, Smith AH. Lung Arsenic exposure through drinking water Bangladesh. Joseph H. Graziano is with the Department of and kidney cancer mortality associated with arsenic in among the Bangladeshi population is an im- Environmental Health Sciences, Mailman School of Public drinking water in Cordoba, Argentina. Int J Epidemiol. 1998;27:561–569. portant public health concern. Yet, there has Health, Columbia University, New York, and the College of Physicians and Surgeons, Columbia University, New York. 8. Tseng W. Effects and dose–response relationships been little research on modifiable risk factors Habibul Ahsan is with the Department of Epidemiology, of skin cancer and blackfoot disease with arsenic. Envi- for arsenic toxicity. The results of our cross- Mailman School of Public Health, Columbia University, ron Health Perspect. 1977;19:109–119. sectional study show that SES—specifically New York, and the Herbert Irving Comprehensive Cancer 9. National Research Council. Subcommittee on Ar- Center, Columbia University, New York. land ownership—modifies the association be- senic in Drinking Water. Arsenic in Drinking Water. Requests for reprints should be sent to Habibul Ahsan, Washington, DC: National Academy Press; 1999. tween arsenic exposure and premalignant MD,MMedSc, Department of Health Studies, University 10. Ahsan H, Chen Y, Kibriya MG, et al. Susceptibility skin lesions, with a higher risk among the of Chicago, 5841 S Maryland Ave, MC 2007, Chicago, IL 60637 (e-mail: [email protected]). to arsenic-induced hyperkeratosis and oxidative stress lower-SES group. Further research is needed This article was accepted February 15, 2006. genes myeloperoxidase and catalase. Cancer Lett. to uncover the biological pathways that may 2003;201:57–65. be responsible for this heterogeneity of effect 11. Ahsan H, Chen Y, Wang Q, Slavkovich V, Contributors Graziano JH, Santella RM. DNA repair gene XPD and and to exploit that knowledge for developing M. Argos analyzed the data, interpreted the results, susceptibility to arsenic-induced hyperkeratosis. Toxicol and evaluating effective interventions. and wrote the article. F. Parvez and Y. Chen assisted Lett. 2003;143:123–131.

830 | Research and Practice | Peer Reviewed | Argos et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

12.Yen IH, Syme SL. The social environment and health: a discussion of the epidemiologic literature. Annu Rev Public Health. 1999;20:287–308. Improving Aging and 13. Ahsan H, Chen Y, Parvez F, et al. Health Effects of Arsenic Longitudinal Study (HEALS): Description of a multidisciplinary epidemiologic investigation. J Expo Public Health Research Anal Environ Epidemiol. 2006;16:191–205. 14 .van Geen A, Ahsan H, Horneman A, et al. Promo- tion of well-switching to mitigate the current arsenic Qualitative and Mixed Methods crisis in Bangladesh. Bull World Health Organ. 2002; Leslie Curry, Renee Shields, and Terrie Wetle, Eds. 80:732–737. Co-published with the Gerontological Society of America 15. Chen Y, Ahsan H, Parvez F, Howe GR. Validity of a food-frequency questionnaire for a large prospec- This new book will enhance researchers’ capacity to design, tive cohort study in Bangladesh. Br J Nutr. 2004;92: conduct , and publish scientifically sound qualitative and 851–859. mixed methods research in aging. With the overall objective 16. Alain G, Tousignant J, Rozenfarb E. Chronic arsenic toxicity. Int J Dermatol. 19 93;32:899–901. of providing a theoretically clear as well practical approach to 17.van Geen A, Zheng Y, Versteeg R, et al. Spatial methodology, the various authors will examine how to choose variability of arsenic in 6000 tube wells in a 25 km2 methods best suited for a particular research question, strate- area of Bangladesh. Water Resour Res. 2003;39:1140. gies for design and analysis, concern of cultural literacy, issues 18. Cheng Z, Zheng Y, Mortlock R, Van Geen A. of alidity and credibility, and effective writing for scientific Rapid multi-element analysis of groundwater by high- journals. resolution inductively coupled plasma mass spectrometry. Anal Bioanal Chem. 2004;379:512–518. 19. Nixon DE, Mussmann GV, Eckdahl SJ, Moyer TP. ORDER TODAY! American Public Health Association Total arsenic in urine: palladium-persulfate vs nickel as PUBLICATION SALES a matrix modifier for graphite furnace atomic absorp- IBSN 0-87553-051-6 WEB: www.apha.org tion spectrophotometry. Clin Chem. 19 91;37: 200 pages softcover 2006 1575–1579. • • E-MAIL: [email protected] $20.96 APHA Members (plus s&h) 20.Durkin MS, Islam S, Hasan ZM, Zaman SS. Mea- $29.95 Nonmembers (plus s&h) TEL: 888-320-APHA sures of socioeconomic status for child health research: FAX: 888-361-APHA comparative results from Bangladesh and Pakistan. Soc Sci Med. 1994;38:1289–1297. 21. Liang K, Zeger S. Longitudinal data-analysis using generalized linear-models. Biometrika. 1986;73:13–22. 22. Pearce N. Effect measures in prevalence studies. Environ Health Perspect. 2004;112:1047–1050. 23. Guha Mazumder DN, Haque R, Ghosh N, et al. Arsenic levels in drinking water and the prevalence of skin lesions in West Bengal, India. Int J Epidemiol. 1998;27:871–877. 24. Haque R, Mazumder DN, Samanta S, et al. Ar- senic in drinking water and skin lesions: dose-response data from West Bengal, India. Epidemiology. 2003;14: 174–182. 25. Loffredo CA, Aposhian HV, Cebrian ME, Yamauchi H, Silbergeld EK. Variability in human metabolism of arsenic. Environ Res. 2003;92:85–91. 26.Vahter M. Genetic polymorphism in the biotrans- formation of inorganic arsenic and its role in toxicity. Toxicol Lett. 2000;112–113:209–217. 27. Hadi A, Parveen R. Arsenicosis in Bangladesh: prevalence and socio-economic correlates. Public Health. 2004;118:559–564. 28.Sikder MS, Maidul ZM, Ali M, Rahman MH. Socio-economic status of chronic arsenicosis patients in Bangladesh. Mymensingh Med J. 2005;14:50–53. 29. Chakraborty AK, Saha KC. Arsenical dermatosis from tubewell water in West Bengal. Indian J Med Res. 19 87;85:326–334. 30. Cuzick J, Evans S, Gillman M, Price Evans DA. Medicinal arsenic and internal malignancies. Br J Can- cer. 19 82;45:904–911. 31. Cuzick J, Sasieni P, Evans S. Ingested arsenic, ker- atoses, and bladder cancer. Am J Epidemiol. 19 92;136: 417–421.

May 2007, Vol 97, No. 5 | American Journal of Public Health Argos et al. | Peer Reviewed | Research and Practice | 831  RESEARCH AND PRACTICE 

Housing Instability Among Current and Former Welfare Recipients

| Robin Phinney, BA, Sheldon Danziger, PhD, Harold A. Pollack, PhD, and Kristin Seefeldt, MPP

The Personal Responsibility and Work Oppor- Objectives. We examined correlates of eviction and homelessness among current tunity Reconciliation Act of 1996 ended the and former welfare recipients from 1997 to 2003 in an urban Michigan community. federal government’s guarantee of cash assis- Methods. Longitudinal cohort data were drawn from the Women’s Employment tance to poor families by replacing the 60- Study, a representative panel study of mothers who were receiving cash welfare year-old entitlement program Aid to Families in February 1997. We used logistic regression analysis to identify risk factors for with Dependent Children with the transition- both eviction and homelessness over the survey period. to-work program Temporary Assistance for Results. Twenty percent (95% confidence interval [CI]=16%, 23%) of respon- Needy Families (TANF). TANF imposed a dents were evicted and 12% (95% CI=10%, 15%) experienced homelessness at cumulative 60-month time limit on receipt of least once between fall 1997 and fall 2003. Multivariate analyses indicated 2 con- federally funded cash benefits for most recipi- sistent risk factors: having less than a high school education and having used illicit drugs other than marijuana. Mental and physical health problems were sig- ents, tied welfare receipt directly to work ac- nificantly associated with homelessness but not evictions. A multivariate screen- tivity, and devolved a great amount of pro- ing algorithm achieved 75% sensitivity and 67% specificity in identifying indi- gram authority to the states.1 viduals at risk for homelessness. A corresponding algorithm for eviction achieved Many studies have explored the impact of 75% sensitivity and 50% specificity. the Personal Responsibility and Work Oppor- Conclusions. The high prevalence of housing instability among our respon- tunity Reconciliation Act on the work behav- dents suggests the need to better target housing assistance and other social ser- ior and welfare status of low-income vices to current and former welfare recipients with identifiable personal prob- women.2,3 Large declines in welfare case- lems. (Am J Public Health. 2007;97:832–837. doi:10.2105/AJPH.2005.082677) loads, increased work activity among single mothers, and reductions in official child poverty rates are indicators of the success of wel- although some of the studies reviewed by Acs misleading in light of evidence indicating a fare reform. However, related research on the and Loprest showed that more than 20% of high prevalence of key risk factors among material well-being of those who no longer welfare leavers continued to experience hous- the broader population of low-income indi- receive welfare (welfare leavers) has docu- ing problems.3 viduals. For example, previous research in- mented high rates of hardships, including lack Despite high rates of housing problems volving the panel data used in our study of health insurance coverage, food insecurity, among current and former welfare recipients, showed that more than one third of all cur- and housing problems.4–8 the characteristics of individuals and families rent and former welfare recipients satisfied Housing problems are particularly acute that are associated with heightened risks of diagnostic screening criteria for at least 1 among low-income families with children. unstable housing conditions have not re- psychiatric disorder in a given year and that Wood and Rangarajan, using 2003 survey ceived much attention. Given the probable re- about two thirds experienced at least 1 dis- data, reported that 16% of unsubsidized cur- lationships between housing-related hard- order over the 6-year study period.10 ,11 Such rent and former welfare recipients in New Jer- ships, material well-being, and labor market mental and physical health problems are sey had experienced eviction, homelessness, outcomes, the lack of knowledge about these likely to contribute to housing instability doubling up with friends or relatives, or fre- characteristics represents a significant gap in among low-income families. quent moves in the previous year.9 In a re- the literature. view of welfare-leaver studies across many The studies published to date have tended PREDICTORS OF HOUSING states, Acs and Loprest found that between to use observational data from samples of INSTABILITY 25% and 50% of welfare leavers reported homeless or otherwise disadvantaged individu- falling behind on housing payments, and be- als. Few studies have compared the character- The homeless have been shown to have tween 6% and 26% reported moving as a re- istics and circumstances of low-income women high levels of physical and mental illness.12 ,13 sult of high housing costs. Smaller percent- experiencing housing problems with those of Physical health problems may foster housing ages experienced eviction (4% to 7%) or women in similar socioeconomic situations instability by depleting economic resources homelessness (1% to 3%). On average, rates who are not experiencing such problems. or interfering with an individual’s ability to of housing problems were somewhat lower Observational studies of severely disad- work steadily.14 ,15 Psychiatric disorders may for welfare leavers than for welfare stayers, vantaged individuals may be especially estrange individuals from family and friends,

832 | Research and Practice | Peer Reviewed | Phinney et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

leaving them with fewer social and material experiencing abuse often leave their primary African American, single, heads of house- resources to draw upon in times of need. In- residence to escape the abuser and may seek holds, and between 18 and 54 years of age. terpersonal problems may place people with refuge in emergency shelters.13 Battered Response rates at the 5 waves were 86%, mental illness at greater risk for eviction or women are also vulnerable to eviction that 92%, 91%, 91%, and 93%, respectively. We homelessness if they are less able to negotiate arises from partners’ abusive behavior di- used data from the 536 WES respondents conflicts with landlords or to address con- rected toward other tenants, landlords, or who completed all 5 interviews. Because tentious living situations.16 ,17 the physical property itself.30 there was little evidence that attrition from Substance use and related disorders are also Women who have been under criminal the sample was nonrandom, we did not use risk factors for homelessness,18 ,19 because they justice supervision may be more likely to ex- sample weights.35 may deplete social and material resources. It perience housing difficulties. Employers are Given the panel design, data were available has been shown that drug use, particularly use less likely to hire individuals with criminal on whether a woman had been evicted or ex- of crack, heroin, and cocaine, is more pro- records31; jail time may erode human capital perienced a period of homelessness between nounced in homeless populations.13 ,19–22 Crimi- or social networks and exacerbate physical or each survey wave. All independent variables, nal offenses related to the possession, use, and mental illness.32 Diminished social networks with the exception of criminal conviction, distribution of illicit substances are explicit resulting from criminal conviction may also were measured as baseline characteristics re- criteria for eviction from public housing and reduce the chance that individuals can rely ported at wave 1 in fall 1997. The dependent loss of housing voucher aid.23 These offenses on friends or family for social or material sup- variables were based on the respondents’ ex- also constitute grounds for eviction from pri- port in times of need, thus heightening the periences after the baseline characteristics vate housing in many jurisdictions. risk of housing instability. Criminal offending had been measured. We analyzed correlates Demographic attributes are related to may foster social contacts that heighten spe- associated with a respondent’s report that housing instability as well. On average, cific risks such as domestic violence. It may (1) she was evicted from her residence at homeless women are disproportionately also signal other individual characteristics and some time in the period starting after the fall young and non-White.12 ,13,22 Young adults behaviors that may interfere with an individ- 19 97 interview and ending with the fall 2003 may be vulnerable to housing instability be- ual’s ability to maintain secure and stable interview and (2) she experienced a spell of cause they have not developed the economic housing.24 In addition, individuals convicted homelessness over the same 6-year period. and social resources to help them obtain and of drug-related felonies are no longer eligible We used multiple logistic regression analy- retain housing.24 Discrimination may amplify for some forms of housing assistance and sis to examine the baseline characteristics as- housing instability among members of mi- public aid.23,33 sociated with each outcome. We included 7 nority groups.20,25,26 Racial differences with Finally, human capital—work experience, employment barriers using definitions devel- respect to wealth may place these individu- work skills, and education—affects an individ- oped in earlier research involving this data als at greater risk for housing instability, be- ual’s ability to find and retain employ- set15 : low educational attainment (did not cause wealth or savings can help people ment,15 , 3 4 to navigate the housing assistance graduate from high school), low human capi- maintain stable housing during periods of system or complicated evictions proceedings, tal (low levels of work experience or low lev- financial hardship.27 or to secure affordable housing through hous- els of use of specific skills on previous jobs), Marriage and cohabitation may be protec- ing searches. Human capital deficiencies may and whether a respondent had a criminal tive factors against eviction and homelessness. foster housing instability if those who are job- conviction, met diagnostic screening criteria In comparison with families headed by mar- less or working in low-paying jobs have diffi- for 1 of 3 measured psychiatric disorders ried couples, single-mother families are much culty making monthly rental payments. (major depression, generalized anxiety disor- more likely to be poor and to have fewer eco- der, and posttraumatic stress disorder), had a nomic resources upon which to call in times METHODS physical health problem, had used “hard” of need. Similarly, cohabitation may con- drugs (defined as stimulants, cocaine, crack, tribute to housing stability by increasing We analyzed data from the Women’s Em- heroin, hallucinogens, or inhalants) at any household economic resources. Marriage and ployment Study (WES), a longitudinal survey time in her life before the fall 1997 interview, cohabitation may also reflect otherwise unob- of single mothers who received cash welfare in or had experienced severe domestic abuse at served individual characteristics such as inter- one Michigan urban county in February 1997. any time up to the fall 1997 interview. personal skills.22 In contrast, cohabitation is Trained interviewers conducted in-person We also examined sociodemographic char- less stable than marriage.28,29 Thus, women interviews with these women in the fall of acteristics, including whether a respondent who cohabit may be more at risk for housing 19 97, 1998, 1999, 2001, and 2003; inter- was married or cohabiting, her race and age, problems than are single women were a rela- views averaged about 1 hour at the first wave the number of children residing with her, and tionship to end and they experience difficulty and about 1.5 hours at the final wave. Re- the percentage of years she had spent on wel- finding a new residence. spondents were selected with equal probabil- fare from 18 years of age to the 1997 inter- Domestic violence may increase the risk of ity from all women who received TANF bene- view. All of these variables other than crimi- eviction and homelessness, because women fits in February 1997 and were White or nal conviction were measured at the first

May 2007, Vol 97, No. 5 | American Journal of Public Health Phinney et al. | Peer Reviewed | Research and Practice | 833  RESEARCH AND PRACTICE 

TABLE 1—Variable Definitions and Prevalence of Characteristics Among Sample The results of our multivariate analysis are Respondents: Women’s Employment Study, 1997–2003 shown in Table 3. (In a related analysis, we used 4 survey waves and estimated fixed ef- Sample fects and conditional logistic regression mod- Dependent variables els for eviction and homelessness over the Reported eviction in at least 1 wave after 1997 interview, % 19.8 6-year survey period. In that analysis, we Reported episode of homelessness in at least 1 wave after 1997 interview, % 12.3 used the contemporaneous value of each Independent variables variable rather than the value at the first in- No high school degree at 1997 interview, % 29.9 terview. Findings are available on request.) Human capital barrier (worked less than 20% of the time between age 18 y and 1997 interview or had 26.4 Women who had less than a high school edu- performed 4 or fewer of 9 specific skills on a job before 1997 interview), % cation and those with a history of hard drug Criminal conviction before 1999 interview, % 4.5 use were more likely to have been evicted at Mental health problem in 12 months before 1997 interview (met diagnostic screening criteria using 36.0 some time during the study period. the CIDI short-form for 1 or more of the following disorders: posttraumatic stress disorder, A greater number of variables were sig- depression, and generalized anxiety disorder), % nificant in the logistic regression examining Physical health problem in 12 months before 1997 interview (had age-specific physical limitation or 54.4 the correlates of homelessness than in the self-reported fair or poor health), % logistic regression examining evictions. Hav- Domestic violence (experienced severe abuse at any time before 1997 interview), % 52.8 ing less than a high school education, hav- Hard drug use (use of cocaine, crack, stimulants, heroin, or other hard drugs at any time before 17.4 ing a criminal conviction, experiencing a 1997 interview), % mental or physical health problem, experi- Married at 1997 interview, % 10.3 encing domestic violence, using hard drugs, Cohabiting at 1997 interview, % 14.7 being African American, and being between African American (coded as 1 vs 0 for White), % 54.7 the ages of 18 and 24 years in 1997 were Aged 18–24 y at 1997 interview, % 25.0 all associated with being homeless at least No. of children in household at 1997 interview, mean 2.24 once after fall 1997. The relationship be- Percentage of years on welfare from age 18 y until 1997 interview 59.3 tween physical health problems and homelessness was particularly strong and statisti- Note. CIDI =Composite International Diagnostic Interview. cally significant (adjusted odds ratio [OR] = 3.19, P <.01). To determine the relative magnitude of each interview. Table 1 includes definitions of all or experienced homelessness. Mental and of the independent variables, we computed the dependent and independent variables. physical health barriers and episodes of do- probability of eviction or homelessness during mestic violence were also significantly asso- the survey period for a representative respon- RESULTS ciated with evictions and homelessness. As dent with no health problems or barriers to of the 1997 survey, about one quarter of work activity and median demographic charac- Table 2 shows descriptive statistics for the women with housing problems reported that teristics (African American, not married or entire sample as well as for women classified they had used hard drugs at some point in cohabiting, older than 24 years, caring for 2 according to their experiences of eviction and their lives, in comparison with about one children, and 59% of years on welfare as an homelessness. Twenty-four percent of respon- sixth of those not experiencing these prob- adult). With these characteristics, the baseline dents (n=130) had unstable housing situa- lems. Those experiencing housing problems predicted probability that a representative re- tions at some time between 1997 and 2003 were more than twice as likely as those not spondent would be evicted at some point after (95% CI=20%, 28%); 19.8% (n=106) had experiencing housing problems to have been the 1997 interview was 8.3%; the probability been evicted at least once (95% CI=16%, convicted of a crime. of homelessness was 1.6%. 23%), and 12.3% (n=66) had been home- At the same time, many respondents who We calculated the predicted probability of less at least once (95% CI=10%, 15%). reported no housing difficulties also reported a respondent experiencing an eviction or Bivariate comparisons indicated that barriers traditionally associated with housing homelessness if she had one of the character- women who experienced 1 of these housing risks. More than 30% reported mental health istics that was significant in the regression problems were more disadvantaged than barriers at the first wave, and approximately but was otherwise identical to the representa- other respondents. Among those evicted or half reported physical health barriers. Also, tive respondent relative to the baseline pre- homeless between 1997 and 2003, approxi- half had experienced domestic violence at diction (8.3% for eviction, 1.6% for home- mately one half had not completed high some time in their life prior to fall 1997, and lessness). Varying the independent variables school, as compared with only about one nearly 20% reported some history of hard one at a time allowed us to determine the ex- quarter of those who had not been evicted drug use in 1997 or preceding years. tent to which the probability of eviction or

834 | Research and Practice | Peer Reviewed | Phinney et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 2—Employment Barriers, Sociodemographic Characteristics, and Experiences of homelessness changed in response to Eviction and Homelessness: Women’s Employment Study, 1997–2003 changes in the characteristics of the representative respondent. Her probability of experi- Evicted During 1997–2003 Homeless During 1997–2003 encing an eviction increased to 12.6% if she Total No Yes No Yes had used hard drugs and to 18.9% if she had (N=536) (n=430) (n=106) (n=470) (n=66) not completed high school. Employment barriers, % 100.0 80.2 19.8 87.7 12.3 Having less than a high school education or No high school degree 29.9 25.6 47.2*** 27.0 50.0*** having a criminal conviction increased the Human capital barrier (low work 26.4 26.2 27.4 25.4 33.9 probability of homelessness to 3.6% and experience or skills) 4.5%, respectively. In addition, the probability Criminal conviction 4.5 3.5 8.5** 3.4 12.1*** increased to 5.1% if she had physical health Mental health problem 36.0 33.6 46.2** 33.4 54.6*** problems, to 2.7% if she had mental health Physical health problem 54.4 51.4 66.7*** 50.6 81.5*** problems, and to 3.0% if she had used hard Domestic violence 52.8 50.0 64.2*** 49.8 74.2*** drugs. If she had experienced domestic vio- Hard drug use 17.4 15.7 24.8** 16.0 27.7** lence, her probability of homelessness in- Sociodemographic characteristics creased to 3.6%, and it fell to 0.9% if she was Married, % 10.3 11.2 6.6 10.6 7.6 White. The probability increased to 3.0% if Cohabitating, % 14.7 15.1 13.2 14.5 16.7 she was between the ages of 18 and 24 years. African American, % 54.7 55.6 50.9 53.8 60.6 We calculated sensitivity and specificity Age 18–24 y, % 25.0 24.7 26.4 23.8 33.3* curves for both homelessness and eviction to No. of children in household, mean 2.24 2.22 2.32 2.25 2.15 examine how well our model correctly distin- Years on welfare as an adult, % 59.3 58.2 63.8** 58.3 66.2** guished between respondents with and without housing problems. For homelessness, the Note. Domestic violence and hard drug use were assessed over a respondent’s lifetime, prior to the first interview in fall 1997. area under the receiver operating characteris- Criminal conviction was assessed prior to 1999. As a result of missing data, values in some columns do not reflect the total sample size. tic curve of 0.79 indicated a rather tight *P<.10; **P<.05; ***P<.01. model fit (an area of 1.0 indicates a perfectly discriminatory model). For eviction, the area under the receiver operating characteristic curve was 0.69. Screening individuals with a predicted probability of homelessness exceed- TABLE 3—Adjusted Odds Ratios (AOR; With 95% Confidence Intervals [CIs]) for Eviction ing 0.11 would result in a sensitivity of 75% and Homelessness (n=523): Women’s Employment Study, 1997–2003 and a specificity of 67%. This same cutoff AOR for AOR for led to a positive predictive value of 24% and Eviction (95% CI) Homelessness (95% CI) a negative predictive value of 95%. In the No high school degree 2.57*** (1.56, 4.24) 2.21** (1.19, 4.10) case of eviction, screening individuals with a Human capital barrier (low work experience or skills) 0.70 (0.41, 1.21) 1.15 (0.60, 2.20) predicted probability of eviction exceeding Criminal conviction 1.78 (0.71, 4.51) 2.79* (0.99, 7.89) 0.15 would result in a sensitivity of 75% but Mental health problem 1.39 (0.86, 2.25) 1.69* (0.93, 3.07) a much lower specificity, 50%. This cutoff Physical health problem 1.39 (0.84, 2.29) 3.19*** (1.54, 6.60) led to a positive predictive value of 27% Domestic violence 1.44 (0.87, 2.36) 2.23** (1.16, 4.29) and a negative predictive value of 89%. Hard drug use 1.59* (0.90, 2.81) 1.83* (0.90, 3.72) Our results reveal that many factors associ- Married 0.49 (0.19, 1.25) 1.18 (0.39, 3.55) ated with work outcomes are also associated Cohabiting 0.79 (0.39, 1.58) 1.82 (0.80, 4.17) with housing outcomes among current and African American 0.78 (0.48, 1.29) 1.83* (0.97, 3.47) former welfare recipients, although the rela- Aged 18–24 y 1.07 (0.63, 1.86) 1.83* (0.94, 3.56) tive importance of these factors differs for No. of children in household 1.10 (0.92, 1.32) 0.87 (0.68, 1.11) each of the various outcomes. In other analy- Percentage of years on welfare as an adult 1.53 (0.55, 4.27) 2.38 (0.60, 9.38) ses (data not shown but available from the χ2 42.6*** 64.6*** authors on request), we found no consistent pattern between employment and eviction or Note.Although 13 respondents had missing data on 1 or more of the independent variables, decreasing the total number of homelessness. That is, respondents who expe- respondents included in the logistic regression analysis to 523, we conducted a sensitivity analysis and confirmed that the missing data did not appear to bias the regression results. rienced one of these negative housing events *P<.10; **P<.05; ***P<.01. were not necessarily less likely to be working, to have experienced job loss, or to have lower

May 2007, Vol 97, No. 5 | American Journal of Public Health Phinney et al. | Peer Reviewed | Research and Practice | 835  RESEARCH AND PRACTICE 

household incomes. Work provides incom- on self-reported information collected by applicants with a history of exposure to do- plete protection against housing instability. trained survey interviewers. These measures mestic violence (74% of the homeless indi- were not based on clinical or diagnostic ex- viduals in our sample had such a history). DISCUSSION aminations. We did use standard epidemio- Exploring housing problems among those logical measures validated in other surveys.37 with physical or mental health problems Other studies, including those involving the Second, our sample was drawn from non- may be particularly important. same data used here,8,36 have documented Hispanic White and African American moth- Current federal housing programs are not improvements in the average socioeconomic ers 18 to 54 years of age who received cash entitlements; relatively few low-income status of single mothers after welfare reform. welfare in fall 1997 in 1 urban Michigan households receive housing assistance. A However, these studies have not focused on county. Although the WES represents well few states use TANF dollars to provide sup- the substantial minority of these women who the population from which it was drawn, it is plemental housing assistance to families experience housing instability. We found that not nationally representative or representative who have left welfare, although under cur- about one fifth of low-income WES mothers of a population of welfare recipients in 2005. rent rules receipt of such assistance for had been evicted and about one eighth had In particular, because of its urban Michigan more than 4 months would count against been homeless at some point between 1997 location and its cohort design, the WES does the 60-month federal time limit.38 A New and 2003. Risks of these housing problems not provide representative data on the youn- Jersey study showed that rates of eviction were much higher within specific subgroups: gest subgroups of current TANF recipients, and homelessness were markedly lower 50% of those with criminal convictions, immigrant or noncitizen-headed households, among current and former welfare recipi- 38.8% of high school dropouts, 34.4% of or households in rural areas. ents who received a housing subsidy or hard drug users, and 30.4% of those ex- Other data sets, such as the National Sur- lived in public housing than among those posed to domestic violence had experienced vey of America’s Families, the National Longi- living in unsubsidized housing.9 In addition, at least 1 of the 2 housing insecurity prob- tudinal Survey of Youth, and the National the positive effect on employment rates and lems assessed. Survey of Drug Use and Health, are more na- earnings was greater among welfare recipi- Although our 6.5-year panel investigation tionally representative. However, the first 2 ents in the Minnesota Family Investment does not permit extensive causal analysis, are (repeated) cross-sectional data sets, and Program who received housing assistance our regression analyses demonstrated that the third does not provide detailed informa- in combination with other forms of assis- several baseline characteristics were associ- tion on mental health, health, drug use, expe- tance than among those who did not receive ated with subsequent episodes of housing riences of domestic violence, and some of the housing assistance.39 instability and are thus potentially important other correlates of housing instability that we Our findings have identified a set of risk targets of practice and policy interventions. analyzed in our study. The numbers of cur- factors that can help service providers target Our results revealed a strong association rent and former welfare recipients in the social services and housing assistance to between 4 personal problems—health limita- WES compare favorably to the numbers in those at high risk for eviction and homeless- tions, psychiatric disorders, hard drug use, these national samples. Also, descriptive sta- ness. Some of the risk factors that might be and domestic violence—and periods of tistics from previous WES investigations of used for such targeting are similar to those homelessness. Many studies of homelessness substance use and psychiatric disorders sug- used to allocate assistance with employ- and health have been unable to disentangle gest close agreement between the WES and ment. However, our findings also indicate the causal relationship between these vari- national data when overlapping data items differences between determinants of hous- ables. In our analysis, these problems were are available.10 ing instability and those related to work. measured in fall 1997 and thus were a pre- Some current and former welfare recipients cursor to the spells of homelessness assessed Policy Implications who work still face significant risks of hous- over the subsequent 6 years. Among the 4 Our results indicate a number of risk facing instability and may thus require assess- problems, only hard drug use was associated tors that should be given more attention by ment and services. Such findings underscore with future evictions, suggesting that differ- policymakers and practitioners. For example, that wages and hours worked are not the ent factors contribute to different unstable it is likely that screening welfare recipients only measures of well-being to consider housing outcomes. for health, mental health, drug use, and do- when examining the transition from welfare mestic violence problems could identify to work. Increased attention to housing as Study Limitations many of the women facing the greatest risks an outcome can improve the well-being of This study involved 2 principal limitations. of homelessness. Almost half of our respon- welfare recipients. First, our data regarding drug use, psychiatric dents who reported eviction or homelessness disorders (assessed using the Composite Inter- lacked a high school degree, and nearly national Diagnostic Interview screening bat- 30% had low skill levels or low levels of About the Authors Robin Phinney is a doctoral candidate in the Gerald R. Ford teries for the 12-month prevalence of psychi- work experience. Welfare caseworkers might School of Public Policy, University of Michigan, Ann Arbor. atric disorders), and health status were based investigate housing concerns among welfare Sheldon Danziger and Kristin Seefeldt are with the

836 | Research and Practice | Peer Reviewed | Phinney et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Gerald R. Ford School of Public Policy, University of subsidies for TANF recipients: evidence from New Symposium on Homelessness Research. Washington, DC: Michigan. Harold A. Pollack is with the School of Social Jersey. Available at: http://www.mathematica-mpr.com/ US Dept of Health and Human Services; 1999: Service Administration, University of Chicago, Chicago, Ill. publications/PDFs/wfnjhousing.pdf. Accessed August 2.1–2.31. Requests for reprints should be sent to Robin Phinney, 8, 2005. 28. Bumpass L, Sweet J. National estimates of cohabi- Gerald R. Ford School of Public Policy, University of 10.Pollack HA, Danziger SH, Seefeldt K, Jayakody R. tation. Demography. 1989;26:615–625. Michigan, Weill Hall Suite 5100, 735 S State St, Ann Arbor, Substance abuse among welfare recipients: trends and MI 48109 (e-mail: [email protected]). 29. Smock P. Cohabitation in the United States: an ap- policy responses. Soc Serv Rev. 2002;76:256–274. This article was accepted May 12, 2006. praisal of research themes, findings, and implications. 11. Corcoran ME, Danziger SK, Tolman R. Long term Annu Rev Sociol. 2000;26:1–20. employment of African-American and white welfare 30. Menard A. Domestic violence and housing. Vio- recipients and the role of persistent mental health Contributors lence Against Women. 2001;7:707–720. problems. Women Health. 2004;39:21–40. R. Phinney was the main contributor to the statistical 31. Holzer H, Stoll M, Wissoker D. Job performance analysis, participated in the initial drafting of the article, 12. Shlay AB, Rossi PH. Social science research and and retention among welfare recipients. Soc Serv Rev. and contributed to editorial revisions. S. Danziger, contemporary studies of homelessness. Annu Rev So- 2004;78:343–369. H.A. Pollack, and K. Seefeldt contributed to the statisti- ciol. 19 92;18:129–160. cal analysis, participated in the initial drafting of the 13. Burt M, Aron LY, Lee E, Valente J. Helping Amer- 32. Western B, Kling JR, Weiman DF. The labor mar- article, and contributed to editorial revisions. ica’s Homeless: Emergency Shelter or Affordable Housing? ket consequences of incarceration. Crime Delinq. 2001; Washington, DC: Urban Institute Press; 2001. 47:410–427. Acknowledgments 14 .Wright J, Weber E. Homelessness and Health. New 33. Jayakody R, Danziger SH, Pollack HA. Welfare York, NY: McGraw-Hill; 1987. reform, substance use, and mental health. J Health Polit This research was supported in part by grants from the Policy Law. 2000;25:623–651. Charles Stewart Mott Foundation, the Joyce Foundation, 15. Danziger SK, Kalil A, Anderson NJ. Human capital, the John D. and Catherine T. MacArthur Foundation, the physical health, and mental health of welfare recipi- 34.Piliavin I, Entner Wright BR, Mare RD, Westerfelt Substance Abuse Policy Research Program of the Robert ents: co-occurrence and correlates. J Soc Issues. 2000; AH. The dynamics of homelessness. Available at: http:// Wood Johnson Foundation (grant 047841), and the Na- 56:635–654. www.irp.wisc.edu/publications/dps/pdfs/dp103594.pdf. tional Institute of Mental Health (grant R24-MH51363). Accessed August 8, 2005. 16. Mojtabai R. Perceived reasons for loss of housing and continued homelessness among homeless persons 35. Cadena B, Pape A. The extent and consequences with mental illness. Psychiatr Serv. 2005;56:172–178. of attrition in the Women’s Employment Study. Avail- Human Participation Protection able at: http://fordschool.umich.edu/research/poverty/ This study was approved by the institutional review 17.Lamb HR, Bachrach LL. Some perspectives on pubs.php. Accessed December 8, 2005. board of the University of Michigan. Written consent was deinstitutionalization. Psychiatr Serv. 2001;52: 36. Danziger SH, Gottschalk P. Diverging Fortunes: obtained from all respondents before each interview. 103 9–1045. Trends in Poverty and Inequality. Washington, DC: Pop- 18.Vangeest J, Johnson T. Substance abuse and ulation Reference Bureau; 2004. homelessness—direct or indirect effects? Ann Epidemiol. References 2002;12:455–461. 37. Kessler RC, Chiu W, Demler O, Walters E. Preva- lence, severity, and comorbidity of twelve-month DSM-IV 1. Administration for Children and Families, US Dept 19. Caton CLM, Hasin D, Shrout PE, et al. Risk fac- of Health and Human Services. Fact sheets: welfare. disorders in the National Comorbidity Survey replica- tors for homelessness among indigent urban adults tion. Arch Gen Psychiatry. 2005;62:617–627. Available at: http://www.acf.hhs.gov/news/facts/tanf. with no history of psychotic illness: a case-control html. Accessed August 8, 2005. study. Am J Public Health. 2000;90:258–263. 38. Sard B, Waller M. Housing Strategies to Strengthen Welfare Policy and Support Working Families. Washing- 2. Blank RM, Haskins R, eds. The New World of 20. Bassuk EL, Buckner JC, Weinreb LF, Browne A, ton, DC: Brookings Institution; 2002. Welfare. Washington, DC: Brookings Institution; 2001. Bassuk SS, Dawson R. Homelessness in female-headed 3. Acs G, Loprest P. Leaving Welfare: Employment families: childhood and adult risk and protective fac- 39. Miller C, Knox V, Gennetian L, Dodoo M, Hunter J, and Well-Being of Families That Left Welfare in the Post- tors. Am J Public Health. 19 97;87:241–248. Redcross C. Reforming Welfare and Rewarding Work: Final Report on the Minnesota Family Investment Pro- Entitlement Era. Kalamazoo, Mich: WE Upjohn Institute 21. Bassuk EL, Buckner JC, Perloff JN, Bassuk SS. gram, Vol. 1. Effects on Adults. New York, NY: Man- of Employment Research; 2004. Prevalence of mental health and substance use disor- power Demonstration Research Corp; 2000. 4. Danziger SH, Heflin CM, Corcoran ME, Oltmans E, ders among homeless and low-income housed mothers. Wang H-C. Does it pay to move from welfare to work? Am J Psychiatry. 19 98;155:1561–1564. JPolicy Anal Manage. 2002;21:671–692. 22. Jencks C. The Homeless. Cambridge, Mass: Har- 5. Meyer B, Sullivan J. The effects of welfare and tax vard University Press; 1994. reform: the material well-being of single mothers in the 23. Department of Housing and Urban Development v 1980s and 1990s. J Public Economics. 2004;88: Rucker et al., 535 US 125 (2002). 13 87–1420. 24.Wright B, Caspi A, Moffitt T, Silva P. Factors asso- 6. Iceland J, Bauman K. Income poverty and mate- ciated with doubled-up housing—a common precursor rial hardship: how strong is the association? Available to homelessness. Soc Serv Rev. 19 9 8;72:92–111. at: http://www.npc.umich.edu/publications/working_ 25. Yinger J. Housing discrimination and residential papers. Accessed August 8, 2005. segregation as causes of poverty. In: Danziger SH, 7. Moffitt R, Winder K. Does it pay to move from Haveman RH, eds. Understanding Poverty. New York, welfare to work: a comment on Danziger, Heflin, NY: Russell Sage Foundation; 2001:359–391. Corcoran, Oltmans, and Wang. J Policy Anal Manage. 26. Hopper K, Milburn NG. Homelessness among 2005;24:399–409. African-Americans: a historical and contemporary per- 8. Danziger SH, Wang H-C. Does it pay to move spective. In: Baumohl J, ed. Homelessness in America. from welfare to work? Reply to Robert Moffitt and Phoenix, Ariz: Oryx Press; 1996:123–131. Katie Winder. J Policy Anal Manage. 2005;24: 27.Rosenheck R, Bassuk EL, Salomon A. Special 411–417. populations of homeless Americans. In: Fosburg LB, 9. Wood RG, Rangarajan A. The benefits of housing Dennis DL, eds. Practical Lessons: The 1998 National

May 2007, Vol 97, No. 5 | American Journal of Public Health Phinney et al. | Peer Reviewed | Research and Practice | 837  RESEARCH AND PRACTICE 

Educational Inequalities in Initiation, Cessation, and Prevalence of Smoking Among 3 Italian Birth Cohorts

| Bruno Federico, MSc, Giuseppe Costa, MD, and Anton E. Kunst, PhD

The association between smoking and low so- Objectives. We examined socioeconomic inequalities in initiation and cessation cioeconomic status (SES) has become increas- rates of smoking and the resultant inequality in smoking prevalence among 3 ingly stronger in almost all industrialized consecutive Italian birth cohorts. 1–4 countries. This is associated with a shift in Methods. We used data from the 1999–2000 Italian National Health Interview the social distribution of smoking over time: Survey, which included 28958 men and 29769 women who were born between higher-SES groups take up the habit before 1940 and 1969. The association between smoking variables and level of educa- lower-SES groups, but ultimately, this pattern tion was assessed with logistic regression and life table analyses. reverses. The transition has been completed Results. Inequalities in the lifetime prevalence of smoking increased across the in the United States and in several northern 3 birth cohorts in Italy. At age 40, lower-educated persons in the youngest cohort European countries, but it is still under way reported on average 1 to 5 years of additional exposure to regular smoking compared with higher-educated persons. Inequalities in smoking prevalence increased in southern Europe.1,5 among both men and women because of widening inequalities in initiation rates. Despite this evidence, antitobacco mea- Among women, growing inequalities in cessation rates also played a role. sures in industrialized countries have so far Conclusions. The relative contribution of initiation and cessation to socioeco- failed to address the widening social inequal- nomic inequalities in smoking rates varied by both gender and birth cohort. For ity in smoking over time, because they have the youngest birth cohort, policies that address inequalities in smoking should mainly focused on decreasing the overall focus on both initiation and cessation. (Am J Public Health. 2006;96:838–845. prevalence of smoking. In a recent review, doi:10.2105/AJPH.2005.067082) Platt et al. suggested that a mix of interventions—such as increased taxation, availability of nicotine-replacement therapy, and mea- whether they varied by age) in the 3 indica- In the case of missing values, the survey sures that address the underlying economic tors associated with tobacco use—probability used an automatic procedure of data imputa- and psychosocial determinants of smoking ini- of taking up the habit, probability of quitting, tion10 for the following variables: gender, tiation and cessation—may reduce inequality and prevalence of smoking—for each birth co- age, occupation, education, municipality size, in smoking prevalence.6 Amos suggested that hort and for both genders. A second specific and geographic area of residence. The pro- the effectiveness of antitobacco interventions objective was to estimate whether differences portion of missing values for the variables as- would be improved if gender differences and in either initiation or cessation contributed to sociated with smoking were very low in our stage of the smoking epidemic (i.e., phase of the inequalities in lifetime prevalence of study: 2% for smoking status, 2% for age at diffusion of smoking within a population) smoking and whether their relative contribu- smoking initiation, and 7% for age at smok- were taken into account.7 However, it is un- tion varied with gender and birth cohort. ing cessation. The analysis of raw or imputed known whether inequalities can be reduced data showed very similar proportions of cur- by measures that affect initiation of smoking METHODS rent, former, and never smokers. Good levels among adolescents or by measures that affect of agreement also were shown for the me- cessation among adults. We used data from the 1999–2000 Italian dian ages at initiation and cessation of smok- We wanted to disentangle the dynamics of Health Interview Survey of 140011 individu- ing (Istituto Nazionale di Statistica [ISTAT], inequalities in smoking with data from Italy, als who were randomly chosen within strata unpublished data, 1998). a country in which the social distribution of of geographical area, municipality, and house- Education is one of the most widely used smoking has been rapidly changing.3 Our goal hold size (multistage sampling). The nonre- and valid indicators of SES.11 Compared was to (1) describe socioeconomic inequalities sponse rate was 13%. The design of similar with other indicators—such as occupation, in initiation and cessation rates of smoking surveys of civilian noninstitutionalized popu- which can only be applied to those actively and the resultant inequality in smoking preva- lations have been published elsewhere.3,8,9 engaged in work—it has a substantial advan- lence among 3 consecutive Italian birth co- Our study was restricted to individuals who tage because it allows the classification of all horts, and (2) identify changes among these were born during the following decades: individuals in a population. Because the birth cohorts. More specifically, our first ob- 1940–1949, 1950–1959, and 1960–1969, highest level of education is usually acquired jective was to describe the educational differ- which yielded a study sample of 28958 men early in life, its use as a proxy of SES makes ences (how large the differences were and and 29769 women. it less likely that selection mechanisms will

838 | Research and Practice | Peer Reviewed | Federico et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 1—Smoking and Educational Statistics for 3 Italian Cohorts, by Gender about this measure have been published elsewhere.12 More Started Quit We also calculated summary measures No. of Educated,a Smoking Before Smoking Before across all age groups for smoking initiation, Birth Cohort Respondents % Age 20 Years, % Age 30 Years, % cessation, and prevalence with an absolute Men 1940–1949 8715 34.1 45.6 4.2 measure of level of education. To summarize 1950–1959 9618 48.5 49.1 9.5 inequalities in smoking initiation and cessa- 1960–1969 10625 52.0 43.1 13.8 tion, we calculated the cumulative probabili- Women 1940–1949 9108 23.8 11.7 3.3 ties of initiation and cessation with the previ- 1950–1959 9663 44.1 25.3 10.3 ously computed age-specific probabilities of 1960–1969 10998 56.5 27.8 16.4 starting and quitting smoking. These measures were estimated with multiple-decre- aPercentage of respondents who had upper-secondary or postsecondary education. ment life tables, and initiation and cessation were the only transition probabilities. For smoking prevalence, we calculated the aver- have an effect (e.g., implying that a person interval. The follow-up of respondents in the age number of years of smoking on the basis attains a lower educational level because he last 5-year interval was on average half of the age-specific prevalence rates of smok- is a smoker). Furthermore, data about in- compared with the other intervals; there- ing. These measures summarize educational come were not available in the Italian fore, we approximately corrected the proba- differences in smoking in concrete terms that Health Interview Survey. We grouped level bility of both initiation and cessation during can be interpreted from a life-course per- of education into 2 categories according to the last interval by multiplying these proba- spective. the highest level successfully completed: bilities by 2. We stratified all analyses by gender and those who had a high-school degree or We used 2 complementary methods to by birth cohort. In the calculation, we took higher (higher-educated) and those who calculate summary measures of the inequal- into account the multistage sampling design had less than a high-school degree (lower- ity in initiation, cessation, and prevalence of the survey, with the municipalities—which educated). The outcome measures were reg- of smoking. First, we used logistic regres- were sampled by strata according to their ular current smoking, age at initiation, and sion to compute the Relative Index of In- population size—being the primary sampling age at cessation. Table 1 shows the propor- equalities (RII) with a 95% confidence in- units. Each observation had a weight that tions of respondents who had a higher edu- terval (CI). Second, we applied the life corresponded with the inverse of the proba- cation, who started smoking before age 20 table technique with the previously calcu- bility of being sampled. We used Stata statis- years, and who quit before age 30 years for lated age-specific probabilities. In the regres- tical software, version 8.2 (Stata Corp, Col- each of the 3 cohorts studied. sion analysis, smoking-related variables lege Station, Tex). For each level of education and 5-year in- were associated with relative position on the terval, we calculated smoking prevalence educational hierarchy. Thus, the RII took RESULTS rates and initiation and cessation probabili- into account the fact that the distribution of ties. The probability of being a smoker, calcu- level of education changed markedly among Figure 1 shows the age-specific probabili- lated from age 10 years, was defined as the the 3 cohorts (Table 1). ties of smoking initiation. These were similar proportion of regular smokers in each age The RII is a ratio that compares the odds among the higher-educated and lower- interval; in this quotient, the numerator was of event for those at the bottom of the edu- educated males who were born between the number of respondents who had ever cational hierarchy compared with those at 1940 and 1949, and they were markedly de- smoked in that interval, and the denomina- the top of it. To compute RIIs, we used a creased among the higher-educated males in tor comprised all respondents. The probabil- finer classification of educational attainment the youngest cohort. Among women, higher ity of smoking initiation, calculated from age by distinguishing 4 ordinal categories on the initiation rates were found among the higher- 10 years, was defined for each age interval basis of highest degree achieved: elementary educated in the older cohorts and among the as the quotient between the number of indi- education (5 years of education), middle lower-educated in the youngest cohort. viduals who started smoking in that age in- school education (8 years), high-school edu- Age-specific probabilities of quitting smok- terval and the number of individuals who cation (12–13 years), and university educa- ing are shown in Figure 2. In all 3 cohorts, were at risk (nonsmokers at the beginning tion (16–18 years). In the regression models, higher-educated men showed an increased of the interval, excluding former smokers). both educational rank and age were entered “risk” for quitting compared with lower- The probability of smoking cessation, calcu- as continuous covariates. The dependent educated men up to age 49 years. Among lated from age 20 years, was defined as the variables were the odds for starting smoking, women, socioeconomic differences increased proportion of smokers at the beginning of the odds for quitting smoking, and the odds in the youngest birth cohort. Again, the each age interval that quit during the same for being a smoker at age 40 years. Details probability of quitting smoking was lower

May 2007, Vol 97, No. 5 | American Journal of Public Health Federico et al. | Peer Reviewed | Research and Practice | 839  RESEARCH AND PRACTICE 

Note.In 2000, data were available up to age 50 years for the 1950–1959 cohort and up to age 40 years for the 1960–1969 cohort. FIGURE 1—Age-specific probabilities of smoking initiation among 3 Italian birth cohorts: 1940–1949 for men (a) and women (b), 1950–1959 for men (c) and women (d), and 1960–1969 for men (e) and women (f).

among lower-SES groups during early The models for smoking initiation and cohort, and a similar shift was observed for adulthood. smoking prevalence at age 40 years high- smoking prevalence. On the contrary, the Figure 3 shows the proportion of individu- light the increasing socioeconomic inequali- likelihood of quitting smoking was higher als who reported smoking on a regular basis ties among younger cohorts of both genders among higher-educated women in all 3 co- for each 5-year interval between age 10 and (Table 2). The RII for smoking initiation was horts, but this difference tended to increase 59 years. Prevalence rates clearly declined 1.07 (95% CI = 0.87, 1.32) for males in the in the younger cohorts. among males in the youngest birth cohort, older cohorts, and it was 5.10 (95% CI = Table 3 shows inequalities with absolute but the decrease was larger among higher- 4.12, 6.31) for the cohort that was born be- differences in the cumulative probability of educated men. Among women, higher preva- tween 1960 and 1969. In all 3 cohorts, smoking initiation and cessation and absolute lence rates were present among the higher- higher-educated individuals were more differences in the number of years having educated in the oldest cohort and among the likely to quit smoking. Among females, smoked. The calculation for these measures is lower-educated in the youngest cohort. The there was a shift in the RIIs for starting extended to age 40 years. The cumulative socioeconomic differences in the prevalence smoking, from values below 1 (indicating a probability of starting smoking was nearly of smoking visibly increased with age among higher probability among the higher-educated) identical for lower-SES (0.52) and higher-SES females in all 3 cohorts. to values larger than 1 in the younger (0.51) men who were born between 1940

840 | Research and Practice | Peer Reviewed | Federico et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Note.In 2000, data were available up to age 50 years for the 1950–1959 cohort and up to age 40 years for the 1960–1969 cohort. FIGURE 2—Age-specific probabilities of smoking cessation among 3 Italian birth cohorts: 1940–1949 for men (a) and women (b), 1950–1959 for men (c) and women (d), and 1960–1969 for men (e) and women (f).

and 1949; cumulative probabilities for similar women in the older cohorts; in the youngest cohorts to positive values in the youngest men in the youngest birth cohort were 0.53 cohort, it was greater among lower-educated birth cohort. and 0.37, respectively. Higher-educated men women. Similar to men, higher-educated in all 3 cohorts had a greater probability of women were more likely than lower-educated DISCUSSION quitting smoking, and this difference tended women to quit smoking in all 3 cohorts, and to increase among higher-educated men in the difference increased from 5.2% among We found increasing inequalities in smok- the youngest cohort. On average, before women who were born between 1940 and ing prevalence among 3 successive birth co- reaching his 40th birthday, a lower-SES man 1949 to 10.5% among women who were horts in Italy, with a distinct gender pattern. who was born between 1940 and 1949 born 2 decades later. When the age span is Among males, the rising inequality was spent 0.8 more years smoking compared with extended to the 60th birthday (1940–1949 mainly the result of widening inequalities in a higher-educated man in the same cohort, cohort), inequalities in cessation of smoking initiation. Among females, this was the result and the educational difference (low minus were somewhat attenuated among women of progressively higher uptake rates among high) among men who were born 2 decades (data not shown). Absolute differences in the lower-SES groups and growing differences in later was 5.1 years. average number of years having smoked cessation rates. Thus, the relative contribu- Among women, the probability of starting changed from negative values (higher preva- tion of initiation and cessation to socioeco- smoking was larger among higher-educated lence among higher-educated) in the older nomic inequalities in smoking prevalence

May 2007, Vol 97, No. 5 | American Journal of Public Health Federico et al. | Peer Reviewed | Research and Practice | 841  RESEARCH AND PRACTICE 

Note.In 2000, data were available up to age 50 years for the 1950–1959 cohort and up to age 40 years for the 1960–1969 cohort. FIGURE 3—Age-specific probabilities of smoking among 3 Italian birth cohorts: 1940–1949 for men (a) and women (b), 1950–1959 for men (c) and women (d), and 1960–1969 for men (e) and women (f).

rates varied according to both gender and TABLE 2—Regression-Based Estimates of Inequalities in Smoking Initiation, Cessation, and birth cohort. Prevalence Among 3 Italian Birth Cohorts, by Gender Evaluation of Data Problems Relative Index of Inequality (95% Confidence Interval) The accuracy of self-reported data on Probability of Probability of Probability of Being smoking has been questioned13 ; addition- Birth Cohort Starting Smoking Quitting Smoking a Smoker at Age 40 Years ally, increased underreporting has been Men seen in Italy.14 ,15 However, this would have 1940–1949 1.07 (0.87, 1.32) 0.77 (0.69, 1.00) 1.21 (1.00, 1.47) affected our estimates of the educational dif- 1950–1959 1.38 (1.11, 1.70) 0.49 (0.37, 0.64) 1.80 (1.42, 2.28) ference only if the misreporting was depen- 1960–1969 5.10 (4.12, 6.31) 0.70 (0.53, 0.92) 4.10 (3.39, 4.96) dent on level of education. In this regard, 1 Women study found no significant differences in the 1940–1949 0.25 (0.20, 0.32) 0.87 (0.58, 1.32) 0.32 (0.25, 0.41) misclassification of smokers between socio- 1950–1959 0.43 (0.35, 0.53) 0.46 (0.33, 0.65) 0.69 (0.56, 0.86) economic groups.16 1960–1969 1.42 (1.18, 1.72) 0.39 (0.29, 0.52) 1.99 (1.61, 2.45) Two other potential sources of error—data and selection bias—are associated with the

842 | Research and Practice | Peer Reviewed | Federico et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 3—Life Table–Based Estimates of Educational Inequalities in Smoking Initiation, Previous studies have reported the chang- Cessation, and Prevalence Among 3 Italian Birth Cohorts, by Gender ing association between smoking and SES by birth cohort in the United States and several Cumulative Probability of Cumulative Probability of Years of Smoking European countries. These studies examined Starting Before Age 40 Yearsa Quitting Before Age 40 Yearsa Before Age 40 Yearsa either initiation or prevalence of smoking. In Birth Lower- Higher- Lower- Higher- Lower- Higher- the United States, the proportion of smokers Cohort Difference Educated Educated Difference Educated Educated Difference Educated Educated was higher among White lower-educated Men 1940–1949 1.0 51.9 50.9 –4.3 16.3 20.6 0.8 14.7 13.9 males than among higher-educated males in 1950–1959 2.6 52.9 50.3 –8.0 24.5 32.5 1.7 14.7 13.0 all birth cohorts from the past century. 1960–1969 15.4 52.6 37.2 –6.5 29.6 36.0 5.1 14.2 9.1 Among White women, higher prevalence Women 1940–1949 –6.9 28.3 35.2 –5.2 14.2 19.4 –3.7 4.4 8.2 rates emerged among the lower-educated 1950–1959 –8.0 29.7 37.8 –9.8 22.6 32.3 –1.9 7.0 9.0 who were born between 1930 and 1939.20 1960–1969 3.4 34.0 30.6 –10.5 31.7 42.2 1.4 8.4 6.9 The same pattern was found among other racial/ethnic groups, with some variations re- a These measures were based on previously calculated age-specific probabilities. garding educational differences (how large the differences were and whether they varied by age). retrospective design of this research. In our were independently drawn and misclassifica- In contrast to individuals who were born study, we reconstructed the lifetime smok- tion of smoking may have occurred in both before 1920, lower-educated Finnish men ing experience of respondents on the basis surveys, prevalence rates matched reason- who were born during the 1920s had higher of questions about the timing of smoking ably well in most cases. The prevalence of initiation and lifetime prevalence rates com- initiation and smoking cessation. Thus, re- ever smoking was larger in retrospective pared with their higher-educated counter- call of either age at starting smoking or age (1999–2000 survey) compared with con- parts. Educational difference among males in at quitting smoking may have been prob- temporaneous (1990–1991 survey) reports, Denmark and Spain appeared 10 and 20 lematic, especially in the case of older re- except among lower-educated males and years later, respectively. Among European spondents. Some studies found high levels higher-educated females in the oldest birth women, the reversal in all these countries was of agreement between retrospective and cohort, which suggests that selective mortal- 1 or 2 decades later.4,21,22 Our data are con- contemporaneous measures of smoking sta- ity played a role among older respondents sistent with these findings and show that the tus, with greater differences emerging with but not among subjects in our study. inversion of the social gradient in smoking longer recall periods.17 However, other There also was a notable contrast among initiation emerged among Italian men and research has shown that retrospective re- those who were born between 1960 and women 2 to 4 decades later than both the US ports on smoking habits produce prevalence 1969. Prevalence estimates from the retro- and northern European countries. This trend rates similar to those from contemporaneous spective data were larger than estimates from coincides with the north-south diffusion of the reports.18 Larger inconsistencies were found the contemporaneous data, with a difference smoking epidemic in Europe.1,5 among light smokers, and among heavy ranging from 9% to 13%. This may have Previous studies of smoking cessation have smokers and nonsmokers, the inconsisten- been the result of smoking habits that were used the quit ratio, which is defined as the cies were smaller.19 not well established in 1990 and 1991, when ratio—at 1 point in time—between former The second potential bias is the selection respondents were in their twenties, or efforts smokers and ever smokers (current+former of respondents for the 1999–2000 survey, to conceal habits and give “desirable” answers smokers). In the United States, quit ratios in- because low-educated smokers may have suf- to the questionnaire. However, the difference creased between 1950 and 1990, but the fered from higher mortality rates. The result in prevalence rates between the 2 surveys increase was lower among the lower-educated would have been an underestimation of the was similar for both educational groups; and Black populations.23 Quitting smoking average number of years having smoked ex- therefore, our estimates of inequalities were has become more frequent during the past perienced by lower-educated individuals. To not substantially biased. 4 decades in most European countries as evaluate the joint effect of recall bias and se- well, especially among higher-educated indi- lective mortality, we compared the smoking Comparison With Previous Studies viduals,24,25 which is similar to what we data recorded in the 2000 survey with data To our knowledge, this study is the first to found in Italy. that was collected in the 1990–1991 Italian describe the socioeconomic differences (how However, the life course perspective and Health Interview Survey. For each survey, large the differences were and whether they the use of cumulative probabilities in our we calculated the proportion of respondents varied by age) in the prevalence, initiation, study have some advantages that quit ratios who were born between 1910 and 1969 and cessation of smoking in Italy, with a focus do not have. We showed there are educa- and who had ever smoked before 1990 on the lifetime smoking trajectories of individ- tional differences in the timing of cessation (data not shown). Although the 2 samples uals in 3 consecutive birth cohorts. by age group, which occurs at earlier ages

May 2007, Vol 97, No. 5 | American Journal of Public Health Federico et al. | Peer Reviewed | Research and Practice | 843  RESEARCH AND PRACTICE 

among those who have higher levels of edu- We have shown that cessation rates in- both gender and birth cohort, inequalities cation. Additionally, we were able to ana- creased among the cohorts born after 1950, were present in both initiation and cessation lyze cessation and initiation rates in parallel but this rate was lower among the lower- of smoking, and they increased among ways and thus, show that the widening of educated individuals. Increased knowledge younger birth cohorts. For the youngest inequalities in smoking prevalence was about the adverse consequences of smoking birth cohort, policies that address inequali- mainly the result of trends in smoking initia- and the different policy measures that have ties in smoking should focus on both initiation (rather than cessation), especially been progressively implemented over a 30- tion and cessation. among men. year span (e.g., banning tobacco advertising, no smoking in public places, health warnings Explanations on cigarette packs)26 may have contributed About the Authors Bruno Federico and Anton E. Kunst are with the Depart- Several factors may have determined the to the decision to quit smoking. However, ment of Public Health, Erasmus Medical Center, Rotter- increasing inequalities in initiation rates higher-educated individuals, who are gener- dam, Netherlands. Bruno Federico is also with the De- across the 3 cohorts. The first reports that ally more health-conscious, seem to have partment of Health and Sport Sciences, University of Cassino, Italy. Giuseppe Costa is with the Department of associated smoking tobacco with cancer were been more responsive to these messages com- Public Health and Microbiology, University of Turin, published during the 1950s, and evidence of pared with lower-educated individuals. Italy. the negative impact of smoking on health It also has been reported that despite simi- Requests for reprints should be sent to Bruno Federico, MSc, Department of Health and Sport Sciences, grew during the following decades. In Italy, lar levels of motivation to stop, lower-SES University of Cassino, viale Bonomi 03043, Cassino (FR) however, health information campaigns were smokers are less successful in their attempts Italy (e-mail: [email protected]). delayed because a comprehensive strategy to quit,32 and they have a lower level of social This article was accepted October 25, 2005. against smoking had not been developed.26 It support and a lower level of confidence in was not until the mid-1980s that local admin- their ability to quit. Inequalities in cessation Contributors B. Federico conducted the data analyses and prepared istrations made a substantial effort to educate of smoking were particularly large among fe- the original draft. A.E. Kunst originated the study and the public about tobacco use prevention.27 males, possibly because lower-SES women supervised all phases of its implementation. G. Costa The gap among educational groups in initia- often perceive smoking as an effective strat- provided data and reviewed drafts of the article. All authors developed research questions and interpreted tion rates became particularly large within the egy for coping with stress and difficult living study findings. cohort that was born during the 1960s, circumstances.33 When disadvantages accu- which makes it plausible that these campaigns mulate over the life course, the risk for be- Acknowledgments were most effective among higher-educated coming a persistent smoker increases.34 We thank the European Network on Smoking Preven- individuals. A similar development occurred Lower-educated individuals reported a tion for financially supporting this study. We are grateful to Francesca Vannoni of the Epidemiology Unit, in the United States, where a significant lower probability of quitting compared with Piedmont Region, Italy, and 2 anonymous reviewers negative correlation between education and higher-educated individuals during early for their valuable comments on previous drafts of the smoking initiation was found only after in- adulthood, but not afterwards, possibly be- article. We also thank Laura Iannucci of Istituto Nazionale di Statistica, who carried out supplementary formation about the hazards of smoking cause of a higher incidence of tobacco-related analyses of the National Health Interview Survey. was diffused.28 health problems with older age. Recent hospi- However, another factor is likely to have talization, development of coronary heart Human Participant Protection played an important role in the large rise of disease, and severely impaired lung func- Data acquired from ISTAT were anonymous, which initiation rates among lower-educated women: tion35,36 are in fact important predictors of ensured individual subjects’ protection. the adoption of smoking as a symbol of inde- smoking cessation and occur most frequently pendence and success, which is influenced by among lower-SES men. References 1. Cavelaars AE, Kunst AE, Geurts JJ, et al. Educa- the mass media, tobacco advertising, and the tional differences in smoking: international comparison. examples of higher-SES women.29 Recently, Conclusions BMJ. 2000;320:1102–1107. tobacco companies have identified lower-SES Policy interventions should be more sensi- 2. Pierce JP, Fiore MC, Novotny TE, et al. Trends in women as a particularly promising “market tive to the needs of the most disadvantaged. cigarette smoking in the United States. Educational differences are increasing. JAMA. 1989;261:56–60. segment,”30 and they are promoting cigarettes “Denormalizing” the use of tobacco—banning 3. Faggiano F, Versino E, Lemma P. Decennial as a way to relieve stress, improve perform- smoking in public places and workplaces, trends of social differentials in smoking habits in Italy. ance, and lose weight and as a means of social prohibiting all direct and indirect forms of to- Cancer Causes Control. 2001;12:665–671. 31 acceptance. Thus, the greater vulnerability bacco promotion, tailoring communication for 4. Schiaffino A, Fernandez E, Borrell C, et al. Gen- of lower socioeconomic groups to these vari- lower-SES groups, and targeting smoking der and educational differences in smoking initiation ous messages and influences combined with cessation services geographically—may even- rates in Spain from 1948 to 1992. Eur J Public Health. 2003;13:56–60. a more skeptical attitude toward the occur- tually decrease inequalities in smoking.37 Al- 5. Edwards R. The problem of tobacco smoking. rence of disease in the future may explain the though the relative contribution of initiation BMJ. 2004;328:217–219. increasing inequalities in smoking initiation and cessation to socioeconomic differences in 6. Platt S, Amos A, Gnich W, et al., Strategies to re- that we found. the lifetime prevalence of smoking varied by duce socioeconomic inequalities in health. In: Bakker MJ,

844 | Research and Practice | Peer Reviewed | Federico et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Mackenbach JP, eds. Reducing Inequalities in Health: A 26. Galeone D. Il contesto normativo nazionale [ To- European Perspective. London, UK, and New York, NY: bacco control: Italian laws]. Ital Heart J. 2001;2 (suppl 1): Routledge; 2002:44–45. 19–21. 7. Amos A. Women and smoking. Br Med Bull. 27. Mangiaracina G. Percorsi sociali e strategie non 1996;52:74–89. istituzionali. Available at: http://www.iss.it/sitp/ofad/ docu/0001.pdf. Accessed August 30, 2005. 8. La Vecchia C, Decarli A, Pagano R. Prevalence of Strategies for Cultural cigarette smoking among subsequent cohorts of Italian 28.Farrell P, Fuchs VR. Schooling and health: the cig- males and females. Prev Med. 1986;15:606–613. arette connection. J Health Econ. 19 82;1:217–230. Competency in Indian 9. Pagano R, La Vecchia C, Decarli A. Smoking in 29. Amos A, Haglund M. From social taboo to “torch Italy, 1994. Tumori. 1996;82:309–313. of freedom”: the marketing of cigarettes to women. Tob Health Care Control. 2000;9:3–8. 10. Barcaroli G, D’Aurizio L, Luzi O, et al. Metodi e by Mim Dixon and Pamela E. Iron software per il controllo e la correzione dei dati. Rome, 30. Barbeau EM, Leavy-Sperounis A, Balbach ED. ISBN# 0-87553-070-2 Italy: ISTAT; 1998. Smoking, social class, and gender: what can public health learn from the tobacco industry about dispari- 11.Winkleby MA, Jatulis DE, Frank E, et al. Socio- ties in smoking? Tob Control. 2004;13:115–120. economic status and health: how education, income, and occupation contribute to risk factors for cardiovas- 31. Cook BL, Wayne GF, Keithly L, et al. One size cular disease. Am J Public Health. 19 92;82:816–820. does not fit all: how the tobacco industry has altered cigarette design to target consumer groups with specific 12. Mackenbach JP, Kunst AE. Measuring the magni- psychological and psychosocial needs. Addiction. 2003; tude of socio-economic inequalities in health: an over- 98:1547–1561. view of available measures illustrated with two examples from Europe. Soc Sci Med. 19 97;44:757–771. 32. Osler M, Prescott E. Psychosocial, behavioural, and health determinants of successful smoking cessa- 13.Patrick DL, Cheadle A, Thompson DC, et al. The tion: a longitudinal study of Danish adults. Tob Control. validity of self-reported smoking: a review and meta- 1998;7:262–267. analysis. Am J Public Health. 1994;84:1086–1093. 33. Graham H. Women’s smoking and family health. 14 . Gallus S, Colombo P, Scarpino V, et al. Smoking Soc Sci Med. 19 87;25:47–56. in Italy, 2002. Tumori. 2002;88:453–456. “It should prove valuable as a guide to 34. Jefferis BJ, Power C, Graham H, et al. Changing 15. La Vecchia C. Smoking in Italy, 1949–1983. Prev social gradients in cigarette smoking and cessation over others in establishing this important di- Med. 1986;15:274–281. two decades of adult follow-up in a British birth co- mension of health care and in reducing 16. Suadicani P, Hein HO, Gyntelberg F. Serum vali- hort. J Public Health (Oxf ). 2004;26:13–18. racial and ethnic disparities.” dated tobacco use and social inequalities in risk of is- 35. Godtfredsen NS, Prescott E, Osler M, et al. Predic- chaemic heart disease. Int J Epidemiol. 1994;23: tors of smoking reduction and cessation in a cohort of Alan R. Nelson, MD, 293–300. danish moderate and heavy smokers. Prev Med. 2001; Chair, Institute of Medicine Committee on 17.Krall EA, Valadian I, Dwyer JT, et al. Accuracy of 33:46–52. Ethnic and Racial Disparities in Health Care recalled smoking data. Am J Public Health. 1989;79: and co-editor of Unequal Treatment: 36.Freund KM, D’Agostino RB, Belanger AJ, et al. 200–202. Predictors of smoking cessation: the Framingham Confronting Racial and Ethnic Disparities in 18.Kenkel D, Lillard DR, Mathios A. Smoke or fog? Study. Am J Epidemiol. 19 92;135:957–964. Health Care (2003). The usefulness of retrospectively reported information 37. Kunst AE, Giskes K, Mackenbach JP. Socio- about smoking. Addiction. 2003;98:1307–1313. Economic Inequalities in Smoking in the European Union. 19.Kenkel DS, Lillard DR, Mathios AD. Accounting Applying an Equity Lens to Tobacco Control Policies. for misclassification error in retrospective smoking Bruxelles, Belgium: ENSP; 2004:52–60. Member: $ 18.85 data. Health Econ. 2004;13:1031–1044. Non-Member: $ 26.95 20. Escobedo LG, Peddicord JP. Smoking prevalence in US birth cohorts: the influence of gender and education. Am J Public Health. 1996;86:231–236. www.aphabookstore.org 21. Laaksonen M, Uutela A, Vartiainen E, et al. Devel- 1-866-320-2742 toll free opment of smoking by birth cohort in the adult population in eastern Finland 1972–97. Tob Control. 19 9 9;8: 1-866-361-2742 fax 161–168. 22. Osler M, Holstein B, Avlund K, et al. Socioeco- nomic position and smoking behaviour in Danish adults. Scand J Public Health. 2001;29:32–39. 23. Gilpin EA, Pierce JP. Demographic differences in patterns in the incidence of smoking cessation: United States 1950–1990. Ann Epidemiol. 2002;12:141–150. 24.Fernandez E, Schiaffino A, Garcia M, Borras JM. Widening social inequalities in smoking cessation in Spain, 1987–1997. J Epidemiol Community Health. 2001;55:729–730. 25. Morabia A, Costanza MC, Bernstein MS, et al. Ages at initiation of cigarette smoking and quit attempts among women: a generation effect. Am J Public Health. 2002;92:71–74.

May 2007, Vol 97, No. 5 | American Journal of Public Health Federico et al. | Peer Reviewed | Research and Practice | 845  RESEARCH AND PRACTICE 

Integrating Disease Control Strategies: Balancing Water Sanitation and Hygiene Interventions to Reduce Diarrheal Disease Burden

| Joseph N.S. Eisenberg, PhD, MPH, James C. Scott, MPH, and Travis Porco, PhD, MPH

In the developing world, more than 1 billion Objectives. Although the burden of diarrheal disease resulting from inade- people continue to lack an adequate supply quate water quality, sanitation practices, and hygiene remains high, there is little of clean water and adequate disposal of exc- understanding of the integration of these environmental control strategies. We 1 reta. Such statistics explain why the overall tested a modeling framework designed to capture the interdependent transmis- global burden of water-, sanitation-, and sion pathways of enteric pathogens. hygiene-related disease remains high2,3 even Methods. We developed a household-level stochastic model accounting for 5 though oral rehydration therapy has led to different transmission pathways. We estimated disease preventable through water reductions in mortality.4 Despite this demon- treatment by comparing 2 scenarios: all households fully exposed to contami- strated need for water, sanitation, and hy- nated drinking water and all households receiving the water quality intervention. giene improvements,3 our understanding of Results. We found that the benefits of a water quality intervention depend on sanitation and hygiene conditions. When sanitation conditions are poor, water integrated control strategies remains poor. quality improvements may have minimal impact regardless of amount of water Part of the reason is that most intervention contamination. If each transmission pathway alone is sufficient to maintain studies have examined 1 intervention in iso- diarrheal disease, single-pathway interventions will have minimal benefit, and lation without considering other potential ultimately an intervention will be successful only if all sufficient pathways are pathways of transmission. eliminated. However, when 1 pathway is critical to maintaining the disease, pub- There is increasing evidence that the effi- lic health efforts should focus on this critical pathway. cacy of household water quality interventions Conclusions. Our findings provide guidance in understanding how to best reduce depends on the level of sanitation within the and eliminate diarrheal disease through integrated control strategies. (Am J Public targeted community.5–7 This dependency Health. 2007;97:846–852. doi:10.2105/AJPH.2006.086207) may explain why, although many household- level water quality intervention studies have pathways have been codified in the F diagram, contaminated through improper management shown impressive reductions in health bur- which classifies transmission pathways as of excreta (poor sanitation). Cairncross et al.24 den,8 results have been highly variable. Some mediated through food, fingers, fomites, flies, extended the F diagram by differentiating be- studies have shown reductions as high as and so on10 (see Bern et al.,11 Huttly et al.,12 tween infection transmission within house- 85%, and others have shown no reduction. and Curtis et al.13 for reviews). The exposure holds and within the public domain. Moreover, estimates of disease reduction factors summarized in the F diagram (e.g., gen- Other studies have addressed the interac- may be inflated because of publication bias eral hygiene behaviors,14 ,15 fecal contamination between different transmission path- (positive results are more likely to be pub- tion,16–18 food contamination,19 and drinking ways, suggesting that the risk associated with lished than are negative results), lack of blind- water storage practices19 )aswell as more dis- water contamination depends on the level of ing (a study design feature in which partici- tal factors (e.g., day-care centers20 and socio- community sanitation.5,7,25 Although implic- pants do not know whether they are involved economic factors21,22)are important to our itly assuming that level of community sanita- in the intervention or nonintervention arm of understanding of these pathways. tion modifies the association between water the study; only 1 of the 15 developing coun- Water may be contaminated through contamination and diarrheal disease, none of try studies reviewed by Fewtrell et al.8 were runoff and may expose individuals through these studies have addressed the observation blinded), and lack of randomization (only 5 drinking water or recreational, bathing, or that the multiple transmission pathways and of the 15 studies reviewed by Fewtrell et al.8 washing activities, and food may be contami- contagious nature of pathogens result in were randomized).9 nated either through infected animals or from risks that are dependent on the disease sta- These interpretive challenges arise in part contact with contaminated water or soil. In- tus of the community.26–29 Many enteric because enteric pathogens are transmitted adequate hygiene may result in contamina- pathogens can be transmitted from infectious through a complex set of interdependent path- tion of fomites in common living spaces23; human excreta to susceptible humans either ways, including both contaminated food and infection may then be transmitted in many directly or indirectly through the environ- water along with household- and community- ways (e.g., through exposures in day-care cen- ment, and thus they are sustained through level person-to-person routes; these various ters or through sexual activity). Soil may be chains of transmission that may pass through

846 | Research and Practice | Peer Reviewed | Eisenberg et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

combinations of pathways. The importance of each pathway depends on the pathogen and specific environmental conditions, and the efficacy of any given exposure-specific intervention strategy depends on the level of pathogen exposure from other pathways. We used simulation modeling to evaluate the effectiveness of water quality interventions under varying community sanitation and hygiene conditions, explicitly acknowledging that rates of infection depend on numbers of current and past infections. We characterized the specifics of this dependency by explicitly modeling transmission pathways, in effect yielding a dynamic version of the F diagram. Specifically, we determined (1) how the efficacy of water quality interventions depends on the level of both household- and community-level transmission and (2) the conditions under which water quality interventions, hygiene and sanitation improvements, or both are effective in reducing the burden of disease in a community.

METHODS FIGURE 1—Diagram illustrating potential transmission routes for enteric (waterborne) Community-Level Model pathogens (a) and how indirect exposure to contaminated drinking water can occur as a In constructing our model, we assumed that result of the multiple and interdependent nature of exposure pathways (b). enteric pathogens can survive in the environment outside of a host; this fact dictates the possible pathways a waterborne pathogen can exploit in completing its transmission cycle communal settings (e.g., washing in rivers or outside source accounts for pathogen intro- (Figure 1a). Because infectious diseases are schools) or in settings where a family member duction from upstream water flow, contami- transmissible, unlike many other conditions contaminates his or her hands in the commu- nated food, or an infectious individual ex- studied in analytic epidemiology, individuals nity and brings that contamination into the posed outside of the community. Any of the may be indirectly at risk of an environmental household. Second, within-household trans- arrows linking households in Figure 2 can exposure; for example, cohort A may be in- mission accounts for the movement of move through any of the paths shown in fected with a pathogen as a result of exposure pathogens between 2 individuals residing in Figure 1. to contaminated water (a water quality issue) the same household. The magnitude of this A prominent feature of this model struc- and in turn may transmit this pathogen to co- transmission pathway is generally thought of ture is that transmission pathways are interde- hort B through a food, hygiene, or sanitation as a function of hygiene. pendent; for example, the rate of infection pathway (Figure 1b). In this manner, hygiene Third, household-to-water transmission from exposure to contaminated water affects and sanitation can modulate the effects of accounts for the contamination of water be- the rate of within- or between-household drinking water contamination, and likewise cause of the inappropriate disposal of feces. transmission. Likewise, the rate of within- or drinking water contamination can modulate The magnitude of this pathway is generally between-household transmission affects the the effects of poor hygiene or sanitation. thought of as a function of sanitation and is rate of pathogens shed into the environment The model used in this study was a often addressed through building latrines. by infectious individuals. Thus, as transmis- household-level model incorporating 5 trans- Fourth, water-to-household transmission ac- sion from person to person increases, the con- mission pathways (Figure 2). First, between- counts for the movement of pathogens to hu- centration of pathogens in the water in- household transmission accounts for the mans as a result of exposure to pathogens in creases, which in turn increases the risk of movement of pathogens from an individual in drinking water. Improving water quality con- exposure to contaminated water. The dy- one household to an individual in another trols this pathway. Finally, external transmis- namic process represented in this relationship household. This transmission can occur in sion of pathogens to the community from an is not accounted for in the standard risk

May 2007, Vol 97, No. 5 | American Journal of Public Health Eisenberg et al. | Peer Reviewed | Research and Practice | 847  RESEARCH AND PRACTICE 

β β φ β β Note. c =between-household transmission; d =within-household transmission; =contamination of water; dw =exposure from contaminated water; e =other sources. FIGURE 2—Schematic for a household-level infection transmission model. models predominant in epidemiology, complete immunity, a property that many according to the total community population wherein risk of infection from 1 transmission types of enteric viruses share. In addition, Ni (Table 1). In addition to these final 2 transmis- pathway is assumed to be independent of rate and N represent the total number of people sion rate parameters, the model included the of infection from other pathways. To repre- living in household i (Ni = Si + Ii + Ri ) and in rate of transmission from the environment Σ β ε sent this conceptual model in a mathematical the community (N= i Ni ), respectively. (other than water), e = r , and the rate of β framework, we used a discrete-event stochas- Seven model parameters require identifica- transmission from water, dw = rW(t), where tic model (a model that estimates probability tion: ρ, the per individual recovery rate; φ, the W(t ) is the number of waterborne pathogens by allowing random variation in 1 or more rate at which infected individuals shed viable at time t. variables over time) structure at the house- pathogens into the water supply; µ, the mor- On the basis of these state variables and hold level.29–31 tality rate for pathogens in the water supply; parameters, 5 events are possible: (1) recovery

We used 3 model-state variables, Si , Ii , and r, the risk of infection per pathogen exposure; from infection, (2) secondary infection from ε Ri ,representing the numbers of susceptible , the number of pathogens in the environ- someone inside the household, (3) secondary individuals, infectious individuals, and individ- ment, not from drinking water but from other infection from someone in the community, β uals immune to further infection, respectively, sources such as food; h , the within-household (4) infection from an environmental source β for household i.For illustrative purposes, rate of transmission; and c , the between- other than drinking water, and (5) infection we chose to model a pathogen that confers household rate of transmission standardized from drinking contaminated water. The first 4

848 | Research and Practice | Peer Reviewed | Eisenberg et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 1—Parameter Values and Units Used in the Simulation Analysis (Figure 3b–e, upper right-hand corners). The findings shown in Figure 3b–e can be inter- Parameter Values Units preted by considering 3 questions: (1) When ρ 10 Days there are zero to low levels of community β φ {0, 0.5, 1.0, 1.5, 2.0} Pathogens/person/day transmission ( c < 0.01), how does household β transmission affect the preventable fraction? c {0, 0.005, 0.01, 0.02, 0.03, 0.04, 0.06, 0.1} No. of transmission events/infected individual β {0, 0.005, 0.01, 0.02, 0.03, 0.04, 0.06, 0.1} No. of transmission events/infected individual (2) When there are zero to low levels of h β ε 30 Pathogens household transmission ( h < 0.01), how does µ1 Pathogens/day community transmission affect the preventa- r 0.000002 Infections/pathogen ble fraction? (3) When there is a high level of β community transmission ( c > 0.08), how Note. ρ=recovery rate; φ=rate at which infected individuals shed pathogens into the water supply; β and β =between- c h does household transmission affect the pre- and within-household transmission rates; ε=level of environmental contamination; µ=pathogen die-off rate in the water supply; r=risk of infection per pathogen exposure. ventable fraction? We discuss each of these issues in turn. In the case of zero to low levels of community transmission, we found that whenever the events are assumed to occur in an exponential contaminated water (A=1). Simulations were household-level transmission rate increased, waiting time pattern; that is, rates remain con- repeated 10 times for each parameter set and the preventable fraction also increased. That stant between events in which the numbers of for both exposed and unexposed scenarios, is, when there was little community transmis- infected individuals change. The hazard for an resulting in a total of 6400 (320×2×10) sion, household transmission acted primarily infection event resulting from drinking water is simulations. Table 1 summarizes the values to amplify the waterborne process, which was dependent on the total number W(t)ofviable for the 7 parameters used in the simulation. the target of our intervention; in addition, pathogens in the water at a specific time; mod- We conducted all simulations and analyses on each time a waterborne case was directly eling the number of pathogens by a differen- a Pentium III PC using MATLAB (Mathworks, prevented, all of the household cases that tial equation led to a time-dependent hazard Natick, Mass). would have resulted from it were also for water-related infections (shown in the on- prevented—resulting in a higher preventable line supplement to this article). RESULTS fraction. Yet, no matter how high the household transmission level is, it cannot result in Simulation Analysis The fraction of disease preventable through self-sustained endemic conditions. In this In this simulation, we chose to use an water treatment varied from a few percentage sense, it does not represent a competing path- event-driven model in which recovery and in- points to more than 75%. This variation was way; in the absence of community transmis- fection events are scheduled for each house- completely explained by 3 transmission path- sion, we found that increased household trans- hold, assuming a simple Poisson process32,33 ways: water contamination (φ), household mission always corresponded to a higher β (details are presented in the online supple- transmission ( h ), and community-level (or preventable fraction. β ment to this article). We ran simulations using between-household) transmission ( c ; Figure 3). Conversely, in the absence of household φ β β all combinations of , c , and h , for a total When water contamination levels were transmission, we discovered that increased of 320 (5 × 8 × 8) parameter sets. To esti- low, the preventable fraction associated with community transmission resulted in an en- mate the efficacy of a water quality interven- water treatment was small. This was true re- tirely different qualitative pattern. In the tion, we examined 2 scenarios: all house- gardless of household or community transmis- case of low levels of household and commu- holds were either fully exposed to sion levels, as depicted by the light contours nity transmission, we found that as commu- contaminated drinking water (A=0) or all in Figure 3a (note that the contour patterns nity transmission began to increase, the pre- households received the water treatment in- shown in Figure 3a ranged from −8.3% to ventable fraction (Figure 3c–e) increased at tervention (A=1). 7.2% and that this range was because of sto- first but eventually decreased. When com- We then defined the fraction of disease chastic variation in the model simulations). In munity transmission levels alone were too preventable through water treatment, known general, as the level of water contamination low to sustain endemicity, each case of wa- − as the preventable fraction, as (IA=1 IA=0)/ increased, so did the preventable fraction, as terborne disease resulted in a finite chain of

IA=1, where IA=1 and IA=0 represent the cu- depicted by the darker contours; however, for cases, and the same amplification mecha- mulative incidence estimates from the scenar- higher levels of contamination, the predicted nism observed for household transmission ios in which (1) there was no intervention preventable fraction was also a function of acted here to increase the preventable frac- and individuals were exposed to contami- transmission levels (Figure 3b–e). tion as well. Eventually, however, commu- β nated water (A = 0) and (2) a water quality When transmission values were high ( c nity transmission alone was able to sustain β intervention was implemented completely and h > 0.06), the preventable fraction was endemic conditions and became an entirely protecting the population from exposure to small regardless of the level of contamination sufficient competing pathway, partially

May 2007, Vol 97, No. 5 | American Journal of Public Health Eisenberg et al. | Peer Reviewed | Research and Practice | 849  RESEARCH AND PRACTICE 

Note. Each contour plot involves a different contamination rate (φ): 0 (a), 0.5 (b), 1.0 (c), 1.5 (d), and 2.0 (e). FIGURE 3—Contours of preventable fractions associated with improving water quality for different rates of household-level and community-level transmission.

replacing water contamination as a source of community-level spread outweighs its amplifi- rehydration therapy has resulted in significant infection and reducing the effectiveness of cation of water-related cases. decreases in mortality, enteric pathogens con- the intervention. A parametric sensitivity analysis suggested tinue to cause a substantial disease burden.3 Finally, we found that when community- that the qualitative features of Figure 3c are Environmental interventions, consisting of level transmission rates were high, increasing robust to varying assumptions of pathogen water quality, sanitation, hygiene, and food- household transmission levels actually re- die-off rates in water (as described in the based interventions, remain as crucial tools in duced the preventable fraction as well, in online supplement to this article). In addition, further decreasing disease burden. contrast to the case of zero to low levels of these features seem to be robust to varying community transmission, in which increasing assumptions of infectivity, duration of infec- Integrative Intervention Strategies household transmission increased the pre- tiousness, and levels of contamination in the Much is known about the natural history of ventable fraction. This occurred because environment. infection transmission and the basic risk fac- household transmission amplified community tors associated with disease, but little is transmission as well as waterborne transmis- DISCUSSION known about how different transmission path- sion. When household transmission increased, ways interact to determine the ultimate effi- the ability of community transmission to sus- Public health policy recommendations are cacy of an intervention. VanDerslice and tain endemic infection levels also increased. urgently needed to provide guidance for de- Briscoe7 provided observational data on the Under these conditions, in which household veloping countries on how to make informed interaction between sanitation and water transmission contributed to a competing decisions regarding the most effective inter- risks; Esrey showed a similar effect using De- sufficient process, Figure 3 suggests that ventions for lowering the incidence of diar- mographic and Health Surveys data,5 as did household transmission’s amplification of rheal diseases. Although promotion of oral Gundry et al.6 in summarizing previously

850 | Research and Practice | Peer Reviewed | Eisenberg et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

published intervention trial results. The range eliminated. To validate this concept, future in- The causal link between infectious individ- of efficacies seen in water intervention trials tervention designs should therefore stratify uals and susceptible individuals, generally illustrates that when sanitation levels are treatment groups according to the intensity of mediated through environmental and social poor, water quality projects may have transmission of alternative pathways. processes, involves a network of contacts. minimal public health effect. For example, a water quality intervention These network contacts include a number of Our model provides a framework for might stratify treatment assignment according important population-level features, such as understanding how to develop and assess to communities with poor versus good sanita- the effect of community sanitation on risk, public health interventions that seek to re- tion. This design would provide important that are missed by an individual analytic ap- duce diarrheal disease. In particular, we information on the degree to which benefits of proach; that is, a system-level perspective is modeled how environmental-, household-, water quality interventions will vary depend- needed when risks manifest themselves and community-level transmissions interact ing on sanitation conditions, a hypothesis put within a causal web of multiple and interde- to determine intervention effectiveness and forth in our model analyses as well as a vari- pendent social, economic, biological, and how this interaction can explain differences ety of other studies.5–7 It should be kept in environmental processes. In the case of en- in the efficacies that have been observed for mind that these model analyses serve only as teric pathogens, these interdependent path- water quality interventions. guides to optimal intervention strategy design. ways are a result of both the ability of the Our analysis quantifies the public health Ultimately, any intervention strategy must be pathogens to exploit a variety of exposure effect that can be obtained by intervening on developed in the context of the perceived and pathways and their contagious properties that 1 transmission pathway, for example, water actual needs of the target population. place susceptible individuals at risk of infec- quality, and shows how this effect depends on In our current model structure, hygiene tion from infectious individuals (Figure 1). the magnitude of other transmission pathways and sanitation are broadly defined as prac- Although the specific model we analyzed such as those associated with sanitation and tices associated with transmission within was designed to examine the presence of a hygiene, a finding consistent with previous households and between households. A more single generic enteric pathogen and to focus studies. Specifically, when community sanita- detailed model structure is needed to inform on a single community, it can be adapted to tion is poor, water quality improvements may specific interventions such as those targeted examine specific pathogens within specific have minimal impact, regardless of the toward hand washing, water storage, or keep- contexts either at a community or a regional amount of water contamination. Under these ing a house clean. The broad definitions used level. The transmission properties we assessed conditions of high community transmission, here, however, illustrate the interdependen- were robust to changing demographic and bi- community-level sanitation must be consid- cies of transmission pathways. ological attributes of the transmission process. ered a necessary intervention and possibly a Increasing transmission can either amplify These changes may alter the quantitative as- sufficient one depending on the level of or attenuate risks, depending on the outcome pects of transmission dynamics, but the quali- water contamination. of interest and on levels of transmission.27,29,34 tative features remain, and the methods we Unfortunately, strikingly few sanitation in- Attenuation occurs when transmission places developed can be extended to assess the effi- tervention studies are available to test the hy- unexposed populations indirectly at risk.34 In cacy of an integrated control and prevention potheses generated in this model analysis.8 our scenario, this takes place through commu- strategy in a particular setting. Additional health impact studies are needed nity transmission, in which unexposed house- to examine the role of sanitation in transmis- holds (those with water treatment) can place Conclusions sion, and future intervention designs need to exposed households (those without water Our analysis contributes to the dialogue on consider the various pathways of transmission treatment) at risk. Thus, community transmis- how to develop optimal integrated control to better identify the set of interventions that sion will attenuate the efficacy of a water strategies by providing insight into the causes are necessary and sufficient to lessen the bur- quality intervention, because indirect cases of the variability observed in the effective- den resulting from diarrheal diseases. may not share the same treatment device as ness of different interventions; for example, their index cases. differing baseline sanitation, hygiene, and Interpretation of Findings Alternatively, amplification of risks occurs water quality conditions may be sufficient to When 1 pathway is critical to maintaining through chains of transmission events that explain observed variations across water diarrheal disease, public health efforts should multiply the effect of an exposure event.27 quality studies. Further simulation studies focus on this critical pathway, whether it in- In our scenario, this takes place primarily can extend the methods we have developed volves improvements in water quality, sanita- through household transmission, in which, for to address more detailed control strategy is- tion, or hygiene. Under conditions in which any given household that is exposed to water sues such as assessing the impact of specific each pathway alone is sufficient to maintain contamination, transmission within the house- sanitation and hygiene interventions in isola- disease at high levels, however, single-path- hold will amplify risk. Amplification of risk tion and in combination with water quality wayinterventions will have minimal benefit, can also occur through community transmis- interventions. We believe that simulation and ultimately an intervention will be success- sion among households with the same treat- methods such as those described here will ful only if all of the sufficient pathways are ment assignment. provide important guidance in efforts to

May 2007, Vol 97, No. 5 | American Journal of Public Health Eisenberg et al. | Peer Reviewed | Research and Practice | 851  RESEARCH AND PRACTICE 

understand how to best reduce and eliminate countries: a systematic review and meta-analysis. 26.Koopman JS, Longini IM Jr, Jacquez JA, et al. As- needless infections and deaths due to diar- Lancet Infect Dis. 2005;5:42–52. sessing risk factors for transmission of infection. Am J Epidemiol. 19 91;133:1199–1209. rheal disease. 9. Juni P, Altman D, Egger M. Systematic reviews in health care: assessing the quality of controlled clinical 27.Koopman JS, Longini IM Jr. The ecological effects trials. BMJ. 2001;323:42–46. of individual exposures and nonlinear disease dynamics in populations. Am J Public Health. 1994;84: 10.Kawata K. Water and other environmental About the Authors 836–842. interventions—the minimum investment concept. Am Joseph N.S. Eisenberg is with the School of Public Health, J Clin Nutr. 1978;31:2114–2123. 28. Eisenberg JN, Lei X, Hubbard AH, Brookhart MA, University of Michigan, Ann Arbor. James C. Scott is with Colford JM Jr. The role of disease transmission and the School of Public Health, University of California, 11. Bern C, Martines J, deZoysa I, Glass RJ. The mag- conferred immunity in outbreaks: analysis of the 1993 Berkeley. Travis Porco is with the California Department of nitude of the global problem of diarrhoeal disease: a Cryptosporidium outbreak in Milwaukee, Wisconsin. Health Services, Richmond, and the Center for Infectious ten year update. Bull World Health Organ. 19 92;70: Am J Epidemiol. 2005;161:62–72. Disease Preparedness, University of California, Berkeley. 705–714. Requests for reprints should be sent to Joseph N.S. 29. Eisenberg JN, Lewis BL, Porco TC, Hubbard AH, 12. Huttly SR, Morris SS, Pisani V. Prevention of diar- Eisenberg, PhD, 611 Church St, Department of Epidemiol- Colford JM Jr. Bias due to secondary transmission in es- rhoea in young children in developing countries. Bull ogy, School of Public Health, University of Michigan, Ann timation of attributable risk from intervention trials. World Health Organ. 19 97;75:163–174. Arbor, MI 48104-3028 (e-mail: [email protected]). Epidemiology. 2003;14:442–450. This article was accepted April 16, 2006. 13. Curtis V, Cairncross S, Yonli R. Domestic hygiene 30. Ball F. Stochastic and deterministic models for SIS and diarrhea—pinpointing the problem. Trop Med Int epidemics among a population partitioned into house- Health. 2000;5:22–32. Contributors holds. Math Biosci. 1999;156:41–67. 14 . Kaltenthaler EC, Drasar BS. Understanding of J.N.S. Eisenberg originated and supervised the study 31. Becker NG, Dietz K. The effect of household dis- hygiene behaviour and diarrhoea in two villages in and led the writing. J.C. Scott completed the analysis. tribution on transmission and control of highly infec- Botswana. J Diarrhoeal Dis Res. 1996;14:75–80. T. Porco assisted in model development and analysis. tious diseases. Math Biosci. 19 95;127:207–219. 15. Huttly SR, Lanata CF. Feces, flies, and fetor: find- 32. Porco TC, Small PM, Blower SM. Amplification ings from a Peruvian shantytown. Rev Panam Salud Acknowledgments dynamics: predicting the effect of HIV on tuberculosis Publica. 19 9 8;4:75–79. This work was funded by the National Institute of Al- outbreaks. J Acquir Immune Defic Syndr. 2001;28: lergy and Infectious Diseases (grant RO1-AI050038). 16. Bukenya GB, Nwokolo N. Transient risk factors 405–498. Note. The funding agency had no role in the design for acute childhood diarrhoea in an urban community 33. Bratley P, Fox BL, Schrage LE. A Guide to Simula- of this study. of Papua New Guinea. Trans R Soc Trop Med Hyg. tion. 2nd ed. New York, NY: Springer-Verlag; 1987. 1990;84:857–860. 34.Halloran ME, Struchiner CJ. Causal inference in 17. Han AM, Moe K. Household feacal contamination Human Participant Protection infectious diseases. Epidemiology. 19 95;6:142–151. No protocol approval was needed for this study. and diarrhoea risk. J Trop Med Hyg. 1990;93: We thank Jamie Bartram for initial conversations 333–336. that stimulated this project and James Koopman for 18.Yeager BA, Huttly SR, Bartolini R, Rojas M, valuable comments on the article. Lanata CF. Defecation practices of young children in a Peruvian shanty town. Soc Sci Med. 1999;49:531–541. References 19. Simango C, Dindiwe J, Rukure G. Bacterial con- 1. Meeting the MDG Drinking Water and Sanitation tamination of food and household stored drinking Target: A Mid-Term Assessment of Progress. Geneva, water in a farmworker community in Zimbabwe. Centr Switzerland: World Health Organization and United Afr J Med. 19 92;38:143–149. Nations Children’s Fund; 2004. 20. Sempértegui F, Estrellá B, Egas J, et al. Risk of di- 2. Black RE, Morris SS, Bryce J. Where and why are arrheal disease in Ecuadorian day-care centers. Pediatr 10 million children dying every year? Lancet. 2003; Infect Dis J. 19 95;14:606–612. 361:2226–2234. 21. Mock NB, Sellers TA, Abdoh AA, Franklin RR. 3. Pruss A, Kay D, Fewtrell L, Bartram J. Estimating Socioeconomic, environmental, demographic and be- the burden of disease from water, sanitation, and hy- havioral factors associated with occurrence of diarrhea giene at a global level. Environ Health Perspect. 2002; in young children in the Republic of Congo. Soc Sci 110:537–542. Med. 19 93;36:807–816. 4. Kosek M, Bern C, Guerrant RL. The global bur- 22. Yeager BA, Lanata CF, Lazo F, Verastegui H, den of diarrhoeal disease, as estimated from studies Black RE. Transmission factors and socioeconomic sta- published between 1992 and 2000. Bull World Health tus as determinants of diarrhoeal incidence in Lima, Organ. 2003;81:197–204. Peru. J Diarrhoeal Dis Res. 19 91;9:186–193. 5. Esrey SA. Water, waste, and well-being: a multi- 23. Stanton BF, Clemens JD, Clements JD. Soiled country study. Am J Epidemiol. 1996;143:608–623. saris: a vector of disease transmission? Trans R Soc 6. Gundry S, Wright J, Conroy R. A systematic re- Trop Med Hyg. 1986;80:485–488. view of the health outcomes related to household 24. Cairncross S, Blumenthal U, Kolsky P, Moraes L, water quality in developing countries. J Water Health. Tayeh A. The public and domestic domains in the 2004;2:1–13. transmission of disease. Trop Med Int Health. 19 9 6;1: 7. VanDerslice J, Briscoe J. Environmental interven- 27–34. tions in developing countries: interactions and their im- 25. Esrey SA, Potash JB, Roberts L, Shiff C. Effects of plications. Am J Epidemiol. 19 95;141:135–144. improved water supply and sanitation on ascariasis, di- 8. Fewtrell L, Kaufmann RB, Kay D, Enanoria W, arrhoea, dracunculiasis, hookworm infection, schistoso- Haller L, Colford JM Jr. Water, sanitation, and hygiene miasis, and trachoma. Bull World Health Organ. 19 91; interventions to reduce diarrhoea in less developed 69:609–621.

852 | Research and Practice | Peer Reviewed | Eisenberg et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

A Comparative Health Survey of the Inhabitants of Roma Settlements in Hungary

| Zsigmond Kósa, MD, PhD, György Széles, MD, PhD, László Kardos, MD, PhD, Karolina Kósa, MD, PhD, Renáta Németh, MSc, Sándor Országh, MSc, Gabriella Fésüs, MSc, Martin McKee, MD, Róza Ádány, MD, PhD, and Zoltán Vokó, MD, PhD

The processes leading to enlargement of the Objectives. We compared the health of people living in Roma settlements with European Union in May 2004 focused un- that of the general population in Hungary. precedented attention on the plight of the Methods. We performed comparative health interview surveys in 2003 to 2004 in rep- Roma, or Gypsy, people in Central and East- resentative samples of the Hungarian population and inhabitants of Roma settlements. 1,2 ern Europe. The Roma, a people who Results. In persons older than 44 years, 10% more of those living in Roma set- moved from northern India into Europe be- tlements reported their health as bad or very bad than did those in the lowest in- tween the 9th and 14th centuries, number come quartile of the general population. Of those who used any health services, between 5 and 10 million people and are the 35% of the Roma inhabitants and 4.4% of the general population experienced European Union’s largest minority group. some discrimination. In Roma settlements, the proportion of persons who thought Within the European Union, most of the that they could do much for their own health was 13% to 15% lower, and heavy smoking and unhealthy diet were 1.5 to 3 times more prevalent, than in the low- Roma population lives in the new member est income quartile of the general population. states—in particular, Romania, Bulgaria, Hun- Conclusions. People living in Roma settlements experience severe social exclusion, gary, and Slovakia.3 The number of Roma liv- which profoundly affects their health. Besides tackling the socioeconomic roots of ing in the United States was estimated to be the poor health of Roma people, specific public health interventions, including health between 200000 and 500000 in the early education and health promotion programs, are needed. (Am J Public Health. 2007; 4 1970s. The first group of Roma to migrate to 97:853–859. doi:10.2105/AJPH.2005.072173) the United States was transported as slaves.5 Researchers have documented in detail the strictly regulated in most European countries. have faced earlier researchers. First, we used poor conditions in which the Roma people live, Fear of legal challenges and a misunderstand- an innovative approach to identify the section the discrimination they face, and the problems ing by researchers of data protection laws in of the Roma population that is most vulner- they confront when trying to access services.6 many countries have resulted in a dearth of able. Second, we incorporated an explicit com- Numerous studies have shown that the Roma reliable statistical data on the number of parison with the majority population. Third, people have high levels of many diseases,7,8 Roma people in Europe. the design, conduct, and interpretation of the but remarkably little systematic research has Another problem is that although many study were fully participative, involving repre- been done on how the health of this popula- studies have documented poor health among sentatives of the Roma population at all stages. tion compares with the majority populations the Roma people, few studies have compared in the countries in which they live.9 their health with that of the majority popula- METHODS Researchers also have considerable diffi- tion. This lack of comparison is an important culty in defining the Roma population. It is omission given the generally poor health in The data used in this study were obtained characterized by great diversity in language Central and Eastern Europe, characterized by from 2 surveys that were designed to be com- and dialect, culture, religion, and social high premature mortality (generally defined parable and were conducted only 6 months class.10 Some Roma people have assimilated as death before age 65 years) mainly as a re- apart. The first survey—the National Health and intermarried with the majority popula- sult of cardiovascular diseases, cancer, Interview Survey—focused on the general tion, although many still live apart from the chronic liver diseases, and accidents. Al- Hungarian adult population. The second was majority population. though life expectancy has been increasing a specific survey of the adult population living The cultural inaccessibility of the Roma for a decade in Hungary, at the time of the in Roma settlements in 3 counties of north- population poses difficulties in research on study, of the current 25 European Union eastern Hungary, the part of the country with their health. The Roma’s strong sense of “oth- member states, only Estonia, Latvia, and the highest Roma population. erness” plays an important role in the inacces- Lithuania had lower figures than Hungary, in sibility.11–13 The widespread fear among ethnic which male life expectancy was 68.4 years in National Health Interview Survey 2003 minorities in Europe that—regardless of their 2003. Hungarian women had the second The National Health Interview Survey, legal status—ethnic statistics will be misused lowest figure of 76.8 years in 2003. 2003, was designed to capture detailed infor- adds to the difficulties.14 For this reason, the We have performed a study that was de- mation on the self-reported health status of gathering of ethnic identity during research is signed to overcome some of the difficulties that the noninstitutionalized Hungarian adult

May 2007, Vol 97, No. 5 | American Journal of Public Health Kósa et al. | Peer Reviewed | Research and Practice | 853  RESEARCH AND PRACTICE 

population, as well as the main behavioral settlement.18 All adults in the households per week or more than 5 standard drinks on and socioeconomic determinants of health.15 selected were interviewed by the interviewers, any day for men), moderate drinking (weekly The study population was randomly se- who were all Roma people who had under- consumption but less than heavy drinking), lected via 2-stage sampling from the Central gone training in survey methods. Interviews occasional drinking (consumption less often Data Processing, Registration and Election Of- were performed from May to August 2004. than weekly), and abstinence.21,22 The instru- fice’s registry. In the first stage, communities ment recommended by the European Health (cities, towns, and villages) were stratified by Questionnaire Interview Surveys (EUROHIS) project of county and by community size. Within coun- The questionnaires used in the 2 surveys WHO for assessing smoking in population sur- ties, communities were chosen with a sam- were almost identical, allowing direct compar- veys was used.20 Social support was measured pling probability proportional to size. Individ- ison of the results.19 using an instrument developed in the Health uals were then selected at random from the The analysis presented here used data on and Lifestyles Survey of England. It consists of selected communities. To maximize the cost- general self-reported health status, including 7 questions about physical and emotional as- effectiveness of the fieldwork, at least 10 indi- functionality and self-perceived health, use pects of social support. These questions com- viduals were chosen from each community of health care services, health behavior (be- bine into a single scale categorizing inform- selected. liefs, perceptions, habits, actions, and so on, ants as having “severe lack,” “some lack,” or In the fall of 2003, fieldworkers with expe- related to health), and socioeconomic status. “no lack” of social support. In this analysis, we rience in interview techniques who had re- We used internationally recommended sur- combined the categories of “some lack” and ceived additional training in health survey vey tools when possible. Most questions were “severe lack.”23 techniques performed the interviews. asked by the interviewer; some sensitive subjects such as alcohol consumption, social sup- Data Analysis Roma Health Survey port, and discrimination were included in a We estimated the prevalence of the key var- The poor living conditions in which some self-administered section of the questionnaire. iables in the 2 target populations. We applied Roma people live, frequently on the outskirts Self-reported health was assessed with a weights to correct for the unequal probability of towns and villages and in substandard standard 5-item question recommended by of selection and for nonresponse, as well as to accommodations, allow relatively straight- the World Health Organization (WHO) to perform poststratification by age, gender, and forward identification of locations in which measure perceived health.20 For this analy- community size in the national survey. We Roma people are concentrated. This study sis, we combined the categories “good” and used the survey analysis module in the statisti- took advantage of this opportunity. Between “very good” as well as “bad” and “very bad.” cal program Stata 6.0 (Stata Corp, College Sta- 2001 and 2003, a detailed environmental Functional limitation was assessed with the tion, Tex) to calculate the 95% confidence in- survey was undertaken in 3 counties in which following question: “Do you have any com- tervals (CIs) of the prevalence estimates after the Roma population is greatest; the research- plaints, injuries, or diseases that limit your taking into account the sampling design. ers identified all such settlements (in which everyday activities, such as working, shop- Mean household equivalent monthly in- the population was almost exclusively Roma) ping, managing your life, playing sports, or come was calculated as the mean total house- and recorded the number of people living in keeping contact with other people?” Height hold income per month divided by the square them.16 Although the size of the overall Roma and weight were self-reported, and body root of the number of persons in the house- population is uncertain, it is estimated that mass index (BMI) was calculated as body hold. Households were divided into quartiles about 6% to 10% of the Roma people live in weight (kg) divided by height in meters based on their mean total household income such settlements.16 ,17 squared (m2). BMI was categorized as per month (ranges in Euros from lowest to In the environmental survey, settlements abnormally thin (BMI < 18.5 kg/m2), normal highest: <202, 203–283, 284–377, >377). with at least 4 households were mapped, and (BMI = 18.5–24.99 kg/m2), overweight We also estimated the prevalence of key vari- the resulting plot was used as the basis for (BMI = 25–29.99 kg/m2), or obese ables for the lowest equivalent monthly in- the health survey reported here. Collectively, (BMI ≥ 30 kg/m2), in accordance with the come quartile in the general population. approximately 62000 persons lived in the WHO guidelines. All analyses were stratified by age and gen- Roma settlements in these counties (of a total Data on cigarette smoking, alcohol con- der. The age categories were defined as 18 to population of the counties of 1877243). The sumption, and social support were derived 29, 30 to 44, and 45 to 64 years. survey sought to capture representative data from answers given to multiple-question mod- on 1000 persons living in these settlements ules. For alcohol, questions focused on fre- RESULTS who were 18 years or older. quency and quantity. Consumption was cate- We used a 2-stage sampling process: towns gorized as heavy drinking (more than 7 Of the planned 7000 interviews, 5072 and villages in the 3 counties with identified standard drinks—a standard drink was defined were completed in the National Health Inter- settlements were selected randomly, and as the equivalent of 12 g of pure ethanol—per view Survey. Of the participants selected, then households were selected with the ran- week or more than 3 standard drinks on any 15% could not be located during the period dom walk method, based on a map of the day for women; more than 14 standard drinks of fieldwork, 8% refused to participate, and

854 | Research and Practice | Peer Reviewed | Kósa et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 1—Major Characteristics of the Socioeconomic Status Among Persons Living in Roma Settlements and in the General Population in Hungary: National Health Interview Survey 2003 and Roma Health Survey 2004

People Living in Roma Settlements People in Lowest Income Quartile General Population Characteristics 18–29 y 30–44 y 45–64 y 18–29 y 30–44 y 45–64 y 18–29 y 30–44 y 45–64 y

Only primary education (8 years), % 73 (68, 77) 80 (75, 84) 87 (82, 91) 43 (36, 50) 44 (38, 50) 50 (45, 56) 16 (14, 19) 17 (15, 20) 29 (26, 32) Actively employed, % 17 (14, 21) 22 (18, 27) 11 (7, 16) 18 (12, 27) 40 (32, 49) 24 (19, 30) 62 (59, 65) 79 (76, 81) 53 (51, 56) Mean household equivalent income 170 (160, 180) 161 (153, 169) 159 (150, 168) 143 (138, 148) 146 (142, 151) 150 (146, 154) 332 (316, 348) 307 (293, 321) 300 (290, 309) per month, Euro Perceived financial status Very bad, % 20 (16, 24) 18 (14, 23) 26 (20, 32) 9 (5, 14) 14 (10, 18) 16 (13, 20) 2 (1.5, 3) 4 (3, 5) 5 (4, 7) Bad, % 40 (35, 45) 43 (38, 49) 39 (33, 46) 31 (25, 38) 34 (29, 40) 36 (31, 40) 13 (11, 16) 16 (14, 18) 20 (18, 22) Living in a 1-room apartment, % 11 (8, 14) 12 (9, 16) 15 (10, 20) 17 (13, 23) 15 (11, 21) 15 (12, 19) 9 (7, 11) 8 (6, 10) 6 (5, 8) Lack of social support, % 24 (20, 28) 27 (22, 32) 24 (19, 30) 16 (11, 22) 20 (15, 25) 23 (19, 28) 11 (9, 13) 15 (13,18) 14 (12, 15)

Note. Numbers are estimated proportions (%) or means in the populations (95% confidence intervals).

4% were unable to participate for other Similarly, at age 30 years and beyond, the 6.7% (95% CI=5.0%, 9.0%) in the lowest reasons. Data from the 4121 persons younger prevalence of functional limitation was higher income quartile of the general population. Of than 65 years were included in the analysis. among women living in settlements than in the the Roma persons who reported discrimina- In the Roma Health Survey, of the 1000 general population because of the very high tion, 69% (95% CI = 62%, 75%) attributed attempted interviews, 969 interviews were frequency of severe functional limitation (Fig- it to their ethnicity or skin color and 18% completed successfully, 12 persons refused to ure 1). The result was similar in the lowest in- (95% CI=13%, 24%) attributed it to their participate, and 19 interviews were incom- come quartile of the general population. social status. The corresponding figures were plete (96.9% response rate). Data from the Among men, practically no difference between 6.0% (95% CI=3.3%, 11%) and 5.0% (95% 936 persons younger than 65 years were in- the Roma and the general population was seen CI=0.7%, 9.1%) in the general population. cluded in the analysis. in the prevalence of any limitation; however, A large difference was seen between the As expected, the participants in the Roma the prevalence of severe functional limitation in Roma population and the general population survey had less education, were less likely to be men aged 30 years or older was highest in the proportion of subjects who thought that employed, had much lower income, worse liv- among the Roma people and among those in they could do much or very much to promote ing conditions, and weaker social support com- the general population with the lowest income. their own health. In the general population, pared with the general population (Table 1). Roma persons were less likely to use health the proportions were 88% (95% CI=86%, The mean household equivalent monthly in- services than was the general population 90%), 80% (95% CI=77%, 82%), and 66% come of Roma people was somewhat higher (Table 2). The difference was especially (95% CI=64%, 69%) in the age groups of than that of people in the lowest income marked in the proportion of persons who 18 to 29, 30 to 44, and 45 to 64 years, re- quartile of the general population. consulted a specialist and in the proportion of spectively, whereas the corresponding figures The self-reported health status of the peo- those who had dental service in the previous were 68% (95% CI=64%, 73%), 53% (95% ple living in Roma settlements was much 12 months. No significant difference was CI=48%, 59%), and 39% (95% CI=33%, worse than the self-reported health status of found in the proportion of persons who had a 46%) in the Roma population and 73% (95% the general population. Of those living in hospital stay in the previous 12 months. De- CI=66%, 79%), 66% (95% CI=60%, 72%), Roma settlements, substantially fewer people spite the existence of a universal screening and 53% (95% CI=48%, 58%) in the lowest reported their health as good or very good, program in Hungary, only 25% of the Roma income quartile of the general population. and many more reported their health as bad women aged 45 to 64 years had undergone Table 3 shows that the prevalence of smok- or very bad—at age 30 to 44 years, 18% mammography within the previous 2 years. ing more than 20 cigarettes per day was 2 to (95% CI=14%, 23%), and at age 45 to 64 Use of health services by the Roma popula- 5 times higher among the Roma population years, 50% (95% CI=44%, 57%), compared tion was similar to use by those in the lowest than in the general population. The preva- with 8% (95% CI=7%, 10%) and 25% income quartile of the general population. lence of smoking was considerably higher (95% CI=23%, 27%), respectively, in the Some kind of discrimination related to among the Roma people older than 30 years general population. In the lowest income health service use was reported by 35% than in the lowest income quartile of the gen- quartile of the general population, these esti- (95% CI=33%, 37%) of the Roma persons eral population. Roma persons were younger mates were 20% (95% CI=15%, 25%) and and 4.4% (95% CI=3.7%, 5.1%) of the gen- than the general population when they started 40% (95% CI=35%, 45%), respectively. eral population. The corresponding figure was smoking, with a mean age at initiation of 16.1

May 2007, Vol 97, No. 5 | American Journal of Public Health Kósa et al. | Peer Reviewed | Research and Practice | 855  RESEARCH AND PRACTICE 

Note. Functional limitation was assessed by the following question: “Do you have any complaints, injuries, or diseases that limit your everyday activities, such as working, shopping, managing your life, playing sports, or keeping contact with other people?” Numbers are estimated proportions (%) in the populations. FIGURE 1—Prevalence of functional limitation in men and women in the Roma settlements and the general population: National Health Interview Survey 2003 and Roma Health Survey 2004

years (95% CI=15.8, 16.3), whereas the in Roma women in all age groups. A stark The problem of ill health among the Roma corresponding figure was 18.3 years (95% contrast was noted between the Roma popula- populations has become especially pressing CI=18.1, 18.6) in the general population. tion and the general population in their diet. because of the priority now being given to im- No large differences were detected in the The proportion of persons who generally used proving the situation of the Roma population overall prevalence of moderate and heavy vegetable oil to cook with and ate fresh fruits in this region, exemplified by the attention drinking between Roma people and the gen- and vegetables daily was much higher in the given to this issue in the negotiations leading eral population, although the pattern varied general population, even compared with those up to Hungary’s European Union accession; by age, with a somewhat higher prevalence of in the lowest income quartile (Table 3). the establishment of the Decade of Roma In- heavy drinking among the Roma men aged clusion, an initiative involving the World Bank, 18 to 29 years and a lower prevalence DISCUSSION European Union, and Open Society Institute among these men at age 30 years and older along with many Roma nongovernmental or- (Table 3). Roma men in all age groups had a Any attempt to assess the health of the Roma ganizations; and related initiatives by organi- higher prevalence of abstinence (Table 3) people in Central Europe faces the fundamental zations such as the United Nations Develop- than did all of the general population. problem noted in the introduction of how to de- ment Programme. Specific programs aiming to The distribution of body weight was fine the population. Yet this difficulty must not improve the health of Roma people exist in broadly similar in the 2 populations, except be used as an excuse to avoid the attempt to Hungary. The latest governmental decree on that obesity tended to be slightly less frequent quantify the burden of ill health that they face. integrating the Roma of Hungary prescribed

856 | Research and Practice | Peer Reviewed | Kósa et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 2—Health Service Use in the Previous 12 Months Among Persons Living in Roma younger than in the National Health Inter- Settlements and in the General Population in Hungary: National Health Interview Survey view Survey, it would have been desirable to 2003 and Roma Health Survey 2004 have stratified the sample with narrow age categories to avoid confounding by age. How- People Living in Roma Settlements General Population ever, the relatively small number of persons 18–29 y 30–44 y 45–64 y 18–29 y 30–44 y 45–64 y involved in the Roma Health Survey pre- Estimated Estimated Estimated Estimated Estimated Estimated cluded doing so, and it is very unlikely that Proportion, Proportion, Proportion, Proportion, Proportion, Proportion, Health Service Use % (95% CI) % (95% CI) % (95% CI) % (95% CI) % (95% CI) % (95% CI) residual confounding by age within the age categories used could have affected our re- Women sults substantially. In addition, although per- Use of any services 65 (58, 71) 62 (55, 70) 78 (69, 85) 74 (70, 77) 73 (69, 76) 80 (78, 83) sons older than 64 were included in the sur- Contact with family physician 70 (64, 76) 71 (63, 77) 86 (78, 91) 62 (59, 65) 62 (59, 65) 77 (75, 79) veys, the very low numbers in the Roma a a a a Consulting a specialist 51 (44, 58) 42 (35, 50) 58 (49, 67) 68 (64, 72) 68 (64, 71) 69 (66, 72) Health Survey precluded their inclusion in a a a a Dental service 44 (38, 51) 33 (26, 41) 32 (24, 41) 61 (57, 65) 50 (46, 54) 39 (36, 43) the analysis. Finally, as in any survey based b Inpatient service 11 (8, 16) 14 (9, 20) 17 (11, 25) 16 (14, 20) 13 (11, 16) 18 (15, 20) on self-report, the data on alcohol consump- a a a a Gynecologist appointment in 90 (85, 93) 82 (75, 87) 62 (53, 71) 90 (87, 92) 92 (89, 94) 86 (84, 87) tion must be interpreted with caution. the previous 5 y The study of those living in the Roma set- Mammography in the 8 (5, 13) 15 (10, 21) 25 (18, 34) 9 (6, 11) 24 (20, 27) 70 (67, 74)a tlements did, however, have some important previous 2 y strengths. The high response rate—facilitated Men by the employment of Roma fieldworkers and Use of any services 42 (35, 49)a 48 (40, 56)a 67 (57, 75) 60 (56, 64)a 61 (57, 65)a 72 (68, 75) the participation of community leaders at all Contact with family physician 48 (41, 55)a 51 (43, 59)a 74 (65, 81) 67 (63, 71)a 68 (64, 71)a 73 (71, 75) stages in the survey—coupled with the sam- Consulting a specialist 22 (16, 28)a 23 (17, 31) 42 (34, 52) 38 (33, 42)a 35 (31, 38) 51 (47, 54) pling method used suggest that the findings Dental service 38 (31, 45) 26 (20, 34)a 21 (14, 29) 42 (37, 46) 40 (36, 44)a 30 (27, 34) are likely to be representative of those living Inpatient servicea 3 (1, 7) 10% (6, 16) 23 (16, 31) 6 (4, 8) 8 (6, 10) 17 (15, 20) in such settlements. The use of identical ques- Note. CI=confidence interval. tionnaires in both surveys ensured compara- a The 95% CIs of the estimates in the Roma population and in the general population do not overlap. bility; pretesting did not give any cause to be- bSpent at least 1 night in a hospital. lieve that questions would be interpreted differently by the 2 populations. several tasks to be implemented for improving The current survey of Roma people had 1 Our comparison determined that people liv- the quality of life of Roma people, such as obvious limitation: it was not representative ing in the Roma settlements have much poorer eliminating or remodeling settlements. In the of the overall Hungarian Roma population. By health than does the general population. At framework of Hungary’s National Public design, it excluded those Roma who have, to ages 45 to 64, their self-reported health status Health Programme, funding has been dedi- various degrees, assimilated with the majority was even worse than in the lowest income cated to research into the health and social population. However, because many people quartile of the general population. The problems of the Roma population as well as to are unwilling to self-define their ethnicity as lifestyles of the people living in Roma settle- various training activities to improve the atti- Roma, this constraint will be very difficult to ments are also less conducive to future health, tude of health and social care workers toward overcome. An inevitable consequence was as measured by rates of smoking and the low this minority. Furthermore, many nongovern- that the Roma Health Survey captured the consumption of fruits and vegetables, even mental organizations work with Roma com- characteristics of the most disadvantaged sec- when compared with the lowest income quar- munities in Hungary, most of them focusing tion of the Roma population. However, the tile of the general population. In addition to on health education. needs of this group are the most important to their high burden of ill health, they face barri- Although previous surveys of the health understand from a policy perspective. ers in accessing health services, particularly and living conditions of the Roma population It is also important to note that the Na- preventive interventions and specialist care. have been done—most notably, a series un- tional Health Interview Survey will have in- This finding is consistent with extensive evi- dertaken by the United Nations Development cluded some people who are Roma. However, dence of such barriers to care reported in Programme24—ours is the first study designed the difficulty with self-defined ethnicity is that qualitative studies, in part reflecting poorer ac- explicitly to compare the health of the Roma it is not possible to exclude them from the cess because Roma settlements are often un- population with that of their majority neigh- sample for the purposes of analysis. It is, how- derserved by essential services and, where bors. The United Nations Development Pro- ever, possible that their inclusion will dilute they exist, are often difficult to staff. However, gramme survey examined living conditions, the true difference between the populations. a further important factor is the high fre- beliefs, and attitudes but few that were di- A further limitation was that because the quency with which Roma respondents de- rectly related to health. age distribution of the Roma population was scribed experiencing direct discrimination

May 2007, Vol 97, No. 5 | American Journal of Public Health Kósa et al. | Peer Reviewed | Research and Practice | 857  RESEARCH AND PRACTICE 

TABLE 3—Prevalence of Health Determinants Among Persons Living in Roma Settlements and in the General Population in Hungary: National Health Interview Survey 2003 and Roma Health Survey 2004

People Living in Roma Settlements People in Lowest Income Quartile General Population 18–29 y 30–44 y 45–64 y 18–29 y 30–44 y 45–64 y 18–29 y 30–44 y 45–64 y Estimated Estimated Estimated Estimated Estimated Estimated Estimated Estimated Estimated Proportion, Proportion, Proportion, Proportion, Proportion, Proportion, Proportion, Proportion, Proportion, Health Determinants % (95% CI) % (95% CI) % (95% CI) % (95% CI) % (95% CI) % (95% CI) % (95% CI) % (95% CI) % (95% CI)

Women Cigarette smoking Never 40 (34, 47) 22 (17, 29) 32 (24, 41)a 43 (34, 52) 36 (28, 45) 55 (48, 61)a 53 (49, 57) 47 (42, 51) 56 (53, 59) Former 9 (6, 14) 12 (8, 18) 13 (8, 20) 7 (3, 13) 8 (5, 14) 13 (9, 19) 12 (9, 15) 12 (9, 14) 16 (14, 19) Moderateb 26 (20, 32) 17 (12, 23) 12 (7, 19) 26 (19, 34) 31 (24, 40) 19 (14, 24) 24 (21, 27) 24 (21, 28) 19 (17, 22) Heavyc 25 (20, 32) 49 (42, 57)a 44 (35, 54)a 25 (18, 32) 25 (18, 32)a 14 (10, 19)a 12 (10, 15) 18 (15, 21) 9 (7, 10) Alcohol consumption Abstinent 83 (77, 88) 82 (76, 88)a 91 (84, 95)a 81 (73, 88) 66 (58, 74)a 71 (65, 77)a 57 (53, 61) 55 (51, 59) 61 (57, 65) Occasional 14 (10, 20) 13 (8, 19)a 7 (4, 14) 17 (11, 26) 28 (21, 36)a 19 (14, 25) 36 (32, 40) 35 (31, 39) 26 (23, 29) Moderate 2 (1, 5) 3 (1, 7) 1 (0, 6) 2 (0, 6) 4 (2, 9) 6 (4, 11) 4 (3, 7) 8 (6, 10) 11 (9, 13) Heavy 1 (0, 4) 2 (1, 6) 1 (0, 6) 0 1 (0, 5) 3 (2, 7) 3 (2, 5) 2 (1, 4) 2 (2, 4) Body mass indexd Abnormally thin 10 (7, 15) 8 (5, 14) 4 (2, 10) 12 (7, 20) 5 (2, 10) 3 (1, 6) 13 (10, 16) 4 (3, 6) 3 (2, 4) Normal 71 (65, 77) 52 (44, 60) 41 (32, 51) 64 (55, 72) 49 (41, 57) 33 (27, 39) 68 (64, 71) 57 (53, 61) 35 (32, 38) Overweight 14 (10, 20) 28 (22, 36) 34 (26, 44) 16 (11, 24) 25 (18, 33) 37 (31, 44) 14 (12, 17) 25 (22, 29) 38 (35, 41) Obese 4 (2, 8) 11 (7, 17)a 21 (15, 30) 8 (4, 14) 21 (15, 29)a 27 (22, 34) 5 (4, 8) 14 (12, 17) 25 (22, 28) Diet Use of vegetable oil 26 (21, 33)a 30 (23, 37)a 28 (21, 37)a 54 (44, 63)a 48 (39, 56)a 48 (41, 54)a 70 (66, 73) 63 (59, 67) 61 (57, 64) Consumption of fruits and vegetables daily 32 (26, 39) 36 (29, 44)a 36 (27, 45)a 45 (37, 53) 55 (46, 63)a 69 (62, 75)a 55 (50, 59) 66 (63, 70) 77 (74, 79) Consumption of fruits and vegetables weekly 49 (43, 56) 45 (38, 53) 42 (33, 51)a 43 (35, 52) 37 (29, 45) 21 (16, 27)a 40 (36, 44) 29 (26, 33) 18 (16, 21) Consumption of fruits and vegetables

Note. CI=confidence interval. aThe 95% CIs of the estimates in the Roma population and in the lowest equivalent income quartile of the general population do not overlap. bOccasional or daily but fewer than 20 cigarettes per day. cAt least 20 cigarettes per day. dBMI was calculated as body weight (kg) divided by height in meters squared (m2). BMI was categorized as abnormally thin (BMI<18.5 kg/m2), normal (BMI=18.5–24.99 kg/m2), overweight (BMI=25–29.99 kg/m2), or obese (BMI≥30 kg/m2). eAlcohol consumption was categorized as heavy drinking (more than 7 standard drinks—a standard drink was defined as the equivalent of 12 g of pure ethanol—per week or more than 3 standard drinks on any day for women; more than 14 standard drinks per week or more than 5 standard drinks on any day for men), moderate drinking (weekly consumption but less than heavy drinking), occasional drinking (consumption less often than weekly), and abstinence.

858 | Research and Practice | Peer Reviewed | Kósa et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

when seeking care, again consistent with many Requests for reprints should be sent to Zoltán Vokó, 5. Hancock I. The Pariah Syndrome: An Account of qualitative studies and accounts in the media. MD, PhD, Department of Preventive Medicine, Faculty of Gypsy Slavery and Persecution. Ann Arbor, Mich: Public Health, Medical and Health Science Centre, Uni- Karoma; 1975. Although this analysis has provided a basic versity of Debrecen, H-4028 Debrecen, Kassai út 26/b, 6. Zoon I. On the Margins: Roma and Public Services description of how the health of the population Hungary (e-mail: [email protected]). in Romania, Bulgaria, and Macedonia. New York, NY: living in Roma settlements in Hungary differs This article was accepted October 14, 2005. Open Society Institute; 2001. from that of the majority population, it is of 7. Koupilová I, Epstein H, Holcík J, Hajioff S, McKee M. Health needs of the Roma population in the Czech and course only a first step in understanding the Contributors Z. Kósa and R. Ádány had the original idea for the com- Slovak Republics. Soc Sci Med. 2001;53:1191–1204. substantial inequality in health between the parative Roma Health Survey. Z. Kósa participated in 8. Kósa K, Lénárt B, Ádány R. Health status of the Roma and the majority population. Further the questionnaire and sampling design and wrote the ar- Roma population in Hungary [in Hungarian]. Orv Hetil. analyses are beyond the scope of this initial arti- ticle. G. Széles was involved in the questionnaire devel- 2002;143:2419–2426. opment, designed and performed the sampling, and in- cle, but the simple comparison of health, health 9. Hajioff S, McKee M. The health of the Roma peo- terpreted the results. L. Kardos performed the analysis. ple: a review of the published literature. J Epidemiol behavior, and health service use of Roma peo- K. Kósa was responsible for the environmental health Community Health. 2000;54:864–869. survey of the settlements that served as a basis for the ple with that of persons in comparable socio- 10. Crowe DM. A History of the Gypsies of Eastern Roma Health Survey and wrote the article. R. Németh Europe and Russia. London, England: I.B. Tauris Pub- economic conditions in the general population was involved in the questionnaire development, de- lishers; 1995. identified specific Roma disadvantages in health signed and performed the sampling, and performed the 11.Fonseca I. Bury Me Standing—The Gypsies and behavior and ethnic discrimination. analysis. S. Országh was involved in the questionnaire development, managed the data, and did the program- Their Journey. New York, NY: Vintage; 1996. It is also relevant to report on some find- ming. G. Fésüs interpreted the results. M. McKee inter- 12. Ladányi J, Szelényi I. Who is gypsy? [in Hungarian] ings of the environmental survey that served preted the results and wrote the article. R. Ádány de- Kritika. 19 97;(Dec):3–6. as a basis for identifying the Roma settle- signed and performed the sampling and interpreted the 13. Ladányi J, Szelényi I. On the objectivity of ethnic results. Z. Vokó was involved in the questionnaire devel- ments. These settlements were often charac- classification [in Hungarian]. Kritika. 1998;(March):33–35. opment, interpreted the results, and wrote the article. 14 . Krizsán A, ed. Ethnic Monitoring and Data Protec- terized by illegal garbage deposits and an tion: The European Context. Budapest, Hungary: INDOK absence of drainage, gas mains, and paved Acknowledgments (Human Rights Information and Documentation Cen- roads. Some settlements had no electricity or The National Health Interview Survey, 2003, was ter); 2001. water mains. Many settlements are built on funded by the Hungarian Ministry of Health in the 15. National Health Interview Survey 2003, Hungary— framework of the National Public Health Programme. Executive Update. Budapest, Hungary: Johan Béla Na- ground that becomes waterlogged after rain- The Roma Health Survey was funded from Hungarian tional Center for Epidemiology; 2004. Available at: fall. These settlements pose substantial health research grants from the National Research and Devel- http://www.oek.hu/oekfile.pl?fid=533. Accessed Feb- hazards to their inhabitants.16 opment Program (NKFP-1B/0013/2002), the National ruary 8, 2007. Scientific Council on Health (ETT: 445/2003), and the In summary, our study provided strong 16. Environmental Health Hazards of Ethnic Minorities Ministry of Environment (KvH-96/167/2000). Living in Colonies [in Hungarian]. Debrecen, Hungary: quantitative evidence of the poor health of a The contribution of fieldworkers and other study School of Public Health, University of Debrecen; 2001. section of the Roma people and highlighted personnel is gratefully acknowledged, as is the support Project Report (KKF/1645/2000) to the Ministry of of the Ministry of Health, Ministry of Environment, and Environment of the Republic of Hungary. the need to develop appropriate multisectoral Ministry of Education of Hungary. 17.Kemény I, Janky B, Lengyel G. Roma in Hungary, interventions that will help them to achieve 1971–2003 [in Hungarian]. Budapest, Hungary: their full health potential. In addition to tack- Human Participant Protection Gondolat Kiadó-MTA Etnikai-nemzeti Kisebbségkutató ling the socioeconomic roots of poor health of The National Health Interview Survey, 2003, and the Intézet; 2004. the Roma people, specific public health inter- Roma Health Survey on which this study was based 18. Milligan P, Njie A, Bennett S. Comparison of two were approved by the Ethical Committee of the Hun- cluster sampling methods for health surveys in devel- ventions, including health education and garian National Scientific Council on Health. Partici- oping countries. Int J Epidemiol. 2004;33:469–476. health promotion programs, are needed. The pants gave written informed consent in both surveys. 19. Questionnaire of the National Health Interview training of community health workers of Roma Survey 2003. Budapest, Hungary: Johan Béla National origin—as recommended in the National Public References Center for Epidemiology; 2004. Available at the Euro- pean Health Interview & Health Examination Surveys 1. European Commission, Directorate General for Health Programme—should continue. Involve- Database: https://www.iph.fgov.be/hishes/print.cfm. Employment and Social Affairs. The Situation of Roma Accessed October 18, 2005. ment of these health workers in public health in an Enlarged European Union. Brussels, Belgium: service can increase the efficiency of public Commission of the European Communities; 2004. 20. Bruin A, Picavet HSJ, Nossikov A. Health Interview Surveys Towards International Harmonization of Methods 2. Ringold D, Orenstein MA, Wilkens E. Roma in an health programs in Roma communities. and Instruments. Copenhagen, Denmark: World Health Expanding Europe: Breaking the Poverty Cycle. Washing- Organization Regional Office for Europe; 1996. ton, DC: World Bank; 2003. Available at: http:// siteresources.worldbank.org/EXTROMA/Resources/ 21. Are Women More Vulnerable on Alcohol Effects? About the Authors roma_in_expanding_europe.pdf. Accessed October 18, Rockville, Md: National Institute on Alcohol Abuse and 2005. Alcoholism; 1999;46. Alcohol Alert. Zsigmond Kósa is the county chief medical officer of 22. Vik PW, Culbertson KA, Sellers K. Readiness to Szabolcs-Szatmár-Bereg County, Hungary. György Széles, 3. McKee M, Adany R, MacLehose L. Health status change drinking among heavy drinking college stu- László Kardos, Karolina Kósa, Gabriella Fésüs, Róza Ádány, and trends in candidate countries. In: McKee M, dents. J Stud Alcohol. 2000;61:674–680. and Zoltán Vokó are with the School of Public Health, MacLehose L, Nolte E, eds. Health Policy and European Medical and Health Science Centre, University of Debrecen, Union Enlargement. Maidenhead, England: Open Uni- 23. Cox BD, Blaxter M, Buckle ALJ, et al. The Health Hungary. Renáta Németh and Sándor Országh are with the versity Press; 2004:24–42. European Observatory on and Lifestyle Survey. London, England: The Health Johan Béla National Center for Epidemiology, Budapest, Health Systems and Policies Series. Promotion Research Trust; 1987. Hungary. Martin McKee is with the London School of 4. Sutherland A. Gypsies: The Hidden Americans. 24. Avoiding the Dependency Trap. New York, NY: Hygiene and Tropical Medicine, London, England. London, England: Macmillan; 1975. United Nations Development Programme; 2002.

May 2007, Vol 97, No. 5 | American Journal of Public Health Kósa et al. | Peer Reviewed | Research and Practice | 859  RESEARCH AND PRACTICE 

Husbands’ Involvement in Housework and Women’s Psychosocial Health: Findings From a Population-Based Study in Lebanon

| Marwan Khawaja, PhD, and Rima R. Habib, PhD

During the 20th century a virtual revolution Objectives. We examined the association between husbands’ involvement in occurred in gender relations, beginning in housework and the psychosocial health of their wives using data on married cou- Western European countries and North ples living in poor neighborhoods in Beirut, Lebanon. America between World War I and World Methods. Data were derived from a cross-sectional survey of 2797 households; War II and then spreading, albeit unevenly, 1652 married couples and their families were included in the analysis. An index to the developing countries of Asia, Latin of husbands’ relative involvement in housework was constructed from 25 items America, and Africa. The profound demo- focusing on division of housework activities. Logistic regression was used to as- graphic changes, particularly the sustained sess associations between husbands’ involvement in housework and wives’ self- decline in human fertility, that have swept the rated mental health status, marital dissatisfaction, and unhappiness. globe during the past 100 years or so have Results. Husbands’ involvement in housework was negatively associated with wives’ psychological distress, marital dissatisfaction, and overall unhappiness after been pivotal in redefining the roles of men adjustment for relevant risk factors. In comparison with wives whose husbands and women as well as the notion of family. were highly involved in housework, wives whose husbands were minimally in- Women’s educational levels have increased volved were 1.60 times more likely to be distressed, 2.96 times more likely to be dramatically during this period, but the paral- uncomfortable with their husbands, and 2.69 times more likely to be unhappy. lel trend toward women becoming more in- Conclusions. Our results showed a significant association between husbands’ volved in the workforce has been the primary involvement in housework and their wives’ psychosocial health. (Am J Public marker of the shift toward relatively more Health. 2007;97:860–866. doi:10.2105/AJPH.2005.080374) egalitarian gender relations. In the early years of the 21st century, we have continued to witness great transforma- with positive effects on their health,9 others relative involvement in household work using tions in gender relations; in some cases estab- have shown that the increase in women’s a detailed list of household tasks. Informed lished institutions and roles are being workload and their dual role have negative12 by the literature on gender perspectives on adapted, and in others new ones are being or neutral13 health effects. women’s health, this study has important impli- created. Notwithstanding these remarkable Much less attention has been focused on cations for public health and clinical practice. changes, women remain largely responsible men’s relative involvement in housework and for household tasks regardless of their em- its implications for the psychosocial health of METHODS ployment status or educational level,1 a situa- their wives. In this area, recent studies have tion with clear implications for their health compared families in which wives are home- Sample and well-being. For example, women with makers and families in which they work.7,11,14 In the Urban Health Study, conducted multiple roles may suffer from elevated stress Findings have demonstrated that “family de- during 2002–2003, trained interviewers and strain as a result of an excess of responsi- mands,” including household chores, are gen- who were women collected data from 15- to bilities and a lack of leisure time. erally associated with homemakers’ health sta- 59-year-old women who were or had been Several scholars have investigated women’s tus and that manual or unskilled workers are married and who resided in 1 of 3 urban involvement in paid and domestic labor and in poorer health than other working women. communities in the greater Beirut area of the resultant effects on their health.2–4 Consid- We examined the association between hus- Lebanon. We selected the study communi- erable attention has been directed toward the bands’ involvement in housework and women’s ties—Nabaa, Hay el Sellom, and the Burj mental health implications of the work envi- psychosocial health using data from a commu- Barajneh refugee camp—using practical and ronment and the “double burden” of paid nity sample of married women, predominantly substantive criteria such as overall conditions work and housework in the context of West- homemakers, in a developing country setting of poverty, poor infrastructure, presence of ern countries.5–11 Results from the available characterized by patriarchy as well as impover- rural immigrants, and proximity to the city of studies are inconclusive. Whereas some ana- ishment. Adopting a categorical approach to Beirut. lysts have argued that involvement in paid measuring division of household labor, we con- As judged by household incomes and sub- labor has generated more control for women, structed a novel but simple index of husbands’ jective assessments of economic well-being,

860 | Research and Practice | Peer Reviewed | Khawaja and Habib American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

these communities can be considered poor sample who were or had been married were Finally, we measured happiness with the according to Lebanese standards. For exam- interviewed in the spring and summer of question “Taking everything into account, do ple, the median yearly household income 2003 to obtain data directly from the women you consider yourself very happy, happy, among the communities’ residents was about themselves. A total of 1869 women were suc- somewhat happy, not happy, or not happy at $4200 in 2002; in comparison, the cessfully interviewed. all?” In a manner similar to that for the mari- Lebanese national average in 1997 was After wives’ and husbands’ records had tal satisfaction item, we grouped responses to nearly $12000.15 Similarly, according to the been matched, a representative sample of this question into 2 categories: happy (coded latest official data published by the Ministry 16 91 married women was selected for the as 0) and unhappy (coded as 1). of Social Affairs, the majority of residents analysis. The matching process involved se- consider themselves poor and indicate that lecting eligible married women (aged 15 to Independent Variables they could not raise $100 if the need arose.15 59 years old) from the women’s data file and Our main independent variable was an Unlike in middle-class Lebanese house- then matching them with their husbands in index of husbands’ involvement in households, people in these disadvantaged commu- the household roster file (which contained in- work. We constructed the index by compar- nities carry out their daily household chores formation on all members of each household). ing husbands’ and wives’ involvement in 25 without help from domestic servants.16 This is because only women aged 15 to 59 different household tasks. We originally ad- Women in these communities continue to as- years old were asked questions about the out- ministered index items in a Likert-scale for- sume the primary responsibility for home- come variables. The study sample of matched mat inquiring whether or not the respondent making and traditionally feminine tasks such couples was extracted from an initial random usually performed a given housework task. as cooking, cleaning, doing the laundry, and sample of 2797 households. For each item, response options were as fol- attending to children’s needs. In contrast, men Different instruments were used at the 2 sur- lows: (1) never, (2) sometimes, (3) most of are the main breadwinners and are in charge vey stages. With the exception of involvement the time, and (4) always. We used these re- of traditionally masculine tasks such as home in housework and income levels, all of the data sponse categories to develop the index in repairs and car maintenance.17 included in our analyses were obtained from several steps. First, we created a data set con- There were also important differences be- the questionnaires administered in the second taining only households with married couples. tween the study communities with respect to stage. Overall response rates were 88.3% and This household-level data set included the socioeconomic status. Burj Barajneh was the 91.1% for the surveys administered at the first 25 items pertaining to husbands and the most disadvantaged of the communities in and second stages, respectively. same items pertaining to wives. terms of living conditions, particularly income Second, we created 25 new variables levels; the primary reason is that Palestinian Outcome Measures comparing husbands’ and wives’ responses refugees, who make up most of the commu- Three outcome variables—mental health to the household task items. In the case of nity’s residents, have largely been “socially ex- status, marital satisfaction, and overall each couple, each of these variables was cluded” in Lebanon, with little or no official happiness—were used to reflect different di- computed as a 6-category indicator of the access to the formal labor market or to mensions of women’s psychosocial health (in husband’s and wife’s involvement: wife only public-sector services. The communities also contrast to their overall well-being18 ). We used (0), mostly wife (1), alternating (2), mostly differed in their ethnic and religious composi- the 12-item General Health Questionnaire husband (3), husband only (4), or neither tions. The populations of Nabaa and Hay el to measure mental health status.19 Women (5). Thus, this indicator reflected the extent Sellom were predominantly Lebanese; 90% whose total score was greater than 4 were of the husband’s involvement in each task, of Burj Barajneh’s residents were Palestinian classified as having poor mental health (i.e., in with scores ranging from 1 (low) to 4 (high). refugees. Nabaa was 80% Christian, whereas psychological distress).20 For example, if a wife “always” did the cook- nearly all of the inhabitants of Hay el Sellom We assessed marital satisfaction with a ing and her husband “never” cooked, the and Burj Barajneh were Muslims. question asking whether wives felt (1) very husband involvement indicator for this item The survey was conducted in 2 stages. comfortable, (2) comfortable, (3) somewhat was coded as 0 (i.e., “wife only”). A couple First, approximately 3000 households were comfortable, (4) not comfortable, or (5) not was defined as “alternating” (coded as 2) if randomly selected from a sampling frame comfortable at all with their husband. Given the husband’s and wife’s levels of involve- constructed specifically for this study; mem- the low frequencies for some of these cate- ment were similar. However, because the bers of 2797 of these households were suc- gories (e.g., only 2.1% of women indicated index was designed to indicate division of cessfully interviewed in the spring of 2002. that they were not comfortable at all with responsibilities between spouses, an item At this stage, all household-level data (e.g., in- their husband), we divided responses into 2 was excluded if neither the husband nor come) were collected through face-to-face in- groups: satisfied (“very comfortable” or the wife performed the task in question terviews conducted with a proxy respondent “comfortable” responses; coded as 0) and (coded as 5). (i.e., any adult in the selected household was dissatisfied (“somewhat comfortable,” “not Table 1 displays the distribution of hus- eligible to answer these questions). Second, all comfortable,” or “not comfortable at all” re- bands’ and wives’ responsibilities for the 15- to 59-year-old women in the household sponses; coded as 1). household tasks assessed. There was a clear

May 2007, Vol 97, No. 5 | American Journal of Public Health Khawaja and Habib | Peer Reviewed | Research and Practice | 861  RESEARCH AND PRACTICE 

TABLE 1—Percentage Distribution of Husbands’ and Wives’ Involvement in Household Tasks: household income levels. Demographic and Urban Health Study, Beirut, Lebanon, 2003 health status risk factors included age group (15–29 years, 30–44 years, 45–59 years), Wife Mostly Mostly Husband current smoking status (yes or no), chronic Task Only, % Wife, % Alternating, % Husband, % Only, % Neither, % health problems (yes or no), and health prob- Washing clothes 91.4 4.7 1.3 0.1 0.1 2.4 lems in the past 2 months (yes or no). Finally, Cleaning bathroom 90.3 4.4 1.6 0.1 1.6 3.4 community of residence (Nabaa, Hay el Sel- Cleaning kitchen 85.1 10.4 2.4 0.2 0.1 1.8 lom, Burj Barajneh) was included as an index Ironing 84.6 5.5 2.3 0.4 0.8 6.4 of social context. This variable was of interest Washing dishes 83.7 11.6 2.7 0.4 0.4 1.6 because, as mentioned, residents of Burj Preparing food 78.1 15.6 4.9 0.2 0.1 1.0 Barajneh were a socially excluded group, and Cleaning rooms 75.9 16.8 5.1 0.1 0.1 1.9 thus the women residing there might be dis- Managing expenses 53.3 8.7 21.4 3.7 11.3 1.6 advantaged with respect to psychosocial Buying personal items (e.g., clothing, shoes, 48.5 13.6 23.7 1.5 4.0 8.8 health status. perfume) Shopping for home needs (e.g., food, soap) 41.0 22.2 22.0 5.5 7.7 1.5 Statistical Analyses Helping with schoolwork 28.8 4.7 6.7 1.5 2.8 55.4 We initially calculated univariate descrip- Following up on children’s schooling 25.7 7.2 18.0 4.5 7.0 37.6 tive statistics for the variables included in our Buying drinking water 22.4 4.4 11.0 4.5 21.8 36.0 data and then conducted bivariate analyses Accompanying someone on an errand 15.1 15.6 48.2 4.2 6.5 10.4 using χ2 tests to examine the associations Paying bills 13.1 5.8 22.2 9.9 45.3 3.7 between psychosocial health status and in- Providing transportation for a family member 11.7 8.3 28.3 5.3 25.1 21.2 dependent variables. Next, we used binomial Providing care for a 4- to 14-year-old son or 9.4 13.3 40.3 0.4 0.4 36.4 logistic regression models to assess the asso- daughter ciations between the outcome measures and Buying water for domestic and personal hygiene 8.4 1.5 4.7 2.1 12.7 70.6 the index of husbands’ involvement in house- Caring for sick family member 8.2 12.5 77.1 0.8 0.5 1.0 work, adjusting for socioeconomic status, so- Performing house maintenance 6.5 1.9 7.7 6.8 63.3 13.9 cial capital, and other relevant demographic Providing care for a 0- to 3-year-old son or 5.3 9.6 25.7 0.2 0.2 59.0 and health risk factors. Stata (Stata Corp, Col- daughter lege Station, Tex) was used in conducting all Caring for elderly family member 4.5 1.2 6.6 0.5 0.0 87.2 analyses.25 Buying fuel 4.1 0.8 3.3 2.5 29.5 59.8 Caring for disabled family member 0.9 0.2 1.0 0.0 0.0 97.9 RESULTS Performing car maintenance 0.4 0.0 0.8 0.7 34.9 63.1 Table 2 shows the univariate distributions of all independent variables included in the division of household labor, with more than quintiles indicating low to high values of analysis, together with their associations with 70% of couples reporting that only the wife husbands’ involvement, but we combined the 3 outcome variables. Overall, 32.3% of performed in-house chores such as cooking the 3 middle categories after they yielded women reported distress, 18.7% reported and washing clothes and dishes. In contrast, essentially identical results with respect to marital dissatisfaction, and 13.2% reported husbands were more involved in typically the outcome variables. being unhappy. In the majority of cases male activities such as car maintenance We adjusted for several socioeconomic, (60.9%), husbands’ level of housework in- (however, wives were involved in these ac- demographic, and health risk factors in our volvement was moderate; in the remaining tivities as well). Other researchers have re- analyses. We used 3 indicators as proxy instances, husbands were almost equally di- ported similar gender typing of household measures for socioeconomic status: yearly vided between low (19.4%) and high (19.7%) tasks.21,22 household income (in quartiles), adjusted for levels of involvement. Finally, item responses were summed for household size using the Office of Economic In terms of income distribution, 26.4% of each couple, yielding an overall index of Cooperation and Development’s equivalence households fell in the highest quartile and husbands’ involvement in housework. This scale23; women’s educational level (no educa- 23.4% in the lowest quartile. More than a index, ranging from 0 to 90, had a high in- tion, elementary school, intermediate school, quarter (27.4%) of the women had less than ternal consistency (Cronbach α=0.77). Its secondary education or more); and labor an elementary school (6 years) education; distribution was highly skewed, however, and force participation (yes or no).24 only 11.2% had at least a secondary educa- thus we decided to transform it into a cate- Sums of amounts from a detailed list of 14 tion. Fewer than a fifth (17.7%) of women gorical variable. We divided the index into different sources were used in determining were in the labor force. More than half

862 | Research and Practice | Peer Reviewed | Khawaja and Habib American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 2—Characteristics of Sample of Married Women Aged 15–59 Years: Urban Health Finally, labor force participation and commu- Study, Beirut, Lebanon, 2003 nity of residence were associated with distress but not with marital dissatisfaction or Total, Distressed, Dissatisfied With Unhappy, unhappiness. Independent Variable No. (%) % Marriage, % % We created multiple logistic regression Husband–wife involvement index scorea models in an effort to uncover the associa- High 321 (19.4) 25.9 10.0 7.2 tions between husbands’ involvement in Medium 1006 (60.9) 29.8 17.0 12.4 housework and our outcome variables Low 325 (19.7) 46.5 32.3 21.2 (Table 3). Relative to high levels of involve- Income group ment in housework, low levels of involvement High 446 (26.4) 26.9 17.0 9.2 were significantly associated with distress Medium high 436 (25.8) 27.8 16.0 11.0 (odds ratio [OR]=1.60; 95% confidence in- Medium low 413 (24.4) 35.8 21.2 15.7 terval [CI]=1.11, 2.30), marital dissatisfaction Low 395 (23.4) 40.5 21.7 17.5 (OR=2.96; 95% CI=1.86, 4.72), and un- Education happiness (OR=2.69; 95% CI=1.53, 4.71) Secondary or above 188 (11.2) 18.1 9.7 6.9 among wives. A gradient in husband’s house- Intermediate 311 (18.5)25.7 12.3 8.0 work involvement was evident for marital Elementary 721 (42.9) 33.4 19.4 14.2 dissatisfaction and unhappiness but not for None 461 (27.4) 40.8 26.0 17.6 distress. However, women whose husbands Labor force participation were moderately involved in housework were Yes 298 (17.7) 39.6 22.8 15.4 1.66 (95% CI=1.09, 2.53) and 1.90 (95% No 1389 (82.3) 31.0 18.1 12.7 CI=1.14, 3.15) times more likely, respec- Age group, y tively, than women whose husbands were 15–29 445 (26.3)26.1 13.3 10.8 highly involved to report marital dissatisfac- 30–44 894 (52.9)31.9 16.9 12.2 tion and unhappiness. 45–59 352 (20.8)42.1 30.8 18.8 Smoking, reports of health problems, and Smoking status low income levels were significantly associ- Nonsmoker 1017 (60.1) 29.3 16.3 10.5 ated with the 3 measures of psychosocial Smoker 674 (39.9) 37.2 22.7 17.2 health. Odds ratios for smokers (vs nonsmok- Health problem in past 2 mo ers) were 1.35 (95% CI = 1.08, 1.69) for dis- No 852 (50.4) 25.2 14.8 9.6 tress, 1.37 (95% CI = 1.05, 1.78) for marital Yes 839 (49.6) 39.8 22.9 16.8 dissatisfaction, and 1.62 (95% CI = 1.19, Chronic health problem 2.19) for unhappiness. The corresponding No 1189 (70.4) 28.3 16.0 10.8 odds ratios for reported health problems Yes 499 (29.6)42.7 25.8 19.0 were 1.64 (95% CI = 1.30, 2.05), 1.31 (95% Community of residence CI=1.00, 1.73), and 1.40 (95% CI = 1.02, Hay el Sellom 483 (28.6) 28.0 17.5 13.9 1.92). Women at the lowest income level Nabaa 614 (36.3) 29.6 20.4 13.4 were more likely than women at the highest Burj Barajneh 594 (35.1) 39.1 18.3 12.5 income level to report distress (OR = 1.55; 95% CI = 1.08, 2.15), marital dissatisfaction a We constructed the index by comparing husbands’ and wives’ involvement in 25 different household tasks. (OR = 1.50; 95% CI = 1.00, 2.23), and unhappiness (OR = 2.34; 95% CI = 1.46, 3.76). Patterns of associations for the remaining (52.9%) were in the reproductive age range Bivariate analyses revealed that scores independent variables were mixed. Labor of 30 to 44 years. A high proportion (39.9%) on the husband involvement index were force participation and community of resi- of women reported smoking, and about half strongly associated with the 3 measures of dence were associated only with distress. reported having had health problems in the psychosocial health assessed. Education, age, Women who were employed were more past 2 months (49.6%). Likewise, a relatively smoking status, reported health problems in likely than those who were not employed to large percentage (29.6%) reported chronic the previous 2 months, and reported chronic report distress (OR=1.49; 95% CI=1.12, health problems. Finally, the percentage of health problems were also associated with 1.99), and distress was more common among women residents was smaller in Hay el Sel- these outcomes. However, income, although women residing in Burj Barajneh than among lom (28.6%) than in Nabaa (36.3%) or Burj associated with distress and unhappiness, was those residing in Hay el Sellom (OR=1.44; Barajneh (35.1%). not associated with marital dissatisfaction. 95% CI=1.08, 1.94).

May 2007, Vol 97, No. 5 | American Journal of Public Health Khawaja and Habib | Peer Reviewed | Research and Practice | 863  RESEARCH AND PRACTICE 

TABLE 3—Adjusted Odds Ratios (ORs; With 95% Confidence Intervals [CIs]) for Distress, DISCUSSION Marital Dissatisfaction, and Unhappiness: Urban Health Study, Beirut, Lebanon, 2003 Our main finding was that involvement of Distress Marital Dissatisfaction Unhappiness a husband in housework is strongly associated Independent Variable OR (95% CI) P OR (95% CI) P OR (95% CI) P with the psychosocial health of his wife. Husband–wife involvement Women, predominantly full-time homemak- index scorea ers, whose husbands were highly involved in High 1.00 1.00 1.00 housework were in better mental health, hap- Medium 1.09 (0.80, 1.48) .573 1.66 (1.09, 2.53) .018 1.90 (1.14, 3.15) .013 pier, and more satisfied with their marriage Low 1.60 (1.11, 2.30) .011 2.96 (1.86, 4.72) ≤.001 2.69 (1.53, 4.71) .001 than other women. This association persisted Income group after adjustment for other relevant risk factors. High 1.00 1.00 1.00 Our results concur with those of previous Medium high 0.98 (0.71, 1.36) .923 0.96 (0.65, 1.42) .827 1.14 (0.70, 1.84) .595 studies from Western countries examining as- Medium low 1.33 (0.96, 1.84) .086 1.25 (0.85, 1.84) .259 1.80 (1.14, 2.86) .013 sociations between division of household labor 26,27 Low 1.55 (1.11, 2.17) .010 1.50 (1.00, 2.23) .050 2.34 (1.46, 3.76) ≤.001 and women’s psychological health. One Education study showed that perceptions of fairness in Secondary or above 1.00 1.00 1.00 the distribution of household tasks are a Intermediate 1.46 (0.92, 2.32) .106 1.15 (0.63, 2.10) .649 0.82 (0.39, 1.69) .592 stronger determinant of psychological distress 6 Elementary 1.92 (1.27, 2.92) .002 1.73 (1.00, 2.95) .046 1.59 (0.85, 2.95) .147 than amount of housework performed. An- None 2.10 (1.35, 3.30) .001 1.77 (1.00, 3.12) .048 1.65 (0.86, 3.19) .133 other study showed that a disproportionate di- Labor force participation vision of household duties is more detrimental No 1.00 1.00 1.00 to women’s mental health than is an overload 27 Yes 1.49 (1.12, 1.99) .005 1.32 (0.95, 1.84) .102 1.20 (0.82, 1.77) .342 of work. A number of recent investigations Age group, y have reported that equitable division of house- 15–29 1.00 1.00 1.00 hold labor is one of the most important deter- 28–31 30–44 1.15 (0.87, 1.51) .320 1.15 (0.80, 1.62) .447 0.93 (0.63, 1.38) .713 minants of women’s psychosocial health. 45–59 1.37 (0.96, 1.97) .082 1.78 (1.17, 2.72) .007 1.17 (0.72, 1.89) .530 We found that socioeconomic variables Smoking status were significantly associated with distress but Nonsmoker 1.00 1.00 1.00 that their associations with happiness and Smoker 1.35 (1.08, 1.69) .008 1.37 (1.05, 1.78) .020 1.62 (1.19, 2.19) .002 marital satisfaction were mixed. Household Health problem in past 2 mo income was associated with the 3 outcome No 1.00 1.00 1.00 measures but only for the lower income Yes 1.64 (1.30, 2.05) .001 1.31 (1.00, 1.73) .050 1.40 (1.02, 1.92) .040 groups: the lower a woman’s family’s income, Chronic health problem the more she reported being distressed, un- No 1.00 1.00 1.00 happy, and dissatisfied with her marriage. Yes 1.31 (1.03, 1.68) .029 1.30 (0.97, 1.73) .077 1.48 (1.06, 2.05) .020 There is a vast literature showing a negative Community of residence relationship between income level and dis- 32–35 Hay el Sellom 1.00 1.00 1.00 tress. Women’s inability to sufficiently Nabaa 1.02 (0.74, 1.39) .911 1.05 (0.73, 1.52) .786 0.94 (0.62, 1.42) .786 support their family financially may trigger Burj Barajneh 1.44 (1.08, 1.94) .015 0.82 (0.57, 1.18) .284 0.61 (0.41, 0.91) .017 negative thoughts such as feelings of fear and worries over family and children, in turn a We constructed the index by comparing husbands’ and wives’ involvement in 25 different household tasks. contributing to poor mental health.33 Our results also revealed a consistent association between women’s educational level and 2 measures of psychosocial health, dis- Education and reported chronic health women not reporting chronic health prob- tress and marital dissatisfaction. These associ- problems were associated with distress and lems, those reporting such problems were ations are in agreement with previous find- unhappiness. Women with no education were more likely to be distressed (OR=1.31; 95% ings showing positive relationships between more likely to be distressed (OR=2.10; 95% CI=1.03, 1.68) and unhappy (OR=1.48; depressive symptoms and low levels of educa- CI=1.35, 3.30) and to be dissatisfied with 95% CI=1.06, 2.05). Finally, older women tion.32,35–40 In a study focusing on quality of their partner (OR=1.77; 95% CI=1.00, (45–59 years) were more likely than younger life among Korean women, educational level 3.12) than were those with at least a second- women (15–29 years) to report marital dis- was positively associated with women’s ary education. Similarly, in comparison with satisfaction (OR=1.78; 95% CI=1.17, 2.72). confidence, psychological well-being, and

864 | Research and Practice | Peer Reviewed | Khawaja and Habib American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

stability.32 In our study, the association be- these associations reached significance for of household labor and women’s psychosocial tween education and unhappiness was nei- both distress and unhappiness. Women often health. A related problem is that we were un- ther statistically significant nor consistent. relate feelings of unhappiness and distress to able to adjust for prior mental health status. We found that women in the labor force health problems.33,40 In contrast, research has As a result, we were not able to verify were more distressed than women not in the shown that health problems can be a reflec- whether the effects associated with division labor force, but there was no significant asso- tion of distress among women.47 For instance, of household labor were an artifact of a selec- ciation with either unhappiness or marital dis- depression has been found to be an indepen- tion process through which women suffering satisfaction. This result was in contrast to the dent risk factor for cardiac problems.48 from health problems may have induced their findings of studies showing that lack of em- Finally, our data showed that women living husbands to do housework. Finally, our data ployment and economic inactivity lead to in- in Burj Barajneh were more likely to be dis- were collected from residents of 3 underprivi- creased distress among women.33–35,40,41 tressed than women living in the other 2 com- leged urban communities, limiting our ability Working women who are of high social status munities. This finding was expected because to generalize findings to other populations of and in good mental health are more likely to the refugees residing there faced economic dis- women in Lebanon or elsewhere. be physically healthy, to earn more money, advantages resulting from the legal restrictions In conclusion, this study demonstrates that and to have a higher level of decisionmaking imposed on them. Another factor that might our index of husbands’ relative involvement authority in their job.5,7,11,14 Such results may have contributed to high distress among resi- in household work is a good predictor of help explain our finding of a positive associa- dents of Burj Barajneh was their substandard women’s psychosocial health. In particular, tion between distress and labor force partici- physical environment, with crowded living con- after adjusting for other relevant variables, we pation; that is, the women of our study were ditions and a lack of recreation facilities. found that when husbands were less involved expected to hold either low-status positions To our knowledge, this is the first study to in housework, their wives were more likely to involving little job satisfaction or relatively use community-based data to investigate asso- be distressed, unhappy, and dissatisfied with high-status positions with stressful workplace ciations between husbands’ involvement in their marriage. conditions. housework and wives’ psychosocial health in We found that older women were signifi- the patriarchal context of the Middle East. We About the Authors cantly more likely than younger women to constructed a novel index of husbands’ and Marwan Khawaja is with the Center for Research on Popu- report that they were dissatisfied with their wives’ involvement in housework based on a lation and Health, American University of Beirut, Beirut, marriage but not to report that they were dis- detailed list of household tasks. This index, Lebanon. Rima R. Habib is with the Department of Envi- ronmental Health, American University of Beirut, Beirut. tressed or unhappy. As shown in the literature, which reflected the ways in which involve- Request for reprints should be sent to Marwan 34,36,42 tiredness gradually intensifies with age. ment in household tasks was divided, enabled Khawaja, PhD, Center for Research on Population and The older women become, the more physically us to ground our analysis in a relational per- Health, Faculty of Health Sciences, American University of Beirut, Box 11-0236 Riad El-Solh, Beirut 1107 2020, and psychologically vulnerable they feel. This spective and document the links between hus- Lebanon (e-mail: [email protected]). fact cannot be fully attributed to the changes bands’ relative contributions to housework This article was accepted July 5, 2006. in hormone levels that occur with the aging and their wives’ psychosocial health. process. Other factors such as general living However, our study involved some impor- Contributors conditions, including economic security, also tant limitations. First, a proxy respondent (in M. Khawaja originated the study, performed the statistical analysis, and prepared the first draft. R. R. Habib play an important role.34 most cases a woman) answered questions supervised data collection and drafted sections of the Our results showed that smoking was about household chores. Thus, reports of article. Both authors conceptualized ideas, interpreted strongly associated with distress, marital dis- household chores may have been influenced findings, and reviewed and edited drafts of the article. satisfaction, and unhappiness, largely echoing by social desirability with respect to what the literature on the association between are suitable household tasks for men and Acknowledgments smoking and psychological health.36,43–45 women.49 Second, given the context of wide- This study was part of a larger multidisciplinary research project on urban health sponsored by the Center There was uncertainty, however, concerning spread illiteracy in our study communities, for Research on Population and Health at the American the direction of causality between smoking we focused on “categorical” measurements of University of Beirut. Support was provided by grants and psychosocial health. Life characteristics household chores rather than amount of time from the Wellcome Trust, the Mellon Foundation, and the Ford Foundation. and psychosocial health may also contribute spent performing a task. However, previous 43 to smoking. Studies have shown that smok- studies have shown that division of household Human Participant Protection ing sometimes functions as a coping mecha- labor is more important than time spent on This study was approved by the institutional review nism helping people deal with pressure and such labor in determining women’s psychoso- board of the American University of Beirut. Participants anxiousness in times of financial difficulties, cial health.27,28 provided informed consent. isolation, or family problems.43,45,46 Third, our survey’s cross-sectional design Our findings concur with previous studies limited our ability to establish causality. We References 1. Lennon MC, Rosenfield S. Relative fairness and reporting associations between health prob- are able to conclude only that in general the division of housework: the importance of options. lems and psychosocial health. In our study, there is a strong association between division Am J Sociol. 1994;92:506–531.

May 2007, Vol 97, No. 5 | American Journal of Public Health Khawaja and Habib | Peer Reviewed | Research and Practice | 865  RESEARCH AND PRACTICE 

2. Arber S. Class, paid employment and family roles: 20.Harrison J, Barrow S, Gask L, Creed F. Social morbidity among low-income women in Aleppo, making sense of structural disadvantages, gender and determinants of GHQ score by postal survey. J Public Syria. Soc Sci Med. 2002;54:1419–1427. health status. Soc Sci Med. 19 91;32:425–436. Health Med. 1999;21:283–288. 39. Duran B, Sanders M, Skipper B, et al. Prevalence 3. Lahelma E, Rahkonen O. Health inequalities in 21. Blair SL, Lichter DT. Measuring the division of and correlates of mental disorders among native Amer- modern societies and beyond. Soc Sci Med. 19 97;4 4: household labour: gender segregation of housework ican women in primary care. Am J Public Health. 2004; 721–722. among American couples. J Fam Issues. 19 91;12: 94:71–77. 91–113. 4. Arber S, Khlat M. Introduction to ‘social and eco- 40.Bromberger JT, Harlow S, Avis N, Kravitz HM, nomic patterning of women’s health in a changing 22. Lammi-Taskula J. Combining work and fatherhood Cordal A. Racial/ethnic differences in the prevalence of world.’ Soc Sci Med. 2002;54:643–647. in Finland. In: Harvey CDH, ed. Walking a Tightrope: depressive symptoms among middle-aged women: the 5. Lennon MC, Rosenfield S. Women and mental Meeting the Challenges of Work and Family. Aldershot, Study of Women’s Health Across the Nation (SWAN). health: the interaction of job and family conditions. England: Ashgate; 2000:1–24. Am J Public Health. 2004;94:1378–1385. J Health Soc Behav. 19 92;33:316–327. 23. Mowafi M, Khawaja M. Poverty. J Epidemiol Com- 41.Bromberger JT, Matthews KA. Employment status 6. Glass J, Fujimoto T. Housework, paid work, and munity Health. 2005;59:260–264. and depressive symptoms in middle-aged women: a depression among husbands and wives. J Health Soc 24. International Labour Office. Resolution concern- longitudinal investigation. Am J Public Health. 1994;84: Behav. 1994;35:179–191. ing statistics of the economically active population, 202–206. 7. Artazcoz L, Borrell C, Benach J. Gender inequali- employment, unemployment and underemployment. 42. Dennerstein L, Dudley EC, Hopper JL, Guthrie JR, ties in health among workers: the relation with family Bull Labor Stat. 1983;3:9–15. Burger HG. A prospective population-based study of demands. J Epidemiol Community Health. 2001;55: 25. Stata, Version 8.0 for Windows [computer pro- menopausal symptoms. Obstet Gynecol. 2000;96: 639–647. gram]. College Station, Tex: Stata Corp; 2003. 351–358. 8. Lee C, Powers JR. Number of social roles, health, 26. Mirowsky J. Depression and marital power: an 43. La Rosa E, Consoli SM, Le Clesiau H, Soufi K, and well-being in three generations of Australian equity model. Am J Sociol. 19 85;91:557–592. Largue G. Psychological distress and stressful life ante- women. Int J Behav Med. 2002;9:195–215. 27. Bird CE. Gender, household labour, and psycho- cedents associated with smoking: a survey consulting a 9. Walters V, Mcdonough P, Strohschein L. The in- logical distress: the impact of the amount and division preventive health center. Presse Med. 2004;33: fluence of work, household structure and social, per- of housework. J Health Soc Behav. 1999;40:32–45. 919–926. sonal and material resources on gender differences in 44. Benjet C, Wagner FA, Borges GG, Medina-Mora health: an analysis of the 1994 Canadian National 28. Noor NM. The relationship between wives’ esti- ME. The relationship of tobacco smoking with depres- Population Health Survey. Soc Sci Med. 2002;54: mates of time spent doing housework, support, and sive symptomatology in the Third Mexican National 677–692. wives’ well-being. J Community Appl Soc Psychol. 19 97; 7:413–423. Addictions Survey. Psychol Med. 2004;34:881–888. 10. Riley AL, Keith VM. Work and housework condi- 45. Cassidy K, Kotynia-English R, Acres J, Flicker L, tions and depressive symptoms among married 29. Baxter J. The joys and justice of housework. Soci- Lautenschlager NT, Almeida OP. Association be- women: the importance of occupational status. Women ology. 2000;34:609–631. tween lifestyle factors and mental measures among Health. 2003;38:1–17. 30. Stevens D, Kiger G, Riley P. Working hard and community-dwelling older women. Aust N Z J Psychi- hardly working: domestic labour and marital satisfac- 11. Borrell C, Muntaner C, Benach J, Artazcoz L. atry. 2004;38:940–947. Social class and self-reported health status among men tion among dual-earner couples. J Marriage Fam. 2001; and women: what is the role of work organisation, 63:504–526. 46. Jun HJ, Subramanian SV, Gortmaker S, Kawachi I. Socioeconomic disadvantage, parenting responsibility, household material standards and household labour? 31. Des Rivieres-Pigeon C, Saurel-Cbuzolles MJ, and women’s smoking in the United States. Am J Public Soc Sci Med. 2004;58:1869–1887. Romito P. Division of domestic work and psychological Health. 2004;94:2170–2176. 12.Gove WR. Gender differences in mental and distress 1 year after childbirth: a comparison between physical illness: the effects of fixed roles and nurturant France, Quebec and Italy. J Community Appl Soc Psy- 47.Krantz G, Östergen PO. Women’s health: do com- roles. Soc Sci Med. 1984;19:77–91. chol. 2002;12:397–409. mon symptoms in women mirror general distress or specific disease entities? Scand J Public Health. 1999; 13. Ahmad-Nia S. Women’s work and health in Iran: 32. Jho MY. Study on the correlation between depres- 27:311–317. a comparison of working and non-working mothers. sion and quality of life for Korean women. Nurs Health Soc Sci Med. 2002;54:753–765. Sci. 2001;3:131–137. 48. Bauer M, Whybrow PC. Depression and other psychiatric illnesses associated with medical conditions. 14 .Hunt K, Annandale E. Just the job? Is the relation- 33. Heilemann MV, Coffey-Love M, Frutos L. Per- Curr Opin Psychiatry. 1999;12:325–329. ship between health and domestic and paid work ceived reasons for depression among women of Mexi- gender-specific? Sociol Health Illn. 19 93;15:632–664. can descent. Arch Psychiatr Nurs. 2004;18:185–192. 49.Press JE, Townsley E. Wives’ and husbands’ 15. Lebanese Socio-Economic Status: Reality and 34. Beutel ME, Weidner K, Schwarz R, Brähler E. housework reporting: gender, class, and social desir- Prospects. Beirut, Lebanon: Ministry of Social Affairs; Age-related complaints in women and their determi- ability. Gend Soc. 1998;12:188–218. 2004. nants based on a representative community study. Eur J Obstet Gynecol Reprod Biol. 2004;117 : 2 0 4–212. 16. Habib RR, Zohry A, Nuwayhid IA, Najdi F. Older adults in the division of domestic labor in communities 35. Chen Y, Subramanian SV, Acedevo-Garcia D, on the outskirts of Beirut. Eur J Ageing. 2006;3: Kawachi I. Women’s status and depressive symptoms: 137–145. a multilevel analysis. Soc Sci Med. 2005;60:49–60. 17. Habib RR, Nuwayhid IA, Yeretzian JS. Paid work 36. Bosworth HB, Bastian LA, Kuchibhatla MN, et al. and domestic labor in disadvantaged communities on Depressive symptoms, menopausal status, and climac- the outskirts of Beirut, Lebanon. Sex Roles J Res. 2006; teric symptoms in women at midlife. Psychosom Med. 55:321–329. 2001;63:603–608. 18. Diener E, Suh E, Lucas R, Smith H. Subjective 37. Eaton WW, Muntaner C, Bovasso G, Smith C. well-being: three decades of progress. Psychol Bull. Socioeconomic status and depressive syndrome: the 1999;125:276–302. role of inter- and intra-generational mobility, government assistance, and work environment. J Health Soc 19. El-Rufaie OF, Daradkeh TK. Validation of the Behav. 2001;42:277–294. Arabic versions of the thirty- and twelve-item General Health Questionnaire. Br J Psychiatry. 1996;169: 38. Maziak W, Asfar T, Mzayek F, Fouad FM, 662–664. Kilzieh N. Socio-demographic correlates of psychiatric

866 | Research and Practice | Peer Reviewed | Khawaja and Habib American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Social Inequalities in Perinatal Mortality in Belo Horizonte, Brazil: The Role of Hospital Care

| Sônia Lansky, MD, PhD, Elisabeth França, MD, PhD, and Ichiro Kawachi, MD, PhD

Brazil has a persistently high infant mortality Objectives. We examined the contribution of hospital type and quality of care 1 rate (22.5 deaths/1000 live births in 2003) to perinatal mortality rates in the city of Belo Horizonte, Brazil. that disproportionately affects the disadvan- Methods. We used a cohort study of all births (40953) and perinatal deaths (826) taged population. Most of the country’s infant in Belo Horizonte in1999. After adjusting for maternal education and birthweight, and perinatal deaths are because of condi- we compared mortality rates according to hospital category—defined by a hos- tions originating during the perinatal period pital’s relation to the national Universal Public Health System (SUS)—and quality that are considered preventable through ac- of care. We used the Wigglesworth Classification to examine perinatal deaths. cess to quality health care. Although there are Results. After we controlled for birthweight and maternal education, the high- important regional disparities, most births est perinatal death rates were observed in private and philanthropic SUS- contracted hospitals (relative to private, non-SUS-contracted hospitals). Hospital (97%) take place in hospitals, and 77% are quality was also directly associated with perinatal death rates. Mortality rates assisted by doctors. A high proportion of peri- were especially high for normal-birthweight babies born in private SUS-contracted natal and infant deaths occur within the first hospitals. Intrapartum asphyxia was the leading cause of preventable death. hours after birth (30% in the first 24 hours Conclusions. In a class-segregated health care system, such as Brazil’s, dis- of life), which suggests the importance of the parities in quality of care between SUS-contracted and non-SUS-contracted hos- level of hospital care. pitals contribute to the unacceptably high rates of perinatal mortality. (Am J Pub- Brazil’s Universal Public Health System (Sis- lic Health. 2007;97:867–873. doi:10.2105/AJPH.2005.075986) tema Único de Saúde, or SUS), which covers the medical expenses of almost 80% of the country’s population, relies on private hospi- hospitals) and quality of hospital care. Our ul- maternal education (from birth and death cer- tals contracted to SUS (37%), as well as hospi- timate goal was to provide public health poli- tificates), birthweight (from chart review when tals run by the philanthropic sector (27%) and cymakers with information that can guide the available or from birth or death certificates), the government (36%).2 Private hospitals not planning and implementation of measures to cause of death (from charts and review of contracted to SUS (non-SUS) provide care for improve the health care system and reduce birth and death certificates; deaths were cate- the remaining minority who can afford private disparities in infant and perinatal mortality. gorized by the Wigglesworth Classification),8 health insurance or direct payment. Conse- and hospital category (from death certificates quently, there is a clear association between METHODS and chart review for perinatal deaths and socioeconomic status and the type of health from birth certificates). facility used. Hospital type can therefore be a This study is based on a 1999 cohort Each hospital was categorized according marker for socioeconomic status,3,4 and it can study involving surveillance of all births to its relation to the SUS system and by its also be an indicator of health care quality.5 So- (n=40953) and perinatal deaths (n=826) in quality of care. There were 20 hospitals con- cioeconomic disparities in the quality of hospi- the city of Belo Horizonte.6 Perinatal deaths tracted to SUS (hereafter called SUS hospital care may in turn explain perinatal mortal- comprise fetal deaths, defined as all stillbirths tals; 12 private SUS hospitals, 4 philan- ity differentials. Few studies have examined with birthweights of 500 g or more or gesta- thropic SUS hospitals, and 4 public SUS socioeconomic inequalities in perinatal mortal- tion age of 22 weeks or more, and early neo- hospitals) and 7 private non-SUS hospitals. ity in Brazil, however, and quality of hospital natal deaths, defined as all infant deaths up to Quality assessment was conducted only in care has not yet been systematically assessed. 7 days of life in which the infant weighed Belo Horizonte hospitals (n = 24); each hos- We analyzed the role of hospital quality at 500 g or more at birth or had a gestational pital received a standardized score of 0 to the time of delivery and birth and its contri- age of 22 weeks or more.7 For our analysis, 2000 (assigned by Costa et al.9), which re- bution to the high perinatal mortality rates in data were collected by hospital chart review lated to its structural ability to assist the the city of Belo Horizonte. Situated in the and linkage of individual records to the Na- mother and the baby.9 Ten hospitals were more developed southeast region of Brazil, tional Live Birth Information System and the scored 1000 or lower, indicating that they Belo Horizonte is the country’s fourth largest National Death Information System, yielding lacked the conditions for such basic health city, with 2.2 million inhabitants. We focused 775 perinatal deaths in 27 hospitals. Infor- care as neonatal resuscitation (low quality); on the differential in perinatal mortality rate mation gathered by 1 of the authors (S.L.) 7 hospitals were scored between 1001 and between hospital categories (SUS vs non-SUS and by trained medical students included 1500 (intermediate quality), while 7 were

May 2007, Vol 97, No. 5 | American Journal of Public Health Lansky et al. | Peer Reviewed | Research and Practice | 867  RESEARCH AND PRACTICE 

TABLE 1—Distribution of Births and Perinatal Deaths by Selected Variables: Belo Horizonte, scored above 1500 (adequate quality). Fur- Brazil, 1999 ther details of the development and validation of the scoring system have been de- Variable Births, no. (%) Deaths, no. (%) Perinatal Mortality Rate Rate Ratio (95% CI) scribed previously.9 Birthplace Maternal education (< 4 years, 4–7 Hospital 40 075 (97.9) 753 (97.2) 18.7 1.0 years, 8–11 years, or ≥ 12 years) was used Other 85 (0.2) 19 (2.5) 223.5 11.9 (7.8, 17.8) as an indicator of socioeconomic status. Missing data 793 (1.9) 0 (0.0) ...... Using the Wigglesworth system,8 we classi- Total 40 953 (100.0) 775 (99.6) 18.9 1.0 (0.9, 1.1) fied the causes of perinatal death as an- Birthweight,a g tepartum, severe congenital malformation, 500–1499 818 (2.0) 382 (50.7) 466.9 87.8 (75.0, 102.9) immaturity (i.e., gestational period less than 1500–2499 3648 (8.9) 177 (23.5) 48.5 9.1 (7.5, 11.2) 37 weeks), intrapartum asphyxia, and other ≥2500 36 487 (89.1) 194 (25.8) 5.3 1.0 specific causes. Total 40 953 (100.0) 753 (100.0) 18.3 3.5 (3.0, 4.1) We analyzed perinatal death rates according to hospital category, adjusting for 2 major Pregnancy confounders: maternal education and birth- Singleton 40 093 (97.9) 678 (90.1) 16.9 1.0 weight. Multivariable regression analysis was Multiple 838 (2.0) 56 (7.4) 66.8 4.0 (3.0, 5.1) carried out to determine the association be- Missing data 22 (0.1) 19 (2.5) ...... tween hospital category and perinatal death. Total 40 953 (100.0) 753 (100.0) 18.3 1.1 (1.0, 1.2) We excluded 231 (29.8%) antepartum Delivery deaths (those that happened before the onset Cesarean 17 002 (41.6) 252 (32.5) 14.8 1.0 of labor) because hospital obstetric care dur- Vaginal 23 922 (58.3) 497 (62.0) 20.7 1.4 (1.2, 1.6) ing labor could not affect birth outcomes in Missing data 29 (0.1) 4 (5.5) ...... these cases. We also excluded nonhospital Total 40 953 (100.0) 753 (100.0) 18.4 1.2 (1.1, 1.4) births (n = 85 [0.2%]) and deaths (n = 19 Maternal education, y [2.5%]) and 3 deaths (0.4%) that took place <4 514 (1.3) 40 (5.3) 77.8 7.2 (5.1, 10.2) in nonmaternity hospitals. In the case of new- 4–7 21 536 (52.6) 358 (47.5) 16.6 1.5 (1.3, 1.9) born transfers between hospitals (17 of the 8–11 11 338 (27.7) 122 (16.2) 10.8 1.0 deaths [2.2%]), death was attributed to the ≥12 5616 (13.7) 74 (9.8) 13.2 1.2 (0.9, 1.6) hospital of birth. Data entry, processing, and Missing data 1949 (4.8) 159 (21.2) ...... analyses were conducted with the software Total 40 953 (100.0) 753 (100.0) 18.4 1.7 (1.4, 2.1) programs Epi Info 6.0 (Centers for Disease Hospital SUS category Control and Prevention, Atlanta, Ga) and Private non-SUS 8970 (21.9) 95 (12.6) 10.6 1.0 Stata version 8 (StataCorp LP, College Sta- Private SUS 16 194 (39.5) 220 (29.2) 13.6 1.3 (1.0, 1.6) tion, Tex). Philanthropic SUS 8816 (21.6) 166 (22.1) 18.8 1.8 (1.4, 2.3) Public SUS 6180 (15.1) 272 (36.1) 44.0 4.2 (3.3, 5.2) RESULTS Missing data 793 (1.9) 0 (0.0) ...... Total 40 953 (100.0) 753 (100.0) 18.4 1.7 (1.4, 2.1) The vast majority of the births took place in hospitals, although 19 (2.5%) of the deaths Hospital quality of careb occurred outside a hospital, either at home, Low 18 206 (44.5) 260 (35.3) 14.3 0.6 (0.5, 0.7) on the streets, during transfer to a hospital, Intermediate 10 558 (25.7) 239 (32.4) 22.6 0.9 (0.8, 1.1) Adequate 9688 (23.7) 238 (32.3) 24.6 1.0 or at another health facility (Table 1). This Missing data 2501 (6.1) 0 (0.0) ...... information could be ascertained only by hospital chart surveillance, because the ba- Total 40 953 (100.0) 737 (100.0) 18.0 1.3 (1.1, 1.4) bies’ birth and death certificates were regis- Note. CI=confidence interval; SUS=Universal Public Health System (Sistema Único de Saúde); non-SUS=private hospitals tered as if they were born and had died in not contracted to SUS. Nineteen domiciliary deaths (2.5%) and 3 nonmaternity hospital deaths (0.4%) were excluded for all the hospitals. A total of 10.9% of all live variables except birthplace. aBirthweight was missing for 4 deaths (0.5%) and 22 live births (0.05%). births were low birthweight (<2500 g), and bQuality care assessment for 24 Belo Horizonte hospitals. Hospital quality of care was determined by giving each hospital a 2.0% were very low birthweight (<1500 g). 9 standardized score (assigned according to Costa et al. ) between 0 to 2000, related to its structural ability to assist mothers By contrast, 74.2% of the deaths occurred and babies. Hospitals with scores of ≤1000 were considered low quality, 1001–1500 intermediate quality, and >1500 adequate quality. among babies born with low birthweight and 50.7% among babies with very low

868 | Research and Practice | Peer Reviewed | Lansky et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

birthweight. However, a quarter of all perina- mortality rate (77.8/1000). A higher propor- perinatal mortality was observed for multiple tal deaths (25.8%) occurred among babies tion of missing data on maternal education births and vaginal delivery. with normal birthweight. was observed for perinatal death (21.2%) When the Wigglesworth Classification was Risk of perinatal mortality increased as than for live births (4.8%). used, significant differences in causes of peri- birthweight decreased, but mortality rates Most of the births (76.2%) and deaths natal death by type of hospital were observed were still high for babies with normal birth- (87.4%) occurred in SUS hospitals. While pri- (Table 2). While antepartum deaths and deaths weight (5.3/1000 live births) and for those vate SUS hospitals accounted for 39.5% of from immaturity prevailed in private non-SUS weighing between 1500 g and 2500 g births, deaths were more concentrated in hospitals, intrapartum asphyxia was much (48.5/1000). A gradient in perinatal mortal- public SUS hospitals (36.1%). The crude peri- more common in SUS hospitals. Rates for as- ity rates according to maternal education was natal mortality rate was highest in public SUS phyxia were 2.0 (public SUS hospitals) to 4.0 found (Table 1). Mothers with 4 to 7 years of hospitals and lowest in private non-SUS hos- (private SUS hospitals) times higher than that schooling accounted for 52.6% of live births pitals. Compared with private non-SUS hospi- seen at private non-SUS hospitals; these rates and 47.5% of infant deaths. The babies of tals, the crude rate ratios for perinatal mortal- were especially high for normal-birthweight the few mothers with less than 4 years of ity ranged from 1.3 (private SUS hospitals) to babies in private SUS and philanthropic SUS schooling experienced the highest perinatal 4.2 (public SUS hospitals). A higher risk of hospitals. Severe congenital malformation represented 7% to 10% of the perinatal deaths, and these rates were higher for low- TABLE 2—Perinatal Mortality (per 1000 Live Births) by Cause of Death, Birthweight, and birthweight babies in public SUS and philan- Maternal Education According to Hospital Category: Belo Horizonte, Brazil, 1999 thropic SUS hospitals than in other types of Hospital Category hospitals. Other causes (such as infection in full-term babies), although small in number, Private Private Public Philanthropic non-SUS SUS SUS SUS Total were also more frequent in SUS hospitals. Public SUS hospitals had the highest per- a Cause of death (Wigglesworth Classification) centage of low-birthweight (21.6%) and very- Antepartum low-birthweight (5.6%) babies, but they also <2500 g 25.1 35.2 58.1 39.5 40.5 had the lowest perinatal death rates for very- ≥ 2500 g 1.0 1.1 3.1 1.3 1.4 low-birthweight and normal-birthweight ba- Severe congenital malformation bies. Private SUS hospitals had the lowest <2500g 7.6 8.8 13.6 15.4 11.4 percentage of low-birthweight babies (6.1%), ≥ 2500 g 0.4 0.5 0.4 1.0 0.4 but they had the highest birthweight-specific b Immaturity mortality rates for babies weighing 500 g to <2500 g 31.6 33.2 48.0 25.1 34.3 1499 g or weighing 2500 g or more. Among ≥ 2500 g 0.0 0.6 0.2 0.3 0.3 babies weighing 1500 g to 2499 g, perinatal Asphyxia death rates were similar for the different types <2500 g 18.5 42.0 50.2 32.8 36.7 of SUS hospitals, but these rates were at least ≥ 2500 g 0.9 3.6 2.7 3.3 2.7 twice as high as those for the private non-SUS Other hospitals. Among normal-birthweight babies, <2500 g 1.1 2.0 1.4 1.9 1.6 perinatal mortality rates at SUS hospitals ≥ 2500 g 0.0 0.6 0.4 1.0 0.5 were 3.0 times higher (public SUS) to 4.1 Birthweight and maternal education times higher (private SUS) than in private c Birthweight, g non-SUS hospitals (Table 2). 500–1499 288.7 679.0 349.9 398.3 376.3 In SUS hospitals, 68.6% of mothers had 1500–2499 17.3 36.4 39.1 35.3 32.1 less than 8 years of schooling, compared with ≥ 2500 1.2 5.1 3.8 4.6 3.9 private non-SUS hospitals, where 85.3% of d Maternal education, y the mothers reported 8 or more years of <8 6.2 11.9 23.3 13.3 14.1 schooling. For both less-educated (< 8 years ≥ 8 6.6 7.9 21.7 8.5 8.8 of schooling) and more-educated (≥ 8years) Note. SUS=Universal Public Health System (Sistema Único de Saúde); non-SUS=private hospitals not contracted to SUS. mothers, perinatal mortality rates were aEach cause of death is categorized as low birthweight (<2500 g) or normal birthweight (≥2500 g). higher in public SUS hospitals (23.3/1000 b Defined as gestational period less than 37 weeks. and 21.7/1000, respectively) than in private cExcluded are 8 live births (0.02%) and 2 deaths (0.3%) with birthweight missing as well as 231 antepartum deaths (30.6%). dExcluded are 1863 births for which maternal education was missing (4.6%), 73 deaths for which maternal education was non-SUS hospitals (6.2/1000 and 6.6/1000). missing (14.0%), 793 births with missing hospital category (1.9%), and 231 antepartum deaths (30.6%). With private non-SUS hospitals used as the reference, rate ratios varied from 1.2

May 2007, Vol 97, No. 5 | American Journal of Public Health Lansky et al. | Peer Reviewed | Research and Practice | 869  RESEARCH AND PRACTICE 

TABLE 3—Perinatal Mortality Rate (MR) and Relative Risk (RR) by Hospital Category, Maternal Education, and Birthweight: Belo Horizonte, Brazil, 1999

Low Birthweight (<2500 g) Normal Birthweight (≥2500 g ) Maternal Education, <8 y Maternal Education, ≥8 y Maternal Education, <8 y Maternal Education, ≥8 y Hospital Description MR RR (95% CI ) MR RR (95% CI ) RRa (95% CI ) MR RRa (95% CI ) MR RR (95% CI ) RRa (95% CI)

SUS category Private SUS 93.4 1.8 (0.8, 4.5) 85.1 1.5 (0.9, 2.5) 1.6 (1.1, 2.3) 6.1 NAb (. . .) 3.9 2.7 (1.2, 5.8) NA (. . .) Philanthropic SUS 74.2 1.4 (0.7, 2.8) 55.8 1.0 (0.6, 1.8) 1.2 (0.7, 1.8) 4.9 NA (. . .) 3.7 2.5 (1.0, 6.1) NA (. . .) Public SUS 98.2 1.8 (0.9, 3.6) 96.3 1.7 (1.1, 2.6) 1.7 (1.2, 2.5) 3.1 NA (. . .) 3.7 2.5 (0.9, 7.4) NA (. . .) Private non-SUS 52.6 1.0 (. . . ) 55.3 1.0 (. . .) 1.0 0.0 1.0 (. . .) 1.5 1.0 (. . .) NA (. . .) Total 84.5 1.6 (0.8, 3.3) 67.8 1.2 (0.8, 1.8) 1.3 (1.0, 1.9) 4.8 NA (. . .) 2.7 1.9 (0.9, 3.7) NA (. . .) Quality of careb Low 87.2 1.0 (0.8, 1.5) 71.9 1.1(0.7, 1.9) 1.1 (0.8, 1.4) 5.8 2.3 (1.2, 4.3) 4.5 3.0 (1.3) 2.5 (1.5, 4.0) Intermediate 81.0 1.0 (. . .) 62.9 1.0 (. . .) 2.6 1.0 (. . .) 1.5 1.0 (. . .) NA (. . .) Adequate 100.7 1.2 (0.9, 1.7) 74.7 1.2 (0.8, 1.8) 1.2 (0.9, 1.6) 4.9 1.9 ( 0.9, 4.2) 2.2 1.5 (0.6, 3.7) 1.7 (0.9, 3.1) Total 85.3 1.1(0.8, 1.4) 71.1 1.1 (0.8, 1.7) 1.1 (0.9, 1.3) 4.8 1.9 (1.0, 3.5) 2.7 1.8 (0.8, 4.0) 1.8 (1.1, 3.0)

Note. CI=confidence interval; SUS=Universal Public Health System (Sistema Único de Saúde). NA=mean not possible to calculate (RR=0). Birthweights are categorized by years of maternal education. Excluded from the data are 231 antepartum deaths (29.8%) for hospital SUS category, 227 antepartum deaths (30.8%) for hospital quality-of-care category, 19 domiciliary deaths (2.5%), and 3 nonmaternity hospital deaths (0.4%). Also excluded are 73 deaths (14.0%) and 1863 births (4.6%) with missing maternal education and 793 births (1.9%) with missing hospital category. aAdjusted relative risk (Mantel-Haenszel). bQuality care assessment for 24 Belo Horizonte hospitals. Hospital quality of care was determined by giving each hospital a standardized score (assigned according to Costa et al.9) between 0 to 2000, related to its structural ability to assist mothers and babies. Hospitals with scores of ≤1000 were considered low quality, 1001–1500 intermediate quality, and above 1500 adequate quality.

(more-educated mothers at private SUS hos- death. When we compared all SUS hospitals DISCUSSION pitals) to 3.5 (less-educated mothers at public and non-SUS hospitals, the difference in peri- SUS hospitals). natal mortality rate was 51.3% higher in the The high perinatal mortality rate (19/1000) When data were stratified for both mater- former (rate ratio [RR] for perinatal mortality in Belo Horizonte is paradoxical given that, as nal education (<8 years vs ≥8 years of rate=2.1); when we compared less-educated in the rest of the country, most births took schooling) and birthweight, private SUS hos- and more-educated mothers, the difference in place in hospitals and were assisted by physi- pitals showed the highest mortality rates for rate was 31.2% higher among the former cians. It is also of major concern that 19 normal-birthweight babies in both strata of (RR for perinatal mortality rate=1.4). The deaths (2.5%) still took place outside of hos- maternal education (Table 3). Once again, pri- population attributable risk for perinatal pitals, even though there was no shortage of vate non-SUS hospitals showed the lowest deaths was 42.4% for SUS hospitals and obstetrical beds in the city. We also observed mortality rates. Public SUS hospitals had the 36.9% for low maternal education. important differences in data quality—for ex- highest mortality rates for low-birthweight ba- Regarding hospital quality, a high propor- ample, there was a higher prevalence of miss- bies. Relative risks for low-birthweight babies, tion of the births (44.5%) took place in lowing information on maternal education for adjusted by maternal education, were 1.5 quality hospitals, and deaths were more babies that died than for live births. (private non-SUS hospitals) and 1.7 (public equally distributed between the remaining 2 Outcomes for SUS hospitals were worse SUS hospitals), with private non-SUS hospitals categories (Table 1). The crude perinatal mor- than for private non-SUS hospitals, which had used as the reference. It was not possible to tality rate was lower in low-quality hospitals. the lowest mortality rates. This finding could estimate the relative risk for the normal-birth- However, when we adjusted for birthweight indicate inadequate capacity to intervene dur- weight group adjusted by maternal education, and maternal education and excluded an- ing labor and birth and after birth, for low- because the death rate in the reference cate- tepartum deaths (Table 3), perinatal mortality birthweight as well as for normal-birthweight gory (private non-SUS hospitals) was zero. rates were actually higher in low-quality hos- babies. Important differences in birthweight- When we controlled for birthweight pitals for normal-birthweight babies, with an specific mortality rates between SUS hospitals (model 2) and both birthweight and maternal adjusted relative risk of 2.5 compared with and non-SUS hospitals were observed. Within education (model 3) using a multiple logistic intermediate-quality hospitals. Low-quality the very-low-birthweight stratum (500–1499 g), regression analysis (Table 4), private SUS hospital status was also an independent risk rates were high for all types of SUS hospitals, and philanthropic SUS hospitals remained factor for perinatal death in the multiple logis- but especially for private SUS hospitals, where independently associated with perinatal tic regression model (Table 4). almost 70% of the babies died. This situation

870 | Research and Practice | Peer Reviewed | Lansky et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 4—Relative Risk (RR) and 95% Confidence Intervals (CIs) for Perinatal Deaths and within the most developed countries, dispari- Selected Variables: Belo Horizonte, Brazil, 1999 ties in mortality are mainly encountered among newborns weighing less than 750 g.13 Model 1 Model 2 Model 3 Among hospitals connected to the SUS Variables RR (95% CI) RRa (95% CI) RRa (95% CI) system, public SUS hospitals had the lowest Group 1 variables mortality rates for very-low-birthweight and Hospital SUS category normal-birthweight babies and the highest Private SUS 1.3 (1.0, 1.6) 2.9 (2.1, 3.7) 3.1 (2.2, 4.4) proportion of low-birthweight babies and Public SUS 4.3 (3.4, 5.4) 1.9 (1.4, 2.5) 1.4 (1.0, 2.0) very-low-birthweight deaths (68.2%). How- Philanthropic SUS 1.8 (1.4, 2.3) 2.2 (1.6, 3.0) 2.1(1.5, 2.9) ever, although public SUS hospitals were not Private non-SUS 1.0 1.0 1.0 associated with perinatal death in the final Birthweight, g logistic regression model, they showed the 500–1499 . . . 185.5 (149.1, 230.8) 202.9 (159.0, 259.0) highest mortality rates among low-birthweight 1500–2499 . . . 10.0 (8.1, 12.4) 9.4 (7.4, 11.9) babies in the stratified analysis. By contrast, ≥2500 . . . 1.0 1.0 private SUS hospitals had the lowest propor- Maternal education, y tion of low-birthweight babies, but the highest <4 ...... 3.4 (2.0, 5.7) mortality rates for very-low-birthweight and 4–7 ...... 0.8 (0.6, 1.1) normal-birthweight babies. This is an impor- 8–11 ...... 0.7 (0.5, 1.0) tant discrepancy from what would be ex- ≥12 ...... 1.0 pected for health facilities that deal mainly Group 2 variables with low-risk pregnancies and babies. Non- Hospital quality of careb timely health care access and low-quality care Low 0.6 (0.5, 0.7) 1.7 (1.4, 2.1) 1.9 (1.5, 2.4) during delivery and the neonatal period could Intermediate 0.9 (0.8, 1.1) 1.1 (0.9,1.4) 0.8 (0.7, 1.1) explain the worse outcomes in these settings. Adequate 1.0 1.0 1.0 These hospitals are associated with low health Birthweight, g care quality, which was an independent risk 500–1499 . . . 199.5 (160.3, 248.2) 198.0 (155.0, 253.0) factor for perinatal mortality as reported in a 1500–2499 . . . 10.3 (8.3, 12.7) 9.1 (7.2, 11.5) previous study.14 Birthweight-specific mortality ≥2500 . . . 1.0 1.0 rates are influenced by access to quality ob- Maternal education, y stetric and neonatal care, particularly among <4 ...... 4.0 (2.4, 6.7) very-low-birthweight babies, but it is also a 4–7 ...... 1.1 (0.8, 1.4) determinant for child survival when birth 8–11 ...... 0.8 (0.6, 1.1) complications occur. Birth complications are ≥12 ...... 1.0 expected—but not predictable—in nearly 15% of all childbirths and occur predominantly Note. SUS=Universal Public Health System (Sistema Único de Saúde).There were no control factors in model 1, we controlled for birthweight in model 2, and was controlled for birthweight and maternal education in model 3. among low-risk and full-term pregnancies. aAdjusted relative risk. Hospital audits conducted by the Perinatal b Quality care assessment for 24 Belo Horizonte hospitals. Hospital quality of care was determined by giving each hospital a Commission of Belo Horizonte City and 2 in- standardized score (assigned according to Costa et al.9) between 0 to 2000, related to its structural ability to assist mothers and babies. Hospitals with scores of ≤1000 were considered low quality, 1001–1500 intermediate quality, and above 1500 dependent studies consistently revealed poor- adequate quality. quality care at private SUS hospitals and the low-quality category hospitals.9,14,15 Most of the mothers were not adequately assisted reflects a disorganized perinatal health care perinatal mortality rates in developed coun- during labor: 80% were not assessed at least system that allows women with high-risk preg- tries in the 1990s. According to the literature, every hour while in labor, partographs (a nancies to deliver their babies in inadequate the leading causes of mortality in this group graph that records the progress of labor and facilities, as well as barriers to accessing inten- of babies are birth complications and intra- assists in identifying when intervention is nec- sive care once the babies are born. partum asphyxia (50%), antepartum death essary) were not used in 75% of the deliver- Almost half of the deaths that took place in (25%), and infection (10%), suggesting that ies, and there was a very low percentage of private SUS hospitals were normal-birthweight differences in risk and access to efficacious corticosteroid use for mothers in premature babies. The observed mortality rates for interventions—such as appropriate obstetric labor or surfactant therapy for premature normal-birthweight babies alone (under- management during pregnancy, labor, and newborns.14 estimated here because antepartum deaths newborn care—contribute to disparities in Previous studies in Brazil provided evi- were excluded) are comparable to the overall perinatal mortality rates.10–12 By contrast, dence of low-quality hospital care during

May 2007, Vol 97, No. 5 | American Journal of Public Health Lansky et al. | Peer Reviewed | Research and Practice | 871  RESEARCH AND PRACTICE 

labor and delivery16–18 as well as barriers to interventions before and during pregnancy, perinatal, and neonatal care. Alternatively, prompt access to hospital care.19 , 2 0 Inadequate after birth, and during pregnancy termina- public services could be increased within the use of corticosteroids for immaturity was ob- tion.13 Important differences between SUS SUS system. In addition, public health policies served in most of the public SUS tertiary-care and non-SUS hospitals were also observed should consider expanding intermediate- hospitals across different states of the for specific conditions, mainly represented by quality hospitals, as these facilities have country.21 Another study showed underuse of congenital or acquired infections in full-term demonstrated outcomes for perinatal mortality surfactant therapy for premature babies, espe- babies. Better management of newborn infec- similar to those of adequate-quality hospitals.14 cially among low socioeconomic groups: use tious diseases—during prenatal or neonatal It is noteworthy that interventions intro- was only 12.5% for SUS patients compared care—could reduce these preventable deaths. duced by the Belo Horizonte Health Depart- with 76.6% among non-SUS patients.22 More One noteworthy point concerns the contri- ment after 1999 decreased early neonatal recently, lower-quality care has been reported bution of rates of cesarean delivery to the mortality by 30% in 2 years, together with for Black women, who are also concentrated differences in hospital perinatal mortality, as smaller decreases in maternal and fetal mor- in public hospitals. This factor contributes to it is the predominant mode of delivery in tality. In this period, there were improve- the fact that infant mortality is 66% higher non-SUS hospitals (33.0% of all births at ments in timely access to hospital admission among Blacks than for Whites in Brazil.16,23,24 SUS hospitals vs 72.0% in non-SUS hospitals during labor and to intensive care units, and Other studies have reported higher neo- in 1999). We found that, compared with 5 private SUS hospitals with low quality natal and infant mortality rates in public SUS vaginal delivery, cesarean delivery was a pro- scores were closed down. Those outcomes hospitals than in private non-SUS hospitals; tective factor for perinatal mortality in both reinforce the importance of hospital care in they have attributed the disparity to hospital non-SUS hospitals (RR = 0.54; 95% confi- reducing perinatal mortality.15 case mix—that is, a higher proportion of poor dence interval [CI] = 0.36, 0.81) and SUS Belo Horizonte’s health system is segre- mothers who use SUS facilities.3,4,25 Residual hospitals (RR = 0.88; 95% CI = 0.74, 1.04), gated, with the more disadvantaged segments confounding could explain in part the high although the difference was not significant of the population relying on SUS facilities mortality rates at public SUS hospitals—as for the latter hospitals, similar to previous while the better off mainly use private well in higher-quality hospitals—because they findings.4,5 Cesarean delivery might therefore non-SUS hospitals; patterns of quality and are referral facilities for high-risk pregnancies be a confounder in the relationship between practices differ greatly, contributing to in- and for seriously ill and very-low-birthweight hospital category and perinatal mortality, be- equalities in health outcomes. Our findings il- babies. Nevertheless, in our analyses, birth- cause the procedure is a marker for high so- lustrate the inverse equity hypothesis: child weight was taken into account precisely be- cioeconomic status. However, when cesarean health inequities increase with greater access cause it is a major predictor for child survival. deliveries were taken into account, there was to medical technology by those of higher still an important differential in perinatal socioeconomic status.31–33 Preventable Causes of Perinatal Deaths mortality between SUS and non-SUS hospi- Brazil must urgently address the paradox of The Wigglesworth Classification highlighted tals (18.7/1000 and 8.5/1000, respectively; persistently high maternal and neonatal mor- disparities between hospitals in perinatal RR=2.2; 95% CI = 1.6, 2.9), suggesting an tality rates that occur in the presence of the causes of death. Intrapartum causes of death effect modification between cesarean deliv- medicalization of birth. While simple, effec- were a major problem at SUS hospitals. As- ery and SUS hospitals. As pointed out by tive, and low-cost practices are poorly de- phyxia, which accounted for 30% of the peri- Barros et al., rates of cesarean delivery are ployed, more elaborate techniques, such as natal mortality, similar to rates in other devel- lower for high-risk women than for lower-risk cesarean delivery and induction of labor, are oping countries,26–28 is highly preventable; it women, and women with the greatest need often misused.30 As pointed out by Daniels requires action based on timely low-cost inter- may still fail to receive it even though rates and colleagues, the country has “achieved a vention during labor, including appropriate of cesarean delivery are high in Brazil.29 rapid economic growth but lagged behind in interpersonal assistance. Every setting should Although the SUS system undoubtedly health improvements,”34(p32) in contrast to be prepared to respond adequately in situa- represents a huge improvement in health care other developing countries, which have priori- tions such as birth complications, which could delivery in Brazil, there is no systematic tized the provision of social services that re- lead to a significant reduction in mortality, monitoring of hospital quality. The system duce mortality and improve quality of life. especially in private SUS hospitals. still depends on private SUS hospitals, which Organization of perinatal care in a regional- Higher rates of severe congenital malfor- account for approximately 30% of the obstet- ized and integrated system and improvement mation at public SUS and philanthropic SUS rics beds and deliver questionable quality of in quality care is fundamental not only at the hospitals may reflect a client selection effect. care. The private sector contracted under SUS community level but at the hospital level as The observed differential between SUS and is a mixture of public and private models well. An ethical and legal approach that guar- non-SUS hospitals could be explained by “with differing expectations and reward sys- antees a standard quality of care for all, as background differences in exposures to risks tems, unduly distorting overall health-care pat- well as access to available technology, is a in the housing and work environment, as terns.”30(p853) Routine audits of hospital quality challenge for the SUS system.35,36 It can occur well as difficulties in accessing efficacious are urgently needed to scale up maternal, only when budget and market constraints, and

872 | Research and Practice | Peer Reviewed | Lansky et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

the common practice of low-quality care for 4. Almeida MF, Novaes HMD, Alencar GP, Accidental home deliveries in southern Sao Paulo, poor people, are overcome.37 Rodrigues LC. Neonatal mortality: socioeconomic fac- Brazil [in Portuguese]. Rev Saude Publica. 2005;39(3): tors, health services risk factors and birth weight in the 366–375. City of São Paulo [in Portuguese]. Rev Bras Epidemiol. 21.Rede Brasileira de Pesquisas Neonatais. Antenatal Conclusions 2002;5:93–107. corticosteroids use and clinical evolution of preterm We point out important disparities in peri- 5. Becerra JE, Atrash HK, Perez N, Saliceti JA. Low newborn infants [in Portuguese]. J Pediatr (Rio J). natal mortality in Belo Horizonte and empha- birth weight and infant mortality in Puerto Rico. Am 2004;80(4):277–284. J Public Health. 19 93;83:1572–1576. size the role of hospital care in producing and 22. Marques S. Neonatal Mortality and Surfactant Use maintaining the unacceptably high rates of 6. Lansky S, França E, Leal MC. Perinatal avoidable in Premature Babies [master’s thesis; in Portuguese]. deaths in Belo Horizonte, Minas Gerais, 1999 [in Por- Goiânia, Brazil: Federal University of Goiás; 2002. perinatal and neonatal mortality in Brazil. We tuguese]. Cad Saude Publica. 2002;18:139–151. argue that, besides intervening on socioeco- 23. Barros FC, Victora CG, Horta BL. Ethnicity and 7. International Classification of Diseases, 10th Revi- infant health in southern Brazil: a birth cohort study. nomic factors that contribute to inequities in sion. Geneva, Switzerland: World Health Organization; Int J Epidemiol. 2001;30:1001–1008. perinatal mortality, it is important to improve 1994. 24. PNUD Brasil. Atlas racial brasileiro. 2004. Avail- the quality of health care delivered to women 8. Keeling JW, MacGillivray I, Golding J, able at: http://www.pnud.org.br/publicacoes/atlas_ and their babies at the health system level. Wigglesworth J, Berry J, Dunn PM. Classification of racial/index.php. Accessed July 7, 2005. perinatal death. Arch Dis Child. 1989;64:1345–1351. 25. Almeida SDM, Barros MBA. Health care and neo- 9. Costa JO, Xavier CC, Proietti FA, Delgado MS. natal mortality [in Portuguese]. Rev Bras Epidemiol. Evaluation of hospital resources for perinatal assistance About the Authors 2004;7(1):22–35. At the time of the study, Sônia Lansky was with the De- in Brazil [in Portuguese]. Rev Saude Publica. 2004; 26.Wen SW, Lei H, Kramer MS, Sauve R. Determi- partment of Society, Human Development and Health, 38(5):701–708. nants of intrapartum fetal deaths in a remote and indi- Harvard School of Public Health, Boston, Mass; the Fed- 10. Bryce J, Boschi-Pinto C, Shibuya K, Black RE, gent population in China. J Perinatol. 2004;24:77–81. eral University of Minas Gerais, Brazil; and the City WHO Child Health Epidemiology Reference Group. Health Department, Belo Horizonte, Brazil. Elisabeth WHO estimates of the causes of death in children. 27.Pattinson RC. Challenges in saving babies— França is with the Department of Social and Preventive Lancet. 2005;365:1147–1152. avoidable factors, missed opportunities and substan- Medicine, Federal University of Minas Gerais, Brazil. 11.Lawn JE, Cousens S, Zupan J, Lancet Neonatal dard care in perinatal deaths in South Africa. S Afr Ichiro Kawachi is with the Department of Society, Human Survival Steering Team. 4 million neonatal deaths: Med J. 2003;93(6):450–455. Development and Health, Harvard School of Public Health, when? Where? Why? Lancet. 2005;365:891–900. 28. Azad K, Abdullah AH, Nahar N. Use of Wig- Boston. glesworth classification for the assessment of perinatal Requests for reprints should be sent to Sônia Lansky, 12. Lawn J, Shibuya K, Stein C. No cry at birth: global mortality in Bangladesh—a preliminary study. MD, PhD, Avenida Afonso Pena 2336 5o andar, Belo estimates of intrapartum stillbirths and intrapartum- Bangladesh Med Res Counc Bull. 2003;29:38–47. Horizonte, Minas Gerais 30130 007, Brazil (e-mail: related neonatal deaths. Bull World Health Organ. 2005; [email protected]). 83(6):409–417. 29. Barros FC, Vaughan JP, Victora CG. Why so many This article was accepted July 12, 2006. 13.Wise P. Disparities in infant mortality. Annu Rev cesarean sections? The need for a further policy Public Health. 2003;24:341–362. change in Brazil. Health Policy Plan. 1986;1:19–29. Contributors 14 . Lansky S, Franca E, Comini CC, Neto LMC, Leal MC. 30. Barros FC, Victora CG, Barros AJD, et al. The S. Lansky originated the study and supervised all as- Perinatal deaths and health care evaluation in mater- challenge of reducing neonatal mortality in middle- pects of its implementation. E. França assisted with the nity hospitals of the Public Health System in Belo Hori- income countries: findings from three Brazilian birth study and analyses. I. Kawachi synthesized analyses zonte, Brazil, 1999 [in Portuguese]. Cad Saude Publica. cohorts in 1982, 1993 and 2004. Lancet. 2005; and writing. All authors helped to conceptualize ideas, 2006;22(1):117–128. 365(9462):847–854. interpret findings, and review drafts of the manuscript. 15.Porto D. Perinatal Health Commission of Belo 31.Victora CG, Vaughan JP, Barros FC, Silva AC, Horizonte, Brazil [in Portuguese]. In: Lotta GS, Barboza Tomasi E. Explaining trends in inequities: evidence Acknowledgments HB, Teixeira MCAC, Pinto V, eds. Twenty Experiences of from Brazilian child health studies. Lancet. 2000;356: This study was funded by the Pan American Health Public Policies and Citizenship. Rio de Janeiro, Brazil: 10 93–1098. Getúlio Vargas Foundation/BNDES/FORD Founda- Organization/World Health Organization (grant AMR/ 32. Gwatkin DR, Bhuiya A, Victora CG. Making tion; 2003:89–102. 99/078643-01) and CNPQ-Brazil (grant 200338/ health systems more equitable. Lancet. 2004;364: 2004-8). 16. Leal MC, Gama SGN, Cunha CB. Racial, sociode- 1273–1280. mographic and prenatal and childcare inequalities in Human Participant Protection Brazil, 1999–2001 [in Portuguese]. Rev Saude Publica. 33. Wise PH, Kotelchuck M, Wilson ML, Mills M. 2005;39(1):100–107. Racial and socioeconomic disparities in childhood mor- This study was approved by the institutional review tality in Boston. N Engl J Med. 19 85;313:360–366. board of the Federal University of Minas Gerais, Brazil 17.Rosa MLGR, Hortale VA. Avoidable perinatal (137/99). deaths and obstetric health care structure in the public 34. Daniels N, Kennedy B, Kawachi I. Justice is good health care system: a case study in a city in Greater for our health. In: Cohen J, Rogers J, eds. Is Inequality Bad for Our Health? Boston, Mass: Beacon Press; References Metropolitan Rio de Janeiro [in Portuguese]. Cad Saude Publica. 2000;16(3):773–783. 2000:3–33. 1. Saúde Brasil 2005—uma Análise da Situação de Saúde. Brasília, Brazil: Ministério da Saúde; 2004: 18. Silva AAM, Coimbra LC, Silva RA, et al. Perinatal 35. Andrade CLT, Szwarcwald CL, Gama SGN, Leal MC. 12 0–134. health and mother–child health care in the municipal- Socioeconomic inequalities and low birth weight and ity of Sao Luiz, Maranhao, Brazil [in Portuguese]. Cad perinatal mortality in Rio de Janeiro, Brazil [in Por- 2. Parto, Aborto e Puerpério—Assistência Humanizada Saude Publica. 2001;17(6):1413–1423. tuguese]. Cad Saude Publica. 2004;20(suppl 1):44–51. a Saúde. Brasília, Brazil: Ministério da Saúde; 2003: 17–25. 19. Leal MC, Gama SGN, Campos M. Factors associ- 36.Wise PH. Reconciling science and politics. Am ated with perinatal morbidity and mortality in a sam- J Prev Med. 19 93;9(6 suppl):7–16. 3. Neto OLM, Barros MBA. Risk factors for neonatal ple of public and private maternity centers in the City and post-neonatal mortality in the Central-West region 37. Victora CG, Wagstaff A, Schellenberg JA, Gwatkin D, of Rio de Janeiro, 1999–2001 [in Portuguese]. Cad of Brazil: linked use of life births and infant deaths rec- Claeson M, Habicht J. Applying an equity lens to child Saude Publica. 2004;20(S1):20–34. ords [in Portuguese]. Cad Saude Publica. 2000;16(2): health and mortality: more of the same is not enough. 477–485. 20. Almeida MF, Alencar GP, Novaes MHD, et al. Lancet. 2003;362(9379):233–241.

May 2007, Vol 97, No. 5 | American Journal of Public Health Lansky et al. | Peer Reviewed | Research and Practice | 873  RESEARCH AND PRACTICE 

Socioeconomic Position, Co-Occurrence of Behavior-Related Risk Factors, and Coronary Heart Disease: the Finnish Public Sector Study

| Mika Kivimäki, PhD, Debbie A. Lawlor, PhD, George Davey Smith, DSc, Anne Kouvonen, PhD, Marianna Virtanen, PhD, Marko Elovainio, PhD, Jussi Vahtera, MD

Coronary heart disease (CHD), a leading Objectives. We examined the associations between socioeconomic position, co- cause of morbidity and mortality in all West- occurrence of behavior-related risk factors, and the effect of these factors on the ern countries, is more prevalent among relative and absolute socioeconomic gradients in coronary heart disease. lower socioeconomic position (SEP) groups Methods. We obtained the socioeconomic position of 9337 men and 39 255 1–7 than among groups that have higher SEP. women who were local government employees aged 17–65 years from employ- Although the evidence of such a socioeco- ers’ records (the Public Sector Study, Finland). A questionnaire survey in nomic gradient in CHD is robust, the extent 2000–2002 was used to collect data about smoking, heavy alcohol consumption, to which this gradient is the result of differ- physical inactivity, obesity, and prevalence of coronary heart disease (myocardial ent distributions of coronary risk factors be- infarction or angina diagnosed by a doctor). tween SEP groups remains controversial. Results. The age-adjusted odds of coronary heart disease were 2.1–2.2 times higher for low-income groups than high-income groups for both men and women, Several epidemiological studies suggested and adjustment for risk factors attenuated these associations by 13%–29%. There that most (60%–95%) of the CHD burden was no further attenuation with additional adjustment for the number of co- can be attributed to established risk factors: occurring risk factors, although socioeconomic disadvantage was associated with smoking, hypertension, diabetes, unfavorable the co-occurrence of multiple risk factors. The absolute difference in coronary cholesterol profile, and physical inactivity; heart disease risk between socioeconomic groups could not be attributed to the appropriately, public health interventions measured risk factors. target these risk factors to reduce the CHD Conclusions. Interventions to reduce adult behavior-related risk factors may epidemic.8–13 However, several studies that not completely remove socioeconomic differences in relative or absolute coro- compare the magnitude of the socioeco- nary heart disease risk, although they would lessen these effects. (Am J Public nomic gradient before and after adjustment Health. 2007;97:874–879. doi:10.2105/AJPH.2005.078691) for these risk factors suggest that they explain only 15%–40% of the association be- might result from the failure of such models to Multivariable adjustment may underesti- tween SEP and CHD.7–14 Thus, the contra- fully account for clustering of individual risk mate the effect of established risk factors diction: most of the population burden of factors. The overall effect of the risk factors when explaining most of the “excess” cases CHD can be attributed to established risk would be underestimated if they were clus- among the lower SEP groups (i.e., the effect of factors, but these same risk factors only ex- tered (i.e., there was a greater than expected established risk factors on the absolute risk plain a small part of the association between number of persons with either no risk factors difference between SEP groups). A recent SEP and CHD. We raise the possibility that or many risk factors), and this clustering was study of 2682 Finnish men in the Kuopio Is- multivariable adjustment for risk factors may substantially more common in low- than high- chemic Heart Disease Risk Factor Study found not have correctly estimated the contribu- SEP groups.17 Underestimation can also occur that although adjustment for smoking, hyper- tion of these risk factors to SEP differences if the risk factors have synergistic effects (i.e., tension, dyslipidemia, and diabetes resulted in in the occurrence of CHD.12 ,15 ,16 the effect of combined risk factors exceeds a modest (24%) attenuation of the relative In the most commonly used approaches, the predicted effects from separate risk factors, socioeconomic gradient of CHD risk, these such as logistic regression and proportional which assumes independence within the par- same risk factors accounted for most (72%) of hazards regression, risk factors are entered into ticular multivariable model).18 In most studies, the absolute socioeconomic gradient, that is, the model to examine how they change the ef- the effect of synergism is examined by includ- the excess risk among those from the lowest fects of SEP indicators on CHD.14 Any change ing interaction terms, but then removing them SEP compared with those in the highest.19 in the effect estimate of SEP’s effects on CHD from the final model if the associated P value The role of behavior-related risk factors in after adjustment for these risk factors is then is large (conventionally >.05). However, most CHD, particularly those that might be modi- used as an indication of the extent to which studies have a limited ability to detect multiple fied through health promotion, is likely to lead the risk factors “explain” the relative socioeco- statistical interactions, and very large data sets to important policy implications. We used a nomic gradient in CHD. Underestimation are required to examine this possibility. large employee sample of people who were

874 | Research and Practice | Peer Reviewed | Kivimäki et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

participating in the Finnish Public Sector in the surveys. We requested the participants’ regression analyses for each of the risk factors Study20,21 to examine the associations between smoking status and the habitual frequency and examined individually as described above. Fi- SEP, co-occurrence of behavior-related risk amount of beer, wine, and spirits intake. Re- nally, we tested the extent of clustered risk fac- factors (smoking, physical inactivity, obesity, sponses to the alcohol questions were trans- tors in the whole population and within each and heavy alcohol consumption),8,11,22–25 and formed into units of alcohol per week.28 Binge SEP group. The expected frequencies were the effect of these factors on the relative and drinking was determined by requesting those predicted given the prevalence of the absolute socioeconomic gradients in CHD. whether the participant had passed out as a risk factors within each SEP group (i.e., assum- result of alcohol consumption more than once ing independence or no clustering). Clustering METHODS during the past 12 months. Physical activity is indicated when individuals are more likely was measured by the metabolic equivalent task to have no or many risk factors and are less The Finnish Public Sector Study focused index29 and was expressed as the sum score of likely to have a single risk factor than would on all local government employees of 10 metabolic equivalent task-hours per day (h/d). be expected if the risk factors were indepen- towns and all employees in 21 public hospi- Self-reported weight and height were used to dent (i.e., the observed-to-expected ratio is >1 tals that provided specialized health care in measure body mass index (kg/m2). for no risk factors, <1 for a single risk factor, the districts where the towns are located.20,21 CHD was measured by using a self- and >1 for 3 and 4 factors). We calculated χ2 These employees cover a wide range of SEPs, administered checklist of common chronic statistics to test the difference in the distribu- from city mayors to semiskilled cleaners. The diseases.30 For each disease, the respondent tions of observed and expected counts within largest groups were nurses and teachers. A was asked to indicate whether or not a physi- each occupational group. total of 48592 (9337 men, 39255 women), cian had diagnosed him or her as having the The associations of SEP, risk factors, and aged 17–65 years responded to a question- disease. Prevalent CHD was determined by the number of co-occurring risk factors with naire survey in 2000–2002 (response 68%). affirmative responses for myocardial infarc- CHD were studied first with age-adjusted lo- Women were slightly overrepresented among tion or angina. The agreement between these gistic models. To estimate the contribution of the respondents (81% women) compared self-assessments and data from medical rec- the risk factors to the association between with eligible employees (n=70961, 76% ords has previously shown to be substantial for SEP and CHD, each risk factor, all their inter- women), but the differences in mean age myocardial infarction and angina (κ>0.70).30 actions on a multiplicative scale, and the (44.7 vs 44.0 years) and SEP (15% vs 17% We coded all of the risk factors as binary number of co-occurring risk factors were performing manual labor) were small. The variables (0 or 1). Risks were defined as ever a added to the model as covariates. We wanted gender and age of respondents were also rep- smoker (current or past smoking), heavy alco- to examine whether greater attenuation was resentative of Finnish public sector employees hol consumption (>21 units of alcohol per achieved if the risk factors were more finely (77% women; mean age 44.6 years).26 How- week or binge drinking),22,23 physical inactivity categorized and linear associations were not ever, the predominance of women did not (<2 metabolic equivalent task h/d),29 and obe- assumed. To do this, we split body mass correspond to the gender distribution of the sity (body mass index>30 kg/m2).25 The age- index, physical activity (metabolic equivalent Finnish general working population (48% fe- adjusted association between SEP and binary task h/d), and alcohol consumption (units per male; mean age 45.5 years).26 risk factors was estimated in a logistic regres- week) into fifths of their distributions and en- Occupational status, income, and education sion analysis that used upper nonmanual labor tered these into the regression model as 3 were used as indicators of SEP. We obtained employees, high-income, and tertiary educa- variables together with 4 indicator variables the participants’ occupational titles from the tion as the reference groups. We counted the that represented past smoking, current smok- employers’ records (1931 different titles)27 number of risk factors on the basis of these bi- ing, high alcohol consumption (>21 units per and used the occupational classification by nary variables. Thus, the participants with all 4 week), and binge drinking. Finally, we esti- Statistics Finland27 to classify individuals into risk factors had a score of 4; those with any 3 mated the absolute risk of CHD associated 3 categories on the basis of these titles: upper risk factors scored 3, and those with no risk with SEP in the whole population and in a nonmanual workers, lower nonmanual work- factors scored 0. We performed a multinomial low-risk group (anyone who was free of all of ers, and manual laborers. Average monthly logistic regression analysis to examine the asso- the measured risk factors) to determine how income figures for men and women were ob- ciation between SEP and co-occurrence of risk many excess cases among the lowest (com- tained by occupational title from Statistics factors (this analysis can assess associations pared with the highest) SEP group would be Finland,27 and the distribution was divided that have a categorical outcome variable, such removed if these risk factors were eliminated. into thirds separately for the men and women as ours).31 The multinomial models were used The analyses were performed separately (referred to as high-, intermediate-, and low- to assess the likelihood of having 1 risk factor, for men and women and for each SEP indica- income groups). Educational level was self- 2 risk factors, and 3 or 4 risk factors versus tor using SAS 8.2 (SAS Institute Inc, Cary, reported in the surveys and was categorized having no risk factors (the reference). The cor- NC) software. The findings were consistent as primary or secondary versus tertiary. responding age-adjusted odds ratios were cal- across all of the 3 SEP indicators, so we re- We assessed 4 behavior-related risk factors culated for the levels of SEP by using the same ported full results for income only (indicator by using standard questionnaire measurements reference groups that were used in the logistic with evenly distributed categories) and have

May 2007, Vol 97, No. 5 | American Journal of Public Health Kivimäki et al. | Peer Reviewed | Research and Practice | 875  RESEARCH AND PRACTICE 

summarized the main findings for occupa- 2.4–3.3 times greater for the men and factors were replaced with more finely cate- tional status and education in the text. women who had 3 or 4 risk factors than for gorized risk variables). No further attenuation those who had no risk factors. Statistical tests in the relative socioeconomic gradient was RESULTS for interaction terms across the risk factors re- found when we used a model that included sulted in no strong evidence of synergism be- both individual risk factors and a score that Information was missing on income for tween the risk factors (P for all interaction represented the total number of risk factors 2227 (5%) of the participants, on any risk terms≥.14). as covariates. Moreover, we used a backward factor for 4044 (8%) of the participants, and Table 1 presents the age-adjusted associa- elimination approach that removed all of on CHD for 4700 (10%) of the participants. tion between SEP and the co-occurrence of the 4-, 3-, and 2-way interaction terms with A total of 39631 employees (82% of all of risk factors. Having a low SEP, compared with P >.05 from a saturated model that included the respondents) had full data for all of these a high SEP, was associated with 1.5–1.7 times all of the risk factors and their interaction variables. They differed slightly from partici- higher odds of having a single risk factor for terms. This approach led to a final model pants who had some missing data in terms of CHDversus having no risk factors. However, that contained no interaction terms, a further gender (81% vs 79% women), mean age SEP had a stronger association with having 3 indication that the interactions between the (44.2 vs 46.8 years), and SEP (15% vs 18% or 4 risk factors (odds ratios 2.4–3.3). The risk factors did not explain the association manual laborers). In spite of this, participants findings were similar when occupational status between SEP and CHD. who had complete data were representative or educational attainment (instead of income) To illustrate what might happen to CHD of all of the respondents (81% women, mean were used as the explanatory SEP variables. risk and absolute SEP gradient if risk factors age 44.7 years, 15% manual). In Table 2, the expected and observed were removed from the population, we In men and women, 31%–45% had no numbers of participants who had 0, 1, 2, and formed a low-risk subgroup that consisted of risk factors, 53%–63% had 1 or 2 risk fac- 3 or 4 risk factors show that the risk factors all the employees who had none of the mea- tors, and ≤1% had all 4 risk factors. There were clustered within all of the SEP groups sured risk factors (Table 4). Comparing this were graded associations between SEP and and that the clustering pattern was similar in subgroup to the entire population suggests each risk factor (except for heavy alcohol con- the groups. These findings were also repli- that CHD risk would have been reduced in sumption). For both genders, the highest risk cated with other SEP indicators. all SEP groups by 6%–48%. However, a occurred for individuals who had the lowest Table 3 shows the multivariable associa- marked socioeconomic gradient in absolute SEP. The risk factors were all positively asso- tion between SEP and CHD. For both gen- risk remained in the subgroup that was free ciated with CHD risk among both genders ders, a simple adjustment for each risk factor of the measured risk factors. The analyses (odds ratios between 1.1 and 2.4). The one entered into the model simultaneously as with other SEP indicators replicated these exception was heavy alcohol consumption, single covariates resulted in a 13%–29% re- findings. which was not associated with an increased duction in the relative SEP gradient of CHD incidence of CHD. The odds of CHD were (20%–23% reduction after the binary risk DISCUSSION

Evidence from a large contemporary popu- TABLE 1—Age-Adjusted Multinomial Regression Models for Risk Factor Clusters, by Income: lation suggests that SEP is associated with the the Finnish Public Sector Study, 2000-2002 co-occurrence of behavior-related risk factors such as smoking, heavy alcohol consumption, Odds Ratio (95% CI) physical inactivity, and obesity. Men and Gender No. 1 vs 0 2 vs 0 3–4 vs 0 women who have low SEP tend to have and Income Participants Risk Factors Risk Factors Risk Factors multiple risk factors more often than those Men who have higher SEP. Although these risk High 2742 1.00 1.00 1.00 factors were clustered in the total sample, no Intermediate 2623 1.48 (1.29, 1.68) 1.67 (1.43, 1.94) 2.33 (1.86, 2.92) evidence was found that clustering was more Low 2644 1.74 (1.52, 1.99) 2.32 (1.99, 2.70) 3.32 (2.66, 4.14) common in low-SEP groups or that the risk Women clusters had a synergistic effect on reported High 12 031 1.00 1.00 1.00 CHD. The effect of adjusting for risk factors Intermediate 11 002 1.16 (1.10, 1.23) 1.28 (1.17, 1.39) 1.59 (1.34, 1.89) on the relative socioeconomic gradient of Low 11 482 1.47 (1.39, 1.55) 2.09 (1.93, 2.26) 2.39 (2.03, 2.82) CHD was to produce a modest (13%–29%) reduction, which is a similar magnitude to re- Note. CI = confidence interval. Risk factors are ex- or current smoking (prevalence for men and women, 47% and 33%), heavy duction found in several other studies.7,14,19 alcohol consumption (defined as >21 units of alcohol per week or binge drinking; 25% and 6%), physical inactivity (27% and 25%), and obesity (body mass index >30 kg/m2; 13% and 11%). Participants with no missing values for any of the risk Furthermore, when we examined absolute factors were included in these models. risk, we found that by removing the behavior- related risk factors, important reductions in

876 | Research and Practice | Peer Reviewed | Kivimäki et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 2—Distributions of Observed (No.) and Expected Numbers (Exp No.) of Participants terms or including both individual risk fac- and the Ratio to Each Other, by Income: the Finnish Public Sector Study, 2000–2002 tors and a score for the number of risk factors) did not result in greater attenuation of High Income Intermediate Income Low Income the relative socioeconomic gradient in CHD Gender and Exp Exp Exp than did simple adjustment for each individ- a a a No. of Risk Factors No. No. Ratio No. No. Ratio No. No. Ratio ual covariate. Men We examined absolute differences in the 0 1028 882 1.16 777 641 1.21 654 532 1.23 prevalence of CHD within a group that was 1 983 1197 0.82 997 1170 0.85 986 1142 0.86 free of all measured behavior-related risk 2 566 554 1.02 615 658 0.94 721 757 0.95 factors in order to illustrate the expected ef- 3–4 165 109 1.52 234 154 1.52 283 283 1.33 fects of removing these risk factors from the Women population. Our findings confirmed the mod- 0 6048 5796 1.04 5015 4741 1.06 4367 4081 1.07 est contribution of these risk factors to the 1 4428 4812 0.92 4272 4653 0.92 4695 5120 0.92 absolute socioeconomic gradient. The low- 2 1318 1293 1.02 1400 1442 0.97 2008 2000 1.00 risk population had a lower prevalence of 3–4 237 130 1.82 315 166 1.90 416 285 1.46 CHD throughout the entire social hierarchy, but the absolute (and relative) socioeco- χ2 Note. P values were the same (P<.001) across all income levels for both genders. P values were on the basis of test with nomic gradient in risk remained. This find- 3 degrees of freedom testing the null hypothesis of no differences in the observed and expected frequencies across all of the number of risk factor categories.Within all of the income groups, the risk factors were clustered with a greater-than-expected ing is in contrast to that of the Kuopio Ische- number of participants who had no risk factors, a lower-than-expected number who had 1 risk factor, and a greater-than- mic Heart Disease Risk Factor study, in expected number who had 3 or 4 risk factors among the men and women. which the absolute socioeconomic gradient aGiven the prevalence of the risk factors within each group and on the basis of the assumption that all of the risk factors were independent of each other. largely disappeared in a subgroup that was free of measured risk factors.19 A potential reason for the discrepancy between these studies involves the selection of risk factors. the prevalence of CHD in all SEP groups adults, aged 18–24 years, reported that Three risk factors: hypertension, dyslipi- would result without removing the socioeco- clustering of behavior-related risk factors demia, and diabetes, which were included in nomic gradient. These findings were replica- was more common among participants who the Kuopio study but not in our study, are ble across the 3 SEP indicators of income, oc- had a history of unemployment and less physiological markers of the underlying cupational status, and education. Thus, common among students.33 Thus, clustering pathophysiological processes that end in according to this study, behavior-related risk of risk factors may be more common among manifest CHD. Only 5% of the CHD events factors do not explain a large part of either individuals who have lower SEP in adoles- occurred among the low-risk population in the relative or absolute socioeconomic gradi- cence and young adulthood, perhaps be- the Kuopio study. Because the elimination ent in prevalent CHD. cause, at these ages, peer pressure related to of these major disease mediators appears Few data have been published on the asso- the initiation or noninitiation of some risk to eliminate much of the absolute socio- ciation between SEP and the clustering of behaviors is very socially patterned. How- economic gradient in CHD, it is likely that risk factors, and those studies were on the ever, the socioeconomic gradient in cluster- any underlying factors, whether socio- basis of much smaller sample sizes than was ing does not appear to be retained in later economic, psychosocial, psychological, early used in our study. Consistent with our find- adulthood. life or genetic, may have their major influ- ing, an investigation of 2900 older women in In this large study, we had the adequate ence through these disease mediators. In our the United Kingdom found evidence of risk ability to test for statistical interactions be- study, with the exception of obesity, all of factor clustering in all of the SEP groups and tween behavior-related risk factors and their the measured risk factors were purely exoge- no evidence that the extent of the clustering effect on CHD, yet we found none for either nous, which reflects the lifestyle of the par- was greater among those from the manual- gender. Participants who had more behav- ticipants. Of the CHD cases, almost 30% labor SEP than among those from the non- ior-related risk factors had greater risk of had none of these behavior-related factors. manual ones.17 CHD, but this effect was consistent with the Therefore, it seems that several etiological By contrast, a study of 3600 Australian additive rather than synergistic effects of pathways to CHD that can be related to SEP adolescents found that both the co-occur- each risk factor. There was no evidence of remain uncovered by these more distal risk rence and clustering of smoking, high levels differences in risk factor clustering by SEP factors. of television watching, overweight, and high and no evidence that the risk factors com- blood pressure were more common in fami- bined synergistically. It is, therefore, not sur- Limitations lies that had a low SEP than in other fami- prising that we found that a more-complex We determined all risk factors from self- lies.32 Similarly, a study of 480 young adjustment (either including interaction reports. Although self-reported height and

May 2007, Vol 97, No. 5 | American Journal of Public Health Kivimäki et al. | Peer Reviewed | Research and Practice | 877  RESEARCH AND PRACTICE 

weight have been shown to be strongly corre- misreporting of weight is similar across SEP misreporting by social class could bias our related with direct measurement, obese individ- groups in our study, it would tend to dilute sults in either direction. uals who self-report tend to underestimate rather than exaggerate the magnitude of the The cross-sectional design of this study is their body mass index.33–36 If this systematic associations we observed. Any variation in open to reverse causality, healthy-worker bias, and survivor bias. If individuals who are diagnosed with disease change to adopt a TABLE 3—Logistic Models Adjusted for Age and Risk Factor, by Income: the Finnish Public healthier lifestyle, the associations of risk fac- Sector Study, 2000–2002 tors with CHD may be underestimated, and the extent to which these risk factors explain Odds Ratio (95% CI),Adjusted For socioeconomic gradients may also be under- Model B + Model B + No. estimated. However, the magnitude of the as- Income and Model A + Risk Interaction Terms Co-occurring Risk sociations that we found between SEP, be- Gradient Change No.a Cases Age (Model A) Factors (Model B) Between Risk Factors Factors (Model C) havior-related risk factors, and CHD are Men similar to those reported in prospective Income studies,37–43 which suggests that the cross- High 2719 54 1.00 1.00 1.00 1.00 sectional nature of the study did not result in Intermediate 2593 60 1.84 (1.25, 2.73) 1.59 (1.07, 2.36) 1.62 (1.09, 2.41) 1.60 (1.08, 2.38) a major bias. The only exception was heavy Low 2583 78 2.24 (1.55, 3.24) 1.88 (1.29, 2.74) 1.93 (1.32, 2.82) 1.91 (1.31, 2.79) alcohol consumption, which was not associ- Change in gradientb 0% –29.0% –25.0% –26.6% ated with CHD, even though it has predicted Women CHD events in several, though not all, pro- Income spective studies.37,40 Replication of our risk High 11 886 70 1.00 1.00 1.00 1.00 cluster analyses with a prospective investiga- Intermediate 10 903 102 1.57 (1.14, 2.16) 1.53 (1.11, 2.11) 1.54 (1.12, 2.12) 1.53 (1.11, 2.11) tion on incident CHD is important, but the Low 11 218 158 2.12 (1.57, 2.84) 1.98 (1.47, 2.67) 1.97 (1.47, 2.68) 1.98 (1.47, 2.67) challenge will be to achieve sufficient statisti- Change in gradientb 0% –12.5% –13.4% –12.5% cal power, which will require a very large cohort that is followed for many years. Note. CI = confidence interval. aIncludes participants who had no missing values for any of the risk factors. Our findings may not be generalized to bPercentage difference in the odds ratios for low income versus high income between the presented model and the age- other populations. However, a socioeconomic adjusted model. gradient in CHD of a similar magnitude has been reported throughout several different European and US populations. In addition, TABLE 4—Age-Adjusted Absolute and Excess Coronary Heart Disease Risk in the Entire the associations between behavior-related risk Sample and a Low-Risk Subsample, by Income: the Finnish Public Sector Study, 2000–2002 factors and CHD risk are similar across these different populations. Thus, it is likely that Men Women our findings could be generalized to most Riskb Excess Riskb Excess developed countries. (per Riskb (per (per Riskb (per Population and Incomea No.a Cases 10000) 10000) No.a Cases 10000) 10000) Conclusions All participants This study has shown that smoking, heavy Income alcohol consumption, physical inactivity, and High 2589 51 132.5 0 11 417 66 62.3 0 obesity do not fully explain the socioeco- Intermediate 2472 57 266.5 134.0 10 334 91 88.0 25.7 nomic gradient in CHD. However, our data, Low 2419 73 334.1 201.6 10 400 140 129.7 67.4 along with previous data, have demonstrated Low risk group (no measured risk factors) that these behavior-related risk factors have Income some explanatory power. Therefore, strategies High 973 11 62.5 0 5753 18 35.8 0 aimed at reducing these risk factors would re- Intermediate 740 9 158.4 95.9 4719 35 73.6 37.8 duce CHD risk in the whole population and Low 609 13 249.6 187.1 3966 52 125.3 89.5 would also attenuate some of the socioeco- Change in gradientc –7.2% +32.8% nomic gradient. Although the more proximal mechanisms through which CHD risk is gen- aIncludes participants with no missing values for any of the risk factors. bAge adjusted. erated are well understood, the determination cPercentage difference in the excess risk for low income versus high income between the low-risk group and the total sample. of these factors (circulating lipid levels, blood pressure, and insulin resistance) is not fully

878 | Research and Practice | Peer Reviewed | Kivimäki et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

understood. Further research is needed to de- 7. Lantz PM, House JS, Lepkowski JM, Williams DR, 26. Statistics Finland Web site. Available at: http:// termine additional ways to eliminate socio- Mero RP, Chen J. Socioeconomic factors, health behav- www.stat.fi/index_en.html. Accessed March 1, 2006. iors, and mortality: results from a nationally representa- economic inequalities in CHD. 27. Classification of Occupations: Handbook 14. tive prospective study of US adults. JAMA. 19 9 8;279: Helsinki: Statistics Finland; 1987. 1703–1708. 28. Kaprio J, Koskenvuo M, Langinvainio H, Romanov K, 8. Marmot MG, Davey Smith G, Stansfeld S, et al. Sarna S, Rose RJ. Genetic influences on use and abuse About the Authors Health inequalities among British civil servants: the of alcohol: a study of 5638 adult Finnish twin broth- Mika Kivimäki is with the Department of Epidemiology and Whitehall II study. Lancet. 19 91;337:1387–1393. ers. Alcohol Clin Exp Res. 19 87;11:349–356. Public Health, University College London, London, En- 9. Mackenbach JP, Bos V, Andersen O, et al. Widen- 29. Kujala UM, Kaprio J, Sarna S, Koskenvuo M. Rela- gland. Debbie A. Lawlor and George Davey Smith are with ing socioeconomic inequalities in mortality in six Western tionship of leisure-time physical activity and mortality. the Department of Social Medicine, University of Bristol, European countries. Int J Epidemiol. 2003;32:830–837. JAMA. 1998;279:440–444. Bristol, England. Anne Kouvonen and Marko Elovainio are 10. Kaplan GA, Keil JE. Socioeconomic factors and with the Department of Psychology, University of Helsinki, 30. Haapanen N, Miilunpalo S, Pasanen M, Oja P, cardiovascular disease: a review of the literature. Circu- Vuori I. Agreement between questionnaire data and Helsinki, Finland. Marianna Virtanen and Jussi Vahtera lation. 19 93;88:1973–1998. are with Finnish Institute of Occupational Health, Helsinki. medical records of chronic diseases in middle-aged and 11.Adler NE, Boyce WT, Chesney MA, Folkman S, Requests for reprints should be sent to Dr. Mika elderly Finnish men and women. Am J Epidemiol. Syme SL. Socioeconomic inequalities in health. No Kivimäki, Department of Epidemiology and Public Health, 19 97;145:762–769. easy solution. JAMA. 19 93;269:3140–3145. University College London, 1-19 Torrington Place, Lon- 31. Hosmer DW, Lemeshow S. Applied Logistic Re- don WC1E 6BT, UK (e-mail: [email protected]). 12. Evans R, Barer M, Marmor T. Why Are Some Peo- gression. New York, NY: John Wiley & Sons; 1989. This article was accepted March 11, 2006. ple Healthy and Others Not? New York, NY: Aldine de 32. Lawlor DA, O’Callaghan MJ, Mamun AA, Wil- Gruyter; 1994. liams GM, Bor W, Najman JM. Socio-economic posi- Contributors 13. Macintyre S. The Black Report and beyond: what tion, cognitive function and clustering of cardiovascular are the issues? Soc Sci Med. 19 97;44:723–745. risk factors in adolescence: findings from the Mater- All of the authors contributed significantly to the origin University study of pregnancy and its outcomes. Psy- and design of the hypothesis, analysis and interpreta- 14 . Marmot MG, Bosma H, Hemingway H, Brunner E, chosom Med. 2005;67:862–868. tion of the data, and writing of the article. Stansfeld S. Contribution of job control and other risk factors to social variations in coronary heart disease in- 33. Raitakari OT, Leino M, Rakkonen K, et al. Clustering cidence. Lancet. 19 97;350:235–239. of risk habits in young adults. The Cardiovascular Risk in Acknowledgments 15. Davey Smith G, Shipley MJ. Confounding of occu- Young Finns Study. Am J Epidemiol. 19 95;142:36–44. The work presented in this article was supported by grants pation and smoking: its magnitude and consequences. 34.Rowland ML. Self-reported weight and height. Am from the Academy of Finland (projects 117604 and Soc Sci Med. 19 91;32:1297–1300. J Clin Nutr. 1990;52:1125–1133. 105195) and the participating towns and hospitals. The work of Debbie A. Lawlor was supported by a United 16. Begg CB. The search for cancer risk factors: when 35. Stevens J, Keil JE, Waid LR, Gazes PC. Accuracy Kingdom Department of Health, Career Scientist Award. can we stop looking? Am J Public Health. 2001;91: of current, 4-year, and 28-year self-reported body 360–364. weight in an elderly population. Am J Epidemiol. 1990; 132:1156–1163. Human Participation Protection 17. Ebrahim S, Montaner D, Lawlor DA. Clustering of risk factors and social class in childhood and adulthood 36. Lawlor DA, Taylor M, Bedford C, Ebrahim S. This study was conducted according to the guidelines in British women’s heart and health study: cross sec- Agreement between measured and self-reported weight of the Helsinki declaration, and the study protocol was tional analysis. BMJ. 2004;328:861–864. in older women. Results from the British Women’s Heart approved by the Ethics Committee of the Finnish Insti- and Health Study. Age & Ageing. 2001;31:16 9–174. tute of Occupational Health. 18. Thompson WD. Effect modification and the limits of biological inference from epidemiologic data. J Clin 37.Powell KE, Thompson PD, Caspersen CJ, Kendrick Epidemiol. 19 91;44:221–232. JS. Physical activity and the incidence of coronary heart References 19.Lynch JW, Davey Smith G, Harper S, Bainbridge K. disease. Annu Rev Public Health. 19 87;8:253–287. 1. Paffenbarger RS, Hyde RT, Wing AL, Lee I-M, Explaining the social gradient in coronary heart dis- 38. Rimm EB, Stampfer MJ, Giovannucci E, et al. Jung DL, Kampert JB. The association of changes in ease: comparing relative and absolute approaches. Body size and fat distribution as predictors of coronary physical-activity level and other lifestyle characteristics J Epidemiol Community Health. 2006;60:435–441. heart disease among middle-aged and older US men. with mortality among men. N Engl J Med. 19 93;328: 20. Kivimäki M, Virtanen M, Vartia M, Elovainio M, Am J Epidemiol. 19 95;141:1117–1127. 538–545. Vahtera J, Keltikangas-Järvinen L. Workplace bullying 39. Corrao G, Bagnardi V, Zambon A, La Vecchia C. 2. Stamler J, Stamler R, Neaton JD, et al. Low risk- and the risk of cardiovascular disease and depression. A meta-analysis of alcohol consumption and the risk factor profile and long-term cardiovascular and noncar- Occup Environ Med. 2003;60:779–783. of 15 diseases. Prev Med. 2004;38:613–619. diovascular mortality and life expectancy: findings for 5 large cohorts of young adult and middle-aged men 21.Vahtera J, Kivimäki M, Pentti J, et al. Organisa- 40. Lloyd-Jones DM, Wilson PW, Larson MG, et al. and women. JAMA. 1999;282:2012–2018. tional downsizing, sickness absence, and mortality: 10- Framingham risk score and prediction of lifetime risk for town prospective cohort study. BMJ. 2004;328:555. coronary heart disease. Am J Cardiol. 2004;94:20–24. 3. Yusuf S, Reddy S, Ounpuu S, Anand S. Global burden of cardiovascular diseases. Part II: variations in 22. Murray RP, Connett JE, Tyas SL, et al. Alcohol 41. Oguma Y, Shinoda-Tagawa T. Physical activity de- cardiovascular disease by specific ethnic groups and ge- volume, drinking pattern, and cardiovascular disease creases cardiovascular disease risk in women: review ographic regions and prevention strategies. Circulation. morbidity and mortality: is there a U-shaped function? and meta-analysis. Am J Prev Med. 2004;26:407–418. 2001;104:2855–2864. Am J Epidemiol. 2002;155:242–248. 42. Emberson JR, Shaper AG, Wannamethee SG, 4. Yusuf S, Hawken S, Ounpuu S, et al. Effect of po- 23.Pletcher MJ, Varosy P, Kiefe CI, Lewis CE, Sidney S, Morris RW, Whincup PH. Alcohol intake in middle age tentially modifiable risk factors associated with myocar- Hulley SB. Alcohol consumption, binge drinking, and and risk of cardiovascular disease and mortality: ac- dial infarction in 52 countries (the INTERHEART early coronary calcification: findings from the Coronary counting for intake variation over time. Am J Epidemiol. study): case-control study. Lancet. 2004;364:937–952. Artery Risk Development in Young Adults (CARDIA) 2005;161:856–863. 5. Magnus P, Beaglehole R. The real contribution of Study. Am J Epidemiol. 2005;161:423–433. 43. Kujala UM, Kaprio J, Sarna S, Koskenvuo M. the major risk factors to the coronary epidemics: time 24. Rimm EB, Klatsky A, Grobbee D, Stampfer MJ. Future hospital care in a population-based series of to end the “only-50%” myth. Arch Intern Med. 2001; Review of moderate alcohol consumption and reduced twin pairs discordant for physical activity behavior. 161:2657–2660. risk of coronary heart disease: is the effect due to beer, Am J Public Health. 1999;89:1869–1872. 6. Greenland P, Knoll MD, Stamler J, et al. Major wine, or spirits? BMJ. 1996;312:731–736. risk factors as antecedents of fatal and nonfatal coro- 25. Kopelman PG. Obesity as a medical problem. nary heart disease events. JAMA. 2003;290:891–897. Nature. 2000;404:635–643.

May 2007, Vol 97, No. 5 | American Journal of Public Health Kivimäki et al. | Peer Reviewed | Research and Practice | 879  RESEARCH AND PRACTICE 

Community Coalitions as a System: Effects of Network Change on Adoption of Evidence-Based Substance Abuse Prevention

| Thomas W. Valente, PhD, Chich Ping Chou, PhD, and Mary Ann Pentz, PhD

Community coalitions are often formed to Objectives. We examined the effect of community coalition network structure help communities mobilize resources and co- on the effectiveness of an intervention designed to accelerate the adoption of ordinate activities that improve the public’s evidence-based substance abuse prevention programs. 1–3 health. Conceivably, coalitions may con- Methods. At baseline, 24 cities were matched and randomly assigned to 3 contribute to all phases of health program deliv- ditions (control, satellite TV training, and training plus technical assistance). We ery, from planning to implementation and surveyed 415 community leaders at baseline and 406 at 18-month follow-up sustainability.4,5 Most important, however, about their attitudes and practices toward substance abuse prevention programs. may be the role of coalitions in assisting com- Network structure was measured by asking leaders whom in their coalition they munities with identifying, planning, and sub- turned to for advice about prevention programs. The outcome was a scale with sequently adopting effective health programs. 4 subscales: coalition function, planning, achievement of benchmarks, and progress in prevention activities. We used multiple linear regression and path In this regard, community coalitions may be analysis to test hypotheses. best served by the promotion of evidence- Results. Intervention had a significant effect on decreasing the density of based programs—those that have been sys- coalition networks. The change in density subsequently increased adoption of tematically evaluated and shown to be effec- evidence-based practices. tive in changing health-related behavior. One Conclusions. Optimal community network structures for the adoption of pub- area in which evidence-based standards and lic health programs are unknown, but it should not be assumed that increasing programs have been well articulated is drug network density or centralization are appropriate goals. Lower-density networks abuse prevention.6,7 Coalitions are particu- may be more efficient for organizing evidence-based prevention programs in larly important to the delivery of drug abuse communities. (Am J Public Health. 2007;97:880–886. doi:10.2105/AJPH.2005. prevention programs because coalitions in- 063644) clude constituents and prevention stakeholders from many perspectives.8 By bringing to- the adoption of health-related behaviors, such information to be circulated throughout the gether representatives from local government, as smoking17 ,18 and contraceptive use.19 , 2 0 coalition. Density also may facilitate diffusion, law enforcement, education, media, parent Network analysis also is used to study inter- because dense networks may reflect a cohe- groups, health agencies, and businesses, coali- organizational relations, because these rela- sive normative environment.3 A network with tions can provide a community forum for tions are believed to affect the delivery of many links is more likely to have members identifying, planning, and adopting prevention health services3,14 and are useful for creating who share common values or beliefs. Thus, a programs that would not otherwise be possi- community capacity.3,16 By adopting a network dense network may reflect a homogenous ble through the efforts of a single agency. perspective in the study of coalitions, we hope coalition, and this homogeneity will facilitate Several features of coalitions affect their to expand the potential of social network anal- information exchange and decisionmaking.20 performance.9 One factor is having a clearly ysis for measuring social capital.21–23 Additionally, centralized networks—those articulated structure in which subcommittees The field of network theory and analysis is with ties directed at 1 or a few members—are make decisions and assign tasks.10 ,11 Other fac- well established, but it has had little applica- expected to facilitate the adoption of evidence- tors include professional representation tion to prevention.24–26 Of the many different based programs. Centralized networks have (whether representatives from various profes- network indices, 2 may have the most poten- hubs that can disseminate information to sions are in the coalition), the variety of key tial for representing a coalition’s structure: many other members quickly. A centralized stakeholder roles represented, participation density and centralization. For example, stud- coalition has leaders who can enact decisions (i.e., the frequency with which members at- ies of the diffusion of innovations have shown more readily, because they have positions of tend meetings), and membership tenure.12 ,13 that network density and network centraliza- power and control.27 Moreover, once central Notably missing in the study of coalition effec- tion are positively associated with faster diffu- members in a centralized network adopt a tiveness is attention to the coalition’s commu- sion of innovations.26 Dense networks provide program, they are able to locate the right nication network, i.e., who is connected more pathways where communication about coalition members to implement that program. to whom and how those connections affect prevention programs can flow compared On the basis of these findings, we expect the outcomes.14–16 Social network analysis has with sparse networks. Conversely, sparse net- adoption of evidence-based practices to be shown how social network properties affect works may not provide enough pathways for greater among dense coalitions than among

880 | Research and Practice | Peer Reviewed | Valente et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

sparse ones, and adoption should be greater previously not existed and by creating more de- Study Participants among centralized networks than among decentralized task work groups where there had Community leaders were identified and re- centralized ones. However, other structural been only a single group. At least 2 previous cruited through a process of snowball sam- characteristics, such as whether a coalition op- studies have shown that the achievement of pling,11 which included 3 criteria: (1) repre- erates as a single group or as multiple subcom- benchmarks is significantly associated with the senting 1 or more prevention stakeholders mittees, may mitigate these relationships.11 adoption of prevention plans.11, 3 9 We used net- (education, law enforcement, parent groups, In a highly structured coalition, denser network analysis methods to measure the coalition youth services, media, local government, busi- works (those with a high volume of connec- structures and explore how the dynamics of the ness, health or medical profession, special or tions) may not facilitate efficiency or progress.28 coalition system affected the coalition’s ability to minority interest group), (2) being—or having First, lower density within a network may re- implement drug abuse prevention programs. On the potential to be—a positive role model for flect more formal collaborations.29 Second, al- the basis of previous research, we hypothesized youth, and (3) willing to participate in a pre- though there is probably a minimum density that the intervention would increase the effi- vention coalition for 2 years. This sampling level within a network needed for coalitions to ciency of existing coalitions by decreasing the process resulted in a list of 1041 potential adopt innovations, once this level is reached— networks’ density and centralization, which in participants (39–179 per city). Among these particularly in structured coalitions—too much turn would positively affect the progress and respondents, 1 community leader in each city density may be a liability. Too much density adoption of prevention planning. was identified and trained annually to serve within a network can create communities with as a site facilitator for STEP, which included too few connections to external information METHODS organizing other leaders for training and and resources, thus making them disadvan- meetings, facilitating data collection, and col- taged.26,30,31 Finally, organizational studies have Our study was part of a larger trial— lecting archival data on meeting process. shown that too much density within a network STEP—that evaluated the dissemination of From the list of potential participants, site fa- can hurt performance.32,33 evidence-based drug prevention programs cilitators identified 709 individuals from the Similarly, networks that are too centralized to 24 cities. Small- to medium-size cities 24 cities who were considered to be active in concentrate power, which may result in less (populations=20000–104000) were re- terms of having attended at least 1 commu- shared decisionmaking and lower commit- cruited from Massachusetts, Colorado, nity or coalition meeting during the previous ment to prevention programs among noncen- Arkansas, Iowa, and Missouri to participate in 12 months. tral members. Centralized networks are re- a 5-year randomized trial. The selected cities Respondents completed both a community ferred to as hierarchical networks, and studies were considered underserved with regard to leader survey and a network survey. Of the have shown that employees in hierarchical or- drug prevention (i.e., few funds for prevention, 709 active leaders, 670 (94.5%) completed ganizations feel less satisfied with their no state incentive grants, and no evidence- either the community leader or network sur- work.34–36 Some researchers have advocated based programs). STEP used relatively low- vey at baseline, and 415 (58.5%) completed for decentralized or horizontal communication cost interactive up-and-down-link satellite tele- both; at 18-month follow-up, data were col- networks as being more appropriate for organ- vision training to deliver 6 evidence-based lected from 406 (57.3%) leaders, and 255 izations that use electronic communication prevention programs over a 3-year period. At leaders (36% of 709 active leaders at base- technology.37 Therefore, although a central- baseline, 67% of the cities had an existing line) had completed surveys at both waves ized network is more efficient,38 a decentral- coalition that ranged in longevity from 2 to of measurement. Thus, there were 821 re- ized one may be more empowering. Thus, the 25 years, 21% had created a prevention coali- spondents at baseline and follow-up, and 255 adoption of new programs may be facilitated tion specifically for STEP, and 12% had only respondents provided data at both waves. in sparser or more decentralized networks. an occasional grouping of community leaders. Four of the 24 communities at baseline Steps Toward Effective Prevention (STEP) dropped out of the study. was a large prevention diffusion trial that in- Research and Measurement Designs cluded a community coalition intervention com- Cities were matched with 2000 US Census Intervention ponent.4,6 We evaluated the effects of this inter- data on demographic variables associated The intervention programs consisted of 6 vention on changing the coalition’s network with risk for drug use (percentage of the pop- interactive televised training segments on evi- density and centralization. We also evaluated ulation that was male, younger than 18 years, dence-based prevention programs administered the mediating effects of network change White, or had income below the federal pov- approximately every 6 months; 3 of these train- (change in network density or centralization) on erty level). Matched cities were then assigned ing segments occurred during the period of our subsequent planning and adoption of evidence- within each state to 1 of 3 conditions: tele- study. Television broadcasts were comple- based prevention programs. The intervention vised prevention training plus technical assis- mented with planning meetings, where skills was designed to increase the efficiency of coali- tance, televised prevention training only, or learned in training were shared with other tion networks in planning and implementing prevention as usual (control). The data in our members who did not participate in the live evidence-based prevention programs by creat- study are from baseline (fall 2001) through broadcast training. Training moved from large ing an organized coalition where one had 18-month follow-up (spring 2003). introductory sessions to smaller audience

May 2007, Vol 97, No. 5 | American Journal of Public Health Valente et al. | Peer Reviewed | Research and Practice | 881  RESEARCH AND PRACTICE 

sessions that targeted those who would actively with whom they were friends. For each com- or follow-up; we used the simple unpaired implement prevention programs; 343 leaders munity, we calculated 2 network-level mea- t test for group comparisons on leader charac- participated in the first session, 196 partici- sures from the advice network using GAUSS teristics (role, tenure, and number of meetings pated in the second session, and 130 partici- software (Aptech Systems, Seattle, Wash). We attended in the last year), density and central- pated in the third session. The topics of the first first calculated density, ization, and planning and adoption outcomes. 3 sessions were (1) identifying risk factors and Finally, we conducted regression analyses l 49 protective factors of drug abuse, (2) organizing (1)D = () . using Stata software. The following model the community, and (3) understanding how to nn–1 was estimated, interact with local media using established where l is the number of links (nominations community approaches for communicating (3) Y =a+b Y +b Tx+b D +b D +e, made) and n is network size (number of 2 1 1 2 3 1 4 2 public health issues and information.40–43 coalition members). Density is determined where Y is 1 of the 5 outcomes in Table 1 at by counting the number of reported links 2 Measures wave 2 and Y is the same outcome at wave 1; and dividing by the maximum number of 1 We used data from 2 surveys. The first sur- Tx represents a treatment community; D and possible links. We also calculated degree 1 vey—the community leader survey—included 27 D represent network-level density at waves 1 centralization, 2 122 items that measured leader attitudes and and 2, respectively; and e is error. The com- ˆ () behaviors regarding community readiness for A DegreeMax– Degree i munity was the unit of analysis. We aggre- (2)C = i=1 , prevention program implementation, individ- D nn2 – 32+ gated the data 2 ways: for only those who ual leader skills and attitudes, and coalition completed both waves 1 and 2 (n=255), and functioning. We used measures of coalition where degree is the number of nominations for all respondents who completed either 50 functioning, planning, and adopting preven- received by each person and n is network wave 1 or wave 2 surveys (n=821). We tion programs. The outcomes consisted of 4 size. Degree centralization varies between first tested intervention effects on each net- scales: organizational functioning (sum of 5 zero and 1, with higher numbers indicating a work score. We then included network density items, 5-point scale from strongly disagree to more centralized network. Both measures and centralization at waves 1 and 2 to test the strongly agree; α=0.83; adapted from Com- are readily available in network analysis mediated effects of network change on change 47 munities United for Prevention21); data-based programs. in program planning and adoption. To in- planning (15 items, 4-point scale from not at crease power, we combined both STEP treat- all to a lot; α=0.87; adapted from Communi- Analysis Plan ment conditions (training and technical assis- ties That Care44,45 and Students Taught We conducted a confirmatory factor analy- tance and training only), which showed no 48 Awareness and Resistance46); benchmark sis with the EQS program to generate an differences in network or outcome measures. achievement (12 items, 4-point scale of overall prevention planning composite score— progress from none to completed; α=0.88; or second-order factor—on the basis of 4 sep- RESULTS adapted from Students Taught Awareness and arate planning scores. We then compared the Resistance39,46); and prevention activity analysis sample with other community lead- Table 1 shows network indicators and progress (14 items, 5-point scale of progress ers who were missing data at either baseline study outcomes for wave 1 and wave 2. The from none to activity completed; α=0.90; adapted from Communities That Care44,45 TABLE 1—Means (SD) for Coalition Network Indicators and Outcomes for Waves 1 and 2 and Students Taught Awareness and Resis- (n=821): STEP,2001–2003 tance13 ). Details of scale development and Wave 1, Mean (SD; Range) Wave 2, Mean (SD; Range) Unpaired tP measurement model analysis have been published elsewhere.11 These 4 scores were ana- Network Indicators lyzed separately and were aggregated to an Density 0.12 (0.06; 0.06–0.33) 0.15 (0.05; 0.05–0.25) 1.09 0.29a overall prevention planning and adoption Centralization 0.41 (0.14; 0.17–0.67) 0.37 (0.14; 0.16–0.61) 0.72 0.48a score as 1 outcome. Outcomes The second survey was the network survey Functioning 3.67 (0.73; 1–5) 3.73 (0.70; 1–5) 1.16 0.12b that had both a roster of all coalition mem- Planning 2.98 (0.55; 1–4) 3.04 (0.53; 1–4) 1.45 0.07b bers and a question that asked each member Achievement 2.11 (0.43; 1–3 2.20 (0.46; 1–3) 2.87 <.001b to name how frequently they talked with Progress 2.16 (0.52; 1–4) 2.13 (0.43; 1–4) 0.74 0.77b each other. The survey also had 3 open- All 4 standardized and combined –0.038 (0.78; –2.01–1.67) –0.037 (0.79; –1.58–1.41) 0.01 0.49b ended nomination questions that asked mem- Note. STEP=Steps Toward Effective Prevention. bers to list up to 7 people to whom they go a2-tailed. for advice about prevention issues, with b1-tailed. whom they discuss prevention issues, and

882 | Research and Practice | Peer Reviewed | Valente et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 2—Effects of Baseline, Baseline Density, and Follow-up Density on Community-Level the community, community-level perceptions Attitudes and Practices Regarding Adoption of Evidence-Based Substance Abuse were consistent over time. Additionally, base- Prevention Programs (n=20): STEP,2001–2003 line density was positively correlated with outcome change, which indicates that some Functioning Planning Achievement Progress Average Across basic level of interpersonal communication All Outcomes and connection is needed for coalitions to Panela function and perform adequately. Table 2 Baseline score .62* .58* .71* .66* .75* shows very similar regression coefficients Wave 1 network density .07 .35* .00 .31 .20 when data were treated cross-sectionally (in- Wave 2 network density –.44* –.44* .04 –.47* –.39* cluding all respondents) and when data were R 2 0.29 0.43 0.47 0.32 0.50 treated as a panel (including only those pres- Cross-sectional panelb ent at both time points). Results for centraliza- Baseline score .74* .57* .66* .65* .73* tion were not significant (data not shown). Wave 1 network density –.13 .48* .12 .54* .25 Figure 1 illustrates the change in the com- Wave 2 network density –.27* –.32* –.08 –.38* –.31* posite outcome by change in density. It also R 2 0.54 0.62 0.35 0.62 0.59 shows that density increased in the control communities; however, these communities Note. STEP=Steps Toward Effective Prevention. Regression model was controlled for the nonindependence of cases on their study condition. had lower increases in program adoption. aRespondents who completed both wave 1 and wave 2; n=255. This indicates that perhaps the absence of an b Respondents who completed either wave 1 or wave 2; n=821. intervention in the control condition led to in- *P<0.05. creased communication among members but no change in their ability to adopt programs, network indicators did not increase signifi- In both the panel (n=255) and cross- whereas an intervention did not change net- cantly between baseline and follow-up, and 2 sectional panel (n=821) results, baseline out- work density but did create an increase in the of the 5 outcomes increased significantly comes were strongly correlated with follow- adoption of evidence-based practices. overall (outcomes were not reported sepa- up outcomes. This suggests that even with a To test this hypothesis, we used path rately by study condition because these re- substantial portion of different individuals in analysis—estimating several regression models sults have been published elsewhere). We used the 1-tailed t test for the outcome changes because we expected the changes to increase. To test for main effects of the intervention, we conducted lagged regression of wave 2 outcomes on a dummy indicator for whether or not the coalition received satellite TV training. There was a significant association that indicated training improved outcomes (data not shown). Table 2 shows standardized regression coefficients for the wave 2 outcomes regressed on their baseline score, baseline density, and wave 2 density. For 3 of the outcomes and the combined score, wave 2 density was significantly and negatively associated with outcomes. Because baseline density was included in the model, the results indicate that density change was negatively associated with outcome change.50 For example, the coefficient for organizational functioning and density at follow-up was β=–0.44, which indicated that organizational function- Note. Coalitions are indicated by Control (0) or Intervention (1). ing was higher for coalitions that decreased FIGURE 1—Change in program adoption outcomes by change in network density with their density (or lower functioning for those Ordinary Least Squares Regression Estimate. with increased density).

May 2007, Vol 97, No. 5 | American Journal of Public Health Valente et al. | Peer Reviewed | Research and Practice | 883  RESEARCH AND PRACTICE 

other organizations that provide access to resources and power, which can be mobilized to adopt evidence-based practices. Too much density indicates that connections are directed within the group and do not provide sufficient pathways for information and behaviors to come from outside the group. Too much density leaves a coalition ineffective at mobilizing the resources it needs to adopt evidence-based prevention programs. To be sure, some density is necessary for the coalition to operate, but too much density can be counterproductive. Coalitions need to balance their efforts between creating a dense, cohesive group versus retaining some connections to outside resources. The association between coalition density and adoption of prevention programs may be Note. STEP=Steps Toward Effective Prevention. The model indicates that the Steps Toward Effective Prevention intervention time dependent. New coalitions might need decreased network density. Decreased network density was associated with increased program adoption. to move “from modest levels of collaboration FIGURE 2—Path model of the effects of treatment, baseline density, and wave 2 density on to increasingly dense and multiplex relation- community-level adoption of prevention programs. ships that can be used to address complex health problems.” 3(p658) Over time, however, this increasing density may become a liability, because it overly insulates the coalition from simultaneously—to test the interaction of the DISCUSSION new ideas or access to new resources. On the intervention with changes in density and out- other hand, communities that have had no comes (Figure 2). We used the EQS pro- Although our results are suggestive that coalition and that build one for the first time gram48 to calculate separate path models for simply increasing network communication or may require network density and centraliza- each outcome variable and the combined out- connectedness, or both, among coalition tion to get prevention planning moving. Coali- come. The model involved simultaneous esti- members will not result in improved adoption leaders must therefore be cognizant of mation of the following 2 equations, tion of evidenced-based practices, caution is the dynamic nature of coalition networks and warranted. First, the results are self-reported networks’ ability to address community con- (4) D =_ +_ Tx+_ D +_ Y +e and 2 1 11 12 1 13 1 1 attitudes and practices and may not reflect cerns, but they must not sacrifice adaptation actual program adoption. Second, network for cohesion. (5) Y =_ +_ Tx+_ D +_ D +_ Y +e , 2 2 21 22 1 23 2 24 1 2 measures depend considerably on the ques- Cohesion, shared mission and goals, and where Tx indicates a treatment community; D1 tion used to measure the network. In this common values are the hallmark of commu- 51 and D2 are network-level densities at wave 1 case, we asked community leaders to indicate nity coalitions. These factors may not trans- and wave 2, respectively; and Y1 and Y2 are to whom they went for advice about preven- late into the successful adoption of prevention the outcome variables measured at waves 1 tion. Other network questions may have elic- programs without leadership, however. That and 2, respectively. These path models are ited different network structures and perhaps leadership can be authoritarian or egalitarian, considered to be saturated models with perfect different results. which might be reflected in either centralized fit to the data. Figure 2 shows that the inter- More communication, in the absence of or decentralized network structures. Either vention was negatively associated with density promotions (satellite TV in this case) that pro- one might be more successful at adopting at wave 2, which indicates that being in the vide information about evidence-based pro- programs but for different reasons.52 We did control condition increased density. Results grams, does not lead to increased adoption not find support in this study for an associa- also indicate that wave 1 density was positively of evidence-based practices. Thus, the public tion between centralization and adoption, and associated with wave 2 density, and wave 1 health system needs to continue informing we hope future research will shed more light outcomes were positively associated with wave coalitions and community planners about on this relationship. 2 outcomes, as expected. Wave 2 density was evidence-based practices. The systems perspective prompted us to negatively associated with wave 2 outcomes, Our results are consistent with Granovet- measure the structure of interpersonal which indicates that decreasing density was ter’s strength of weak ties theory.31 Communi- interaction—who goes to whom for advice— associated with lower program adoption. ties that are less dense may have weak ties to rather than rely only on frequency of

884 | Research and Practice | Peer Reviewed | Valente et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

interactions as a measure of communication. 4. Pentz MA. Form follows function: designs for pre- 20.Rogers EM. Diffusion of Innovations. 5th ed. New The communication scale included in the sur- vention effectiveness and diffusion research. Prev Sci. York, NY: Free Press; 2003. 2004;5:23–29. vey did not change significantly between 21. Moore S, Shiell A, Hawe P, Haines V. The privi- leging of communitarian ideas: citation practices and baseline and follow-up and was not associ- 5. Saxe L, Reber E, Hallfors D, et al. Think globally, act locally: assessing the impact of community-based the translation of social capital into public health re- ated with adoption, density, or changes in substance abuse prevention. Eval Program Plann. 19 97; search. Am J Public Health. 2005;95:1330–1337. density. Thus, the system perspective we used 20:357–366. 22. Pearce N, Smith GD. Is social capital the key to uncovered significant network dynamics that 6. Pentz MA. Evidence-Based Prevention: Character- inequalities in health? Am J Public Health. 2003;93: were not apparent in individual reports. istics, impact, and future direction. J Psychoactive Drugs. 122–129. 2003;35(special suppl):143–152. The main finding from our study is that we 23. Lomas J. Social capital and health: implications for 7. Center for Substance Abuse Prevention (CSAP). public health and epidemiology. Soc Sci Med. 1998;47: should not assume increased communication CSAP’s prevention portal: model programs, 2002. 1181–1188. in the form of network density will always Available at: http://modelprograms.samhsa.gov/ 24. Scott J. Network Analysis: A Handbook. Newbury benefit coalition functioning. In this case, it template.cfm. Accessed May 16, 2005. Park, Calif: Sage; 2000. was associated with decreased ability to adopt 8. Hawkins DJ, Catalano RF, Arthur MW. Promoting 25. Wasserman S, Faust K. Social Networks Analysis: evidence-based programs. System-level think- science-based prevention in communities. Addictive Methods and Applications. Cambridge, UK: Cambridge Behav. 2002;27:951–976. University Press; 1994. ing and measures helped us reexamine naïve 9. Butterfoss FD, Goodman RM, Wandersman A. expectations about how community coalitions 26.Valente TW. Network Models of the Diffusion of Community coalitions for prevention and health pro- Innovations. Cresskill, NJ: Hampton Press; 1995. function and how to improve their capacity motion: factors predicting satisfaction, participation and 27.Freeman L. Centrality in social networks: concep- planning. Health Educ Q. 1996;23:65–79. for the adoption of programs that work. tual clarification. Soc Networks. 1979;1:215–239. 10. Hays CE, Hays S.Pl, DeVille JO, Mulhall PF. Ca- 28. Guimera R, Arenas A, Díaz-Guilera A, Giralt F. pacity for effectiveness: the relationship between coali- Dynamical properties of model communication net- tion structure and community impact. Eval Program About the Authors works. Physical Review E Stat Nonlin Soft Matter Phys. Plann. 2000;23:373–379. Thomas W. Valente, Chih Ping Chou, and Mary Ann Pentz 2002;66(2 pt 2):026704. are with the Institute for Prevention Research, Department 11. Jasuja GK, Chou CP, Bernstein K, Wang E, 29. Singer HH, Kegler MC. Assessing interorganiza- of Preventive Medicine, Keck School of Medicine, Univer- McClure M, Pentz MA. Using structural characteristics tional networks as a dimension of community capacity: sity of Southern California, Alhambra. of community coalitions to predict progress in adopting illustrations from a community intervention to prevent Requests for reprints should be sent to Thomas W. Valente, evidence-based prevention programs. Eval Program lead poisoning. Health Educ Behav. 2004;31:808–821. PhD, Keck School of Medicine, 1000 S Fremont Ave, Plann. 2005;28:173–184. Bldg A, Rm 5133, Alhambra CA 91803 (e-mail: tvalente@ 30. Gans H. The Urban Villagers: Group and Class in 12.Kegler M, Steckler A, McLeroy K, Malek SH. Fac- usc.edu). the Life of Italian-Americans. New York, NY: Free Press; tors that contribute to effective community health pro- This article was accepted November 2, 2005. 19 62. motion coalitions: a study of 10 Project ASSIST coalitions in North Carolina. Health Educ Res. 1998;13: 31.Granovetter M. The strength of weak ties. Am J 225–238. Sociol. 1973;78:1360–1380. Contributors 32. Oh H, Chung MH, Labianca G. Group social capi- 13. Mansergh G, Rohrbach L, Montgomery SB, T. W. Valente originated the study and conducted the tal and group effectiveness: the role of informal social- Pentz MA, Johnson CA. Process evaluation of commu- network analysis. C.P. Chou conducted the path analy- izing ties. Acad Manage J. 2004;47:860–875. sis. M.A. Pentz designed and implemented the STEP nity coalitions for alcohol and other drug prevention: trial. All authors wrote the article. comparison of two models. J Community Psychol. 1996; 33. Uzzi B. Social structure and competition in inter- 24:118–135. firm networks: the paradox of embeddedness. Admin Sci Q. 19 97;42:35–67. 14 .Kwait J, Valente TW, Celentano DD. Interorgani- Acknowledgments zational relationships among HIV/AIDS service organi- 34. Shaw ME. Communication networks. In: This study was funded by the National Institute on Drug zations in Baltimore: a network analysis. J Urban Berkowitz L, ed. Advances in Experimental Social Abuse (grants R01 DA012524 and P50 DA 16094). Health. 2001;78:468–487. Psychology. New York, NY: Academic Press; 1964: 111–147. 15. Stuart TE. Network positions and propensities to collaborate: an investigation of strategic alliance forma- 35. Roberts KH, O’Reilly III CA. Some correlates of Human Participant Protection tion in a high-technology industry. Admin Sci Q. 1998; communication roles in organizations. Acad Manage J. All procedures were reviewed and approved by the 43:668–698. 1979;22:42–57. University of Southern California institutional review 36. Flap H, Völker B. Goal specific social capital and board. 16.Wickizer T, Von Korff M, Cheadle A, et al. Acti- vating communities for health promotion: a process job satisfaction: effects of different types of networks evaluation method. Am J Public Health. 19 93;83: on instrumental and social aspects of work. Soc Net- References 122–129. works. 2001;23:297–320. 1. Institute of Medicine. Assuring the Health of the 17. Ennett ST, Bauman KE. Peer group structure and 37. Barabasi AL. Linked: How Everything Is Connected Public in the 21st Century. Washington, DC: National adolescent cigarette smoking: a social network analysis. to Everything Else and What It Means for Business, Sci- Academy Press; 2002. J Health Soc Behav. 19 93;34:226–236. ence and Everyday Life. New York, NY: Plume; 2003. 2. Berkowitz B, Wolff T. The Spirit of the Coalition. 18. Alexander C, Piazza M, Mekos D, Valente TW. 38. Malone T. The Future of Work: How the New Order Washington, DC: American Public Health Association; Peer networks and adolescent cigarette smoking: an of Business Will Shape Your Organization, Your Manage- 2000. analysis of the national longitudinal study of adolescent ment Style and Your Life. Boston, Mass: Harvard Busi- health. J Adolesc Health. 2001;29:22–30. ness School Press; 2004. 3. Provan KG, Nakama L, Veazie MA, Teufel-Shone NI, 39.Pentz MA. Community organization and school Huddleston C. Building community capacity around 19.Valente TW, Watkins S, Jato MN, Van der Straten A, liaisons: how to get programs started. J School Health. chronic disease services through a collaborative in- Tsitsol LM. Social network associations with contracep- 1986;56:382–388. terorganizational network. Health Educ Behav. 2003; tive use among Cameroonian women in voluntary as- 30:646–662. sociations. Soc Sci Med. 19 97;45:677–687. 40. Bandura A, Jeffrey RW, Wright CL. Efficacy of

May 2007, Vol 97, No. 5 | American Journal of Public Health Valente et al. | Peer Reviewed | Research and Practice | 885  RESEARCH AND PRACTICE 

participant modeling as a function of response induction aids. J Abnorm Psychol. 1974;83:56–64. Fighting Global Blindness 41. Hawkins JD, Catalano RF. Communities that Care: Action for Drug Abuse Prevention. San Francisco; Calif: Improving World Vision Through Cataract Elimination Jossey-Bass; 1992. By Sanduk Ruit, MD, Charles C.Wykoff, MD, D.Phil., MD, Geoffrey C.Tabin, MD 42. Pentz MA, Dwyer JH, MacKinnon DP, et al. A multi-community trial for primary prevention of adoles- noperated cataract is the cause of millions of cases cent drug abuse: effects on drug use prevalence. JAMA. of visual impairment and blindness in poor 1989;261:3259–3266. U populations throughout both the developing and the 43. Pentz MA, Mihalic SF, Grotpeter JK. The Midwest- developed world. This wonderfully written volume ern prevention project. In: Elliot DS, ed. Blueprints for shares the experiences of a team of surgeons who have Violence Prevention. Boulder, Colo: University of Colo- rado; 1997:3–43. demonstrated how the surgical procedures can be simplified and made more efficient, accessible, and far 44. Arthur MW, Hawkins DJ, Catalano RF, Olson JJ. less expensive. It is a step–by–step manual to solving Diffusion Project: Fall 1998 Community Key Informant Interview. Seattle, Wash: University of Washington; the problem where adequate surgeons can be trained 1998. to follow suit. Subject matter ranges from ways to increase demand among those who need surgery to the 45. Greenberg M, Osgood W. Technical report on the Community Evaluation Scales (CES). University Park, Pa: organization of surgical services, responsibilities among Pennsylvania State University; 2000. different personnel, efficient layout of clinical facilities, and how to sustain services at the least cost to those 46.Pentz MA, Valente TW. (1993). Project STAR: a substance abuse prevention campaign in Kansas City. who need it most. In: Becker TE, Rogers EM, eds. Organizational Aspects ISBN 0-87553-067-2 • spiral bound • 2006 of Health Communication Campaigns: What Works? $31.50 APHA Members • $45.00 Nonmembers Newbury Park, Calif: Sage Publications; 1993:37–60. 47. Borgatti SP, Everett MG, Freeman LC. UCINET American Public Health TO ORDER: Association for Windows: Software for Social Network Analysis. Har- web www.aphabookstore.org • vard, Mass: Analytic Technologies; 2004. 800 I Street, NW, Washington, DC 20001 email [email protected] • fax 888.361.APHA • 48. Bentler PM. EQS Structural Equations Program www.apha.org phone 888.320.APHA M-F 8am-5pm EST Manual. Encino, Calif: Multivariate Software, Inc; 19 95. 49. Intercooled STATA 8.0 for Windows. College Sta- tion, Tex: STATA; 2003. 50. Valente TW. Evaluating Health Promotion Pro- grams. New York, NY: Oxford University Press; 2002. 51. Goodman RM, Speers MA, McLeroy K, et al. Iden- tifying and defining the dimensions of community capacity to provide a basis for measurement. Health Educ Behav. 1998;25:258–278. 52. Kadushin C. Why it is so difficult to form effective coalitions. City Community. 2005;4:255–275.

886 | Research and Practice | Peer Reviewed | Valente et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Associations Between Lifestyle and Depressed Mood: Longitudinal Results From the Maastricht Aging Study

| Coen H. van Gool, PhD, Gertrudis I.J.M. Kempen, PhD, Hans Bosma, PhD, Martin P.J. van Boxtel, PhD, Jelle Jolles, PhD, and Jacques T.M. van Eijk, PhD

Depressed mood is presumed to be caused by Objectives. We examined whether healthy lifestyles are associated with ab- a variety of physical, psychological, and so- sence of depressed mood. 1 cioenvironmental factors. For example, un- Methods. A sample of 1169 adult participants in the Maastricht Aging Study pro- healthy lifestyles such as smoking, excessive vided baseline and 6-year follow-up data on smoking, alcohol use, physical ex- alcohol use, low levels of physical exercise, or ercise, body mass index, and mood. We examined associations between lifestyles being overweight or obese may provoke and depressed mood using longitudinal analyses controlling for baseline de- chronic diseases2,3 or worsen one’s health sta- pressive symptoms and covariates. tus over time.4 Chronic diseases frequently Results. Reports of excessive alcohol use at baseline predicted depressed mood coincide with increased symptoms of depres- at follow-up (relative risk [RR] = 2.48; 95% confidence interval [CI] = 1.08, 5.69), sion,5 and feelings of depression may in turn and reports of more than 30 minutes of physical exercise per day at baseline 6 were associated with an absence of depressed mood at follow-up (RR=0.52; 95% result in unhealthy lifestyles. Potentially in- CI=0.29, 0.92). Reports of being engaged in physical exercise throughout the 6-year tensifying this “downward spiral,” unhealthy follow-up period were also associated with absence of depressed mood (RR=0.56; lifestyles might elicit or exacerbate feelings 95% CI=0.34, 0.93). 7,8 of depression, and depression may subse- Conclusions. In this relatively healthy population sample, certain lifestyles either quently provoke or worsen the consequences predicted or protected against depressed mood. Adopting or maintaining healthy 9,10 associated with chronic diseases. However, lifestyles might be a starting point in preventing or treating depressed mood over it is not unequivocally clear how unhealthy time. (Am J Public Health. 2006;96:887–894. doi:10.2105/AJPH.2004.053199) lifestyles and the emergence of depressed mood (i.e., a clinically relevant level of depressive symptoms11 ) are associated over time. the emergence of depressed mood, this com- professional occupations and knowledge and Although research has consistently estab- mon and debilitating condition might be pre- experience required25); women and older indi- lished that there is a cross-sectional associa- vented or treated in the future through pro- viduals were oversampled to ensure adequate tion between smoking and depressed moting healthy lifestyles. We sought to representation of these groups in follow-up mood,12–14 little evidence is available regard- determine whether healthy lifestyles are asso- measurements. ing whether there is a longitudinal associa- ciated, over time, with absence of depressed Between 1993 and 1995 (baseline), 1823 tion, that is, whether smoking precedes or mood in the general population. respondents returned the questionnaire and follows depressed mood. It has been shown underwent cognitive and physical examina- that, in general, heavy alcohol use is associ- METHODS tions.26 Six years after completing their base- ated with depressed mood.15 ,16 Moreover, de- line assessments, these 1823 participants pressed mood is more often secondary to al- Design and Study Population were invited to take part in follow-up exami- coholism than primary (i.e., clinicians more We used data from the longitudinal Maas- nations. A total of 294 respondents refused often treat individuals with alcoholism who tricht Aging Study, an ongoing investigation further participation, 116 had died, and 37 have also developed depressed mood as a examining determinants of normal cognitive had been lost to follow-up; thus, 1376 partici- secondary reason for treatment than vice aging. Questionnaires were sent to 3449 indi- pants underwent reassessments. Ultimately, versa).15 Physical activity seems to help coun- viduals, aged 24 to 81 years, who were free of 1169 (33.9%) of the 3449 respondents were teract prevalent depressive symptoms and medical conditions that interfered with their included in our study sample (207 respon- protect against subsequent depression, but normal cognitive functioning at their entry into dents were lost to subsequent analyses owing longitudinal studies are necessary to further the study. This population was drawn from the to incomplete data on relevant variables at unravel this association.17 ,18 The relation be- Registration Network Family Practices,23,24 a baseline or follow-up). tween being overweight or obese and being primary care research sampling frame consist- depressed is controversial; different studies ing of 9919 individuals whose native language Measures have revealed negative, positive, and no asso- is Dutch. Our randomly recruited sample was Baseline depressive symptoms were as- ciations between these conditions.19–22 stratified according to general ability level (de- sessed with the self-report Symptom Checklist If healthy lifestyles are associated with the fined as level of occupational achievement, in- 90 Depression Scale.27,28 The 16 items on this absence of depressed mood or protect against cluding degree of complexity associated with instrument are rated in 5 categories ranging

May 2006, Vol 97, No. 5 | American Journal of Public Health van Gool et al. | Peer Reviewed | Research and Practice | 887  RESEARCH AND PRACTICE 

from no complaint (1) to maximal complaints Mean numbers of minutes spent daily on (junior vocational training), or high (senior vo- (5). Scores can range from 16 to 80, with physical exercise at baseline and follow-up cational or academic training). higher scores indicating a higher number of were computed on the basis of the number The questionnaire asked respondents to in- depressive symptoms. At follow-up, we used of hours participants reported spending each dicate whether they needed assistance in the the self-report Center for Epidemiologic Stud- week, on average, engaging in light activities following IADLs as a result of their physical ies Depression Scale (CES-D),29 an instrument such as ball sports, aerobic exercise, walking, condition: grocery shopping, housekeeping, that has demonstrated good psychometric and biking. These measures were used to preparing meals, maintaining personal hy- qualities in epidemiological studies involving group participants into the following cate- giene, and getting dressed. If respondents older populations.30 The CES-D’s 20 items gories: those not engaging in physical exercise answered no on these questions, their IADL are rated in 4 categories ranging from no (the reference category), those engaging in status was not considered to be impaired. If complaint (0) to maximal complaints (3). physical exercise for up to 30 minutes per they responded yes to 1 or more of these Scores can range from 0 to 60, again with day on average, and those engaging in physi- items, their IADL status was considered im- higher scores indicating a higher number of cal exercise for more than 30 minutes per paired. Finally, in an interview conducted by depressive symptoms. We used a CES-D day on average. Transitions in physical exer- a trained research assistant, respondents were threshold score of 16 or above in screening cise over time were categorized as (1) respon- given the opportunity to indicate whether a for depressed mood.29 Strong correlations be- dent still engages in physical exercise, (2) respon- medical doctor had ever diagnosed them with tween these 2 depression instruments have dent initiated physical exercise, (3) respondent 1 or more of 37 chronic diseases. been found,31 and the predictive validity of discontinued physical exercise, and (4) re- both scales has been reported elsewhere.32 spondent still does not engage in physical ex- Statistical Analyses On the basis of respondents’ reports of ercise (reference category). After comparing individuals who did not their current and former smoking behavior, Members of the study staff assessed re- take part in the study or were lost to follow- they were grouped into the following cate- spondents’ body weight and height at base- up with study participants (using t tests and gories at baseline and follow-up: current line and follow-up. We used body mass χ2 analyses), analyzing the characteristics of smoker (the reference category), former index (weight in kilograms divided by height the sample at baseline, and comparing study smoker, and never smoker. Transitions in in meters squared) cutoff scores of 27.8 variables at baseline and follow-up, we exam- smoking behavior over time were categorized or above for men and 27.3 or above for ined associations between sociodemographic as (1) respondent still does not smoke, (2) re- women to distinguish between respondents variables and baseline lifestyle domains and spondent quit smoking, (3) respondent initi- who were (at least) overweight (the refer- follow-up depressive symptoms. In addition, ated smoking, and (4) respondent still smokes ence category) and those who were not.33 we examined associations between transitions (reference category). These values corresponded with weights in lifestyle domains and depressed mood at Mean alcohol consumption at baseline and that were roughly 20% or more above the follow-up. We used cross-sectional and longi- follow-up was calculated according to partici- desired weights listed in the 1983 Metropol- tudinal techniques in our analyses, specifi- pants’ reports of the number of glasses of al- itan Life Insurance Company tables. Transi- cally paired-samples t tests, χ2 analyses, and cohol (representing approximately 10 g of al- tions in overweight status over time were McNemar tests; analyses of variance; and cohol in conformance with the unit of alcohol categorized as (1) respondent is still not multivariate logistic regression models in system) they drank per day on average (more overweight, (2) respondent is no longer which transitions in the various lifestyle do- than 10 glasses, between 7 and 10 glasses, overweight, (3) respondent became over- mains were independent variables and follow- between 3 and 6 glasses, 1 or 2 glasses, weight, and (4) respondent is still over- up depressed mood was the outcome vari- none) and the average number of days per weight (reference category). able. SPSS software (SPSS Inc, Chicago, Ill) week they consumed alcohol (every day, 5 or Previous research has shown that age, gen- was used in analyzing the data. The un- 6 days, 3 or 4 days, 1 or 2 days, less than 1 der, marital status, educational level, ability to healthy lifestyle components served as refer- day). These measures were used to group engage in instrumental activities of daily liv- ence categories, and longitudinal analyses participants into the following categories: ing (IADLs), and chronic disease are associ- were adjusted for baseline depressive symp- nondrinkers (the reference category), those ated with both lifestyle34–36 and depres- toms and covariates. consuming up to 2 drinks per day on aver- sion.10 , 3 7–39 We included these variables in age, and those consuming 3 or more drinks our analyses as covariates. Data on marital RESULTS per day on average (excessive alcohol use). status, age, gender, and educational level Transitions in alcohol use over time were were obtained from the questionnaire com- As can be seen in Table 1, attrition analy- categorized as (1) respondent still drinks pleted by the respondents. Marital status was ses at baseline demonstrated that the 2280 alcohol, (2) respondent initiated alcohol use, categorized as widowed, not married or no individuals who either were lost to follow-up (3) respondent quit drinking alcohol, and longer married, and married or living with a (n = 654) or did not take part in the study (4) respondent still does not drink alcohol partner. Educational level was categorized as (n = 1626) were significantly older (mean = (reference category). low (primary education at most), intermediate 55.1 years, SD = 17.6) than the 1169 study

888 | Research and Practice | Peer Reviewed | van Gool et al. American Journal of Public Health | May 2006, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 1—Baseline Characteristics of Sample Participants and Comparisons on Key respondents (mean = 48.9 years, SD= 14.7). Variables at Baseline and Follow-Up: Maastricht Aging Study In addition, they were more likely to be female (55.8% vs 47.6%) or widowed (12.3% Nonrespondent vs 5.0%), to be at low levels of education or Lost to Follow-Up Sample (n=1169) (51.6% vs 37.2%), to report impairments in (n=2280a) Baselineb Follow-Upc IADLs (20.9% vs 7.4%), and to be over- Age group, y, % weight (39.8% vs 32.9%). Finally, they re- 24–44 32.2 41.9*** . . . ported more chronic diseases (mean = 1.9, 45–64 32.3 40.7 . . . SD=1.8, and mean = 1.3, SD = 1.3, respec- 65–81 35.5 17.4 . . . tively), fewer minutes of physical activity per Mean age (continuous), y (SD) 55.1 (17.6) 48.9 (14.7)*** . . . day (mean = 13.6, SD = 27.3, and mean = Gender, % 18.2, SD = 28.3, respectively), and more Male 44.2 52.4*** . . . symptoms of depression (mean = 22.0, SD= Female 55.8 47.6 . . . 7.9, and mean = 20.5, SD = 6.1, respectively). Marital status, % Table 1 also presents data on lifestyle at Married/living together 74.2 81.3*** . . . follow-up. Between baseline and follow-up, the percentage of respondents who reported Not married/no longer married 13.5 13.7 . . . smoking decreased significantly (from Widowed 12.3 5.0 . . . 26.7% to 22.0%). The percentage of re- Education level, %d spondents who reported not drinking alcohol High 21.6 28.8*** . . . at all increased significantly during this pe- Intermediate 26.8 34.0 . . . riod (from 13.8% to 15.5%; Table 1), as did Low 51.6 37.2 . . . the percentage reporting that they consumed IADL status, % 3 or more drinks per day on average (from Not impaired 79.9 92.6*** . . . 6.8% to 8.3%). Impaired 20.9 7.4 . . . The percentage of respondents who re- No. of chronic diseases, % ported not engaging in physical exercise in- None 23.6 33.4*** . . . creased significantly from 49.4% at baseline 1 27.3 33.8 . . . to 62.8% at follow-up. In addition, average 2 or more 49.1 32.8 . . . number of minutes spent daily on physical Mean no. of chronic diseases (continuous) (SD) 1.9 (1.8) 1.3 (1.3)*** . . . exercise decreased significantly from 18.2 Smoking status, % (SD = 28.3) at baseline to 10.6 (SD = 20.5) Never smoker 34.4 35.0 37.6*** at follow-up. Not only did the average body Former smoker 36.6 38.3 40.4 mass index in the sample exhibit a significant Current smoker 29.0 26.7 22.0 increase from 26.5 kg/m2 (SD=4.1) to 2 Average daily alcohol intake, % 27.0 kg/m (SD=4.2) between baseline and None 24.0 13.8*** 15.5* follow-up, the percentage of overweight re- Up to 2 drinks 68.9 79.4 76.2 spondents also increased significantly (from 3 or more drinks 7.1 6.8 8.3 32.9% at baseline to 39.6% at follow-up). Mean no. of drinks per week (continuous) (SD) 6.1 (8.9) 6.1 (9.5) 6.2 (8.9) Finally, 14.0% of the respondents had a score Average amount of time spent daily on physical exercise, % of 16 or above (the threshold score for depressed mood) on the CES-D Scale at follow- More than 30 min 15.6 22.2*** 12.1*** up (Table 1). Up to 30 min 23.3 28.4 25.1 Results of univariate longitudinal analyses None 61.1 49.4 62.8 focusing on follow-up depressive symptoms, Mean no. of minutes of physical exercise per day 13.6 (27.3) 18.2 (28.3)*** 10.6 (20.5)*** stratified according to baseline sociodemo- (continuous) (SD) graphic variables and lifestyle domains, are Overweight, % shown in Table 2. Respondents aged 65 No 60.2 67.1** 60.4*** through 81 years had higher follow-up mean Yes 39.8 32.9 39.6 depression scores than respondents in the Mean body mass index (continuous) (SD) 27.1 (4.4) 26.5 (4.1)** 27.0 (4.2)*** other age categories. Also, women had sig- Continued nificantly higher mean depression scores at follow-up than men, and respondents at low

May 2006, Vol 97, No. 5 | American Journal of Public Health van Gool et al. | Peer Reviewed | Research and Practice | 889  RESEARCH AND PRACTICE 

TABLE 1—Continued respondents who reported engaging in physical exercise at baseline as well as at follow-up Depressive symptomatology decreased their risk of being depressed at Mean Symptom Checklist 90 score (continuous) (SD) 22.0 (7.9) 20.5 (6.1)*** . . . follow-up by 44% compared with respon- Mean CES-D score (continuous) (SD) ...... 7.9 (6.6) dents who did not engage in physical exercise CES-D score <16, % ...... 86.0 throughout the 6-year period (RR=0.56; CES-D score ≥16, % ...... 14.0 95% CI=0.34, 0.93; Table 4). Note. IADL=instrumental activity of daily living; CES-D=Center for Epidemiologic Studies Depression Scale. Continuous variables were compared via paired samples t tests and univariate analyses of variance; subcategories were compared via χ2 analyses; DISCUSSION and changes in lifestyle categories between baseline and follow-up were tested with McNemar tests for paired observations. aNonresponse and loss to follow-up numbers varied from 2280 for sociodemographic characteristics, smoking behavior, alcohol intake, physical exercise, and Symptom Checklist 90 Depression Scale score to 654 for body mass index and number In assessing longitudinal associations be- of chronic diseases. tween lifestyle domains and depressed mood, b Significance values refer to comparisons with nonrespondents or individuals lost to follow-up. we found that excessive alcohol use at base- cSignificance values refer to differences from baseline. dEducational level was categorized as low (primary education at most), intermediate (junior vocational training), or high line (compared with abstinence) predicted (senior vocational or academic training). depressed mood at follow-up and that engag- *P<.05; **P<.01; ***P<.001. ing in more than 30 minutes of physical exercise on average per day at baseline (compared with not exercising) was associated with an education levels had higher follow-up depres- Also, each glass of alcohol consumed on aver- absence of depressed mood at follow-up. In sion scores than respondents at intermediate age per day at baseline was associated with a addition, we found that those who initiated al- or high education levels. 17% increased risk of depressed mood at cohol use were at reduced odds of depressed In addition, respondents with impairments follow-up (RR=1.17; 95% CI=1.03, 1.32). mood at follow-up relative to steady non- in IADLs at baseline had higher follow-up Both of these analyses were adjusted for co- drinkers and that those who persistently en- mean depression scores than respondents variates (Table 3). gaged in physical exercise were less likely to with no IADL impairments. Respondents re- Respondents who reported engaging in be depressed at follow-up than those who per- porting 2 or more chronic diseases at baseline physical exercise for more than 30 minutes sistently did not engage in physical exercise. had higher follow-up depression scores than per day on average at baseline had a 48% Our results did not show any longitudinal respondents reporting no chronic diseases or lower risk of being depressed at follow-up associations between smoking behavior and no more than 1 chronic disease, and those than respondents who reported not engaging depressed mood. The increase between base- who reported not engaging in physical exer- in physical exercise at baseline (RR=0.52; line and follow-up in the percentage of re- cise at baseline had higher follow-up scores 95% CI=0.29, 0.92). Also, each minute of spondents who reported never having than those in the other physical activity physical exercise per day reported at baseline smoked (Table 1) indicates some inconsis- groups. Finally, respondents who were over- was associated with a 1% decreased risk of tency in questionnaire responses over the weight at baseline had higher follow-up mean depressed mood at follow-up (RR=0.99; 6-year period, and this may have diminished depression scores than respondents who were 95% CI = 0.98, 1.00). Both analyses were the reliability of our data on smoking behav- not overweight at baseline (Table 2). adjusted for covariates (Table 3). ior and weakened the observed associations Post hoc interaction analyses revealed that Table 4 shows relative risks and 95% con- between smoking behavior and depressive neither gender nor age had a modifying effect fidence intervals derived from longitudinal symptoms. on the associations between the baseline life- multivariate logistic regression models with Our finding of a significant longitudinal style domains assessed and follow-up depres- transitions in lifestyle domains as determi- predictive effect of excessive alcohol use at sive symptoms (data not shown). Hence, nants of follow-up depressed mood. Analyses baseline on the presence of depressed mood analyses were not carried out separately for adjusted for covariates demonstrated that re- at follow-up is in accord, to some extent, with women and men or for different age groups. spondents who initiated alcohol use between the results of Aneshensel and Huba.8 In their Table 3 shows relative risks (RRs) and baseline and follow-up were at lower risk of study involving 742 adults in the Los Ange- 95% confidence intervals (CIs) derived from being depressed at follow-up than respon- les metropolitan area, they found contradic- longitudinal multivariate logistic regression dents who reported no alcohol use at baseline tory cross-sectional and longitudinal effects models with baseline lifestyle domains as de- or follow-up (RR=0.18; 95% CI=0.04, of alcohol use on depression. They inferred terminants of follow-up depressed mood. In 0.76; Table 4). Post hoc analyses revealed that high levels of alcohol use were associ- comparison with respondents who reported that none of the 47 respondents who initiated ated with low scores for depressive symp- no alcohol use at baseline, those who re- alcohol use between baseline and follow-up toms (cross sectional) but that high initial ported excessive alcohol use at baseline were exceeded 11 alcoholic drinks per week on levels of alcohol use were associated with roughly 2.5 times as likely to be depressed at average (data not shown). Finally, analyses subsequent increased depressive symptoms follow-up (RR=2.48; 95% CI=1.08, 5.69). adjusted for covariates demonstrated that (longitudinal).

890 | Research and Practice | Peer Reviewed | van Gool et al. American Journal of Public Health | May 2006, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 2—CES-D Depression Scores at Follow-Up (n=1169) and Comparisons Stratified by However, we are uncertain about the valid- Baseline Sociodemographic and Lifestyle Variables: Maastricht Aging Study ity of the observed association between alcohol use initiation and absence of depressed b Mean Depressed Nondepressed mood at follow-up. Because of the relatively Depression (n=164), (n=1005), small size of the group initiating alcohol use No. Scorea (SD) Mean (SD) or % Mean (SD) or %c (n=47) compared with the other groups, the Age, y 50.8 (14.5) 48.6 (14.7)* fact that nonrespondents reported more de- 24–44 490 6.9 (6.5)***12.4 87.6 pressive symptoms at baseline than respon- 45–64 476 8.4 (6.9)14.7 85.3 dents, and the fact that none of the respon- 65–81 203 8.9 (6.0)16.3 83.7 dents initiating alcohol use between baseline Gender and follow-up exceeded an average of 2 Male 612 7.0 (5.7)*** 9.2 90.8*** glasses of alcohol per day, we cannot rule Female 557 8.9 (7.3) 19.4 80.6 out that this finding was a statistical artifact Marital status attributable to selection bias. Married/living together 951 7.9 (6.7) 13.9 85.7 Our finding of a significant longitudinal Not married/no longer married 160 7.6 (6.1) 14.4 85.6 protective effect of baseline physical exercise Widowed 58 8.6 (6.0) 15.5 84.5 (at recommended levels) on subsequent de- Education level pressed mood is in line with the results of d High 337 6.8 (6.1)*** 11.3 88.7** Strawbridge and colleagues.17 In a 5-year Intermediate 398 7.5 (6.4) 10.8 89.2 follow-up investigation, they found that high Low 434 9.1 (7.0) 19.1 80.9 levels of physical activity were associated IADL status with absence of depression and were protec- Not impaired 1082 7.6 (6.4)*** 12.8 87.2*** tive against subsequent depression among Impaired 87 11.3 (7.8) 28.7 71.3 19 47 community-dwelling adults aged 50 No. of chronic diseases 1.5 (1.5) 1.2 (1.3)* through 94 years. They used a measure of None 390 6.8 (6.6)*** 11.5 88.5* physical activity focusing on usual frequency 1 396 7.8 (6.3) 12.9 87.1 (never, sometimes, often) of exercise, taking 2 or more 383 9.0 (6.7) 17.8 82.2 part in active sports, taking long walks, and Smoking behavior swimming. Never smoked 409 7.4 (6.1) 12.0 88.0 Furthermore, the protective effect of physi- Formerly smoked 448 8.1 (6.6) 14.3 85.7 cal exercise on subsequent depression re- Currently smokes 312 8.2 (7.2) 16.3 83.7 ported by Strawbridge et al.17 was confirmed Average daily alcohol intake in our analyses examining the effects of physi- None 161 8.8 (7.7) 18.0 82.0 cal exercise behavior over time: Respondents Up to 2 alcoholic drinks 928 7.7 (6.4) 13.1 86.9 who reported engaging in physical activity at 3 or more alcoholic drinks 80 7.7 (6.4) 16.2 83.8 both baseline and follow-up were at 44% Mean no. of alcoholic drinks per week (SD) 6.1 (10.6) 6.0 (9.3) lower risk of subsequent depression than Average amount of time spent daily on physical exercise those who reported not engaging in physical More than 30 min 259 6.5 (5.4)*** 7.7 92.3** Up to 30 min 332 7.9 (6.3) 14.5 85.5 exercise at either baseline or follow-up. Thus, None 578 8.5 (7.2) 16.6 83.4 it can be cautiously suggested not only that Mean no. of minutes of physical exercise per day (SD) 12.3 (24.3) 19.2 (29.0)** physical exercise may be an effective element 40 Overweight in the treatment of depression but that No 784 7.5 (6.4)** 13.1 86.9 maintenance of regular physical exercise over Yes 385 8.6 (7.0) 15.8 84.2 a relatively long period of time may protect Mean body mass index, kg/m2 (SD) 26.4 (4.1) 26.5 (4.1) against the emergence of clinically relevant levels of depressive symptoms. The inhibitory Note. CES-D=Center for Epidemiologic Studies Depression Scale; IADL=instrumental activity of daily living. Continuous effect of exercise on levels of inflammatory variables were assessed via t tests and univariate analyses of variance; categorical variables were assessed via χ2 analyses. aSignificance values refer to subcategory comparisons. and cardiovascular risk factors may be part of bCES-D score of 16 or above. the explanatory pathway through which exer- c Significance values refer to differences between depressed and nondepressed participants. cise protects against depression,41 in that the dEducational level was categorized as low (primary education at most), intermediate (junior vocational training), or high (senior vocational or academic training). presence of high levels of these risk factors *P<.05; **P<.01; ***P<.001. has also been associated with the presence of high levels of depressive symptoms.42,43

May 2006, Vol 97, No. 5 | American Journal of Public Health van Gool et al. | Peer Reviewed | Research and Practice | 891  RESEARCH AND PRACTICE 

TABLE 3—Multivariate Logistic Regression Models for Baseline Lifestyle Domains as in residual confounding by these broader so- Determinants of Depressed Mood at Follow-Up (n=1169) cioeconomic conditions. These limitations should be considered in interpreting our re- Depression at Follow-Up sults and formulating public health recom- a b Adjusted RR 1 Adjusted RR 2 mendations. Baseline Lifestyle Domain No. (95% CI) (95% CI) We believe that our study involved a num- Smoking behavior ber of strengths as well. We comprehensively Currently smokes 312 Reference Reference examined the effects of 4 lifestyle domains Formerly smoked 448 0.89 (0.57, 1.40) 0.88 (0.56, 1.40) on depressive symptoms using longitudinal Never smoked 409 0.73 (0.46, 1.17) 0.67 (0.41, 1.09) analysis techniques. The pivotal findings of Average daily alcohol intake this study were that (1) excessive alcohol use None 161 Reference Reference at baseline predicted depressed mood at Up to 2 alcoholic drinks 928 0.92 (0.55, 1.54) 1.15 (0.68, 1.96) follow-up; (2) physical exercise at recom- 3 or more alcoholic drinksc 80 1.49 (0.68, 3.24) 2.48 (1.08, 5.69)* mended levels predicted absence of depressed Mean no. of alcoholic drinks per day (continuous)d 1169 1.07 (0.95, 1.21) 1.17 (1.03, 1.32)* mood at follow-up; and (3) physical exercise Average amount of time spent daily on physical exercise over a relatively long period of time was asso- None 578 Reference Reference ciated with absence of depressed mood. Up to 30 min 332 0.86 (0.57, 1.30) 0.87 (0.56, 1.33) Our results indicate that the potential More than 30 minc 259 0.43 (0.24, 0.76)** 0.52 (0.29, 0.92)* downward spiral mentioned in the introduc- Mean no. of minutes of physical exercise per day (continuous)d 1169 0.99 (0.98, 1.00)** 0.99 (0.98, 1.00)* tion might be halted through adoption or Overweight maintenance of healthy lifestyles, which Yes 385 Reference Reference could prevent the deterioration or even oc- No 784 0.90 (0.62, 1.33) 1.03 (0.69, 1.54) currence of depressed mood over time and, Body mass index (continuous)d 1169 0.99 (0.95, 1.04) 0.98 (0.93, 1.02) in turn, the worsening of chronic disease symptoms. In addition to the role of behav- Note.RR=relative risk; CI=confidence interval. ior change, adoption or maintenance of aAdjusted for baseline depressive symptomatology. bAdjusted for baseline depressive symptomatology, age, gender, marital status, educational level, instrumental activities of healthy lifestyles might be facilitated by the daily living status, and number of chronic diseases. creation of health-promotive environments46 c Significance levels refer to differences from reference category. in both homes and workplaces (e.g., through dValues refer to the odds of subsequent depressed mood associated with each 1-unit increase in the continuous lifestyle variable. *P<.05; **P<.01. offering only low-fat, high-fiber meals in company and school cafeterias or lowering sales taxes on memberships in health and This study was limited by the considerable ever, is the risk of inaccurate recollection of physical fitness clubs), reducing barriers to attrition because of nonresponse and loss to past events and response bias. These factors, engaging in healthy behaviors and motivat- follow-up. The 6-year follow-up period may if present, may have led to a certain degree ing people to engage in these behaviors. have been a key source of selection bias, po- of distortion of our findings. Future research could assess the effects of tentially resulting in the sample included here Finally, a pair of suboptimal conditions of implementing such health-promotive envi- being less representative of the overall study our study need to be mentioned. First, CES-D ronments on lifestyle alterations and subse- population than desired. However, this is a scores were not available at baseline. There- quent changes in both overall health and frequently encountered problem in longitu- fore, we used Symptom Checklist 90 Depres- mental health. dinal aging studies and is difficult to sion Scale baseline scores to control our sta- 10 , 4 4 avoid. In addition, according to Kempen tistical analyses for initial level of depressive About the Authors and Van Sonderen, attrition has more ad- symptoms. Because these 2 instruments have Coen H. van Gool, Gertrudis I.J.M. Kempen, Hans Bosma, verse effects in the case of descriptive meas- been shown to be valid and highly corre- and Jacques T.M. van Eijk are with the Department of urements than in the case of measures fo- lated,31 we believe that our analyses were Health Care Studies, Section Medical Sociology, and the Care and Public Health Research Institute, Universiteit Maastricht, cusing on longitudinal associations, such as adequately controlled; however, this may be Maastricht, the Netherlands. Martin P.J. van Boxtel and those used in our study.45 open to debate. Second, we did not adjust for Jelle Jolles are with the Department of Psychiatry and Neu- Bias may also have been introduced broad socioeconomic factors, such as unem- ropsychology and the European Graduate School of Neu- roscience, Universiteit Maastricht. through the use of mainly self-report mea- ployment, major life events, and work stress- Requests for reprints should be sent to Coen H. van Gool, sures. In general, research involving the use ors, that might be associated with both un- PhD, Department for Public Health Forecasting, National of self-reported measures is inexpensive, easy healthy lifestyles and depression. Thus, Institute for Public Health and the Environment, PO Box 1, NL-3720 BA Bilthoven, The Netherlands (e-mail: to conduct, and not affected by interrater control for educational level alone may not [email protected]). variability. A downside of such research, how- have been sufficient and could have resulted This article was accepted December 20, 2005.

892 | Research and Practice | Peer Reviewed | van Gool et al. American Journal of Public Health | May 2006, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 4—Multivariate Logistic Regression Models for Transitions in Lifestyle Domains 8. Aneshensel CS, Huba GJ. Depression, alcohol use, and smoking over 1 year: a four-wave longitudinal Between Baseline and Follow-Up as Determinants of Depressed Mood at Follow-Up causal model. J Abnorm Psychol. 19 83;92:134–150. (n=1169) 9. Depression Guideline Panel. Depression in Primary Care, Volume 1: Detection and Diagnosis. Rockville, Md: Depression at Follow-Up Agency for Health Care Policy and Research; 1993. Adjusted RR 1a Adjusted RR 2b AHCPR publication 93-0550. Lifestyle Transition No. (95% CI) (95% CI) 10.van Gool CH, Kempen GI, Penninx BW, et al. Im- pact of depression on disablement in late middle aged Smoking behavior and older persons: results from the Longitudinal Aging Still does not smoke 826 0.78 (0.50, 1.22) 0.74 (0.46, 1.17) Study Amsterdam. Soc Sci Med. 2005;60:25–36. Quit smoking 86 0.89 (0.40, 1.95) 0.90 (0.40, 2.00) 11. Haringsma R, Engels GI, Beekman AT, Spinhoven P. Initiated smoking 31 1.09 (0.37, 3.20) 1.23 (0.42, 3.60) The criterion validity of the Center for Epidemiological Studies Depression Scale (CES-D) in a sample of self- Still smokes 226 Reference Reference referred elders with depressive symptomatology. Int J Alcohol intake Geriatr Psychiatry. 2004;19:558–563. Still drinks alcohol 941 0.63 (0.37, 1.09) 0.80 (0.45, 1.41) 12. Anda RF, Williamson DF, Escobedo LG, et al. Initiated alcohol use 47 0.17 (0.04, 0.73)* 0.18 (0.04, 0.76)* Depression and the dynamics of smoking: a national perspective. JAMA. 1990;264:1541–1545. Quit drinking alcohol 67 1.35 (0.60, 3.01) 1.29 (0.57, 2.91) Still does not drink alcohol 114 Reference Reference 13. Carney RM, Rich MW, Tevelde A, et al. Major depressive disorder in coronary artery disease. Am J Physical exercise Cardiol. 1987;60:1273–1275. Still engages in physical exercise 328 0.50 (0.31, 0.82)** 0.56 (0.34, 0.93)** 14 .Frederick T, Frerichs RR, Clark VA. Personal Initiated physical exercise 107 0.64 (0.32, 1.27) 0.66 (0.32, 1.33) health habits and symptoms of depression at the com- Discontinued physical exercise 263 0.78 (0.49, 1.24) 0.80 (0.50, 1.30) munity level. Prev Med. 19 88;17:173–182. Still does not engage in physical exercise 471 Reference Reference 15.Freed EX. Alcohol and mood: an updated review. Int J Addict. 1978;13:173–200. Body mass index Still not overweight 661 0.92 (0.61, 1.40) 1.05 (0.68, 1.62) 16. Aneshensel CS. An application of log-linear models: the stress-buffering function of alcohol use. J Drug No longer overweight 45 1.40 (0.57, 3.41) 1.46 (0.58, 3.67) Educ. 19 83;13:287–301. Became overweight 123 1.07 (0.56, 2.05) 1.20 (0.62, 2.32) 17. Strawbridge WJ, Deleger S, Roberts RE, et al. Still overweight 340 Reference Reference Physical activity reduces the risk of subsequent depression for older adults. Am J Epidemiol. 2002;156: Note.RR=relative risk; CI=confidence interval. Significance values refer to differences from reference category. 328–334. aAdjusted for baseline depressive symptomatology. 18. Goodwin RD. Association between physical activ- bAdjusted for baseline depressive symptomatology, age, gender, marital status, educational level, instrumental activities of ity and mental disorders among adults in the United daily living status, and number of chronic diseases. States. Prev Med. 2003;36:698–703. *P<.05; **P<.01. 19. Bin Li Z, Yin Ho S, Man Chan W, et al. Obesity and depressive symptoms in Chinese elderly. Int J Geri- atr Psychiatry. 2004;19:68–74. Contributors 2. Gohlke H. Lifestyle modification—is it worth it? 20.Roberts RE, Strawbridge WJ, Deleger S, et al. Are C. H. van Gool and G. I. J. M. Kempen formulated the Herz. 2004;29:139–144. the fat more jolly? Ann Behav Med. 2002;24: hypothesis. C.H. van Gool analyzed the data, inter- 3. Meigs JB, Hu FB, Rifai N, et al. Biomarkers of en- 16 9–180. preted the findings, and drafted the article. G.I.J.M. dothelial dysfunction and risk of type 2 diabetes melli- 21.Ross CE. Overweight and depression. J Health So- Kempen, H. Bosma, M.P.J. van Boxtel, J. Jolles, and tus. JAMA. 2004;291:1978–1986. cial Behav. 1994;35:63–79. J.T.M. van Eijk assisted with interpretation of findings 4. Penninx BW, Leveille S, Ferrucci L, van Eijk JT, and critical revision of the article. 22. Carpenter KM, Hasin DS, Allison DB, et al. Rela- Guralnik JM. Exploring the effect of depression on tionships between obesity and DSM-IV major depres- physical disability: longitudinal evidence from the es- sive disorder, suicide ideation, and suicide attempts: tablished populations for epidemiologic studies of the Acknowledgments results from a general population study. Am J Public elderly. Am J Public Health. 1999;89:1346–1352. This work was supported in part by a grant from the Health. 2000;90:251–257. Dutch Ministries of Education and Health and Welfare, 5. Bisschop MI, Kriegsman DM, Deeg DJ, et al. The via the Steering Committee for Gerontological Research. longitudinal relation between chronic diseases and de- 23. Metsemakers JF, Höppener P, Knottnerus JA, et al. pression in older persons in the community: the Longi- Computerized health information in the Netherlands: a tudinal Aging Study Amsterdam. J Clin Epidemiol. registration network of family practices. Br J Gen Pract. Human Participant Protection 2004;57:187–194. 19 92;42:102–106. This study was approved by the medical ethics commit- 6. van Gool CH, Kempen GI, Penninx BW, et al. Re- 24.Jolles J, Houx PJ, van Boxtel MP, et al., eds. Maas- tee of the University Hospital Maastricht. Participants lationship between changes in depressive symptoms tricht Aging Study: Determinants of Cognitive Aging. provided written informed consent. and unhealthy lifestyles in late middle aged and older Maastricht, the Netherlands: Neuropsych Publishers; persons: results from the Longitudinal Aging Study 19 95. References Amsterdam. Age Ageing. 2003;32:81–87. 25. Valentijn SA, van Boxtel MP, van Hooren SA, et 1. Jorm AF. The epidemiology of depressive states in 7. National Academy on an Aging Society. Depres- al. Change in sensory functioning predicts change in the elderly: implications for recognition, intervention sion: a treatable disease. Available at: http://www. cognitive functioning: results from a 6-year follow-up and prevention. Soc Psychiatry Psychiatr Epidemiol. agingsociety.org/agingsociety/pdf/depression.pdf. Ac- in the Maastricht Aging Study. J Am Geriatr Soc. 2005; 19 95;30:53–59. cessed April 5, 2006. 53:374–380.

May 2006, Vol 97, No. 5 | American Journal of Public Health van Gool et al. | Peer Reviewed | Research and Practice | 893  RESEARCH AND PRACTICE 

26.van Boxtel MP, Buntinx F, Houx PJ, et al. The Health and Nutrition Examination Survey. Arch Intern relation between morbidity and cognitive perform- Med. 2004;164:1010–1014. ance in a normal aging population. J Gerontol. 1998; 44.Koster A, Bosma H, van Lenthe FJ, Kempen GI, 53A:M146–M154. Mackenbach JP, van Eijk JT. The role of psychosocial 27. Derogatis LR, Lipman RS, Covi L. SCL-90: an factors in explaining socio-economic differences in mo- outpatient psychiatric rating scale—preliminary report. bility decline in a chronically ill population: results Psychopharmacol Bull. 1973;9:13–27. from the GLOBE study. Soc Sci Med. 2005;61: 28.Arrindell WA, Ettema JHM. Dimensional struc- 123–132. ture, reliability and validity of the Dutch version of the 45. Kempen GI, Van Sonderen E. Psychological attrib- Symptom Checklist (SCL-90). Ned Tijdschr Psychologie. utes and changes in disability among low-functioning, 19 81;43:381–387. older persons: does attrition affect the outcomes? J Clin 29.Radloff LS. The CES-D Scale: A self-report de- Epidemiol. 2002;55:224–229. pression scale for research in the general population. 46. Stokols D. Translating social ecological theory into Appl Psychol Meas. 1977;1:385–401. guidelines for community health promotion. Am J Health 30.Beekman AT, Deeg DJ, Van Limbeek J, et al. Cri- Promotion. 1996;10:282–298. terion validity of the Center for Epidemiologic Studies Depression Scale (CES-D): results from a community- based sample of older subjects in the Netherlands. Psychol Med. 19 97;27:231–235. 31.Weissman MM, Sholomskas D, Pottenger M, Prusoff BA, Locke BZ. Assessing depressive symptoms in five psychiatric populations: a validation study. Am J Epidemiol. 1977;106:203–214. 32. Dohrenwend BP. Mental Illness in the United States: Epidemiological Estimates. New York, NY: NOW Available on CD-Rom Praeger; 1980. 33. Williamson DF, Kahn HS, Remington PL, et al. The 10-year incidence of overweight and major weight gain in US adults. Arch Intern Med. 1990;150: 665–672. Control of Communicable 34. Enright PL, McBurnie MA, Bittner V, et al. The Diseases Manual 6-min walk test: a quick measure of functional status in elderly adults. Chest. 2003;123:387–398. Edited by David L. Heymann, MD 35. Ruchlin HS. Prevalence and correlates of alcohol use among older adults. Prev Med. 19 97;26:651–657. 36.Henderson PN, Rhoades D, Henderson JA, et al. Protection for you and Smoking cessation and its determinants among older American Indians: the Strong Heart Study. Ethn Dis. your community at your fingertips. 2004;14:274–279. 37.Beekman AT, Penninx BW, Deeg DJ, et al. De- pression and physical health in later life: results from the Longitudinal Aging Study Amsterdam (LASA). J Af- ISBN 0-87553-035-4 fect Disord. 19 97;46:219–231. hardcover ❚ 2004 38. Dong C, Sanchez LE, Price RA. Relationship of $30.00 APHA Members $43.00 Nonmembers obesity to depression: a family-based study. Int J Obes ORDER TODAY! plus shipping and handling Relat Metab Disord. 2004;28:790–795. American Public Health 39. Schoevers RA, Beekman AT, Deeg DJ, Geerlings ISBN 0-87553-034-6 Association MI, Jonker C, Van Tilburg W. Risk factors for depres- softcover ❚ 2004 Publication Sales Web: www.apha.org sion in later life: results of a prospective community $23.00 APHA Members based study (AMSTEL). J Affect Disord. 2000;59: E-mail: [email protected] $33.00 Nonmembers 127–137. Tel: 888-320-APHA plus shipping and handling FAX: 888-361-APHA 40. Dunn AL, Trivedi MH, Kampert JB, Clark CG, Chambliss HO. Exercise treatment for depression: efficacy and dose response. Am J Prev Med. 2005;28:1–8. 41. Abramson JL, Vaccarino V. Relationship between physical activity and inflammation among apparently healthy middle-aged and older US adults. Arch Intern Med. 2002;162:1286–1292. 42. Danner M, Kasl SV, Abramson JL, Vaccarino V. Association between depression and elevated C-reactive protein. Psychosom Med. 2003;65:347–356. 43. Ford DE, Erlinger TP. Depression and C-reactive protein in US adults: data from the Third National

894 | Research and Practice | Peer Reviewed | van Gool et al. American Journal of Public Health | May 2006, Vol 97, No. 5  RESEARCH AND PRACTICE 

Trends in Characteristics and Country of Origin Among Foreign-Trained Nurses in the United States, 1990 and 2000

| Daniel Polsky, PhD, Sara J. Ross, MPA/ID, Barbara L. Brush, PhD, RN, and Julie Sochalski, PhD, RN

Foreign-trained nurses have been a small but Objectives. We describe long-term trends in the characteristics of foreign- steady part of the nursing workforce in the trained new entrants to the registered nurse (RN) workforce in the United States. United States for more than 50 years. Short- Methods. Using the 1990 and 2000 US Census 5% Public Use Microdata Sam- ages in the US registered nurse (RN) work- ple files, we compared trends in characteristics of US- and foreign-trained new force, which have occurred cyclically over this entrants to the RN labor force (n=40827) and identified trends in the country of period, typically provide the impetus for em- origin of the foreign-trained new entrants. ployers to seek foreign workers.1–3 With the Results. Foreign-trained RNs grew as a percentage of new entrants to the RN current nursing shortage and reported RN va- workforce, from 8.8% in 1990 to 15.2% in 2000. Compared with US-trained RNs, cancy rates of 10% to 15% in hospitals and foreign-trained RNs were 3 times as likely to work in nursing homes and were nursing homes, there is currently a height- more likely to have earned a bachelor’s degree. In 2000, 21% of foreign-trained RNs originated from low-income countries, a doubling of the rate since 1990. ened interest in foreign-trained RNs.3–5 Conclusions. Foreign-trained RNs now account for a substantial and growing Estimates of the number of RNs based on 6 proportion of the US RN workforce. Our findings suggest foreign-trained RNs the Current Population Survey show that the entering the United States are not of lower quality than US-trained RNs. However, number of foreign-born RNs employed in the growth in the proportion of RNs from low-income countries may have negative United States (11.5% of RNs employed in the consequences in those countries. (Am J Public Health. 2007;97:895–899. doi:10. United States in 2000) grew 4 times faster than 2105/AJPH.2005.072330) the number of US-born RNs during the 1990s and accounted for one third of the growth in the nursing workforce between 2001 and based on self-report. We validated the the end of their formal education was lower 2003.5 These trends have raised concerns that weighted totals of the census sample of RNs than their age in the year they entered the the aggressive recruitment of nurses from over- with the weighted totals from the National United States. Our main results were not seas will not be met with equally vigorous as- Sample Survey of Registered Nurses sensitive to alternative indirect definitions of surance of the quality and skills of the immigrat- (NSSRN).18 RNs in the labor force totaled foreign-trained RNs. ing nurses7,8 and that recruiting nurses from 2176 851 from the 2000 census and RN characteristics in this study consisted vulnerable low-income countries may have 2201 813 from the 2000 NSSRN; there of demographics, income, education, practice negative health consequences for the popula- were 1747309 RNs from the 1990 census setting, and geographic location. Income in tions of those countries.9–16 and 1740030 from an average of the 1988 19 9 0, which is converted into 2000 US dol- We used the US decennial censuses to and 1992 NSSRN surveys. These nearly lars on the basis of the consumer price compare the sociodemographic and employ- identical totals validate the use of the census index, included wages, salaries, and self- ment characteristics of US- and foreign- for describing the RN labor force. For this employment income. Education represented trained new entrants to the RN workforce. study of new entrants to the RN workforce, highest degree attained rather than highest We also present trends from 1990 to 2000 we excluded RNs with more than 10 years of nursing degree. Work setting was based on a in the characteristics of the country of origin US work experience, which resulted in a categorization of census industry codes. of foreign-trained entrants. Finally, we de- study sample of 40827 individuals (21797 Years of total work experience was derived scribe the short- and long-term implications of in 1990 and 19030 in 2000). from age minus years of advanced schooling nurse migration for the nursing workforce Foreign-trained RNs are defined as RNs minus 18 (the median age for completion of needs in the United States and in developing born in a foreign country who did not enter secondary education). Years of foreign work countries. the United States until their formal training experience was the difference between age was complete. Because the census does not at arrival in the United States and age when METHODS specifically ask the location of a nurse’s train- training was complete. Our estimates of ing, we indirectly determined foreign-trained work experience actually estimated maxi- Using the 5% Public Use Microdata Sample RNs by using their country of birth, education mum potential work experience because we files17 from the decennial census of 1990 and level, and year of entry into the United States. could not account for time away from nurs- 2000, we identified the number of RNs in More precisely, in this study, foreign-trained ing or identify those who started nursing as the labor force. Occupation in the census is RNs were those born abroad whose age at a second career.

May 2007, Vol 97, No. 5 | American Journal of Public Health Polsky et al. | Peer Reviewed | Research and Practice | 895  RESEARCH AND PRACTICE 

We used country of birth from the census 9.1% of the nursing workforce (data not to have a bachelor’s degree, the gap had nar- to assign originating continent, country in- shown). In 1990, there were only 113000 rowed. come level, and country nurse-to-population foreign-trained RNs, representing 6.5% of the The annual income of foreign-trained en- ratio. Per capita country income was based on nursing workforce. These figures yield a 40% trants was considerably higher than that of gross national income in 2000 from the increase in the number of foreign-trained RNs US-trained entrants. In 2000, foreign-trained World Bank’s 2001 World Development Indi- as a proportion of total RNs. nurses earned $44000, compared with cators.19 Values for gross national income Table 1 provides demographic and employ- $33000 for US-trained nurses. Using an in- were all converted to US dollars by the World ment characteristics of new entrants to the come regression to control for age, work Bank Atlas Method.20 Nurse-to-population US RN workforce by location of training and hours, and state of residence, we estimated ratios were obtained from the World Health over time. From 1990 to 2000, the share of that half of this difference could be accounted Organization.21 Income level and nurse-to- new entrants that were foreign trained grew for by the fact that foreign-trained entrants population ratio of country of origin were cat- appreciably. There were 469249 RNs work- were older, worked more hours, and were lo- egorized into low, middle, and high groups. ing in 1990 who had joined the workforce cated in states with higher nurse wages. The The low and high groups each contained during the previous decade; of these, 8.8% trend in incomes shows that from 1990 to 25% of all countries; the remaining countries were foreign trained. By 2000, that share had 2000, foreign-trained entrants experienced were in the middle group. For those RNs who nearly doubled, with 15.3% of the 406869 an 8.5% real growth in income, whereas the we determined were foreign trained, we as- new entrants foreign trained. growth for US-trained entrants was flat. This sumed that the country of birth was the same The characteristics of the foreign-trained trend can be partially explained by the more as the country of training. group also changed. The percentage of rapid growth among foreign-trained entrants foreign-trained entrants with at least a bache- in age and hours worked. It may also be in- RESULTS lor’s degree declined slightly, from 69.1% to fluenced by differences between the 2 groups 67.8%, whereas among US-trained entrants, that were not specifically captured in the cen- Using 2000 census data, we calculated that the percentage with at least a bachelor’s de- sus data, such as differences in the shifts there were 181000 foreign-trained RNs work- gree increased from 59.6% to 64.3%. Thus, worked (e.g., whether foreign nurses were ing in the United States in 2000, representing while US-trained entrants still were less likely more likely to work the more highly compen- sated evening and night shifts). Of particular note is the growth in the re- TABLE 1—Demographic and Employment Characteristics of New Entrants to the US cruitment of foreign nurses to long-term care Registered Nurse Workforce: US Census, 1990 and 2000 facilities. Although nursing homes have tradi- US-Trained Nurses Foreign-Trained Nurses tionally been more likely to employ foreign nurses than have other health care settings, Characteristic 1990 Census 2000 Census 1990 Census 2000 Census employment of foreign-trained new entrant New entrants, weighted no. 427875 344803 41374 62066 RNs in nursing homes has grown by nearly 3 Mean age, y (SD) 27.5 (3.1) 27.5 (3.0) 35.0 (7.3) 37.0 (7.7) times in the past decade. Historical difficulty Men, % 5.8 9.9 8.6 14.2 in maintaining adequate staffing levels may Race, % have influenced the growth in numbers of White 89.6 83.6 23.3 30.1 foreign-trained nurses in these settings.22 Fur- Black 7.0 8.2 13.5 16.1 thermore, wages in nursing homes grew more Asian 2.0 4.3 60.5 48.7 rapidly in 2000 than in hospitals, which may Other 1.4 3.9 2.7 5.1 have contributed to the relative income Education, % growth of foreign- versus US-trained nurses.23 Associate’s 40.4 35.7 30.8 32.2 The geographic location of new entrants is Bachelor’s 48.2 54.3 55.8 57.1 displayed in Figure 1. The geographic distri- Master’s or above 11.4 10.0 13.3 10.7 bution of US-trained entrants in 1990 and Mean US work experience, y (SD) 6.2 (2.8) 6.1 (2.8) 4.4 (3.4) 5.4 (3.0) 2000 look similar. The geographic distribu- Mean total work experience, y (SD) 6.2 (2.8) 6.1 (2.8) 13.4 (7.4) 15.5 (7.8) tion of foreign-trained entrants in 1990 and Mean income, $ (SD) 32776 (15862) 32931 (17743) 40273 (20392) 43703 (25872) 2000 shows a dramatic shift from the Mid- Hours worked per year 1764 1782 1862 1898 Atlantic and Pacific census divisions to the Setting, % South and Midwest census divisions. Hospital 79.8 72.2 83.4 63.8 The characteristics of the continent or Nursing home 4.1 6.6 6.3 17.5 country of origin of the foreign-trained en- Other 16.0 21.3 10.3 18.7 trants are shown in Table 2. As of 2000, the majority of foreign-trained RNs came from

896 | Research and Practice | Peer Reviewed | Polsky et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

FIGURE 1—Location of US- and foreign-trained new entrant registered nurses, in 1990 (a) and 2000 (b).

Asia, predominantly from the Philippines. classified by nurse-to-population ratio. The country’s excess demand for nurses will de- Nearly 25% came from North or South share of foreign-trained nurses from countries pend, in part, on whether this practice main- America (Canada, Mexico, the Caribbean is- with a high ratio of nurses increased by 66%, tains the quality of the RN workforce. Com- lands, South and Central America), of whom whereas the share from countries with a low pared with their US counterparts, foreign- more than half were from Canada; Europe ratio grew by 30%. trained entrants are more likely to have a and Africa contributed 11% each. This distri- bachelor’s degree and are likely to have as bution reflects a relative shift away from for- DISCUSSION much work experience. Even though educa- eign-trained RNs from Asia. In addition, from tion and work experience in foreign countries 1990 to 2000, there was a doubling of the Foreign-trained RNs are assuming a larger with less-advanced health care technology share of new foreign-trained RNs from Can- role in the provision of nursing care in the may not be directly comparable to education ada and a tripling in the share from Africa. United States. There were nearly 181000 and work experience in the United States, Aside from a shift in continent of origin, foreign-trained RNs working in the United there is no evidence from our results to sug- there was also a shift in the economic profile States in 2000, which represents 9.1% of the gest that the foreign-trained RNs are lower- of the source countries within the continent. nursing workforce. The percentage of foreign- quality RNs. This conclusion is further sup- In 1990, 18% of new foreign-trained RNs trained RNs increased by 40%, from 6.5% of ported by our somewhat surprising finding came from high-income countries and 11% RNs in 1990. New entrants with foreign train- that the foreign-trained entrants had a substan- came from low-income countries. By 2000, ing grew from 8.8% of new entrants in 1990 tially higher average income than US-trained there was a shift away from middle-income to 15.2% in 2000. Other studies have af- entrants. The contribution of differences in countries, with a higher percentage from both firmed this trend and suggest that it has accel- skills and of other factors to this higher income high-income (25.4%) and low-income coun- erated since 2000.5,8 should be more fully investigated. tries (20.7%). The same shift from the middle From a US perspective, the wisdom of Upward trends in the recruitment of for- can be seen when the countries of origin are relying on foreign-trained nurses to meet the eign nurses by large developed countries may

May 2007, Vol 97, No. 5 | American Journal of Public Health Polsky et al. | Peer Reviewed | Research and Practice | 897  RESEARCH AND PRACTICE 

TABLE 2—Characteristics of New has been cited as a binding constraint in pro- low incomes and low stocks of nurses. The 24 Foreign-Trained Entrants to the US grams to address global health priorities. impact of nurse emigration to the United Registered Nurse Workforce: US Responding to this issue by imposing re- States on the countries sending them requires Census, 1990 and 2000 strictions on nurse migration from vulnerable closer examination, because this could have a countries may address a short-term crisis, but large impact not only on the future ability of Foreign-Trained Nurses it could ultimately shrink the local supply of the United States to rely on overseas nurse re- 1990 2000 nurses. The prospect for higher wages over- cruitment but also on global health. Characteristic Census Census seas attracts talented local workers to the field New entrants in the labor force 41 374 62 066 of nursing, which may increase the local and Continent/country, % worldwide nursing supply. The implications of About the Authors policy changes should be carefully considered Daniel Polsky is with the School of Medicine and the Europe 10.9 11.5 Wharton School, University of Pennsylvania, Philadelphia. Asia 62.8 52.1 to avoid unintended consequences. At the time of the study, Sara J. Ross was with the School of Americas (excluding Canada) 15.3 10.9 An important limitation of this work was Medicine, University of Pennsylvania, Philadelphia. Bar- that we did not observe foreign training, ex- bara L. Brush is with the University of Michigan School of Canada 6.1 13.6 Nursing, Ann Arbor. Julie Sochalski is with the School of perience, and RN licensure directly. Foreign Africa 4.1 11.1 Nursing, University of Pennsylvania, and the Leonard training and experience were inferred on the Davis Institute of Health Economics, Philadelphia. Oceania 0.7 0.9 Requests for reprints should be sent to Daniel Polsky, a basis of age, education, and year of entry into Per capita country income, % PhD, 423 Guardian Dr, Blockley Hall, Rm 1212, Phila- the United States. Whether a nurse was li- Low (≤$755) 11.1 20.7 delphia, PA 19104 (e-mail: [email protected]). censed was based on self-report, but our total This article was accepted November 24, 2005. Middle ($755–$9265) 70.9 54.0 estimated number of RNs corresponded High (≥$9266) 18.0 25.4 closely to estimates from the NSSRN, where Contributors Country nurse-to-population sampling is based on state licensure records. D. Polsky led the conceptualization and research design, ratio,b % guided the analytic process, and reviewed research However, the NSSRN cannot be accurately findings. S.J. Ross aided in the research design and car- Low (≤63/100000) 9.9 12.9 used as a benchmark for accounting for the ried out the analysis. B.L. Brush provided critical re- Middle (64–478/100000) 73.4 59.4 number of foreign-trained RNs in the United view of and expert consultation on the research find- High (≥479/100000) 16.7 27.8 ings. J. Sochalski guided the research aims and provided States, because the NSSRN undercounts for- a critical review of research findings. All authors con- aMean income is based on gross national income, eign nurses by sampling the same number of tributed to the writing of the article. from the World Bank’s 2001 World Development nurses from all states (foreign nurses typically 19 Indicators. locate in large states) and by identifying the Acknowledgments bData obtained from the World Health Organization.21 foreign trained by the location-of-training The University Research Foundation at the University of Pennsylvania provided financial support for this question (nonrespondents to this question study. may be predominantly foreign trained). exacerbate existing problems of access to The United States has eased its nursing Human Participant Protection health care for many smaller and poorer shortage by, among other strategies, recruiting The University of Pennsylvania institutional review countries that are net exporters of nurses and more foreign-trained nurses. Monitoring the board determined this project to be exempt. already have low nurse-to-population ratios. characteristics of these nurses provides US We found that, compared with 1990, new policymakers and workforce managers with References 1. Feldstein PJ. The shortage of nurses. In: Feldstein PJ, foreign-trained RNs in 2000 were twice as information on the impact of allowing this ed. Health Policy Issues: An Economic Perspective on likely to originate from low-income countries trend to continue unabated. Such information, Health Reform. 2nd ed. Washington, DC: Association of and 30% more likely to originate from coun- though, tells only part of the story. Although University Programs in Health Administration; 1999: 229–238. tries with a low supply of nurses. Even a the levels of education and work experience 2. Goodwin HJ. The nursing shortage in the United small increase in the proportion of US RNs of foreign-trained nurses are not lower than States of America: an integrative review of the litera- from countries with low numbers of nurses those of US-trained nurses, these features only ture. J Adv Nurs. 2003;43:335–350. may represent a large proportion of those address the technical dimensions of care. Cul- 3. Brush BL, Berger AM. Sending for nurses: foreign countries’ stock of nurses. For example, the tural competence, defined here as the ability nurse migration, 1965–2002. Nurs Health Policy Rev. 2002;1:103–115. 11.1% of foreign-trained RNs who entered to use a cultural understanding of patients in 4. Buerhaus PI, Staiger DO. Trouble in the nurse the United States from Africa between 1990 caring for them, which substantially affects the labor market? Recent trends and future outlook. Health 8 and 2000 alone represents more than 1% of quality of care and is not easily captured in Aff. 1999;18:214–222. the entire stock of African nurses. Deteriora- any data sources, could likewise be changing 5. Buerhaus PI, Staiger DO, Auerbach DI. New signs tion of the nursing stock of these poor coun- and having a noteworthy—and currently un- of a strengthening US nurse labor market? Health Aff. 2004 2004;23:526–533. tries may ultimately threaten their capacity to measured—effect on the delivery of patient 6. Bureau of Labor Statistics. Current Population train nurses and replenish their nursing work- care in the United States. Furthermore, the Survey. Available at: http://www.bls.gov/cps. Accessed force. The lack of a skilled nursing workforce source countries are increasingly those with June 2005.

898 | Research and Practice | Peer Reviewed | Polsky et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

7. Buchan J, Parkin T, Sochalski J. International Nurse Mobility: Trends and Policy Implications. Geneva, Switz- erland: World Health Organization; 2003. 8. Brush BL, Sochalski J, Berger AM. Imported care: The End of Polio? recruiting foreign nurses to US health care facilities. Health Aff. 2004;23:78–87. Behind the scenes of the campaign to vaccinate 9. Kingma M. Nursing migration: global treasure hunt or disaster-in-the-making? Nurs Inq. 2001;8: every child on the planet. 205–212. Tim Brookes with Omar A. Khan, MD, MHS 10.Prystay C. US solution is Philippine dilemma: as recruiters snap up more nurses, hospitals in Manila are We are close to eradicating polio from the earth. Through a scrambling. Wall Street Journal. November 30, collaborative effort on the part of many health organizations, 2002:A8. including CDC, WHO, Rotary International, and many 11.Wickett D, McCutcheon H. Issues of qualification assessment for nurses in a global market. Nurse Educ others, there has been a global campaign to vaccinate every Today. 2002;22:44–52. child on the planet against this dreaded disease. Travel with 12. Muula AS, Mfutso-Bengo JM, Makoza J, Chatipwa E. our authors as they take part in one of the last vaccination The ethics of developed nations recruiting nurses from campaigns, and read about the successes and obstacles developing countries: the case of Malawi. Nurs Ethics. affecting the final goal. With numerous photos of campaigns 2003;10:433–438. in various parts of the world, including an 8-page insert of 13. Schubert C. Nurses disappearing from developing color photos. nations. Nat Med. 2003;9:979. 14 . Aiken L, Buchan J, Sochalski J, Nichols B, Powell M. Trends in international nurse migration. Health Aff. ORDER TODAY! American Public Health Association 2004;23:69–77. ISBN: 978-0-87553-186-1 PUBLICATION SALES 15.Willetts A, Martineau T. Ethical International Re- 208 pages, softcover, 2007 WEB: www.apha.org cruitment of Health Professionals: Will Codes of Practice $35.00 APHA Members (plus s&h) E-MAIL: [email protected] Protect Developing Country Health Systems? Liverpool, England: Liverpool School of Tropical Medicine; 2004. $39.95 Nonmembers (plus s&h) TEL: 888-320-APHA 16. Dugger CW. An exodus of African nurses puts FAX: 888-361-APHA infants and the ill in peril. New York Times. July 12, 2004:1. 17.US Census Bureau. 5% Public Use Microdata Sample Files. Available at: http://ftp2.census.gov/ census_2000/datasets/PUMS/FivePercent. Accessed June 2005. 18. Health Resources and Services Administration. National Sample Survey of Registered Nurses. Available at: http://datawarehouse.hrsa.gov/nursingsurvey.htm. Accessed June 2005. 19. World Development Indicators 2001. Washington, DC: World Bank; 2001. 20.World Bank Atlas Method. Available at: http:// web.worldbank.org. Accessed June 2005. 21. The World Health Report 2006: Working Together for Health. Geneva, Switzerland: World Health Organi- zation; 2006. 22. Priester R, Reinardy JR. Recruiting immigrants for long-term care nursing positions. J Aging Soc Policy. 2003;15(4):1–19. 23. Spratley E, Johnson A, Sochalski J, Fritz M, Spencer W. The Registered Nurse Population, March 2000: Findings From the National Sample Survey of Registered Nurses. Washington, DC: US Department of Health and Human Services; 2001. 24. Burkhalter H. Shortage of African doctors and nurses: we must support primary health infrastructure. Washington Post. May 25, 2004:A17.

May 2007, Vol 97, No. 5 | American Journal of Public Health Polsky et al. | Peer Reviewed | Research and Practice | 899  RESEARCH AND PRACTICE 

Globalization, Binational Communities, and Imported Food Risks: Results of an Outbreak Investigation of Lead Poisoning in Monterey County, California

| Margaret A. Handley, PhD, MPH, Celeste Hall, RN, Eric Sanford, MD, Evie Diaz, Enrique Gonzalez-Mendez, MD, Kaitie Drace, BA, Robert Wilson, PhD, Mario Villalobos, PhD, and Mary Croughan, PhD

In the past 25 years, the United States and Objectives. Although the burden of lead poisoning has decreased across de- many other countries have witnessed dra- veloped countries, it remains the most prevalent environmental poison worldwide. matic decreases in environmental exposure to Our objective was to investigate the sources of an outbreak of lead poisoning in lead, largely resulting from environmental Monterey County, California. policies that mandated that lead be removed Methods. An investigation in 3 county health department clinics in Monterey from gas, paint, and other manufactured County, California, was conducted between 2001 and 2003 to identify risk factors products, and through the modification of for elevated blood lead levels (≥10 µg/dL) among children and pregnant women. mining and other extraction processes.1–4 Results. The prevalence of elevated blood lead levels was significantly higher The success of these programs is evidenced in 1 of the 3 clinics (6% among screened children and 13% among prenatal pa- by studies that have reported substantial de- tients). Risk factors included eating imported foods (relative risk [RR]=3.4; 95% confidence interval [CI]=1.2, 9.5) and having originated from the Zimatlan area clines in blood lead levels (BLLs) in popula- of Oaxaca, Mexico, compared with other areas of Oaxaca (RR=4.0; 95% CI=1.7, tion samples of children in the United States 5 9.5). Home-prepared dried grasshoppers (chapulines) sent from Oaxaca were over the past 10 years. found to contain significant amounts of lead. Despite these gains, lead exposure contin- Conclusions. Consumption of foods imported from Oaxaca was identified as ues to be one of the most prevalent and a risk factor for elevated blood lead levels in Monterey County, California. Lead- harmful sources of environmental poisoning contaminated imported chapulines were identified as 1 source of lead poisoning, in much of the world. Recent estimates by the although other sources may also contribute to the observed findings. Food trans- World Health Organization indicate that the port between binational communities presents a unique risk for the importation percentage of children aged younger than 6 900–906. doi:10.2105/AJPH.2005.074138) years with elevated BLLs, defined as 10 or more micrograms of lead per deciliter of blood [µg/dL],4,6 is well over 20% in several that allow for strong cultural ties to be main- who were seeing “way too many cases of regions, including Latin America.4 The World tained over vast distances can also lead to the lead poisoning” among their patients who Health Organization emphasizes that the risk importation of environmental hazards. We in- were from Oaxaca. Although there had been of lead toxicity remains as high as in previous vestigated an outbreak of lead poisoning in prior lead case investigations conducted by centuries for children in many countries. Monterey County, California, in which a local- the Monterey County Health Department, The insidious attack of lead on the develop- ized problem of lead poisoning was found to be the more common sources of lead exposure ing nervous system significantly limits the associated with contaminated imported foods in California, such as lead-contaminated ability of children worldwide to climb out of among community residents from the southern water from old pipes, peeling lead-based the poverty in which so many grow up. Mexican state of Oaxaca (Figure 1). paint, lead-contaminated soil, occupational Globalization and the regular transnational In 2000, one of the community doctors sources, lead-glazed ceramics, or home movement of capital, goods, and people allow (E. S.) affiliated with the Department of Fam- health remedies containing lead, had not migrants to maintain strong ties with their com- ily and Community Medicine at the Univer- been identified in the majority of these inves- munities of origin. In California, thousands of sity of California, San Francisco, asked for tigations (Donna Staunton, Public Health migrants travel back and forth to small commu- help in investigating a large number of lead Nurse, written communication, May 2005). nities in Mexico and other parts of Latin Amer- poisoning cases where he worked. The clinic, ica. This connectedness enables families to in Seaside, Calif, is one of Monterey County METHODS communicate regularly, eat and transport foods Health Department’s 3 community-based pri- prepared in their hometowns, and preserve lan- mary care clinics. There were no epidemio- We used epidemiological methods common guage and customs while living in separate logical data from the County or State Health in outbreak investigations, which incorporated countries, all of which creates binational com- Department that could answer the questions case–control and cohort designs. Qualitative munities.7–8 Unfortunately, the same conditions of the doctor and nurse at the Seaside clinic methods, including focus groups, were used to

900 | Research and Practice | Peer Reviewed | Handley et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

2001. Abstracted information included sex, race/ethnicity, birthplace, BLL, and answers to lead-risk screening questions. We matched subjects on age at first lead screening (3 groups: 0 to <1 years; 1 to ≤3 years; and 4 to 6 years) to remove the possible confounding for children who were tested outside the 12- to 36-month screening window. We also abstracted prenatal and primary care charts of mothers of cases and controls to determine whether children were either born in or had families from Oaxaca, Mexico.

Prenatal Lead Testing and Assessment of Risk Factors In August 2002 the Monterey County Health Department Clinics instituted a policy of prenatal blood lead testing during the routine prenatal testing done at the first visit. Between November 2002 and August 2003, a subset of the women at the Seaside and Salinas clinics who were screened for lead was also interviewed about risk factors for lead poisoning. Together, these clinics saw more than 95% of all prenatal patients who at- FIGURE 1—State of Oaxaca, Mexico, with Zimatlan District and State Capital (Oaxaca City) tended county clinics for prenatal care. shaded. Women were asked during a routine visit with a trained comprehensive perinatal services coordinator if they were interested in the study, complement and inform the epidemiological Food and Drug Administration 2003; sample and those who gave written informed consent findings.9,10 Initial steps focused on estimating 030675), and by M.V. were interviewed about risk factors for lead the prevalence of elevated BLLs among chil- poisoning. Women approached for the study dren screened for lead in Monterey County Screening Children for Lead in Monterey were those who could return for a scheduled Health Department Primary Care Clinics, in- County visit. All women who received the lead testing cluding the clinic in Seaside and 2 others in We analyzed laboratory data for children were counseled about their test result and the neighboring communities of Salinas and aged from birth to 6 years to estimate the given prevention education that was tailored Marina. A case–control study in Seaside to de- prevalence of elevated BLLs among screened to their literacy and language background. termine if pediatric lead cases were concen- children. State and county policies require all Risk factors included area of birth, consump- trated among the families from Oaxaca, Mex- children to be screened for lead at 12 and tion of imported foods, use of lead-glazed ce- ico, followed. Focus groups with families of 36 months (with all children to be screened ramics, and occupational and environmental lead-poisoned children, Oaxacan community at least once by age 72 months), in accor- exposures to lead. Because we were interested healers (curanderas), and pregnant women with dance with California’s Child Health and in examining whether women from different elevated BLLs were then conducted in Sea- Disability Prevention Program guidelines.6,11 areas of Oaxaca had different risk profiles for side, to better understand community percep- Children are tested for lead through either a elevated BLLs, we examined as risk factors tions about lead poisoning and to inform the venous or capillary blood sample.12 both town of birth and the town where the next study design. The next study examined family of the woman or her partner lived. risk factors for lead among pregnant women Case–Control Study Examining Risk We next began a prospective investigation who entered care at 2 of the primary care clin- Factors in Seaside of women receiving prenatal care. We be- ics in Monterey County that saw the majority We reviewed charts of pediatric elevated lieved that if we found a similar prevalence of prenatal patients in the clinic system. BLL cases and a sample of controls (with of elevated BLLs and similar risk factors, we Food testing was conducted by West Coast BLLs <10 µg/dL) identified from the labora- would be able to identify some of the poten- Analytic Service, the State of California Food tory database as Seaside clinic patients tially preventable exposures occurring in this and Drug Administration (State of California, screened between January 1997 and June community. Also, because we were beginning

May 2007, Vol 97, No. 5 | American Journal of Public Health Handley et al. | Peer Reviewed | Research and Practice | 901  RESEARCH AND PRACTICE 

to suspect a food-borne source of lead, we Case–Control Study Examining Risk participated. The study population was almost wanted to understand food preparation and Factors in Seaside entirely Latina (95%) and born in Mexico importation practices to better develop lead The results of the case–control analysis (87%). Sixty-six women were from the state poisoning prevention messages.14–16 are summarized in Table 1. More than 70% of Oaxaca: 32 were from the Zimatlan area (90/126) of pediatric cases were born in the (12 from Santa Ynez Yatzeche, 15 from San RESULTS United States, although the majority of cases Pablo Huixtepec, and 5 from Zimatlan) in the born in Mexico were from Oaxaca. Almost western central valley of Oaxaca, and 34 Prevalence of Elevated Blood Lead all cases (95%) were Latino, compared with were from towns in the eastern central valley Levels 70% of controls. Close to 40% of cases had or from the Pacific coast of Oaxaca. These 2 For children tested between January 1997 chart documentation of 1 or more venous- areas of Oaxaca represent culturally distinct and June 2001, with 1 observation per child confirmed BLLs equal or greater than 20 communities—the Zimatlan-area Oaxacans are per year, the prevalence of elevated BLLs at µg/dL. We believed that the high proportion Zapotecans, and the Oaxacans from the east- Seaside was 3 times that of the clinics in Sali- of pediatric cases with a BLL of 20 µg/dL ern valley are primarily Miztecans.7 nas and Marina (6.0% compared with 1.7% or higher suggested an acute exposure, The prevalence of elevated BLLs in the pre- each) and significantly higher than popula- which was consistent with our understanding natal patients study population was 12%; 18% tion-based estimates in the United States over of the unique foods and importation prac- of the women screened in Seaside and 1% of the same period (2.2%).5 Seaside accounted tices that were taking place between vendors the women screened in Salinas had elevated for 47% of pediatric lead tests but 80% of from Seaside and family members living in BLLs. Table 2 presents demographic and risk the pediatric lead cases in Monterey County. Oaxaca. factor characteristics of women in the study, by Lead screening rates for children in this target BLL, and associated relative risk (RR) estimates age group exceeded 97% in each of the 3 Prenatal Testing and Assessment of and 95% confidence intervals (CI). Women clinics.13 These findings suggested that the Risk Factors with elevated BLLs were more likely to be lead problem in Monterey County was con- A total of 214 women enrolled in the Oaxacan born (96%) (compared with women centrated in Seaside, so the investigation was study between November 2002 and August born elsewhere in Mexico) and from Seaside focused there. 2003. All women approached for the study (96%); more likely to eat imported foods from Mexico, including foods produced locally or home-grown by family members, such as TABLE 1—Demographic Characteristics of Pediatric Cases and Controls: Seaside Family pumpkin seeds, tortillas, chocolate, and chapu- Health Center, Monterey County, Calif, 1997–2001 lines (home-prepared dried grasshoppers; 84%); and more likely to report having a Cases (BLL ≥10 µg/dL), Controls (BLL <10 µg/dL), friend or relative “with lead in their blood” No. (%; n=146) No. (%; n=285) Matched Odds Ratio (95% CI) (28%). In an analysis restricted to the 66 Age (matched) women from Oaxaca, the RR for women born ≤1 year 10 (7) 36 (12) NA in the Zimatlan area compared with other 2 to 3 years 98 (67) 183 (64) NA areas of Oaxaca was 4.0 (95% CI=1.7, 9.5). 4 to 6 years 38 (26) 66 (24) NA Women from the Zimatlan area reported simi- Male 67 (46) 149 (52) 0.8 (0.6, 1.2) lar patterns of imported food consumption as Ethnicitya women from other areas of Oaxaca. Latino 139 (95) 192 (70) 8.4 (3.8, 18.7) Women with elevated BLLs were not sig- Other 7 (5) 81 (30) NA nificantly more likely than women with nor- Birthplaceb mal BLLs to be younger, more recently ar- Mexico 35 (28) 12 (5) 7.6 (3.8, 15.4) rived immigrants (immigration within 1 year United States 90 (72) 236 (95) NA compared with more than 1 year or ≤5 years Region of birthc compared with 6 or more years), prepare Oaxaca, Mexico 26 (74) 4 (33) 5.7 (1.4, 23.9) food in lead-glazed ceramics or to have used Other Mexico 9 (26) 8 (67) NA lead-glazed ceramics growing up, live in older housing, live in remodeled houses, live with Notes. BLL=blood lead level; CI=confidence interval; NA=not applicable. Charts were reviewed and abstracted for 146 of the 157 cases (93%) and 285 of the 296 controls (96%). someone who uses lead in their job, or to an=419. Ethnicity information was missing for 12 controls.“Other” includes 1 Asian and 6 White cases, and 33 White, 21 have been out of the United States in the pre- African American, 15 Asian, and 12 other controls. vious year. bn=378. For 20 cases and 33 controls, birthplace information was missing. Only 1 case and 5 controls were from countries outside the United States or Mexico. In an analysis restricted to the 54 Oaxa- cRestricted to the 47 children born in Mexico (35 cases and 12 controls). can-born women screened at Seaside, almost half (n=23; 44%) had elevated BLLs.

902 | Research and Practice | Peer Reviewed | Handley et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 2—Patient Characteristics, Responses to Lead Screening Questions, and Relative Among Oaxacan-born women, women from Risk Estimates for Elevated Blood Lead Levels (BLLs): Monterey County, Calif, 2003 the Zimatlan area were more likely to have elevated BLLs than women from other parts ≥ All Patients, BLL <10 µg/dL, BLL 10 µg/dL, Relative Risk of Oaxaca. Nineteen of the 32 Zimatlan-born No. (%; n=214) No. (%; n=189) No. (%; n=25) (95% CI) women had elevated BLLs (59%), compared Patient characteristics with 4 of the 22 women (18%) from other Age areas of Oaxaca screened at the Seaside clinic ≤20 years 42 (20) 35 (19) 7 (28) 1.6 (0.7, 3.6) (RR=3.3; 95% CI=1.3, 8.3). By comparison, 21 years and older 172 (80) 154 (81) 18 (72) the prevalence of elevated BLLs among a Ethnicity women from other parts of Mexico who Latina 205 (96) 180 (95) 25 (100) inestimable participated in the study and were screened Other 9 (4) 9 (5) 0 (0) at Seaside was less than 1% (1/161)—this Region of birthb woman reported that she was not born in Oaxaca, Mexico 66 (31) 42 (22) 24 (96) 44 (6.1, 318.0) Oaxaca but that her family lived in Zimatlan. Other Mexico 121 (69) 120 (78) 1 (4) In Salinas, no women screened for lead were Region of Oaxaca from Zimatlan, but the 1 woman with an ele- Zimatlan, Oaxaca 32 13 19 4.0 (1.7,9.5) vated BLL was from another area of Oaxaca. Other Oaxaca 34 29 5 Recent arrival to United Statesc In the 6 months after the introduction of ≤1 year in United States 47 (24) 39 (20) 8 (32) 1.5 (0.7, 3.2) prenatal lead testing by the county health de- 2 or more years in United States 148 (76) 131 (80) 17 (68) partment, we estimated the prevalence of ele- Clinic location vated BLLs among all new prenatal patients Seaside 135 (63) 111 (59) 24 (96) 14.0 (1.9, 101.8) seen in the Salinas and Seaside clinics. This Salinas 79 (17) 78 (41) 1d (4) was done to determine how representative Responses to lead screening questions the subset of interviewed study participants How often do you eat food that you was of all women entering care in the clinics prepare or store in clay pottery?e during this period. Between August 2003 Any reported frequency 15 (7) 12 (6) 3 (12) 1.8 (0.6, 5.2) and February 2004, we conducted a chart re- Never 195 (93) 173 (94) 22 (88) view of all women who entered prenatal care How often did you use clay pottery at these 2 clinics. Demographic information, growing up? including birthplace and lead test results, Any reported frequency 158 (81) 135 (79) 23 (92) 2.7 (0.7, 11.2) were abstracted for 420 of the 450 women Never 38 (19) 36 (11) 2 (8) (93%) entering prenatal care. In this sample, Do you have a friend or relative who has 402 women had complete lead test results had a high level of lead in their and demographic data. The prevalence of ele- blood? vated BLLs was 13% in Seaside and 0% in Yes 19 (9) 12 (6) 7 (28) 4.0 (1.0, 8.2) Salinas. All 27 cases were from Oaxaca, al- No 193 (91) 175 (94) 18 (72) though the town or area of origin was not re- Do you live or stay in a house that has recently been renovated? corded in the charts. Women in the chart re- Yes 51 (24) 49 (26) 2 (8) 0.3 (0.7, 1.2) view study were similar to women in the No 161 (56) 139 (54) 22 (92) interview study with regard to age, ethnicity, Do you live in house with peeling paint? and time in the United States. Yes 27 (13) 24 (13) 3 (12) 0.9 (0.3, 2.9) No 187 (87) 165 (87) 22 (88) Confirmation of Food-Borne Lead Do you live or stay with someone who Contamination uses lead in their job? In September 2003, a 2-year-old boy Yes 10 (5) 10 (5) 0 (0) inestimable screened as part of his Child Health and Dis- No 202 (95) 177 (95) 25 (100) ability Prevention Program visit in Seaside had Have you lived or traveled outside the a BLL of 36 µg/dL, after having a much United States in the past year? lower lead test result the preceding November Yes 45 (21) 38 (20) 7 (28) 1.5 (0.6, 3.3) (4 µg/dL). The child was born in the United No 168 (79) 150 (80) 18 (72) States, but his parents had previously immi- Continued grated from San Pablo Huixtepec in Oaxaca. The child’s mother provided samples of foods

May 2007, Vol 97, No. 5 | American Journal of Public Health Handley et al. | Peer Reviewed | Research and Practice | 903  RESEARCH AND PRACTICE 

TABLE 2—Continued no indication that they were produced com- mercially. Several chapulin samples had ex- How often do you eat ____ from Mexico? tremely high amounts of lead. Additional cha- f Imported foods pulin samples that were boiled but not Any reported frequency 130 (61) 109 (58) 21 (84) 3.4 (1.2, 9.5) seasoned contained 580 ppm and 3 ppm of Never 84 (39) 80 (32) 4 (16) lead. Eating foods with lead levels like these Imported pumpkin seeds would result in daily intake levels that ex- Any reported frequency 59 (28) 48 (25) 11 (44) 2.1 (1.0, 4.3) ceeded the Food and Drug Administration’s Never 155 (72) 141 (75) 14 (56) provisional tolerable intake level for children Imported tortillas of 6 µg of lead per day by many orders of Any reported frequency 65 (30) 45 (24) 20 (80) 9.2 (3.6, 23.4) magnitude (based on an estimate of 5 to 15 Never 149 (70) 144 (76) 5 (20) ounces consumed daily).18 Additional samples Imported mole of pumpkin seeds and tortillas had no detect- Any reported frequency 41 (19) 36 (19) 5 (20) 1.1 (0.4, 2.6) Never 173 (81) 153 (81) 20 (80) able lead (less than 1 ppm), and 1 sample of Imported chocolate mole (a prepared and condensed sauce) con- Any reported frequency 51 (24) 40 (21) 11 (44) 2.5 (1.2, 5.2) tained 40 ppm. Never 163 (76) 149 (79) 14 (56) Imported tamarind candy DISCUSSION Any reported frequency 21 (10) 21 (11) 0 (0) NA Never 193 (90) 168 (89) 25 (100) We found a significant public health Imported chapulinesa problem of lead poisoning among a group Any reported frequency 37 (17) 24 (13) 13 (52) 5.2 (2.6, 10.4) of immigrants who have come to Seaside, Never 177 (83) 165 (87) 12 (48) Calif, from Oaxaca, Mexico. The high prevalence of elevated BLLs in both children and Note. NA indicates data is not applicable. Chapulin=seasoned dried grasshopper. Screening questions were adapted from Centers for Disease Control and Prevention–recommended lead screening questions for children.Additional pediatric pregnant women over several years, as well as screening questions were asked, including questions about living in a house built before 1960, the use of home health the association between elevated BLLs and at remedies, eating nonfood items, and living with someone who uses lead in their hobby, but the frequency of positive least 1 food that is widely eaten within this responses for these questions was too small to calculate risk estimates. aRelative risk estimates were inestimable for ethnicity because of zero cell frequencies for non-Latina cases with community, suggests an even greater problem elevated BLLs. among the Oaxacan community in Seaside, bRestricted to the 187 women born in Mexico.Values in brackets indicate the birth regions within Oaxaca. Relative risks and possibly among other communities that presented for Oaxaca-born compared with born elsewhere in Mexico (n=187), and for Zimatlan-born compared with born elsewhere in Oaxaca (n=66). have not yet been identified. The majority of c n=195. Eighteen women born in the United States and 1 for whom time in United States was missing were excluded. Oaxacans who live in Seaside are from indige- Relative risks for <5 years compared with >5 years in the United States were almost identical (2.0; 95% CI=0.7, 5.0). d nous Zapotecan communities that are distinct This woman indicated that her parents were from Oaxaca, although she was born elsewhere in Mexico. en=210. Examples of clay pottery were shown to study subjects during the interview. Data for past use of clay pottery were from other Oaxacans who live in Monterey restricted to the 196 women born outside the United States. County, and it is possible that they may have f “Any imported food” was coded from any frequency reported to the food-specific questions. Imported herbs and herbal unique sources of lead exposure that have not remedies were also asked about but were only reported by 7 women. been determined. The fact that reports of eating imported foods, including tortillas (with a received from her family in Oaxaca for testing. The investigation of the child’s home by the RR higher than 9), were associated with ele- Tortillas and a bag of chapulines that she had Monterey County Childhood Lead Prevention vated BLLs among prenatal patients needs to received 2 weeks previously, through an im- Program did not find common lead-containing be substantiated with wider testing of im- porter that relayed foods directly from families sources such as home health remedies, lead- ported foods. That several chapulin samples living in Oaxaca to families in Seaside, were glazed pottery, or food sources known to obtained in Seaside and in Zimatlan con- tested. The mother reported that the child had have lead such as tamarind candy (Donna tained high levels of lead suggests that the consumed about a small baby food jar’s worth Staunton, Public Health Nurse, written com- home community in Oaxaca has significant of chapulines in the previous few days (about munication, May 2005). ongoing lead exposure. There have been no 4 ounces). Food analyses indicated that the Between August 2003 and June 2005 we studies of this area to date to determine the chapulines contained 2300 ppm of lead, and collected and tested several samples of chapu- extent of lead poisoning or to pinpoint the the tortillas contained less than 0.1 ppm. The lines obtained from vendors in Seaside who lead sources so that widespread lead poison- mother was counseled to remove the food reported that the samples were from Zimat- ing can be prevented. source, and a state-wide health alert was is- lan, from vendors in the marketplace in Zi- Although there is strong evidence that the sued by the California Department of Health matlan, and from fields in the Zimatlan area origins of the elevated BLLs in Seaside are Services with regard to imported chapulines.17 (Table 3). Products were not labeled and had associated with lead contamination of food in

904 | Research and Practice | Peer Reviewed | Handley et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 3—Lead Levels in Chapulin Samples Originating From Zimatlan, Oaxaca in an area of Wales contaminated with mine tailings found similar lead levels in local Sample Origin Date Collected Lead Level species of grasshopper to those found in the Collected in Seaside chapulin samples collected in this study, Seaside market—reported to be from Zimatlan, Oaxaca August 2003 40 ppm which raises the possibility that lead could be October 2003 none detected bioaccumulated in locally harvested foods, in- 26 Family of child with elevated BLL—imported from family October 2003 2300 ppm cluding chapulines, in Oaxaca. These poten- in San Pablo Huixtepec, Oaxaca tial routes of exposure will be examined in Collected in Zimatlan future studies we are developing with rela- Zimatlan market February 2004 13 ppm tives of case families in Zimatlan. March 2005 157 ppm We are not aware of any studies that have March 2005 2500 ppm previously identified an outbreak of lead poi- March 2005 none detected soning from home-prepared and locally pro- August 2005 1500 ppm duced foods. Clearly, there is an extensive sys- Field outside Zimatlan (boiled but not seasoned) December 2003 580 ppm tem of transporting foods between binational February 2004 3 ppm communities that remains outside the international surveillance systems designed to iden- Note. Chapulin=seasoned dried grasshoppers. BLL=blood lead levels. tify food contamination in imported products. aSeasoned dried grasshoppers. These epidemiological findings highlight the importance of the collection of community- level data that is disaggregated both geograph- Oaxaca, there are also 3 lines of evidence show- of lead-glazed ceramics among women in our ically and within ethnic groups,27,28 because ing that the exposures are occurring in Sea- study is consistent with studies in Mexico.22 without them, the elevated BLLs in Seaside side: (1) the high proportion of pediatric lead If lead-glazed ceramics had been a strong among Oaxacans would have been diluted in cases that were born in the United States risk factor for elevated BLLs in the women in county-level prevalence assessments. They also (70%); (2) the data that indicate that recent im- the prenatal study, it would have been likely demonstrate the importance of community- migration or recent travel outside the United to affect all Oaxacan-born women, rather oriented primary care that is engaged with bi- States were not associated with elevated BLLs than primarily those from the Zimatlan area. national communities,29,30 and of the increas- among prenatal women studied; and (3) the There is no known local production of leading relevance of globalization in the examina- food test results (from foods obtained in Sea- glazed ceramics in this area, with the majority tion of health outcomes. We were able to side) and the case report of the child who had of lead-glazed ceramics produced in eastern identify the lead problem in Seaside as a rean initially low BLL followed by an extremely Oaxaca and distributed to other parts of sult of the insightfulness of the clinicians who high one after eating lead-contaminated foods Oaxaca and Mexico.23 Lead-glazed ceramics noticed the cases with high lead levels and in Seaside that were sent by his family in Oax- are available, however, in local markets in were aware of the binational nature of the aca. We believe that the lead exposures in the Monterey County and some women did re- lives of Oaxacans living in Monterey County. Seaside community result from both ongoing port using them for preparing certain types of An understanding of the different indigenous exposures from imported foods that contain foods. We suspect that use of lead-glazed ce- groups, towns of origin, the fluidity of migra- lead and from past exposures that may have re- ramics does contribute to the elevated BLLs tion patterns, and of the importance of food sulted in the storage of lead in bone and other in this population but that it is not the pri- importation in this community was critical to organs that is later released into the blood, in- mary source of lead exposure. It is possible the development of the outbreak investigation. cluding during pregnancy.19 that lead-glazed ceramics are being used in The problem of elevated BLLs in Monterey That elevated BLLs were not strongly as- Zimatlan in a manner that is increasing the County reflects a binational problem. It is our sociated with current or past use of lead- amount of lead in food, and it is also possible hope that by determining the origins of the glazed clay ceramics among prenatal patients that ingredients themselves are contaminated lead exposure in this community, many more is surprising because their use has been asso- with lead, through food preparation practices cases of lead poisoning will be averted, both ciated with higher mean BLLs in studies in (such as the drying of ingredients or grinding among migrant communities and among their Mexico.20–21 Our outbreak investigation was and milling practices).24 communities of origin. focused on risk factors for elevated BLL Environmental contamination is also a pos- (rather than for mean BLL), which might par- sible source of lead because the Zimatlan Val- tially account for this difference. However, ley has a history of silver mining,25 in which About the Authors our focus group interviews confirm that lead-contaminated mine tailings have been At the time of the study, Margaret A. Handley, Kaitie Drace, Robert Wilson, and Mary Croughan were with the women do not widely use Mexican pottery in dispersed throughout the area. One study that Department of Family Medicine and Community Medicine, the United States, and our finding of past use examined soil, plant, and invertebrate samples University of California, San Francisco. Celeste Hall and

May 2007, Vol 97, No. 5 | American Journal of Public Health Handley et al. | Peer Reviewed | Research and Practice | 905  RESEARCH AND PRACTICE 

Evie Diaz were with the Monterey County Health Depart- National Health and Nutrition Examination Surveys. Still Linger. Rockville, Md: Food and Drug Administra- ment, Salinas, Calif, and the Department of Family and JAMA. 1994;272:284–291. tion; 1998. Community Medicine, University of California, San Fran- 2. Grosse SD, Matte TD, Schwartz J, Jackson RJ. Eco- 19.Tellez-Rojo MM, Hernandez-Avila M, Lamadrid- cisco. Eric Sanford was with the Monterey County Health nomic gains resulting from the reduction in children’s Figueroa H, et al. Impact of bone lead and bone re- Department, Salinas, and the Department of Family Medi- exposure to lead in the United States. Environ Health sorption on plasma and whole blood lead levels during cine Collaborative Research Network, University of Califor- Perspect. 2002;110:563–569. pregnancy. Am J Epidemiol. 2004;160:668–678. nia, San Francisco. Enrique Gonzalez-Mendez was with the Southwest Community Health Clinic in Sonoma 3. Meyer PA, McGeehin MA, Falk H. A global ap- 20. Hernandez Avila M, Romieu I, Rios C, Rivero A, County, Santa Rosa, Calif, and the Department of Family proach to childhood lead poisoning prevention. Int J Palazuelos E. Lead-glazed ceramics as major determi- Medicine Collaborative Research Network, University of Hyg Environ Health. 2003;206:363–369. nants of blood lead levels in Mexican women. Environ Health Perspect. 19 91;9 4:117–120. California, San Francisco. Mario Villalobos was with la 4. Gordon B, Mackay R, Rehfuess E. Inheriting the Universidad Nacional Autonoma de Mexico, Mexico City. World: The Atlas of Children’s Health and the Environ- 21.Rothenberg SJ, Perez Guerrero IA, Perroni Requests for reprints should be sent to Margaret A. ment. Geneva, Switzerland: World Health Organization; Hernandez E, et al. Sources of lead in pregnant women Handley, PhD, Assistant Professor, Department of Family 2004. in the Valley of Mexico [in Spanish]. Salud Publica Mex. and Community Medicine, University of California, 1990;32:632–643. San Francisco, San Francisco General Hospital, 995 5. Meyer PA, Pivetz T, Dignam TA, Homa DM, Potrero Ave, Ward 83, San Francisco, CA 94110 (e-mail: Schoonover J, Brody D; Centers for Disease Control 22. Ettinger AS, Tellez-Rojo MM, Amarasiriwardena C, [email protected]). and Prevention. Surveillance for elevated blood lead et al. Effect of breast milk lead on infant blood lead This article was accepted December 20, 2005. levels among children—United States, 1997–2001. levels at 1 month of age. Environ Health Perspect. 2004; MMWR Surveill Summ. 2003;52:1–21. 112 : 13 81–1385. 6. Screening Young Children for Lead Poisoning: Guid- 23. Hernandez-Serrato MI, Mendoza-Alvarado LR, Contributors ance for State and Local Public Health Officials. Atlanta, Rojas-Martinez R, et al. Factors associated with lead M. A. Handley originated the study, supervised all as- Ga: Centers for Disease Control and Prevention; 1997. exposure in Oaxaca, Mexico. J Expo Anal Environ Epi- pects of its implementation and data analysis, and led demiol. 2003;13:341–347. the writing. C. Hall and E. Sanford originated the 7. Fox J, Rivera-Salgado G, eds. Indigenous Mexican study, conducted data collection, and provided writing Migrants in the United States. La Jolla, Calif: Center for 24. Richter E, El-Sharif N, Fischbein A, et al. Re- support. E. Diaz and K. Drace conducted data collec- US–Mexico Studies and Center for Comparative Immi- emergence of lead poisoning from contaminated flour tion. E. Gonzalez-Mendez and M. Croughan originated gration Studies, University of California, San Diego; in a West Bank Palestinian village. Int J Occup Environ the study and provided writing support. R. Wilson con- 2004. Health. 2000;6:183–186. ducted data analysis. M. Villalobos conducted sample 8. Grieshop J. The envois of San Pablo Huixtepec, 25. Alvarez LR. Geografia General del Estado de Oax- analyses. Oaxaca: food, home, and transnationalism. Hum Organ. aca. Oaxaca, Mexico: Ediciones Carteles;1994:456. 2006;65:400–406. 26.Milton A, Johnson MS, Cook JA. Lead within Acknowledgments 9. Borkan JM. Mixed methods studies: a foundation ecosystems on metalliferous mine tailings in Wales and The authors were supported by grants from the Univer- for primary care research. Ann Fam Med. 2004;2:4–6. Ireland. Sci Total Environ. 2002;299:177–190. sity of California Institute for Mexico and the United 10.Creswell JW, Fetters MD, Ivankova NV. Designing 27. Fielding JE, Frieden TR. Local knowledge to en- States, the Monterey Commission on Families and a mixed methods study in primary care. Ann Fam Med. able local action. Am J Prev Med. 2004;27:183–184. Children, the California Academy of Family Physicians, 2004;2:7–12. and la Universidad Nacional Autonoma de Mexico 28. Chen JY, Diament AL, Kagawa-Singer M, Pourat N, (grant UNAM-PAPIIT 0113905). 11. Updated Guidance for Screening for and Manage- Wold C. Disaggregating data on Asian and Pacific Is- We acknowledge the contributions of the Monterey ment of Childhood Lead Poisoning. Sacramento Calif: lander women to assess cancer screening. Am J Prev County Health Department and the staff at Seaside Childhood Health and Disability Prevention Program, Med. 2004;27:139–145. State of California Department of Health Services; Family Health Center, Alisal Health Center, and Marina 29. Mullan F, Epstein L. Community-oriented primary 2003:1,§704. Health Center, in Monterey County, Calif. care: new relevance in a changing world. Am J Public We also acknowledge the personal contributions of 12. Dignam TA, Evens A, Eduardo E, et al. High- Health. 2002;92:1748–1755. the following individuals: Susan McNelly, who at the intensity targeted screening for elevated blood lead 30.Westfall JM, Mold J, Fagnan L. Practice-based time of the study was with the Monterey County Health levels among children in 2 inner-city Chicago commu- research—“Blue Highways” on the NIH roadmap. Department; Dean Schillinger, University of California, nities. Am J Public Health. 2004;94:1945–1951. JAMA. 297(4):403–406. San Francisco, Division of General Internal Medicine, 13. Status Report: Quality Improvement Plan for Mon- San Francisco General Hospital; and Pilar Fernández- terey County Health Department’s Division of Primary Lomelín from Laboratorio de Análisis Físicos y Quími- Care. Salinas, Calif: Monterey County Health Depart- cos del Ambiente, Instituto de Geographia, for the diges- ment Division of Primary Care; 2002. tions and analyses of samples and her guidance in the analytic work of the undergraduate chemistry students 14 . Gardella C. Lead exposure in pregnancy: a review at la Universidad Nacional Autonoma de Mexico—Belen of the literature and argument for routine prenatal Ramírez, Pedro Bazán, and Claudia Merino; and Raul screening. Obstet Gynecol Surv. 2001;56:231–238. Aguirre Gomes, Instituto de Geographia. 15. Klitzman S, Sharma A, Nicaj L, Vitkevich R, Note. The findings and conclusions are those of the Leighton J. Lead poisoning among pregnant women authors and do not necessarily represent the views of in New York City: risk factors and screening practices. the funding agencies. J Urban Health. 2002;79:225–237. 16. [No authors listed.] Screening for elevated blood Human Participant Protection lead levels. American Academy of Pediatrics Commit- This study was approved by the institutional review tee on Environmental Health. Pediatrics. 19 9 8;101: board at the University of California, San Francisco, and 1072–1078. the Monterey County Division of Primary Care. 17. Health Warning Related to Consumption of Chapu- lines (Grasshoppers) From Oaxaca, Mexico. Sacramento: References State of California Department of Health Services; No- vember 13, 2003. 1. Pirkle JL, Brody DJ, Gunter EW, et al. The decline in blood lead levels in the United States. The 18.Farley D. FDA Consumer Alert: Dangers of Lead

906 | Research and Practice | Peer Reviewed | Handley et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Family History of Diabetes, Awareness of Risk Factors, and Health Behaviors Among African Americans

| Kesha Baptiste-Roberts, PhD, Tiffany L. Gary, PhD, Gloria L.A. Beckles, MD, MSc, Edward W. Gregg, PhD, Michelle Owens, PhD, Deborah Porterfield, MD, and Michael M. Engelgau, MD, MS

Diabetes is a major contributor to morbidity Objectives. We examined the role of family history of diabetes in awareness of and mortality and generates large direct as diabetes risk factors and engaging in health behaviors. 1–3 well as indirect costs. The prevalence of di- Methods. We conducted a cross-sectional analysis of 1122 African American 4 abetes among US adults has increased sub- adults without diabetes who were participants in Project DIRECT (Diabetes In- stantially over the past several decades and terventions Reaching and Educating Communities Together). stood at 8.7% in 2002.2,5 The burden of Results. After adjustment for age, gender, income, education, body mass index, type 2 diabetes disproportionately affects Af- and perceived health status, African Americans with a family history of diabetes rican Americans. For example, data from na- were more aware than those without such a history of several diabetes risk fac- tionally representative samples show that tors: having a family member with the disease (relative risk [RR]=1.09; 95% con- White men are one half to one fifth as likely fidence interval [CI]=1.03, 1.15), being overweight (RR=1.12; 95% CI=1.05, 1.18), not exercising (RR=1.17; 95% CI=1.07, 1.27), and consuming energy-dense foods as African American men to have or develop (RR=1.10; 95% CI=1.00, 1.17). Also, they were more likely to consume 5 or more diabetes, and African American women are servings of fruits and vegetables per day (RR=1.31; 95% CI=1.02, 1.66) and to have approximately twice as likely as White been screened for diabetes (RR=1.21; 95% CI=1.12, 1.29). 6,7 women to have or develop the disease. It is Conclusions. African Americans with a family history of diabetes were more estimated that 33% to 50% of people with aware of diabetes risk factors and more likely to engage in certain health behav- 8,9 type 2 diabetes are not diagnosed. As a re- iors than were African Americans without a family history of the disease. (Am J sult, many patients may already have early Public Health. 2007;97:907–912. doi:10.2105/AJPH.2005.077032) complications of the disease at the time of their clinical diagnosis. In addition to older age and being over- diabetes in the development of type 2 dia- METHODS weight, family history is a well-known risk betes, insulin resistance, and obesity.13–18 factor for type 2 diabetes, with risk esti- One study examined the impact of family Study Population mates (relative risks [RRs]) ranging from 2 history of diabetes on glycemic control,19 We conducted a cross-sectional study in- to 6 depending on study design and case and a few studies have examined whether volving African American adults without dia- definition.10 Family histories reflect both in- family history of diabetes influences per- betes who were participating in Project herited genetic susceptibilities and shared ceived susceptibility and protective health DIRECT (Diabetes Interventions Reaching environments, which include cultural factors behaviors.20,21 However, these studies have and Educating Communities Together) at such as preferences, values, and perceptions been conducted among nonminority popula- baseline (i.e., before implementation of any and behavioral factors such as diet and tions. To our knowledge, the specific role interventions). Project DIRECT, a multiyear physical activity.11 Thus, family history of di- that family history of diabetes plays among community-based project based in North Car- abetes may be a useful tool to identify indi- African Americans in terms of their being olina, is designed to improve diabetes detec- viduals at increased risk of the disease and aware of diabetes risk factors and engaging tion, quality of care, self-care services, and risk target behavior modifications that could po- in protective health behaviors has not yet factors among African Americans; health pro- tentially delay disease onset and improve been explored. motion, outreach, and diabetes care are the 3 health outcomes. For example, individuals In this study, we conceptualized our data in main intervention components. The methods with impaired glucose tolerance could be the context of the Health Belief Model,22 a used in the project have been described in de- encouraged to make lifestyle changes, given psychological model that attempts to explain tail elsewhere.23–25 Briefly, the target popula- that results from randomized clinical trials and predict health behaviors by focusing on tion consisted of civilian, noninstitutionalized indicate that losing weight, reducing fat in- people’s attitudes and beliefs. We hypothe- adults 18 years or older who resided in se- take, and increasing physical activity can re- sized that individuals with a family history of lected areas of Raleigh and Greensboro, NC. sult in a 58% reduction in the incidence of diabetes would be more aware of risk factors A multistage area probability sample design diabetes.12 and more likely to engage in healthy behav- was used in which area segments were se- Previous research has focused primarily iors than would individuals without such a lected from census files and then sample hous- on examining the role of family history of family history. ing units were selected for screening.

May 2007, Vol 97, No. 5 | American Journal of Public Health Baptiste-Roberts et al. | Peer Reviewed | Research and Practice | 907  RESEARCH AND PRACTICE 

Lead letters were sent to each sampled ad- pregnancy), and behavioral variables (physical (Stata Corp, College Station, Tex) survey com- dress to inform the residents of the purpose activity, attempting weight loss, and participa- mands. We conducted descriptive analyses and legitimacy of the survey. Trained field in- tion in screening for diabetes) were self- (i.e., means and frequencies) for sociodemo- terviewers visited each sample housing unit reported. Weight was measured, and height graphic, risk factor awareness, and health and selected eligible participants according to was obtained through a self-report. Body mass behavior variables. We used the t test and specific criteria. Eligible participants were index (BMI; weight in kilograms divided by χ2 test to determine statistical differences on asked to complete an in-person interview. The height in meters squared) categories, classified these variables between participants with and overall interview response rate was 87%. The according to National Institutes of Health without a family history of diabetes. weighted response rate was higher in Greens- guidelines, were optimal or underweight (less We conducted multiple logistic regression boro (88.9%) than in Raleigh (84.4%). The than 25 kg/m2), overweight (25–29.9 kg/m2), analyses to evaluate the relationship between final study sample included 2310 participants and obese (30 kg/m2 or above).26 having a family history of diabetes and dia- (2210 African Americans, 65 Whites, and 35 Participants were asked whether any of betes risk awareness variables. Initially we ad- members of other races). In this analysis, we their immediate family members (mother, fa- justed for sociodemographic factors (age, used data from the baseline assessment con- ther, sisters, brothers) had diabetes. Possible gender, education, and income), and in subse- ducted in 1997 and evaluated the 1585 indi- responses were “yes,” “no,” and “don’t know.” quent models we also adjusted for BMI and viduals who identified themselves as African Individuals were classified as having a family perceived health status. In addition, we evalu- American and reported not having diabetes. history of diabetes if they reported that any ated the relationship between health behaviors Participants with missing data on socio- first-degree relative (parent or sibling) had and having a family history of diabetes while demographic (not including income), family diabetes and as not having a family history adjusting for sociodemographic factors, BMI, history, or dietary variables were excluded, if they answered “no” or “don’t know” to all and perceived health status. Because of the yielding a final study population of 1122. parts of the question. A modified version of high prevalence of many of the outcome vari- Sample sizes varied slightly for a few variables the Block questionnaire was used in assessing ables, we used Zhang and Yu’s method29 to es- as a result of missing values. Missing data fruit, vegetable, and fat intakes.27 We calcu- timate relative risks from odds ratios generated were most frequent for income (n=1023) and lated daily servings of fruits and vegetables by our logistic regression models. diabetes screenings (n=1078). and total fat intake (in grams) using equa- Responses regarding attempts at weight tions derived by Block et al.27 (as outlined RESULTS loss were included only in analyses involving elsewhere28). participants who were overweight or obese As a means of assessing awareness of risk The participants were predominantly (n=732). Only participants who were not at- factors for diabetes, participants were asked women (62%), and 35% were 50 years or tempting to lose weight (n=779) were asked whether each of a series of 7 factors “defi- older. Sixty-three percent had a high school whether they were trying to maintain their nitely does not increase,” “probably does not education or less, 63% were employed, and weight. Participants included in the analyses increase,” “probably increases,” or “definitely 61% had an annual income below $25000. were similar to those not included because of increases” a person’s chance of developing Fifteen percent rated their health as fair or missing data for variables of interest with re- diabetes; responses were assigned codes of 1 poor. Overall, 20% of the participants had a spect to all of the variables examined except to 4, respectively (0 was assigned for “don’t mother with diabetes, 11% had a father with that higher percentages of those included re- know”). We investigated awareness separately diabetes, and 18% had a sibling with diabetes ported consuming 5 or more servings of fruits for each of the risk factors assessed and, in (in total, 36% of the participants had a family and vegetables per day (P=.036), engaging addition, calculated an overall awareness history of diabetes). Sixty-five percent were in physical activity (P=.040), and having score as the sum of all of a participant’s re- overweight or obese, and 15% had received been screened for diabetes (P=.031). A sponses (range: 0 to 28). Participants were advice from a doctor to lose weight. Thirty- lower percentage of the included participants classified as being aware of a risk factor if two percent were attempting to lose weight, had received advice from a doctor to lose they indicated that it “probably increases” or and 36% were trying to maintain their weight (P=.004). “definitely increases” a person’s chance of de- weight. Approximately one third (34%) were veloping diabetes. For our logistic regression sedentary. Study Variables and Measures analyses, we created awareness score tertiles Participants reported an average of 3.8 Data on sociodemographic characteristics and compared those in the second and third daily servings of fruits and vegetables, and (age, gender, education, income), health vari- tertiles with those in the first tertile. Also, we 23% reported consuming 5 or more servings ables (self-rated health status, advice from a split scores at the median19 to classify partici- a day. Mean total intake of fat per day was doctor to lose weight, family history of dia- pants as “not aware” or “aware.” 87.8 g. Approximately two thirds of the par- betes), awareness of risk factors for diabetes ticipants (66%) had been screened for dia- (older age, overweight, family members with Statistical Analyses betes. Levels of awareness of the risk of dia- diabetes, insufficient exercise, minority race/ To account for the complex survey design, betes imposed by being overweight and by ethnicity, energy-dense diet, diabetes during we conducted all analyses using Stata version 8 having family members with the disease both

908 | Research and Practice | Peer Reviewed | Baptiste-Roberts et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 1—Selected Baseline Characteristics, by Family History of Diabetes: African In the multivariate analyses, having a fam- American Project DIRECT Participants (n=1122), 1997 ily history of diabetes was initially significantly associated with greater awareness of Family History, No Family History, the risks of being overweight, having family Characteristics No. (%) (n=410) No. (%) (n=712) P members with diabetes, not exercising Sociodemographic variables enough, and consuming a energy-dense diet Age < 50 y 255 (64.0) 458 (65.2) .711 (Table 2). These associations persisted after Male gender 127 (32.3) 282 (41.6) .003* adjustment for gender, age, income, and edu- High school education or less 265 (63.8) 450 (63.3) .872 cation as well as further adjustment for BMI Yearly income < $25000 241 (62.0) 407 (61.5) .879 and perceived health status. Analyses involv- Health variables ing awareness score tertiles showed that hav- Health status fair or poor 81 (19.7) 92 (12.8) .005* ing a family history was associated with BMI optimal or underweight 121 (29.8) 269 (37.8) .009* greater awareness among participants in the Doctor advised weight loss 78 (18.6) 96 (13.3) .020* second (RR=1.19; 95% confidence interval Awareness of risk factors [CI]=0.99, 1.41) and third (RR=1.53; 95% Older age 239 (60.1) 405 (58.0) .523 CI=1.20, 1.90) tertiles than among partici- Overweight 354 (87.2) 547 (77.0) <.001 pants in the first tertile. Having a family his- Family members with diabetes 343 (83.6) 533 (76.0) .002* tory of diabetes was also associated with hav- Insufficient exercise 285 (69.4) 424 (60.1) .004* ing a summary awareness score above the Minority race/ethnicity 196 (49.9) 311 (44.0) .098 median; after all stages of adjustment, the rel- Energy-dense diet 296 (73.4) 472 (66.6) .021* ative risk was 1.20 (95% CI=1.10, 1.29). Diabetes during pregnancy 96 (23.3) 165 (23.2) .967 After adjustment for age, gender, income, Awareness score 19+ (median = 19) 245 (59.7) 328 (47.4) .001 education, BMI, and perceived health status, Behavioral variables participants with a family history of diabetes Sedentary 144 (34.1) 252 (34.7) .850 were more likely to consume 5 or more Attempting weight loss 138 (48.5) 189 (42.8) .164 servings of fruits and vegetables per day Attempting to maintain weight 150 (55.9) 252 (50.5) .181 (RR = 1.31; 95% CI = 1.02, 1.66) and to Less than 5 daily servingsa of fruits and vegetables 298 (73.1) 566 (80) .021* have been screened (RR = 1.21; 95% Daily total fat intake < 73 g 104 (24.7) 176 (24.2) .844 CI=1.12, 1.29) than those with no family Screened for diabetes 301 (74.6) 413 (61.2) <.001 history (Table 3). In addition, awareness of diabetes risk factors was associated with Note. DIRECT=Diabetes Interventions Reaching and Educating Communities Together. Sample sizes vary for some variables as a result of missing values. health behaviors independent of family his- aUS Department of Agriculture food pyramid definitions. tory. Adjusted analyses showed that individu- *P<.05. als who were aware of diabetes risk factors were more likely to be attempting weight loss (RR=1.22; 95% CI=1.04, 1.39), en- exceeded 75%, but percentages were below such a history to be aware of 4 of the 7 dia- gaging in physical activity (RR = 1.19; 95% 50% with respect to the risks associated with betes risk factors assessed: being over- CI=1.02, 1.36), consuming 5 or more having diabetes during pregnancy and be- weight, having family members with dia- servings of fruits and vegetables per day longing to a minority racial/ethnic group (the betes, not exercising enough, and eating an (RR = 1.27; 95% CI = 1.11, 1.41), and partici- percentage was below 25% in the case of dia- energy-dense diet. When the median sum- pating in diabetes screenings (RR = 1.25; betes during pregnancy). mary awareness score was used to define 95% CI = 1.08, 1.42). The unadjusted analyses showed that women awareness level, those with a family history were more likely than men to have a family again were more aware of risk factors for DISCUSSION history of diabetes (Table 1), and those with a diabetes (59.7% vs 47.4%). In addition, sig- family history were more likely than those nificant differences between the 2 groups Our results suggest that having a family without a family history to rate their health as were observed for 2 health behaviors. history of diabetes is associated with better fair or poor (19.7% vs 12.8%). Higher propor- Those with a family history of diabetes were awareness of diabetes risk factors, more daily tions of participants with a family history of dia- more likely than those without a family his- consumption of fruits and vegetables, and betes were overweight and had received advice tory to eat 5 or more servings of fruits and participation in diabetes screening. The rate from a doctor to lose weight. vegetables per day (26.9% vs 20.4%) and of awareness of diabetes risk factors among Participants with a family history of dia- to have been screened for diabetes (74.6% our participants with a family history of the betes were more likely than those without vs 61.2%). disease exceeded that observed by Pierce et

May 2007, Vol 97, No. 5 | American Journal of Public Health Baptiste-Roberts et al. | Peer Reviewed | Research and Practice | 909  RESEARCH AND PRACTICE 

TABLE 2—Relationships Between Family History of Diabetes and Awareness of Diabetes more servings of fruits and vegetables per day. Risk Factors: African American Project DIRECT Participants (n=1122), 1997 Also, in our univariate analyses, we found that those with a family history were more likely Model 1, Model 2, Model 3, to receive advice from a doctor to lose weight, Crude Risk Ratio Adjusted RRa Adjusted RRb Adjusted RRc but we could not determine whether they Risk Factor (95% CI) (95% CI) (95% CI) (95% CI) were counseled specifically on physical activ- Older age 1.04 (0.93, 1.14) 1.04 (0.92, 1.15) 1.04 (0.93, 1.14) 1.03 (0.92, 1.14) ity. Murff et al.31 examined such counseling by Overweight 1.13 (1.07, 1.18) 1.12 (1.05, 1.18) 1.13 (1.07, 1.18) 1.12 (1.05, 1.18) health care providers and reported no signifi- Family members with diabetes 1.10 (1.04, 1.15) 1.10 (1.03, 1.15) 1.10 (1.04, 1.15) 1.09 (1.03, 1.15) cant differences according to family history. Insufficient exercise 1.16 (1.05, 1.25) 1.18 (1.07, 1.27) 1.16 (1.05, 1.25) 1.17 (1.07, 1.27) Finally, we used parents and siblings (i.e., Minority race/ethnicity 1.13 (0.98, 1.29) 1.15 (0.99, 1.32) 1.13 (0.98, 1.29) 1.15 (0.99, 1.32) first-degree relatives) to define family history, Energy-dense diet 1.10 (1.02, 1.18) 1.09 (1.00, 1.17) 1.10 (1.02, 1.18) 1.09 (1.00, 1.17) whereas some studies have included both Diabetes during pregnancy 1.01 (0.78, 1.27) 1.05 (0.80, 1.36) 1.07 (0.81, 1.38) 1.07 (0.81, 1.37) first- and second-degree relatives (e.g., aunts Awareness score comparison and uncles) in their definition. Results from Tertile 2 vs tertile 1 1.15 (0.95, 1.35) 1.20 (0.99, 1.41) 1.20 (0.99, 1.41) 1.19 (0.99, 1.41) studies involving first-degree family members Tertile 3 vs tertile 1 1.57 (1.26, 1.91) 1.53 (1.20, 1.91) 1.53 (1.19, 1.90) 1.53 (1.20, 1.90) only and those combining first-, second-, and third-degree relatives (e.g., first cousins) have Note. DIRECT=Diabetes Interventions Reaching and Educating Communities Together; CI = confidence interval RR = relative risk. 32,33 aAdjusted for gender, age, income, and education. been similar. bAdjusted for gender, age, income, education, and body mass index. Our univariate analyses showed that women c Adjusted for gender, age, income, education, body mass index, and perceived health status. were more likely than men to report having a family history of diabetes. This result was similar to that of Annis et al., who found that more women than men reported having a TABLE 3—Relationships Between Family History of Diabetes and Selected Behavioral first-degree relative with diabetes and that Variables: African American Project DIRECT Participants (n=1122), 1997 women were more likely to report on female Model 1, Model 2, Model 3, than on male relatives with diabetes.34 Fur- a b c Crude Risk Ratio Adjusted RR Adjusted RR Adjusted RR thermore, a recent study showed that, in com- Behavioral Variable (95% CI) (95% CI) (95% CI) (95% CI) parison with men, women were slightly more Physically active (vs sedentary) 1.01 (0.90, 1.11) 1.02 (0.90, 1.13) 1.02 (0.90, 1.13) 1.03 (0.91, 1.14) likely to regard family history as important to Attempting weight loss 1.17 (0.93, 1.43) 1.13 (0.90, 1.49) 1.13 (0.90, 1.49) 1.13 (0.89, 1.40) their own health and were more likely to col- 35 Attempting to maintain weight 1.11 (0.95, 1.26) 1.07 (0.91, 1.23) 1.05 (0.88, 1.21) 1.06 (0.88, 1.22) lect family medical information. 5+ daily servingsd of fruits and 1.32 (1.05, 1.64) 1.31 (1.02, 1.65) 1.29 (1.00, 1.62) 1.31 (1.02, 1.66) Several limitations of this study deserve vegetables (vs < 5) comment. First, the study was cross-sectional, Daily total fat intake < 73 g (vs ≥73 g) 1.02 (0.82, 1.24) 1.02 (0.81, 1.25) 1.01 (0.81, 1.24) 1.00 (0.80, 1.24) and thus we are unable to make causal infer- Screened for diabetes 1.22 (1.13, 1.30) 1.22 (1.13, 1.30) 1.22 (1.13, 1.30) 1.21 (1.12, 1.29) ences. Second, most of the data were obtained through self-report, which may have Note. DIRECT=Diabetes Interventions Reaching and Educating Communities Together; CI = confidence interval; RR = relative risk. resulted in biased estimates of measures of aAdjusted for gender, age, income, and education. bAdjusted for gender, age, income, education, and body mass index. association. Third, the questions used to as- cAdjusted for gender, age, income, education, body mass index, and perceived health status. sess awareness of diabetes risk factors were dUS Department of Agriculture food pyramid definitions. developed by the Project DIRECT investiga- tive team, and their reliability and validity are not known as of yet. al.,21 who examined 105 offspring of parents control behaviors, more often than individu- Fourth, many diabetes cases are undiag- with type 2 diabetes. Pierce et al. found that als without a family history of diabetes. In ad- nosed,4 and people may not know the dia- 49% of offspring recognized parental history dition, our findings are not in accord with the betes status of all members of their family, as a risk factor for diabetes, but fewer recog- results of a United Kingdom study showing resulting in misclassification. Most likely, we nized that being overweight (38%) and older that first-degree relatives of people with underestimated the percentage of our partici- (33%) and engaging in little exercise (21%) type 2 diabetes consumed diets higher in fat pants who had a family history of the disease, were additional risk factors. Our results do and cholesterol, increasing their risk of devel- and thus the strength of the associations ob- not support the findings of Forsyth and oping diabetes.30 served was probably attenuated. A study con- Goetsch20; in their study, individuals with a We found that individuals with a family his- ducted by Bensen et al., however, showed family history of diabetes engaged in health- tory of diabetes were more likely than those that accuracy of probands’ reporting of dia- protective behaviors, specifically weight without a family history to consume 5 or betes among family members was quite high,

910 | Research and Practice | Peer Reviewed | Baptiste-Roberts et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

with reports of parents’ diabetes status having their resulting behaviors. No doubt, many of D. Porterfield, and M.M. Engelgau contributed to the somewhat greater sensitivity (87%) than these individuals believe themselves to be at interpretation and discussion of the findings. those of siblings’ status (72%).36 increased risk, and if they become knowl- Fifth, our study population consisted of a edgeable regarding the health behaviors that Acknowledgments sample of African Americans from Raleigh may delay onset of the disease or reduce its This work was funded by the Division of Diabetes and Greensboro, NC, so our results cannot be risk, they may engage in these behaviors. Translation, Centers for Disease Control and Prevention. We thank the Project DIRECT staff, executive extrapolated to all African Americans residing Other people, however, may have fatalistic board, and community members for their work on in the United States. Finally, we were not able attitudes, believing that diabetes is inevitable the study. We also thank the study participants for to distinguish between type 2 diabetes and regardless of what they do. Powe et al. con- their cooperation. type 1 diabetes, and thus we were not able to ceptualized fatalism within the context of the assess any differing effects they had on African American experience as a complex Human Participant Protection diabetes-related health behaviors or aware- psychological cycle characterized by feelings This study was approved by the Committee on Human Research at Johns Hopkins University. All study proce- ness of diabetes risk factors. We were also un- of powerlessness, worthlessness, meaningless- dures were explained to participants, and written in- 38,39 able to evaluate the impact of survival on par- ness, and social despair. Fatalism has formed consent was obtained. ticipant’s knowledge of family diabetes been shown to be associated with poor dia- history. We can posit that perhaps individuals betes self-management, and it may be associ- References who succumbed to diabetes-related complica- ated with decreased engagement in healthy 1. Brancati FL, Kao WH, Folsom AR, Watson RL, tions may have been less knowledgeable behaviors.40 Szklo M. Incident type 2 diabetes mellitus in African about their diabetes and less effective in man- Nevertheless, our study has implications for American and white adults: the Atherosclerosis Risk in Communities Study. JAMA. 2000;283:2253–2259. aging the disease. behavior change that, as mentioned earlier, can 2. King H, Aubert RE, Herman WH. Global burden Despite these limitations, our study had be conceptualized through constructs derived of diabetes, 1995–2025: prevalence, numerical esti- 22 several strengths. First, the sample was popu- from the Health Belief Model. A family his- mates, and projections. Diabetes Care. 1998;21: lation based. Second, extensive data on dia- tory of diabetes may be indicative of perceived 1414 – 14 31. betes-related health behaviors were available. susceptibility to the disease, which would in 3. Harris MI, ed. Diabetes in America. Bethesda, Md: Third, we are unaware of any other published turn influence one’s likelihood of behavior National Diabetes Data Group; 1995. studies examining the interrelationships be- change. Awareness of diabetes risk factors can 4. American Diabetes Association. Diagnosis and classification of diabetes mellitus. Diabetes Care. 2004; tween family history of diabetes, health be- be considered a cue to action. Although Afri- 27(suppl 1):S5–S10. haviors, and diabetes risk awareness among can Americans with a family history of the 5. American Diabetes Association. Screening for African Americans, who are disproportion- disease may be more aware of associated risk type 2 diabetes. Diabetes Care. 2004;27(suppl 1): ately burdened by this disease. factors than those without such a history, ap- S11–S14. The results of the Diabetes Prevention Proj- propriately designed programs focusing on life- 6. Harris MI, Flegal KM, Cowie CC, et al. Prevalence ect and the emphasis on prediabetes interven- style improvements should nevertheless be of diabetes, impaired fasting glucose, and impaired glucose tolerance in U.S. adults: the Third National Health tions have raised the important question of targeted toward this high-risk group to delay or and Nutrition Examination Survey, 1988–1994. Dia- whether knowledge of diabetes risk factors in- prevent the development of diabetes. betes Care. 1998;21:518–524. fluences health behaviors. The present analy- 7. Lipton RB, Liao Y, Cao G, Cooper RS, McGee D. sis, albeit cross-sectional in design, suggests Determinants of incident non-insulin-dependent diabetes mellitus among blacks and whites in a national that a family history of diabetes (but not About the Authors sample: the NHANES I Epidemiologic Follow-up Study. Kesha Baptiste-Roberts and Tiffany L. Gary are with the knowledge of risk factors per se) is associated Am J Epidemiol. 19 93;138:826–839. Department of Epidemiology, Johns Hopkins Bloomberg with awareness of risk factors, consumption of School of Public Health, Baltimore, Md. Gloria L.A. Beckles, 8. Harris MI. Undiagnosed NIDDM: clinical and 5 or more servings of fruits and vegetables a Edward W. Gregg, Michelle Owens, and Michael M. public health issues. Diabetes Care. 19 93;16:642–652. Engelgau are with the Division of Diabetes Translation, day, and increased participation in screening. 9. Harris MI, Klein R, Welborn TA, Knuiman MW. Centers for Disease Control and Prevention, Atlanta, Ga. Onset of NIDDM occurs at least 4–7 years before clin- Although evidence is insufficient to conclude Deborah Porterfield is with the Division of Public Health, ical diagnosis. Diabetes Care. 19 92;15:815–819. that screening is effective in reducing morbid- North Carolina Department of Health and Human Ser- ity and mortality associated with diabetes,37 vices, Raleigh. 10. Harrison TA, Hindorff LA, Kim H, et al. Family Requests for reprints should be sent to Tiffany L. Gary, history of diabetes as a potential public health tool. the rate of screening participation in our study PhD, Department of Epidemiology, Johns Hopkins Am J Prev Med. 2003;24:152–159. should be considered a positive finding. Bloomberg School of Public Health, 615 N Wolfe St, Room 11.Keku TO, Millikan RC, Martin C, Rahkra-Burris TK, We did not observe differences in other E6531, Baltimore, MD 21205 (e-mail: [email protected]). Sandler RS. Family history of colon cancer: what does This article was accepted June 28, 2006. health behaviors according to family history it mean and how is it useful? Am J Prev Med. 2003;24: 170–176. status, which may seem surprising. This lack of differences may be explained in substantial Contributors 12.Tuomilehto J, Lindstrom J, Eriksson JG, et al. Pre- vention of type 2 diabetes mellitus by changes in life- K. Baptiste-Roberts and T.L. Gary originated and de- measure by the disproportionate burden of style among subjects with impaired glucose tolerance. signed the study. K. Baptiste-Roberts performed the N Engl J Med. 2001;344:1343–1350. diabetes borne by African Americans, which statistical analysis, and T.L. Gary was responsible for may influence their perceptions of risk and supervision. G.L.A. Beckles, E.W. Gregg, M. Owens, 13. Goran MI, Coronges K, Bergman RN, Cruz ML,

May 2007, Vol 97, No. 5 | American Journal of Public Health Baptiste-Roberts et al. | Peer Reviewed | Research and Practice | 911  RESEARCH AND PRACTICE 

Gower BA. Influence of family history of type 2 dia- Walker M. Non-diabetic relatives of type 2 diabetic betes on insulin sensitivity in prepubertal children. families: dietary intake contributes to the increased risk NOW Available on J Clin Endocrinol Metab. 2003;88:192–195. of diabetes. Diabet Med. 2001;18:984–990. 14 . Mitchell BD, Zaccaro D, Wagenknecht LE, et al. 31. Murff HJ, Rothman RL, Byrne DW, Syngal S. The DVD and VHS! Insulin sensitivity, body fat distribution, and family dia- impact of family history of diabetes on glucose testing betes history: the IRAS Family Study. Obes Res. 2004; and counseling behavior in primary care. Diabetes Care. 12:831–839. 2004;27:2247–2248. 15. Nyholm B, Nielsen MF, Kristensen K, et al. Evi- 32. Erasmus RT, Blanco BE, Okesina AB, Mesa AJ, dence of increased visceral obesity and reduced physi- Gqweta Z, Matsha T. Importance of family history in cal fitness in healthy insulin-resistant first-degree rela- type 2 black South African diabetic patients. Postgrad tives of type 2 diabetic patients. Eur J Endocrinol. Med J. 2001;77:323–325. 2004;150:207–214. 33. Sargeant LA, Wareham NJ, Khaw KT. Family his- 16.Onyemere KU, Lipton RB. Parental history and tory of diabetes identifies a group at increased risk for early-onset type 2 diabetes in African Americans and the metabolic consequences of obesity and physical in- Latinos in Chicago. J Pediatr. 2002;141:825–829. activity in EPIC-Norfolk: a population-based study. Int 17. Shaw JT, Purdie DM, Neil HA, Levy JC, Turner RC. J Obes Relat Metab Disord. 2000;24:1333–1339. The relative risks of hyperglycaemia, obesity and dys- 34. Annis AM, Caulder MS, Cook ML, Duquette D. lipidaemia in the relatives of patients with type II dia- Family history, diabetes, and other demographic and betes mellitus. Diabetologia. 1999;42:24–27. risk factors among participants of the National Health 18. Srinivasan SR, Frontini MG, Berenson GS. Longi- and Nutrition Examination Survey 1999–2002. Prev tudinal changes in risk variables of insulin resistance Chronic Dis. 2005;2:1–12. syndrome from childhood to young adulthood in offspring of parents with type 2 diabetes: the Bogalusa 35. Centers for Disease Control and Prevention. Heart Study. Metabolism. 2003;52:443–450. Awareness of family health history as a risk factor for disease–United States, 2004. MMWR Morb Mortal 19. Kao WH, Batts-Turner M, Rami T, Brancati FL, Wkly Rep. 2004;53:1044–1047. Gary TL. Parental history of diabetes and glycemic control in urban African Americans with type 2 dia- 36. Bensen JT, Liese AD, Rushing JT, et al. Accuracy betes. Diabetes. 2003;52(suppl 1):A222. of proband reported family history: the NHLBI Family n this moving video, the people of Heart Study (FHS). Genet Epidemiol. 19 9 9;17: 20.Forsyth LH, Goetsch VL. Perceived threat of ill- public health share what they do 141–150. ness and health protective behaviors in offspring of Iand how their work improves and adults with non-insulin-dependent diabetes mellitus. 37. Engelgau MM, Narayan KM, Herman WH. protects the lives of those in their com- Behav Med. 19 97;23:112–121. Screening for type 2 diabetes. Diabetes Care. 2000;23: munities. Color, 8 minutes. Available 1563–1580. 21.Pierce M, Harding D, Ridout D, Keen H, Bradley C. in DVD and VHS. Risk and prevention of type II diabetes: offspring’s 38.Powe BD, Johnson A. Fatalism as a barrier to can- views. Br J Gen Pract. 2001;51:194–199. cer screening among African Americans: philosophical $13.95 APHA Members • $19.95 Nonmembers Stock No. VHS 0-87553-033-8 22. Becker MH. The Health Belief Model and per- perspectives. J Religion Health. 19 95;34:119–125. sonal health behavior. Health Educ Monogr. 1974;2: 39.Powe BD, Weinrich S. An intervention to de- Stock No. DVD 0-87553-044-3 324–473. crease cancer fatalism among rural elders. Oncol Nurs Forum. 1999;26:583–588. American Public Health 23. Engelgau MM, Narayan KM, Geiss LS, et al. A Association project to reduce the burden of diabetes in the African- 40. Egede LE, Bonadonna RJ. Diabetes self-manage- 800 I Street, NW, American community: Project DIRECT. J Natl Med ment in African Americans: an exploration of the role Washington, DC 20001 Assoc. 1998;90:605–613. of fatalism. Diabetes Educ. 2003;29:105–115. www.apha.org 24.Gregg EW, Geiss LS, Saaddine J, et al. Use of diabetes preventive care and complications risk in two ORDER TODAY! African-American communities. Am J Prev Med. 2001; 21:197–202. American Public Health Association Publication Sales 25. Narayan KM, Gregg EW, Fagot-Campagna A, et al. Web: www.apha.org Relationship between quality of diabetes care and pa- E-mail: [email protected] tient satisfaction. J Natl Med Assoc. 2003;95:64–70. Tel: 888-320-APHA 26. Clinical Guidelines on the Identification, Evaluation, FAX: 888-361-APHA and Treatment of Overweight and Obesity in Adults: The Evidence Report. Bethesda, Md: National Institutes of Health; 1998. 27. Block G, Gillespie C, Rosenbaum EH, Jenson C. A rapid food screener to assess fat and fruit and vegetable intake. Am J Prev Med. 2000;18:284–288. 28. Gary TL, Baptiste-Roberts K, Gregg EW, et al. Fruit, vegetable, and fat intake in a population-based sample of African Americans. J Natl Med Assoc. 2004; 96:1599–1605. 29. Zhang J, Yu KF. What’s the relative risk? A method of correcting the odds ratio in cohort studies of common outcomes. JAMA. 1998;280:1690–1691. 30.Adamson AJ, Foster E, Butler TJ, Bennet S,

912 | Research and Practice | Peer Reviewed | Baptiste-Roberts et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Associations Between Body Composition, Anthropometry, and Mortality in Women Aged 65 Years and Older

| Chantal Matkin Dolan, PhD, MPH, Helena Kraemer, PhD, Warren Browner, MD, MPH, Kristine Ensrud, MD, MPH, and Jennifer L. Kelsey, PhD

More than 50% of the US adult population is Objectives. We examined the relation between measures of body size and mor- overweight or obese according to the criteria tality in a predominantly White cohort of 8029 women aged 65 years and older of the National Heart, Lung, and Blood Insti- who were participating in the Study of Osteoporotic Fractures. 1 tute (NHLBI). The NHLBI expert panel de- Methods. Body composition measures (fat and lean mass and percentage body fined overweight as body mass index (BMI; fat) were calculated by bioelectrical impedance analysis. Anthropometric mea- weight in kilograms divided by height in me- sures were body mass index (BMI; kg/m2) and waist circumference. ters squared) from 25.0 to 29.9 kg/m2 and Results. During 8 years of follow-up, there were 945 deaths. Mortality was low- obesity as BMI≥30.0 kg/m2.1 However, ap- est among women in the middle of the distribution of each body size measure. For 2 plying a single set of cutpoints to define over- BMI, the lowest mortality rates were in the range 24.6 to 29.8 kg/m . The U-shaped weight and obesity in different age groups relations were seen throughout the age ranges included in this study and were not attributable to smoking or measures of preexisting illness. Body composition mea- may not be appropriate. Several studies have sures were not better predictors of mortality than BMI or waist girth. suggested that the relative risk of mortality as- Conclusions. Our results do not support applying the National Institutes of sociated with increased BMI is greater among Health categorization of BMI from 25 to 29.9 kg/m2 as overweight in older women, 2–7 younger women than older women. because women with BMIs in this range had the lowest mortality. (Am J Public The shape of the relation between BMI and Health. 2007;97:913–918. doi:10.2105/AJPH.2005.084178) mortality is also controversial. One large prospective study showed a positive linear association between BMI and mortality in women 1988 through community-based listings in were instructed to maintain a normal fluid aged 30 to 55 years who had been followed and around Baltimore, Md; Minneapolis, balance and to abstain from vigorous physical for 16 years8; several other studies of women Minn; and Portland, Ore, and in the Mononga- activity and ingestion of alcohol and caffeine at various ages have reported a U-shaped rela- hela Valley area near Pittsburgh, Pa.20 Women for 12 hours prior to the clinic visit. tion,2,5,9–18 which indicates an elevated mortal- were recruited from voter registration lists Women were weighed while wearing in- ity risk among those with low BMI and those (Pennsylvania and Minnesota), driver’s license door clothing without shoes; weight was mea- with high BMI. Some evidence suggests that and identification card holders (Maryland), sured with a balance beam scale. Height was this nonlinear association may be the result of and health maintenance organization mem- measured with a wall-mounted stadiometer. not controlling for confounding by smoking or bership lists (Minnesota and Oregon). We did BMI was calculated from weight and height preexisting illness,19 but other studies have ob- not enroll women who were Black (because of at visit 2. Waist girth was measured with served a U-shaped distribution even when ad- their decreased risk for hip fracture, the end an inelastic tape measure during the visit 2 justing for these variables.5,9,10,12–16 point of greatest interest to the main study), examination. We used data from the Study of Osteo- who had bilateral hip replacements, or who Lean mass was estimated from BIA as porotic Fractures, a large prospective cohort were unable to walk without assistance. 0.470×(Height2/Resistance)+(0.170× study of predominantly White women aged More than 98% of the participants were Weight)+(0.03×Reactance)+5.7.21 Fat mass 65 years and older, to examine the relation White. Of the 9704 women who entered the was calculated as the difference between total between measures of obesity and mortality study at baseline, 85% of the surviving co- body weight and lean mass. Percentage body during an 8-year-average follow-up period. hort (n=8082) completed a follow-up clinic fat was fat mass expressed as a percentage of Body composition was measured directly by visit at year 2 (visit 2) between January 1989 total weight. bioelectrical impedance analysis (BIA) as well and January 1991 (when they were at least A validation substudy of 205 women as by traditional measures of adiposity, in- 67 years old). Bioelectric impedance meas- demonstrated that estimates from BIA were cluding BMI and waist circumference. urements were made only at visit 2. We in- well correlated with dual x-ray absorptiome- cluded the 8029 women who had complete try (DXA; Hologic QDR 1000, Hologic Inc, METHODS bioelectric impedance measurements so that Waltham, Mass) measures of fat mass (r=0.89) we could estimate lean mass, fat mass, and fat and lean mass (r=0.79).22 These correlations Study Participants and Measurements mass percentage. were consistent across all the age categories Women aged 65 years and older were re- All body composition and body size meas- and were observed despite an average of 2 cruited from September 1986 to October urements were made at visit 2. Participants years’ difference between the BIA and DXA

May 2007, Vol 97, No. 5 | American Journal of Public Health Dolan et al. | Peer Reviewed | Research and Practice | 913  RESEARCH AND PRACTICE 

measures. DXA has been validated as a pre- TABLE 1—Body Composition Measures and Pearson Correlation Coefficients for Association 23 cise measure of body composition. Between Body Composition Measures in Women Aged 65 Years and Older: Study of Study participants were contacted every 4 Osteoporotic Fractures, Baltimore, Md; Minneapolis, Minn; Portland, Ore; Monongahela months, and follow-up for mortality was more Valley Area, Pa; 1986–1997 than 99% complete.24 Because of relatively a small numbers in specific cause-of-death cate- Pearson Correlation Coefficients gories, overall mortality was used as the end Measures Mean ±SD BMI Body Fat Lean Mass Fat Mass Waist point. The average time from visit 2 until the BMI (kg/m2) 26.2 ±4.6 1.0 0.87 0.70 0.92 0.86 end of follow-up for this analysis (November Body fat, % 39.2 ±6.0 1.0 0.51 0.95 0.80 19 97) was 8 years. Lean mass, kg 39.8 ±4.5 1.0 0.73 0.66 Potential Confounding Variables Fat mass, kg 26.6 ±8.6 1.0 0.85 Information on most demographic, lifestyle, Waist, cm 84.1 ±11.5 1.0 and clinical covariates of interest was ob- Note. BMI = body mass index. tained at visit 2 by interview (alcohol con- aP≤.001. sumption, marital status, use of hormones, use of diuretics, and reproductive history) or by examination (muscle strength, including grip Cox regression model with a quadratic term table also shows that all of the measures of strength and femoral neck bone mineral den- because the association between each anthro- body composition and anthropometry were sity with DXA). Femoral neck bone mineral pometric measure and mortality was curvilin- highly correlated with each other. density has been associated with both obesity ear. Proportionality assumptions of the mod- No notable differences in height, use of and mortality.25 els were checked by plotting the log(−log) thiazide diuretics, use of oral estrogen, or al- Some covariates were measured only at survival curves. Interaction terms between cohol consumption were seen between sur- baseline, including walking for exercise, ciga- each body size measure and age were in- vivors and those who died during follow-up rette smoking (never, former, current), educa- cluded, but no interactions were apparent. (Table 2). Those who died during follow-up tion, self-reported health compared with oth- The optimal value (nadir of the curve)27 of were less well educated, were less likely to ers the same age (excellent, good, fair, poor, each body size variable was stable in all age be married or to walk for exercise, had lower very poor), diabetes, and hypertension. groups (66–69, 70–74, 75–79, 80–84, and grip strength, and were more likely to be Analyses ≥85 years), so all age groups were combined smokers, to report their health status as fair Descriptive statistical analyses were per- in the results presented here. We controlled or poor, and to use nonthiazide diuretics. formed to identify potential confounding vari- for the effects of age by including it as a con- Age, smoking status, self-reported health sta- ables for inclusion in multivariate models. For tinuously distributed covariate. tus, grip strength, nonthiazide diuretic use, continuous variables, analysis of covariance To depict the curvilinear associations be- and femoral bone mineral density were in- was used to estimate age-adjusted means and tween the body size measures and mortality, cluded in the multivariate analyses as poten- standard deviations among survivors and each body size measure was categorized into tial confounding variables because they were among those who died during follow-up. For 5 equally sized quintiles (on the basis of the also independently associated with mortality. categorical variables, percentages were ad- distribution in the entire sample at visit 2). justed to the age distribution of the entire Mortality rate ratios were calculated for each Associations Between Body Size and cohort (n=8029) at visit 2 by the direct quintile relative to the lowest quintile. Mortality method.26 Pearson correlation coefficients All statistical analyses were carried out with All measures of body size had a U-shaped were calculated to determine the correlations the SAS version 6.0 (SAS Institute, Cary, NC) relation with mortality, as indicated by a sta- between the anthropometric main variables and EGRET (Statistics and Epidemiology Re- tistically significant quadratic term (P<.05) of interest. search Corp, Seattle, Wash, and Cytel Inc, Cam- in the Cox proportional hazards models, Cox proportional hazards models were bridge, Mass) statistical programming packages. whether adjustment was made for age only or used to estimate the associations between an- for other potential confounding variables as thropometric variables and rate of mortality. RESULTS well. Table 3 and Figure 1 show multivariate- Models were run for all women, adjusting for adjusted mortality rates according to quintile age only; for all women, adjusting for multiple The median age of the cohort at visit 2 was of body size indicator. Table 4 presents the potential confounders; and for nonsmokers 72 years. During the follow-up period, 945 quintiles of the body composition and body only, adjusting for multiple potential con- deaths occurred among the 8029 women who size measures in this cohort. The lowest mor- founders. The censor date was either the date had complete BIA measures at visit 2. Table 1 tality rates consistently occurred in the mid- of death or the end of the follow-up period. gives the means and standard deviations for dle of the distributions of body size indica- Each body size measure was included in a the various body size measures at visit 2. The tors, and the highest mortality rates were at

914 | Research and Practice | Peer Reviewed | Dolan et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 2—Comparison of Selected Characteristics of Women Aged 65 Years and Older Who similar for all measures of body composition Survived (n=7084) and Those Who Died (n=945) During Follow-Up: Study of Osteoporotic and anthropometry. The lowest mortality was Fractures, Baltimore, Md; Minneapolis, Minn; Portland, Ore; Monongahela Valley Area, Pa; consistently observed among women in the 1986–1997 middle of the distributions of the body size measures, with the highest rates at either the Characteristic Survivors Died During Follow-Up lowest or the highest quintiles. These rela- Days to censora 2875 ±227.25 1587.3 ±261.30 tions were observed after an average 8-year Age at visit 2, ya 73.2 ±4.8 76.3 ±6.1 follow-up period across all of the age groups Height at visit 2, cm 159.2 ±5.9 159.0 ±5.8 and were not attributable to the confounding High school education or less,a % 60.7 65.1 effects of smoking. The U-shaped relation was Married, % 47.4 40.3 observed when we excluded women with pre- Smoking status,a % existing illnesses such as hypertension and di- Never smoker 62.2 48.4 abetes, women who died in the first 2 years Past smoker 29.4 33.8 of follow-up, and women who lost more than Current smoker 8.4 17.9 10% of their body weight since the age of 50. Drinks per week of alcohol 1.9 ±4.2 2.0 ±4.0 Our results are consistent with several large Self-reported health status,a % prospective cohort studies that have reported Excellent 33.8 21.4 a U-shaped relation between body size and Good 52.1 51.7 mortality among adult women of various age 2,5,9–18 Fair 13.1 23.3 groups. Both the American Cancer Poor or very poor 1.0 3.6 Society study, which included more than 2 Walks for exercise,a % 52.9 45.3 400000 women aged 30 years and older Grip strength,a kg 18.8 ±4.2 17.7 ±4.5 and a study from Norway of more than 11 Thiazide diuretic use, % 21.0 22.5 900000 women aged 15 to 90 years re- Nonthiazide diuretic use,a % 6.8 15.7 ported U-shaped relations between BMI and Oral estrogen use at visit 2, % 15.1 12.7 mortality. However, neither the American Cancer Society study nor the Norwegian ± Note. Continuous variables are mean SD; except for days to censor and age at visit 2, they were age adjusted by analysis of study adjusted for smoking. The Nurses covariance. Categorical variables were age adjusted to the age distribution of the cohort (n=8029) at visit 2 by the direct method. aDifference between survivors and those who died was significant (P≤.05). Health Study, which includes more than 115 000 women aged 30 to 55 years at baseline (followed for 16 years), reported that either end. The quadratic term was significant from smoking status. Among the nonsmok- after adjusting for smoking, the association for all body size variables when data were ers, the highest mortality consistently oc- between BMI and mortality was linear.19 stratified into 5-year age groups. curred among women in the highest quintile However, other studies of women in age When we estimated the optimal values of body size. groups more comparable to the Nurses (values at which mortality was lowest) for Because of the concern that preexisting ill- Health Study have adjusted for smoking and the body size measures,27 the nadir of the ness could influence the associations between still reported a U-shaped relation between curves (the lowest mortality rates) corre- body size and mortality, analyses were also body size and mortality.9,10,14,16,28 Recently, the sponded to values in the third or fourth adjusted for hypertension and diabetes. Leisure World Cohort (including more than quintiles of the distributions for each of the Again, the U-shaped relation between body 8000 women, mean age 73 years, followed measures. For example, the estimated opti- size measures and mortality was similar to over a 23-year period) reported a reverse- mal value for BMI was 29.2 kg/m2, which is that seen in the unadjusted results (data not J–shaped relation between BMI and mortal- within the range of the fourth quintile of the shown). Furthermore, the U shape was ob- ity, with controls for age at entry and smok- BMI distribution. served for the measures of body size when ing. Although obese women were at higher we excluded women who died within the first risk of mortality than were “normal-weight” Effects of Potential Confounders 2 years of follow-up or those had lost more women, the highest risk of mortality was ob- Among nonsmokers, the patterns of asso- than 10% of their body weight since age 50 served among underweight women, and thus ciations between quintile of body size mea- years. a reverse-J–shaped relation.18 sures and mortality were similar to the re- It is not surprising that the women at low- sults for the entire cohort (Table 3), DISCUSSION est risk for mortality are neither the thinnest confirming that the U-shaped association be- nor the most obese. However, the levels of tween body size measures and mortality is In this cohort of older, predominantly BMI associated with the lowest risk of mortal- not explained by uncontrolled confounding White women, the pattern of mortality was ity in our study merit comment. The BMI

May 2007, Vol 97, No. 5 | American Journal of Public Health Dolan et al. | Peer Reviewed | Research and Practice | 915  RESEARCH AND PRACTICE 

levels for the 2 quintiles at lowest risk were between 24.6 and 29.8 kg/m2, and the optimal value was estimated as 29.2 kg/m2. Ac- cording to the recent NHLBI guidelines, the majority of these women would be classified as overweight and almost obese. The Clinical Guidelines on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults recommends that “all overweight and obese adults (age 18 years of age or older) with a BMI of 25 kg/m2 or higher are considered at risk.”1 Our results suggest that these guidelines are not appropriate for older women and that classifying women over 65 years of age with BMI from 24.6 to 29.8 kg/m2 as overweight and therefore at increased risk for mortality may be incorrect. Note. To depict the curvilinear associations between the body size measures and mortality, each body size measure was Other studies have reported that the associa- categorized into 5 equally sized quintiles (on the basis of the distribution in the entire sample at visit 2). Ratios adjusted for tion between obesity and mortality is different age, smoking, self-reported health, grip strength, nonthiadine diuretic use, and femoral neck bone mineral density. for older and younger women.3,4,28–30 Perhaps FIGURE 1—Quintiles of body composition measures and mortality risk in women aged 65 a certain amount of adiposity confers a sur- years and older: study of osteoporotic fractures; Baltimore, Md; Minneapolis, Minn; vival advantage in elderly women. Some stud- Portland, Ore; Monongahela Valley Area, Pa; 1986–1997 ies have suggested that the association between body size and mortality in older women

TABLE 3—Adjusted Rate Ratios (RRs; With 95% Confidence Intervals [CIs]), by Quintile of Body Composition Measures in Women Aged 65 Years and Older: Study of Osteoporotic Fractures: Baltimore, Md; Minneapolis, Minn; Portland, Ore; Monongahela Valley Area, Pa; 1986–1997

Lean Mass Fat Mass Percentage Body Fat Body Mass Index Waist Girth

Age-adjusted RR (95% CI) First quintile 1.0 1.0 1.0 1.0 1.0 Second quintile 0.75 (0.62–0.91) 0.71 (0.59–0.85) 0.78 (0.65–0.94) 0.74 (0.61–0.89) 0.79 (0.64–0.99) Third quintile 0.67 (0.55–0.82) 0.58 (0.48–0.71) 0.70 (0.58–0.85) 0.62 (0.51–0.75) 0.84 (0.68–1.05) Fourth quintile 0.67 (0.55–0.82) 0.65 (0.53–0.79) 0.60 (0.49–0.74) 0.65 (0.53–0.79) 0.98 (0.79–1.20) Fifth quintile 0.87 (0.72–1.06) 0.81 (0.67–0.99) 0.84 (0.69–1.01) 0.86 (0.71–1.04) 1.09 (0.88–1.34) Multivariate-adjusted RR (95% CI)a First quintile 1.0 1.0 1.0 1.0 1.0 Second quintile 0.88 (0.72–1.080 0.80 (0.66–0.98) 0.88 (0.73–1.07) 0.80 (0.65–0.96) 0.87 (0.69–1.09) Third quintile 0.83 (0.67–1.08) 0.67 (0.54–0.83) 0.77 (0.63–0.94) 0.70 (0.57–0.87) 0.93 (0.75–1.17) Fourth quintile 0.88 (0.70–1.09) 0.72 (0.58–0.89) 0.65 (0.53–0.82) 0.72 (0.58–0.89) 1.10 (0.88–1.36) Fifth quintile 1.16 (0.92–1.45) 0.93 (0.75–1.16) 0.86 (0.70–1.06) 0.89 (0.72–1.10) 1.18 (0.94–1.47) Multivariate-adjusted RR (95% CI)b First quintile 1.0 1.0 1.0 1.0 1.0 Second quintile 0.88 (0.66–1.16) 0.71 (0.54–0.93) 0.89 (0.68–1.16) 0.94 (0.71–1.24) 1.05 (0.78–1.42) Third quintile 0.78 (0.58–1.05) 0.74 (0.56–0.98) 0.77 (0.58–1.02) 0.83 (0.62–1.11) 0.88 (0.64–1.22) Fourth quintile 1.06 (0.79–1.42) 0.80 (0.60–1.06) 0.64 (0.47–0.87) 0.81 (0.60–1.09) 1.14 (0.84–1.53) Fifth quintile 1.32 (0.98–1.79) 1.04 (0.77–1.40) 0.99 (0.75–1.32) 1.20 (0.90–1.60) 1.28 (0.94–1.75)

Note. To depict the curvilinear associations between the body size measures and mortality, each body size measure was categorized into 5 equally sized quintiles (on the basis of the distribution in the entire sample at visit 2). aAdjusted for age, smoking, self-reported health, grip strength, nonthiazide diuretic use, and femoral neck bone mineral density. bFor never smokers adjusted for age, self-reported health, grip strength, nonthiazide diuretic use, and femoral neck bone mineral density.

916 | Research and Practice | Peer Reviewed | Dolan et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 4—Quintiles of Body Composition and Anthropometry Measures of Women Aged 65 older adults) also reported an increased rela- Years and Older: Study of Osteoporotic Fractures: Baltimore, Md; Minneapolis, Minn; tive risk of mortality among the obese but Portland, Ore; Monongahela Valley Area, Pa; 1986–1997 little evidence of increased risk among those classified as overweight. 39 2 Quintile Lean Mass, kg Fat Mass, kg Percentage Body Fat Body Mass Index, kg/m Waist, cm Our results provide evidence of the First ≤36.09 ≤19.49 ≤34.34 ≤22.38 ≤74.1 U-shaped association between measures of Second >36.09–38.34 >19.49–23.63 >34.34–38.02 >22.38–24.56 >74.1–80.0 obesity and mortality in older White women Third >38.34–40.43 >23.63–27.68 >38.02–40.91 >24.56–26.73 >80.0–85.7 and extend these findings to specific measures Fourth >40.43–43.30 >27.68–33.26 >40.91–44.25 >26.73–29.82 >85.7–93.4 of fat and lean mass. Few studies have re- Fifth >43.30 >33.26 >44.25 >29.82 >93.4 ported on the prediction of mortality from body size measures in older women. The Note. To depict the curvilinear associations between the body size measures and mortality, each body size measure was shape of the relation was not attributable to categorized into 5 equally sized quintiles (on the basis of the distribution in the entire sample at visit 2). smoking, preexisting illness, or any other factors measured in this study. The patterns of risk were similar for the different body size is explained either by preexisting poor health unable to walk or with bilateral hip replace- measures. Using more complicated and expen- status2,3,8,11 or weight loss.31–34 We found that ments were excluded. Some error may have sive measures of body size such as BIA did the U-shaped relation between body size and been introduced by the 2-year time difference not provide an advantage over easier and less mortality remained when we adjusted for self- between visit 2 (when body size measurements expensive measures such as BMI and waist reported health status or excluded early deaths were obtained) and baseline (when some con- circumference. Finally, our results do not sup- as well as when we excluded women who had founding variables were measured). We did not port the application of the NHLBI guidelines lost more than 10% of their body weight since have an estimate of total caloric intake or infor- for the classification and treatment of over- they were aged 50 years. mation on dietary patterns during the study or weight to older women with BMIs of 25.0 to Although it is difficult to conclude which earlier in life. Finally, we were unable to exam- 29.9 kg/m2, because these women had the of the measures of body composition and an- ine the association between body size measures lowest rates of mortality for their age. thropometry best predicts mortality, we can and specific causes of death because of rela- draw a few practical conclusions. First, the tively small numbers in individual cause-of- U-shaped relation between body size and death categories. About the Authors mortality is consistent among these various Nevertheless, this is the largest prospective At the time of the study, Chantal Matkin Dolan and Jennifer L. Kelsey were with the Department of Health Research and highly correlated measures. Second, the more study of obesity and mortality that has in- Policy, Stanford University School of Medicine, Palo Alto, specific measures of obesity (BIA-measured cluded estimates of lean mass and fat mass in Calif. Helena Kraemer was with the Department of Psychia- lean mass, fat mass, and percentage body fat) older women. Previous studies have measured try and Behavioral Sciences, Stanford University School of Medicine, Palo Alto. Warren Browner is with the California do not provide an obvious advantage over the only BMI, weight, or weight change or used a Pacific Medical Center Research Institute, San Francisco, more general and less expensive indicators of measure of waist circumference. In addition, and the University of California, San Francisco. Kristine obesity (BMI, waist circumference) for predict- until recently there have been relatively few Ensrud is with the Division of Epidemiology, School of Pub- lic Health, University of Minnesota, Minneapolis. ing mortality. In the absence of a clear advan- studies of the association between obesity or Requests for reprints should be sent to Chantal Matkin tage of the BIA measures in predicting mor- body size and mortality in older women. Dolan, PhD, MPH, PO Box 448, Palo Alto, CA 94302 tality, lower cost and ease of measurement Our results showing minimum mortality (e-mail: [email protected]). This article was accepted May 10, 2006. favor the use of BMI or waist circumference. in the middle of the distribution of body Although this large community-based study composition levels are consistent with the re- Contributors of mortality in older women has many strengths, sults of the National Health and Nutrition C.M. Dolan developed the research proposal, analyzed it has some limitations. We did not enroll a Examination Survey I,13 which reported that the data, and led the writing of the article. H. Kraemer probability sample of a defined population, and a broad range of BMI values was associated assisted in the development of the statistical analysis plan and interpretation of the data. W. Browner and almost all the women were White. We cannot with lower mortality, as well as with other K. Ensrud contributed to the development of the hy- address possible variations in the association studies that have suggested that women clas- pothesis and the interpretation of the data. J.L. Kelsey between obesity and mortality by race or eth- sified as overweight may not be at excess contributed to the development and design of the study, the analysis, and the interpretation of the results. nicity. These results do not address mortality risk for mortality, particularly in older age All authors reviewed and edited drafts of the article. risk among women categorized as underweight groups.16 ,35–39 Furthermore, in a study com- according to NHLBI criteria (BMI<18.0 kg/m2), bining data from 5 prospective cohorts in the Acknowledgments because there were few such women in our United States, more than 80% of deaths at- We would like to thank investigators in the Study of sample (women in the lowest BMI quintile had tributable to excess weight were among those Osteoporotic Fractures Research Group from the fol- ≤ 2 2 40 lowing institutions: BMI 22.38 kg/m ). Also, this cohort is not with a BMI greater than 30 kg/m . A meta- University of California, San Francisco (coordinating representative of all older women, as those analysis of BMI and mortality (not limited to center): S.R. Cummings (principal investigator),

May 2007, Vol 97, No. 5 | American Journal of Public Health Dolan et al. | Peer Reviewed | Research and Practice | 917  RESEARCH AND PRACTICE 

M.C. Nevitt (coinvestigator), K.L. Stone (coinvestigator), weight and mortality among women. N Engl J Med. Mellstrom D. Bone mineral density is a predictor of D.C. Bauer (coinvestigator), D.M. Black (study statisti- 19 95;333:677–685. survival. Calcif Tissue Int. 1998;63:190–196. cian), H.K. Genant (director, central radiology laboratory), 9. Folsom AR, Kaye SA, Sellers TA, et al. Body fat 26. Selvin S. Measures of risk: rates and probabilities. R. Benard, T. Blackwell, W.S. Browner, M. Dockrell, distribution and 5-year risk of death in older women In: Statistical Analysis of Epidemiologic Data. New York, S. Ewing, C. Fox, R. Fullman, D. Kimmel, S. Litwack, [published erratum appears in JAMA. 19 93;269:1254]. NY: Oxford University Press; 1991, 1–35. L.Y. Lui, J. Maeda, P. Mannen, L. Nusgarten, L. Palermo, JAMA. 19 93;269:483–487. M. Rahorst, C. Schambach, J. Schneider, R. Scott, 27. Durazo-Arvizu R, McGee D, Li Z, Cooper R. Es- D. Tanaka, C. Yeung. 10. Harris T, Cook EF, Garrison R, Higgins M, Kannel W, tablishing the nadir of the body mass index mortality University of Maryland: M.C. Hochber (principal in- Goldman L. Body mass index and mortality among non- relationship: a case study. J Am Stat Assoc. 19 97;92: vestigator), L. Makell (project director), R. Nichols, smoking older persons. JAMA. 1988;259:1520–1524. 1312–1319. C. Boehm, L. Finazzo, T. Page, S. Trusty, B. Whitkop. 11.Waaler HT. Height, weight, and mortality. The 28. Rissanen A, Knekt P, Heliovarra M, Aromma A, University of Minnesota: K.E. Ensrud (principal inves- Norwegian experience. Acta Med Scand Suppl. 1984; Reunanen A, Maatela J. Weight and mortality in tigator), K. Margolis (coinvestigator), P. Schreiner (coin- 679:1–56. Finnish women. J Clin Epidemiol. 19 91;4 4:787–795. vestigator), K. Worzala (coinvestigator), S. Love (clinical research director), E. Mitson (clinic coordinator), C. Bird, 12. Durazo-Arvizu R, Cooper RS, Luke A, Prewitt TE, 29. Durazo-Arvizu R, Goldbourt U, McGee DL. Body D. Blanks, F. Imker-Witte, K. Jacobson, K. Knauth, Liao Y, McGee DL. Relative weight and mortality in mass index and mortality [letter]. N Engl J Med. 2000; N. Nelson, E. Penland-Miller, G. Saecker. U.S. blacks and whites: findings from representative 342:286–289. national population samples. Ann Epidemiol. 19 97;7: University of Pittsburgh: J.A. Cauley (principal investi- 30. Singh PN, Lindsted KD, Fraser GE. Body weight 383–95. gator), L.H. Kuller (coprincipal investigator), M. Vogt and mortality among adults who never smoked. Am J (coinvestigator), L. Harper (project director), L. Buck 13. Durazo-Arvizu RA, McGee DL, Cooper RS, Liao Y, Epidemiol. 1999;150:1152–1164. (clinic coordinator), C. Bashada, D. Cusick, G. Engleka, Luke A. Mortality and optimal body mass index in a 31. Harris TB, Launer LJ, Madans J, Feldman JJ. Co- A. Flaugh, A. Githens, M. Gorecki, D. Medve, M. Nasim, sample of the US population. Am J Epidemiol. 1998; hort study of effect of being overweight and change in C. Newman, S. Rudovsky, N. Watson, D. Lee. 147:739–749. weight on risk of coronary heart disease in old age. Kaiser Permanente Center for Health Research, Port- 14 . Sempos CT, Durazo-Arvizu R, McGee DL, Cooper RS, BMJ. 19 97;314:1791–1794. land, Ore: T. Hillier (principal investigator), E. Harris Prewitt T. The influence of cigarette smoking on the (coprincipal investigator), E. Orwoll (coinvestigator), 32. Losonczy KG, Harris TB, Cornoni-Huntley J, et al. association between body weight and mortality. The H. Nelson (coinvestigator), M. Aiken (biostatistician), Does weight loss from middle age to old age explain Framingham Heart Study revisited. Ann Epidemiol. J. Van Marter (project administrator), M. Rix (clinic co- the inverse weight mortality relation in old age? Am J 1998;8:289–300. ordinator), J. Wallace, K. Snider, K. Canova, K Pedula, Epidemiol. 19 95;141:312–321. J. Rizzo. 15. Dey DK, Rothenberg E, Sundh V, Bosaeus I, 33. Rumpel C, Harris TB, Madans J. Modification of Steen B. Body mass index, weight change and mortal- the relationship between Quetelet Index and mortality ity in the elderly. A 15 y longitudinal study of 70 y by weight-loss history among older women. Ann Epi- Human Participant Protection olds. Eur J Clin Nutr. 2001;55:482–492. The institutional review boards at each institution ap- demiol. 19 93;3:343–350. 16. Flegal KM, Graudbard BI, Williamson DF, Gail proved the study. All women provided written informed 34. Diehr P, Bild DE, Harris TB, Duxbury A, Siscov- HM. Excess deaths associated with underweight, over- consent at study entry and at each clinical examination. ick D, Rossi M. Body mass index and mortality in non- weight, and obesity. JAMA. 2005;293:1861–1867. smoking older adults: the cardiovascular health study. References 17. Katzmaryzk PT, Craig CL, Bouchard C. Under- Am J Public Health. 1998;88:623–629. weight, overweight and obesity: relationship with mor- 1. National Heart, Lung, and Blood Institute. Clinical 35. Farrell SW, Braun L, Barlow CE, Cheng YJ, Blair tality in the 13-year follow-up of the Canada Fitness Guidelines on the Identification, Evaluation, and Treat- SN. The relation of body mass index, cardiorespiratory Survey. J Clin Epidemiol. 2001;54:916–920. ment of Overweight: The Evidence Report. Bethesda, Md: fitness, and all-cause mortality in women. Obes Res. National Heart, Lung, and Blood Institute, 1998. NIH 18. Corrada MM, Kawas CH, Mozaffar F, Pahanini-Hill 2002;10:417–423. Publication No. 98-4083. A. Association of body mass index and weight change 36. Haapanen-Niemi N, Miilunpalo S, Pasanen M, with all-cause mortality in the elderly. Am J Epidemiol. 2. Lew EA, Garfinkel L. Variations in mortality by Vuori I, Oja P, Malmberg J. Body mass index, physi- 2006;163:938–949. weight among 750,000 men and women. J Chronic cal inactivity and low level of physical fitness as de- Dis. 1979;32:563–576. 19. Manson JE, Stampfer MJ, Hennekens CH, Willett terminants of all-cause and cardiovascular mortality— WC. Body weight and longevity: a reassessment. a 16 y follow-up of middle-aged and elderly men and 3. Lindsted KD, Singh P. Body mass index and 26- JAMA. 19 87;257:353–358. women. Int J Obes Relat Metab Disord. 2000;24: year risk of mortality among women who never 14 65–1474. smoked: findings from the Adventist Mortality Study. 20. Cummings SR, Black DM, Nevitt MC, et al. Ap- Am J Epidemiol. 19 97;146:1–11. pendicular bone density and age predict hip fracture in 37. Strawbridge WJ, Wallhagen MI, Shema SJ. New women. The Study of Osteoporotic Fractures Research NHLBI clinical guidelines for obesity and overweight: 4. Stevens J, Cai J, Pamuk ER, Williamson DF, Group. JAMA. 1990;263:665–668. will they promote health? Am J Public Health. 2000; Thun MJ, Wood JL. The effect of age on the associa- 90:340–343. tion between body-mass index and mortality. N Engl J 21. Lohman T. Advances in Body Composition Assess- Med. 1998;338:1–7. ment. Champaign, Ill: Human Kinetics Publishers; 1992. 38. Heiat A, Vaccarino V, Krumholz HM. An evidence- based assessment of federal guidelines for overweight 5. Calle EE, Thun MJ, Petrelli JM, Rodriguez C, 22. Ensrud KE, Lipschutz RC, Cauley JA, et al. Body and obesity. Arch Intern Med. 2001;161:1194–1203. Heath CW Jr. Body-mass index and mortality in a pro- size and hip fracture risk in older women: a prospec- spective cohort of U.S. adults. N Engl J Med. 1999;341: tive study. Study of the Osteoporotic Fractures Re- 39. McGee DL, Diverse Populations Collaboration. 10 97–1105. search Group. Am J Med. 19 97;103:274–280. Body mass index and mortality: a meta-analysis based on person-level data from twenty-six observational 6. Stevens J, Cai J, Juhaeri, Thun MJ, Williamson DF, 23. Mazess RB, Barden HS, Bisek JP, Hanson J. studies. Ann Epidemiol. 2005;15:87–97. Wood JL. Consequences of the use of different mea- Dual-energy x-ray absorptiometry for total-body and sures of effect to determine impact of age on the asso- regional-bone mineral and soft-tissue composition. Am 40. Allison DB, Fontaine KR, Manson JE, Stevens J, ciation between obesity and mortality. Am J Epidemiol. J Clin Nutr. 19 9 0;51:110 6–1112. VanItallie TB. Annual deaths attributable to obesity in 1999;150:399–407. the United States. JAMA. 1999;282:1530–1538. 24.Vogt MT, Cauley JA, Scott JC, Kuller LH, Browner 7. Bender R, Jockel KH, Trautner C, Spraul M, WS. Smoking and mortality among older women: the Merger M. Effect of age on excess mortality in obesity. study of osteoporotic fractures. Arch Intern Med. 1996; JAMA. 1999;281:1498–1504. 156:630–636. 8. Manson JE, Willet WC, Stampfer MJ, et al. Body 25. Johansson C, Black D, Johnell O, Oden A,

918 | Research and Practice | Peer Reviewed | Dolan et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

The Latino Paradox in Neighborhood Context: The Case of Asthma and Other Respiratory Conditions

| Kathleen A. Cagney, PhD, Christopher R. Browning, PhD, Danielle M. Wallace, MA

Racial and ethnic disparities in health status Objectives. Evidence indicates that foreign-born Latinos have a health advan- continue to cause concern prompting numer- tage compared with US-born persons of the same socioeconomic status. An ex- ous efforts in clinical practice and policy to en- planation for this paradox has remained elusive. We examined the extent to hance access to care, increase adherence to which this paradox exists for the prevalence of asthma and other respiratory con- treatment regimes, and improve health habits. ditions. We then explored the role of neighborhood social context in under- One anomaly in the disparities literature is the standing any observed advantage. We invoked theories of social organization, col- oft-documented “Latino Paradox.” Also de- lective efficacy, and the urban ethnic enclave. scribed as the Hispanic or Epidemiological Methods. We combined data from the Project on Human Development in Chi- Paradox, these terms refer to the relatively cago Neighborhoods Community Survey with 2 other data sources and used hi- good health and longer life span of foreign- erarchical generalized linear modeling techniques. Results. We found a distinctly graded effect for asthma and other breathing born Latinos when expectations (in most problems among foreign-born Latinos, depending on community composition. analyses, primarily based on their socioeco- Foreign-born Latinos embedded in a neighborhood that had a high percentage nomic status) suggest far greater morbidity and of foreign-born residents experienced a significantly lower prevalence of asthma mortality than observed. The paradox has and other breathing problems; those in communities that had a low percentage been documented for several health outcomes, of foreign-born residents had the highest prevalence overall (even when compared including all-cause mortality, infant mortality, with African Americans). and functional status.1–3 Recent work indicates Conclusions. Foreign-born Latinos have a respiratory health advantage only in that the mortality paradox may be the result of enclave-like settings. Contexts such as these may provide the cohesiveness crit- return migration effects, at least for those of ical for effective prevention. (Am J Public Health. 2007;97:919–925. doi:10.2105/ Mexican origin.4 Evidence related to morbid- AJPH.2005.071472) ity, however, continues to indicate a health advantage for foreign-born Latinos when com- living in communities without a strong immi- We employed theories of social organiza- pared with their US-born counterparts.5 grant presence may be at a disadvantage. tion, collective efficacy,15 ,16 immigrant adapta- In addition to migration and associated Knowledge of the comparative morbidity tion, and the urban ethnic enclave17–19 to un- data artifact explanations, social and cultural experience of US-born and foreign-born derstand how neighborhood social context capital (i.e., community-level social cohesive- Latinos is limited,7 as is understanding of contributes to asthma and other respiratory ness, norms, and practices) have been hy- the role of community context in shaping conditions, by Latino foreign-born status. pothesized as mechanisms responsible for their health. Asthma, in particular, remains Social organization and collective efficacy the relatively beneficial health trajectories of relatively understudied in the Latino popu- theories enabled us to draw out components Latino immigrants.4,6 High levels of social lation.8–12 The prevalence of asthma appears of community life that could be important capital may affect the community’s potential to vary across small areas and to be affected both to the prevalence of respiratory condi- to encourage positive health habits or sanc- by neighborhood characteristics13 ; attention tions and to the social organizational features tion negative ones. This hypothesized health to social context is important to understand- of the ethnic enclave. Structural features of behavior pathway, however, does not ad- ing the prevalence of asthma in population the neighborhood (e.g., residential stability, dress larger forces that operate at the neigh- subgroups.14 The urban ethnic enclave pro- ethnic heterogeneity, and economic status) set borhood level. Neighborhood-level influ- vides a rich example of the form of commu- the stage for neighborhood social processes to ences, such as the availability of social nity social context where networks are emerge. These social processes—collective effi- support or the accessibility and quality of dense and the transmission of health- cacy and social network interaction and social public parks, may have independent effects enhancing information is likely great. We exchange—may have independent effects on on health. Moreover, community characteris- examine the extent to which the Latino health.20 Collective efficacy captures the level tics may condition any Latino advantage; Paradox exists for asthma and other respira- of trust and attachment in the neighborhood that is, foreign-born Latinos may benefit tory conditions. We then extend previous re- that can be drawn on for mutually beneficial from being embedded in immigrant-domi- search by moving beyond individual-level action (i.e., the ability of the community to nated and potentially more supportive con- predictors to examine the social context in come together for the common good). For ex- texts.7 By contrast, foreign-born Latinos which these conditions take root. ample, neighbors may take responsibility for

May 2007, Vol 97, No. 5 | American Journal of Public Health Cagney et al. | Peer Reviewed | Research and Practice | 919  RESEARCH AND PRACTICE 

maintaining safe streets and parks; residents reliable and valid.21 Although this question socioeconomic status, housing density, and are then drawn outdoors where they get exer- does not measure the presence of any 1 of family structure. Two measures of neighbor- cise and fresh air. Neighbors do so because these conditions individually, it does provide a hood social context were derived from these “it’s what we do in our community,” rather comprehensive assessment of respiratory dis- data. Collective efficacy was operationalized by than because it directly benefits someone they orders. Prevalence data from the City of Chi- combining measures of social cohesion and in- know. Social interaction and social exchange, cago suggested that most of these cases may formal social control. Social cohesion was con- by contrast, capture the breadth of potentially be asthma, but we note that our outcome mea- structed from a cluster of conceptually related health-protective social support within a com- sure was meant to capture breathing disorders items from the PHDCN-CS that measured the munity, and these measures speak to direct so- beyond asthma.22,23 The individual-level co- respondent’s level of agreement (on a 5-point cial interaction. For example, sharing health variates derived from the MCIC-MS included scale) with the following statements: (1) People information or driving a neighbor to the doc- age, gender, race/ethnicity, marital status, in- around here are willing to help their neigh- tor are the types of instrumental activities that come, education, home ownership, regular bors, (2) This is a close-knit neighborhood, result from dense network connections. source of care, insurance status, current smok- (3) People in this neighborhood can be trusted, Our aim in applying these theoretical ing behavior, and a physician-indicated weight and (4) People in this neighborhood generally frameworks was to introduce a conceptualiza- problem. We compared Whites, Blacks, and don’t get along with each other. Health-related tion of the enclave experience that is congru- Latinos born in and outside the United States. informal social control was determined from ent with previous work but adds an emphasis We chose to exclude those who identified as the respondent’s level of agreement with the on neighborhood-level social processes. The Puerto Rican (n=110). Although Puerto Ricans following statements: (1) If I were sick I could interdependence of individual and commu- share a language and some common cultural count on my neighbors to shop for groceries nity characteristics may largely explain the elements with the Latinos in our sample, they for me, and (2) You can count on adults in this Latino paradox: the relative health advantage are US citizens. Thus, migration patterns be- neighborhood to watch out that children are of foreign-born Latinos may be contingent on tween the continental United States and Puerto safe and don’t get in trouble. An additional in- features of their residential context. Rico are more fluid. The unique status of formal social control item asked respondents Puerto Ricans renders comparisons difficult how likely it was that people in their neighbor- METHODS under the Latino paradox rubric. hood would intervene if a fight broke out in Decennial Census. Census data made it pos- front of their house. The informal social con- Data and Measures sible for us to construct measures of neighbor- trol items address expectations for beneficial To address our hypotheses we combined 3 hood socioeconomic structure and composi- health-related action as well as neighborhood data sources from the 1990s that provide tion. Three of the 5 neighborhood-level supervision of potentially hazardous conditions individual-level outcomes nested in neighbor- measures came from these data. The first or violent situations. The 7 items were com- hoods, as well as measures that capture measure was a residential stability factor score bined to form a single scale of health-related individual- and neighborhood-level phenom- that included the percentage of residents liv- collective efficacy. ena: (1) the Metropolitan Chicago Information ing in the same house since 1985 and the The social interaction and exchange scale Center Metro Survey (MCIC-MS), (2) the De- percentage of owner-occupied dwellings. The measured the frequency of interaction and cennial Census, and (3) the Project on Human second measure was the logged value of a network-based exchange among neighbors. Development in Chicago Neighborhoods concentrated poverty factor score, which in- By contrast to the generalized assessments of Community Survey (PHDCN-CS). cluded the percentage of female-headed trust, solidarity, and shared expectations for MCIC-MS. The MCIC-MS included a serial households, unemployed residents, and resi- informal social control included in the mea- cross-section of adults aged 18 years and older dents living below the poverty line or receiv- sure of collective efficacy, the social interac- who resided in the 6-county metropolitan Chi- ing public assistance. The third measure was tion and exchange scale was designed to cap- cago area (on average, 3000 respondents per the logged value of the percentage of foreign- ture actual ties between neighborhood wave). To create the individual-level compo- born residents in the neighborhood. residents. Respondents were asked, how often nent of our final analytic data set, we pooled PHDCN-CS. The sampling design of the do you and people in this neighborhood the City of Chicago subsample of the 1995, PHDCN-CS relied on 1990 US Census data (1) have parties or other get-togethers where 19 97, and 1999 waves of the MCIC-MS for Chicago to identify 343 neighborhood other people in the neighborhood are invited, (n=3191). The outcome measure and individ- clusters—groups of 2 to 3 census tracts that (2) visit in each others homes or on the street, ual-level covariates were derived from these contain approximately 8000 people. Major ge- (3) ask each other advice about personal data. The outcome measure was a dichoto- ographic boundaries (e.g., railroad tracks, things such as child rearing or job openings, mous measure derived from the question, has parks, freeways), knowledge of Chicago’s local and (4) do favors for each other? a doctor ever told you that you have asthma, neighborhoods, and cluster analyses of census bronchitis, emphysema, or other breathing data guided the construction of neighborhood Analysis problems? Evidence indicated that this form clusters so that they were relatively homoge- The clustering of respondents within of self-reported health status question is both neous with respect to racial/ethnic mix, Chicago’s neighborhoods rendered standard

920 | Research and Practice | Peer Reviewed | Cagney et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Finally, Figure 1 displays the predicted ordinary least squares techniques inappropri- TABLE 1—Summary Statistics of the probabilities of asthma and other breathing ate because of the likely underestimation of Study Population (N=3191): Chicago, problems for foreign-born Latino residents at standard errors. Our analysis strategy used hi- Ill, 1995, 1997, 1999 erarchical modeling techniques to investigate selected levels of the percentage of foreign- the prevalence of asthma and other breathing Mean (SD) born residents in the community. problems across neighborhoods.24,25 This ap- Outcome proach had several advantages. First, the tech- RESULTS Asthma/breathing problems 0.18 nique allowed for adjustment of standard errors for the effects of clustering within Control variables Approximately 18% of the MCIC-MS re- neighborhoods. Second, it facilitated examina- Female 0.59 spondents had been told by a physician that tion of interactions between the individual- Age 42.99 (16.14) they had asthma or other breathing problems. level foreign-born status and neighborhood- Race/ethnicity This result is similar to contemporary reports level factors. To correct independent White 0.42 of asthma prevalence in Chicago, where data neighborhood-level measures of collective effi- Black 0.36 indicate that physician-diagnosed asthma ap- 27 cacy and social interaction and exchange for Foreign-born Latino 0.10 proaches 20% in some neighborhoods. Just missing data and differences in the reliability under 19% of Whites reported asthma or US-born Latino 0.06 of estimates across neighborhoods, we used other breathing problems compared with Other 0.06 empirical Bayes residuals from a 3-level item- 22.2% of Blacks, 14.8% of US-born Latinos, response model of the component items of Married 0.39 and 7.8% of foreign-born Latinos. Asthma these scales.26 We used the Hierarchical Linear Income, $ also varied by the population composition of and Nonlinear Modeling 6 software program <9999 0.09 the community. Communities with large for- (Scientific Software International, Lincolnwood, 10000–14999 0.07 eign-born populations reported that 16.3% of Ill) to conduct our analyses. 15000–19999 0.08 residents had asthma and other breathing We began by calculating the means and 20000–24999 0.08 problems; communities with medium and standard deviations that described our study 25000–29999 0.11 small foreign-born populations had rates of population, the individual-level component of 16.8% and 22.2%, respectively. 30000–39999 0.16 our analysis (Table 1). The main feature of As shown in Table 1, the sample was pre- 40000–49999 0.14 our analysis was a series of 6 nested hierarchi- dominately women (59%) and middle aged 50000–69999 0.11 cal logit models (Table 2) that combined indi- (mean = 43 years). Whites made up 42% of vidual and neighborhood-level covariates. A 70000–89999 0.07 the sample, African Americans 36%, for- “yes” response to the question about asthma ≥90000 0.08 eign-born Latinos 10%, US-born Latinos and other breathing problems formed the out- Education 6%, and other racial/ethnic groups 6%. Ed- come. The coefficients presented in Table 2 ≤4th grade 0.01 ucation, income, insurance, and access to are odds ratios (with confidence intervals [CIs] 5th–8th grade 0.05 care values were relatively representative of in parentheses). We first analyzed individual- 9th–12th grade, no diploma 0.11 the Chicago population. Approximately 29% level factors, then sequentially introduced resi- High-school graduate 0.16 currently smoked, and 20% had been told dential stability, concentrated poverty, percent by a physician that they had a weight prob- Trade or vocational school 0.07 foreign-born, collective efficacy, and social in- lem (both of which exacerbate asthma and Some college 0.26 teraction and exchange. The interview year other breathing problems). Although we re- College graduate 0.17 was included as a control variable across mod- port mean age and included age as a contin- els. Our final model was as follows: Some graduate study 0.04 uous variable in our multivariate models Graduate degree 0.13 (described later) we did investigate age ef-  ϕij  (1) log  =+ββ01jj (LatForeignBorn ) ij+ Home ownership 0.43 fects of asthma and other breathing prob-  1− ϕij  No regular source of medical care 0.04 lems by group: ≤ 18–35 (n = 1287), 36–54 ββ23jijjij()()LatUSborn++ Black ≥ Q Insurance (n = 1154), 55 (n = 750). If an age effect β + β 4 j(OtherRacce)ij∑ qj Xqij was present, it could mean, for instance, that q=5 Private insurance 0.67 our data contained more cases of emphy- Medicare 0.08 27 βγγ=+ + γ + sema than asthma at later ages. Reports of (2) 00001jjj()()ConPov 02 RStab Medicaid 0.08 γ ++γ asthma and other breathing problems across 03(PercenttForBorn)(jj04 CollEff ) No Insurance 0.17 γ 05()SocIntExj + u 0 j these 3 age groups were strikingly similar at Current smoker 0.29 18.4%, 17.9%, and 18.8%, respectively Weight problem 0.20 (3) βγγ11011jj=+()PercentForBorn + (analysis using the t test indicated no signifi-

γ 12()CollEff j + γ 13()SocIntExjj+ u 1 cant differences between age groups). We

May 2007, Vol 97, No. 5 | American Journal of Public Health Cagney et al. | Peer Reviewed | Research and Practice | 921  RESEARCH AND PRACTICE 

TABLE 2— Six Hierarchical Logit Models of Individual- and Neighborhood-Level Covariates of Asthma and Other Breathing Problems: Chicago, Ill, 1995, 1997, 1999

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Variable OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI)

Individual-level characteristics Year 1.01 (0.95, 1.07) 1.01 (0.95, 1.07) 1.01 (0.95, 1.08) 1.01 (0.96, 1.08) 1.01 (0.96, 1.08) 1.01 (0.96, 1.08) Age 0.99 (0.99, 1.00) 0.99 (0.99, 1.00) 0.99 (0.99, 1.00) 0.99 (0.99, 1.00) 0.99 (0.99, 1.00) 0.99 (0.99, 1.00) Gender (female) 1.41*** (1.17, 1.70) 1.41*** (0.17, 1.70) 1.41*** (1.17, 1.70) 1.41*** (1.17, 1.70) 1.40*** (1.16, 1.69) 1.40*** (1.16, 1.69) Race/ethnicitya Black 1.08 (0.88, 1.34) 1.02 (0.73, 1.43) 1.04 (0.75, 1.45) 1.05 (0.76, 1.46) 1.05 (0.76, 1.45) 1.05 (0.76, 1.46) US-born Latino 0.68*** (0.44, 1.05) 0.64** (0.42, 0.98) 0.63** (0.41, 0.97) 0.64** (0.42, 0.98) 0.63** (0.41, 0.96) 0.63** (0.41, 0.96) Foreign-born Latino 0.40*** (0.24, 0.67) 0.38*** (0.22, 0.64) 0.70 (0.37, 1.35) 0.60 (0.29, 1.24) 0.50** (0.26, 0.98) 0.48** (0.23, 0.99) Other 0.62** (0.40, 0.97) 0.61** (0.39, 0.97) 0.61** (0.39, 0.97) 0.62** (0.39, 0.97) 0.62* (0.39, 0.97) 0.62** (0.39, 0.97) Married 0.98 (0.80, 1.20) 0.97 (0.79, 1.20) 0.97 (0.79, 1.20) 0.97 (0.79, 1.19) 0.98 (0.79, 1.20) 0.98 (0.79, 1.20) Income 1.03 (0.98, 1.08) 1.04 (0.99, 1.09) 1.04 (0.99, 1.09) 1.04 (0.99, 1.09) 1.04 (0.99, 1.09) 1.04 (0.99, 1.09) Education 1.00 (0.94, 1.05) 1.00 (0.95, 1.06) 1.00 (0.95, 1.06) 1.00 (0.95, 1.06) 1.00 (0.94, 1.06) 1.00 (0.94, 1.06) Home ownership 0.94 (0.75, 1.19) 0.91 (0.71, 1.17) 0.90 (0.70, 1.16) 0.90 (0.70, 1.15) 0.89 (0.70, 1.14) 0.89 (0.70, 1.14) No regular source of medical care 0.72 (0.42, 1.23) 0.71 (0.42, 1.22) 0.71 (0.42, 1.22) 0.71 (0.42, 1.20) 0.73 (0.43, 1.24) 0.72 (0.42, 1.23) Insurance statusb Medicare 0.97 (0.66, 1.44) 1.00 (0.67, 1.47) 0.98 (0.66, 1.45) 0.97 (0.66, 1.43) 0.98 (0.67, 1.45) 0.98 (0.67, 1.44) Medicaid 1.12 (0.78, 1.62) 1.12 (0.77, 1.62) 1.11 (0.77, 1.61) 1.12 (0.77, 1.61) 1.12 (0.78, 1.62) 1.12 (0.77, 1.62) No insurance 1.23 (0.94, 1.61) 1.22 (0.94, 1.60) 1.23 (0.94, 1.60) 1.23 (0.94, 1.60) 1.25* (0.96, 1.62) 1.24 (0.96, 1.62) Current smoker 1.26** (1.04, 1.53) 1.25** (1.04, 1.52) 1.25** (1.03, 1.52) 1.25** (1.03, 1.52) 1.24** (1.03, 1.51) 1.24** (1.03, 1.51) Weight problem 1.76*** (1.43, 2.16) 1.74*** (1.41, 2.15) 1.75*** (1.42, 2.15) 1.75*** (1.42, 2.15) 1.73*** (1.41, 2.13) 1.73*** (1.41, 2.14) Neighborhood-level characteristics Constant Residential stability . . . 1.08 (0.98, 1.20) 1.08 (0.98, 1.20) 1.08 (0.97, 1.20) 1.08 (0.98, 1.20) 1.08 (0.98, 1.20) Concentrated poverty . . . 1.00 (0.87, 1.16) 1.01 (0.87, 1.17) 1.01 (0.87, 1.17) 1.02 (0.88, 1.18) 1.02 (0.88, 1.18) Logged percentage foreign born . . . 1.01 (0.88, 1.15) 1.02 (0.90, 1.17) 1.02 (0.90, 1.17) 1.03 (0.91, 1.17) 1.03 (0.91, 1.17) Health-related collective efficacy . . . 0.89*** (0.80, 1.00) 0.90** (0.80, 1.00) 0.91* (0.82, 1.01) 0.88* (0.77, 1.01) 0.89* (0.78, 1.02) Social interaction/exchange ...... 1.05 (0.93, 1.20) 1.05 (0.93, 1.19) Foreign-born Latino Logged percentage foreign born ...... 0.57*** (0.37, 0.86) 0.55*** (0.36, 0.83) 0.64** (0.43, 0.96) 0.62** (0.41, 0.93) Health-related collective efficacy ...... 0.65 (0.37, 1.15) . . . 0.83 (0.46, 1.51) Social interaction/exchange ...... 0.45*** (0.25, 0.80) 0.47** (0.25, 0.86) Intercept 0.23*** (0.20, 0.27) 0.24*** (0.20, 0.30) 0.24*** (0.20, 0.29) 0.24*** (0.20, 0.29) 0.24*** (0.20, 0.29) 0.24*** (0.20, 0.29) Variance component 0.08* 0.06* 0.05 0.04 0.04 0.04

aReference categories=White. bReference category= private insurance. *P<.10; **P<.05; ***P<.01

also focused on the younger age group individual-level factors only and indicates that problems and individuals who had a physi- (18–35 years) in our multilevel models and female gender is predictive of asthma and cian-indicated weight problem were approxi- found comparable results to those models other breathing problems, whereas Latino sta- mately 1.8 times more likely to make such a that included all ages. We still were not able tus, both US and foreign born, is protective report. Model 2 introduced neighborhood- to address the possibility of differential dis- (as is the status of “other” ethnicity). Smoking level variables and indicated that none of the ease composition by age group, but given and a physician-indicated weight problem structural indicators—residential stability, con- similar prevalence rates and model results, both significantly increased the odds of re- centrated poverty, percentage foreign-born— we elected to examine the full sample. porting asthma and other breathing problems; had a significant effect on the likelihood of re- Multilevel results are presented in the 6 smokers were approximately 1.3 times more porting asthma and other breathing problems. models shown in Table 2. Model 1 included likely to report asthma and other breathing However, the effect of collective efficacy is

922 | Research and Practice | Peer Reviewed | Cagney et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

FIGURE 1—Predicted Probability of Asthma and Other Breathing Problems for Foreign-Born Latino Respondents, by Percentage of Foreign-Born Neighborhood Residents: Chicago, Ill, 1995, 1997, and 1999.

significant, which is consistent with previous lective efficacy and social interaction and ex- when they lived in a neighborhood that had a research13 ; each 1-standard-deviation increase change (adding social interaction and ex- high percentage of foreign-born residents. in the collective efficacy score reduced change to the model for the intercept as well). Conversely, their counterparts who lived in asthma and other breathing problems by 11% Collective efficacy remained insignificant, but neighborhoods with a low percentage of (social interaction and exchange was not a both the logged percentage of foreign-born foreign-born residents experienced very high significant predictor of asthma and other residents and social interaction and exchange rates of asthma and other breathing problems breathing problems in separate analyses). variables retained their significance. (approximately 22%)—higher, in fact, than the Models 3 through 6 illustrate the results Figure 1 illustrates—using predicted proba- average level of asthma and other breathing from cross-level interactions between neigh- bilities from Model 6—the relationships among problems for the other groups. Thus, the prob- borhood factors and the foreign-born Latino race/ethnicity, asthma and other breathing ability of asthma and other breathing prob- effect. Model 3 indicated that the percentage problems, and the percentage of foreign-born lems for foreign-born Latinos appears contin- of foreign-born residents in the community residents in the neighborhood (the latter rep- gent on the population composition of the interacts with individual-level Latino immi- resented at high [1.5 SD above the mean], av- community. grant status. As the percentage of foreign- erage [the mean], and low [1.5 SD below the born residents increased, the protective effect mean] values). We focused on the experience DISCUSSION of foreign-born Latino status was enhanced. of foreign-born Latinos, but included the ad- This effect remained robust with the inclusion justed average effect from Model 6 for the The relative health advantage of foreign- of collective efficacy (Model 4), which did not other race/ethnic groups as reference points. born Latinos has been observed across age interact significantly with foreign-born status. We found a distinctly graded effect of the groups, outcomes, and, in some cases, coun- By contrast, Model 5 indicated that social in- prevalence of asthma and other breathing try of origin.2,29,30 Previous research has de- teraction and exchange significantly increased problems that was dependent on the foreign- lineated the individual-level characteristics the protective effect of foreign-born status on born residential composition of the commu- associated with this health advantage but, the likelihood of asthma and other breathing nity. Foreign-born Latinos experienced much to our knowledge, has not examined the problems. Finally, Model 6 included both col- lower rates of asthma (approximately 5%) neighborhood social context in which these

May 2007, Vol 97, No. 5 | American Journal of Public Health Cagney et al. | Peer Reviewed | Research and Practice | 923  RESEARCH AND PRACTICE 

characteristics are embedded nor the inter- may substantially underestimate that benefit care in our models, this type of question may dependence between them. We identified a for immigrants who reside in ethnic enclaves. disproportionately underestimate the preva- health benefit from foreign-born Latino sta- Future analyses will explore differences by lence of asthma for those whose interaction tus: Latinos born outside the United States country of origin, given that migration pat- with the medical care system is intermittent. experienced appreciably lower rates of terns may differ across Chicago neighbor- Third, our study was confined to the City of asthma and other breathing problems. We hoods. For instance, Latinos who reside in Chicago, so the ability to generalize is limited. found, however, that this benefit observed in communities that have a low percentage of Analyses from other urban centers that have the aggregate was heavily dependent on the foreign-born residents may be more vulner- prominent Latino enclaves would allow for a communities in which foreign-born Latinos able to asthma and other respiratory condi- richer understanding of the enclave experi- reside. When embedded in a neighborhood tions. With respect to socioeconomic status, ence and, depending upon availability of data, where the presence of other foreign-born we did not find evidence that foreign-born could address the independent mechanisms residents is high, the risk of asthma and Latinos who lived in communities with a low relevant for a host of respiratory conditions. other breathing problems was abated. Con- percentage of foreign-born residents were The Latino population is the fastest grow- versely, foreign-born Latinos who lived in more disadvantaged when it comes to basic ing and largest population subgroup in the communities where there was a low percent- economic and demographic characteristics. United States.35 Attention to the context in age of foreign-born residents experienced Indeed, if anything, the foreign-born Latinos which Latinos reside could provide important the highest rates of asthma and other breath- in communities of fewer foreign-born resi- insights into trajectories of acute and chronic ing problems overall. Thus, the Latino ad- dents were more advantaged than were their conditions. The divergent experience of vantage with respect to asthma may accrue counterparts in communities that had a foreign-born Latinos illustrates that commu- only when it is socially leveraged. higher percentage of foreign-born residents. nity is critical to shaping health. Research of Consistent with the early theoretical and The characteristics we considered may omit this form contributes to our understanding of empirical research of urban sociologists Shaw an important predictor of asthma and other social capital and the extent to which it aids and McKay,15 homogeneity with respect to eth- breathing problems, but the overall profile in ameliorating threats to respiratory health nicity and immigrant status may increase infor- suggests comparability or marginal advantage and, potentially, to other conditions. Knowl- mation exchange through a common language. for foreign-born Latinos in communities of edge of the community and its characteristics Shared culture or lifestyle behaviors also may fewer foreign-born residents. On a related may provide the foundation for initiatives be at play.31,32 We found divergent experi- note, most Latinos in Chicago are Mexican meant to disseminate information and ad- ences by neighborhood context after the intro- (70.4%),34 so another extension of this work dress concerns about prevention, possible duction of individual-level behavioral factors will focus on the asthma experience among triggers, and treatment of asthma and other associated with asthma and other breathing Mexicans. Preliminary analyses of these data respiratory conditions. problems (e.g., smoking, weight problem). indicate that results from the Mexican sub- Thus, the role of community and cultural sup- sample are consistent with those for all Lati- About the Authors ports may not only influence individual-level nos (results available from first author on Kathleen A. Cagney is with the Departments of Health behaviors but may also affect health in their request). Finally, additional analyses will in- Studies and Comparative Human Development, University own right. Collective efficacy, found to be im- corporate alternative assessments of commu- of Chicago, Chicago, Ill. Christopher R. Browning is with 13 the Department of Sociology, Ohio State University, portant to asthma rates in previous work, was nity, including constructs to assess specific Columbus. Danielle M. Wallace is with the Department not as important to the health of foreign-born characteristics of ethnic enclaves. of Sociology, University of Chicago, Chicago. Latinos as was the level of social interaction Some caveats merit consideration. First, Requests for reprints should be sent to Kathleen A. Cagney, Department of Health Studies, The University of and exchange. Collective efficacy may operate our outcome was a composite measure of res- Chicago, 5841 South Maryland Ave., MC 2007, Chicago, differently for Latinos or it may not have the piratory conditions. Asthma, bronchitis, em- IL 60637 (e-mail: [email protected]). immediate or intimate effect that social net- physema, and other breathing disorders differ This article was accepted May 12, 2006. works provide.33 Communities that have lim- in etiology, so they also may differ in the ex- ited social ties could attenuate the health bene- tent to which they are affected by neighbor- Contributors K.A. Cagney developed the research question, wrote fit of immigrant status and offer further hood factors; the role of neighborhood in the the article, and identified analyses to be conducted. evidence that the Latino paradox is context de- onset or trajectory of these respiratory condi- C.R. Browning contributed to article development and pendent. Importantly, social network interactions could vary in important ways. Future statistical analyses. D.M. Wallace conducted statistical analyses. All authors provided revisions to article drafts. tion and exchange does not explain the effect data collection efforts would benefit from a of percentage foreign born; rather, the effect of decomposition of these conditions so that Acknowledgments both characteristics suggests that neighbor- neighborhood structural and social process Support for this research was provided by the National hood factors may be additive. A further impli- measures could be examined for each. Sec- Institute on Aging and the Office of Behavioral and So- cation of this finding is that analyses of the ond, reports of respiratory conditions are con- cial Sciences Research (grant R01AG022488). We thank Robert Sampson, Felton Earls, and mem- health advantage for foreign-born Latinos fail tingent upon a clinician’s evaluation. Even bers of the Project on Human Development in Chicago to take into account neighborhood context though we controlled for regular source of Neighborhoods for providing access to the community

924 | Research and Practice | Peer Reviewed | Cagney et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

survey, and the Metropolitan Chicago Information Cen- 16. Sampson RJ, Raudenbush SW, Earls F. Neighbor- 26.Raudenbush SW, Bryk AS. Hierarchical Linear ter for access to the metro survey. We also thank David hoods and violent crime: a multilevel study of collec- Models: Applications and Data Analysis Methods. Thou- Meltzer for comments on an earlier version of this arti- tive efficacy. Science. 19 97;277:918–924. sand Oaks, Calif: Sage; 2002. cle and Sandra Thomas for her expertise in describing 17.Waters MC, Eschbach K. Immigration and ethnic 27. Whitman S, Williams C, Shah AM. Sinai Health Chicago’s asthma prevalence. and racial-inequality in the United States. Annu Rev System’s Community Health Survey: Report 1. Chicago, Sociol. 19 95;21:419–446. Ill: Sinai Health System; 2004. Human Participant Protection 18.Portes A, Truelove C. Making sense of diversity: 28. Malik A, Saltoun CA, Yarnold PR, Grammer LC. This research was approved by the institutional review recent research on Hispanic minorities in the United Prevalence of obstructive airways disease in the disad- board at The University of Chicago. States. Annu Rev Sociol. 19 87;13:359–385. vantaged elderly of Chicago. Allergy Asthma Proc. 19. Sanders JM, Nee V. Limits of ethnic solidarity in the 2004;25:169–173. References enclave economy. Am Sociol Rev. 19 87;52:745–773. 29. Elo I, Turra C, Kestenbaum B, Ferguson BR. Mortal- ity among elderly Hispanics in the United States: past ev- 1. Cobas JA, Balcazar H, Benin MB, Keith VM, 20.Browning CR, Cagney KA. Neighborhood struc- idence and new results. Demography. 2004;41:109–128. Chong Y. Acculturation and low-birthweight infants tural disadvantage, collective efficacy, and self-rated among Latino women: a reanalysis of HHANES data physical health in an urban setting. J Health Soc Behav. 30. Abraido-Lanza AF, Dohrenwend BP, Ng-Mak DS, with structural equation models. Am J Public Health. 2002;43:383–399. Turner JB. The Latino mortality paradox: a test of the 1996;86:394–396. 21.Patrick DL, Erickson P. Health Status and Health “salmon bias” and healthy migrant hypotheses. Am J Public Health. 1999;89:1543–1548. 2. Markides KS, Coreil J. The health of Hispanics in Policy: Allocating Resources to Health Care. New York, the southwestern United-States - an epidemiologic par- NY: Oxford University Press; 1993. 31. Fitzpatrick K, Lagory M. Unhealthy Places: The adox. Public Health Rep. 1986;101:253–265. 22. Naureckas ET, Wolf RL, Trubitt MJ, et al. The Ecology of Risk in the Urban Landscape. New York, NY: Routledge; 2000. 3. Patel KV, Eschbach K, Ray LA, Markides KS. Chicago Asthma Consortium: a community coalition Evaluation of mortality data for older Mexican Ameri- targeting reductions in asthma morbidity. Chest. 19 9 9; 32. Sorlie PD, Backlund E, Johnson NJ, Rogot E. Mor- cans: implications for the Hispanic paradox. Am J Epi- 116(4 Suppl 1):190S–193S. tality by Hispanic status in the United States. JAMA. demiol. 2004;159:707–715. 23. Thomas S. Telephone surveys for asthma surveil- 19 93;270:2464–2468. 4. Palloni A, Arias E. Paradox lost: explaining the lance: The Chicago Respiratory Health Survey. Paper 33. Klinenberg E. Heatwave: A Social Autopsy of Dis- Hispanic adult mortality advantage. Demography. presented at: A Public Health Response to Asthma. aster in Chicago. Chicago, Ill: The University of Chicago 2004;41:385–415. February 15–17, 2000; Atlanta, GA. Press; 2002. 5. Morales LS, Lara M, Kington RS, Valdez RO, 24.Diez-Roux AV. Multilevel analysis in public health 34. Guzman B. The Hispanic Population. Washington, Escarce JJ. Socioeconomic, cultural, and behavioral fac- research. Annu Rev Public Health. 2000;21:171–192. DC: US Census Bureau; 2001. tors affecting Hispanic health outcomes. J Health Care 25. Snijders T, Bosker R. Multilevel Analysis: An Intro- 35. Greico EM, Cassidy RC. Overview of Race and His- Poor Underserved. 2002;13:477–503. duction to Basic and Advanced Multilevel Modeling. Lon- panic Origin 2000. Washington, DC: US Census Bu- 6. LeClere FB, Rogers RG, Peters KD. Ethnicity and don, England: Sage; 1999. reau; 2001. mortality in the United States: individual and community correlates. Soc Forces. 19 97;76:169–198. 7. Eschbach K, Ostir GV, Patel KV, Markides KS, Goodwin JS. Neighborhood context and mortality among older Mexican Americans: is there a barrio advantage? Am J Public Health. 2004;94:1807–1812. 8. Sperber K, Ibrahim H, Hoffman B, Eisenmesser B, The Legacy of Katrina and Rita Hsu H, Corn B. Effectiveness of a specialized asthma Health Care Providers Remember—And Look Ahead clinic in reducing asthma morbidity in an inner-city minority population. J Asthma. 19 95;32:335–343. by Lisa Tracy 9. Corn B, Hamrung G, Ellis A, Kalb T, Sperber K. Pat- “This is a book about real public terns of asthma death and near-death in an inner-city ter- health heroes—those who selflessly tiary care teaching hospital. J Asthma. 19 95;32:405–412. put aside their own worries and 10. Diaz T, Sturm T, Matte T, et al. Medication use problems to help others in great among children with asthma in East Harlem. Pediatrics. need. This is an important oral 2000;105:1188–1193.

Waters history of our nation’s most 11. Findley S, Lawler K, Bindra M, Maggio L, significant natural disaster. These Penachio MM, Maylahn C. Elevated asthma and indoor stories send us a vital warning environmental exposures among Puerto Rican children of East Harlem. J Asthma. 2003;40:557–569. about the consequences of poor

y preparedness. We should listen.” 12.Homa DM, Mannino DM, Lara M. Asthma mortality in U.S. Hispanics of Mexican, Puerto Rican, and Georges C. Benjamin, MD, FACP Cuban heritage, 1990–1995. Am J Respir Crit Care Executive Director, APHA Med. 2000;161(2 Pt 1):504–509.

13. Cagney KA, Browning CR. Exploring neighborhood- dd level variation in asthma and other respiratory ORDER TODAY! American Public Health Association diseases—the contribution of neighborhood social con- ISBN: 978-087553-185-4 PUBLICATION SALES text. J Gen Intern Med. 2004;19:229–236. 132 pages, 2-colors WEB: www.apha.org 14 . Gold DR, Wright R. Population disparities in E-MAIL: [email protected] asthma. Annu Rev Public Health. 2005;26:89–113. softcover, 11 x 8.5 Mu 15. Shaw CR, McKay HD. Juvenile Delinquency and $33.00 APHA Members and TEL: 888-320-APHA Urban Areas: A Study of Rates of Delinquents in Relation nonmembers (plus s&h) FAX: 888-361-APHA to Differential Characteristics of Local Communities in American Cities. Chicago, Ill: University of Chicago Press; 1969.

May 2007, Vol 97, No. 5 | American Journal of Public Health Cagney et al. | Peer Reviewed | Research and Practice | 925  RESEARCH AND PRACTICE 

The Relationship Between Neighborhood Characteristics and Self-Rated Health for Adults With Chronic Conditions

| Arleen F. Brown, MD, PhD, Alfonso Ang, PhD, and Anne R. Pebley, PhD

Residence in a socioeconomically deprived Objectives. We sought to determine whether the association between neighborhood neighborhood has been linked to all-cause characteristics and health differs for people with and without a chronic condition. 1–3 4 mortality, functional decline, poorer Methods. We analyzed data from 2536 adults from the Los Angeles Family and 5,6 health status, and higher incidence and Neighborhood Survey and evaluated the relationship between the presence of a prevalence of chronic conditions such as dia- chronic condition at the individual level, neighborhood socioeconomic status (SES), betes, cardiovascular disease, and cancer.2,7–16 and self-rated health. We constructed multilevel models to evaluate the relation- People living in deprived neighborhoods are ship between the neighborhood SES index and self-rated health for people with likely to experience multiple dimensions of and without chronic conditions, after adjustment for other individual characteristics. poor environmental and social quality, includ- Results. Having a chronic condition was associated with substantially poorer ing higher-priced yet lower-quality foods, high self-rated health among participants in a deprived area than among those in a more advantaged area. crime rates, poor-quality housing, limited Conclusions. Residence in a disadvantaged neighborhood may be associated transportation, toxic environments, and lower with barriers to the management of a chronic condition. Further work is needed social cohesion and social support, all of to identify the specific characteristics of disadvantaged areas associated with 13 ,17–22 which may contribute to poorer health. poorer self-rated health for adults with chronic conditions. (Am J Public Health. Adults with chronic conditions may be partic- 2007;97:926–932. doi:10.2105/AJPH.2005.069443) ularly vulnerable to these dimensions of neighborhood deprivation. Models of chronic disease management, an unhealthy lifestyle19 , 2 0 , 2 7 may be associ- of census tracts in Los Angeles County con- such as the Chronic Care Model23–25 and the ated with greater reductions in health status ducted in 2000–2001. The design of Disablement Framework,26 highlight the im- among adults with chronic conditions than LAFANS is presented elsewhere.35,36 Briefly, portance of community resources to the man- among those without a chronic condition. 1652 census tracts in Los Angeles County agement of chronic conditions. Yet research We conducted a cross-sectional analysis of were stratified on the basis of the percentage on chronic conditions tends to emphasize the 2000–2001 Los Angeles Family and of people living in poverty as obtained from clinical care, the health care system, and indi- Neighborhood Survey (LAFANS) to examine 19 97 estimates. Census tracts were classified vidual factors, and only infrequently exam- whether the neighborhood socioeconomic en- as very deprived (90%–100% of residents ines the role of the neighborhood environ- vironment contributes to differences in self- living in poverty), deprived (60%–89%), and ment in the management of chronic disease. rated health among persons with and without not deprived (1%–59%). In a representative There are, however, several mechanisms a chronic condition. We chose to evaluate sample of 65 tracts (20 very deprived, 20 de- through which the neighborhood context self-rated health because it is closely associ- prived, and 25 not deprived), 40 to 50 may differentially affect the health of people ated with several health outcomes, including dwelling units were sampled at random, with with chronic conditions. Among adults with morbidity28,29 and mortality,30,31 and determi- an oversample of households with children. conditions such as diabetes, cardiovascular nants of health ratings have been shown to Within each household, LAFANS randomly disease, arthritis, and asthma, adequate dis- differ between people with and without selected 1 adult (aged 18 years or older), who ease management often requires continuous chronic conditions.32–34 We hypothesized was interviewed in person. These analyses in- clinical follow-up, self-care, and complex that, independent of individual income or ed- clude data only from the randomly sampled medication, dietary, and exercise regimens,24 ucation, lower neighborhood socioeconomic adults in the LAFANS Wave 1 cohort. all of which may be influenced by neighbor- status would be associated with lower self- The main predictors in the analysis were hood factors such as available health care, rated health and that the association would be the socioeconomic status of the neighborhood access to exercise facilities and nutritious strongest for persons with a chronic condition. of residence and the presence of a chronic foods, and environments otherwise con- condition. We assigned each tract a neighbor- ducive to self-management. Thus, the charac- METHODS hood socioeconomic status (SES) index.3 The teristics of local areas, such as limited avail- SES index is the unweighted average of 5 ability or accessibility of health services, For the analyses, we used data from census variables (percentage of individuals 25 infrastructure deprivation, environmental LAFANS Wave 1, a longitudinal study of years or older without a high school degree, stressors, and social interactions that promote families in a stratified probability sample median family income, median home value,

926 | Research and Practice | Peer Reviewed | Brown et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

percentage blue collar individuals, and per- depression” or screening positive for depres- research.32,39,44,45 At the tract level, the main centage unemployed individuals), with the di- sive symptoms on the short form of the Com- predictor was the neighborhood SES index. rection reversed for some variables; it was posite International Diagnostic Interview.40 Be- The models also included cross-level inter- constructed from census tract–level data ob- cause not all the randomly sampled adults in actions between the neighborhood SES index tained from Summary File 3 of the 2000 US the cohort underwent screening for depression, and the presence of a chronic condition. Sep- Census.37 We used an unweighted sum be- we present separate results that include de- arate models were also constructed for each cause a principal components analysis indi- pression as a covariate in the portion of the of the most common chronic conditions (hy- cated that all 5 components contributed sample that was screened for depression. pertension, arthritis, diabetes, and chronic equally to the first factor, which accounted The dependent variable was self-rated lung disease); these models included an inter- for 93% of the variance. Both the composite health, which was measured by the question, action term between the neighborhood SES and the individual census variables were eval- “How would you rate your overall health?” index and the chronic condition. Another uated in separate models, but because we Response categories were poor, fair, good, model included an unweighted sum of were interested in assessing multiple dimen- very good, and excellent. Poorer self-rated chronic conditions. The logistic models in- sions of the neighborhood socioeconomic en- health has been associated with mortality and cluded sampling weights that take into con- vironment and the findings did not differ ap- functional limitations in longitudinal studies.41 sideration both nonresponse and the over- preciably, we used the SES index in these Unadjusted analyses were conducted using sample of poor households and households analyses. analyses of variance (repeated-measures anal- with children.36 The SES index was evaluated in separate ysis of variance [ANOVA]) and t tests. We A potential confounder in research on analyses as either a continuous variable or in evaluated self-rated health as a dichotomous neighborhood effects is “residential selection”: the 3 categories from the original sampling variable (fair or poor vs good, very good, or people in better health may choose to live in frame (“very deprived,” “deprived,” and “not excellent) and constructed weighted logistic more advantaged neighborhoods. Although deprived” census tracts). The results were regression models. Because self-rated health we could not directly assess the effect of se- similar for the 2 strategies; we present the has an ordinal response scale, we conducted lection on health outcomes in these cross- results obtained from the second strategy. a sensitivity analysis that evaluated the out- sectional analyses, we were able to evaluate To test alternative definitions of “neighbor- come as an ordered categorical variable (poor health ratings among people with the greatest hoods,” we constructed the SES index scores or fair, good, very good, and excellent). The residential stability (i.e., those who had lived at both smaller (census block group) and poor and fair categories were combined be- at the same address for 5 or more years). We larger (census tract) levels and compared final cause only 4% of the overall sample rated also evaluated interaction terms between the versions of the models with these 2 different their health as poor. The score test42,43 sup- neighborhood SES index and either individ- definitions. Because we found no substantial ported the proportional odds assumption ual income or individual education. Because χ2 difference in the results and prior literature ( 38 =47, P=.15). Multilevel ordered logistic these individual-level interactions were not suggests a high correlation between census models with random intercepts were con- significant and did not appreciably alter our block and census tract indictors,38 we present structed using Predictive Quasi Likelihood ap- findings, they are not presented in our final the census tract–level analyses. proximation procedures in HLM 5 software results. The primary individual predictor was a re- (Scientific Software International, Lincol- We derived the relative risks and the 95% port of a physician’s diagnosis of 1 or more nwood, Ill). Calculation of the intraclass corre- confidence intervals by bootstrapping with re- chronic conditions, as determined by the lation coefficient in the null (or empty) model placement over 1000 repetitions.46,47 The ex- question, “Has a doctor ever told you that suggested that 16.7% of the variance in self- pected values for each category of chronic con- you have … ?” followed by a list of conditions rated health was between census tracts. After dition and neighborhood poverty were then that included hypertension, arthritis, diabetes, control for sociodemographic and clinical var- calculated. The analyses were performed using and a chronic lung problem (asthma, chronic iables, the variance in self-rated health be- the statistical packages HLM 5, SAS version bronchitis, or chronic obstructive pulmonary tween census tracts was reduced to 10.8%. 9.1 (SAS Institute Inc, Cary, NC), and Stata disease). These conditions, which were in- At the individual level, the model included version 9.0 (Stata Corp, College Station, Tex). cluded in analyses either individually or as an the main predictor (the presence of a chronic unweighted sum of the conditions, were the condition) and individual covariates, includ- RESULTS ones most commonly reported by study par- ing demographic characteristics (age, gender, ticipants and are conditions that generally re- race/ethnicity, household income, education, The sample comprised 2536 adults (require substantial self-care. immigrant status), body mass index, and sponse rate=70%), 848 of whom reported 1 Because depression is associated with health health behaviors (smoking, alcohol use, and a or more of the chronic conditions of interest. status and quality of life,39 we also evaluated it physician visit in the prior year). A separate Compared with the randomly sampled adults as a separate covariate in the regression model. model also included depression as a covari- included in LAFANS, nonrespondents did not Depression was defined as either ever having ate. Many of these characteristics have been differ by race/ethnicity, gender, income, or received a physician’s diagnosis of “major associated with health status in previous education but were more likely to be the head

May 2007, Vol 97, No. 5 | American Journal of Public Health Brown et al. | Peer Reviewed | Research and Practice | 927  RESEARCH AND PRACTICE 

36 TABLE 1—Sample Characteristics of Participants and Census Tracts in the Los Angeles of household. Older people, women, Whites, Family and Neighborhood Survey: Los Angeles, Calif, 2000–2001 and African Americans were all more likely to report a chronic condition (Table 1). People No Chronic Any Chronic with chronic conditions also had lower educa- Conditions (n=1688) Condition (n=848) P tional attainment and annual household in- Demographic characteristics come. Compared with those who did not re- Mean age, y (SD) 36 (12) 46 (17) <.001 port a chronic condition, people with 1 or Female, % 55 66 <.001 more chronic conditions were not more likely Race/ethnicity, % to live in the poorest census tracts, but in the White 24 28 .03 aggregate, their tracts of residence had slightly Latino 60 47 <.001 lower median SES index scores and they were African American 7 14 <.001 more likely to have lived in the same resi- Asian/Pacific Islander 8 7 .46 dence for 5 years or longer. Those with Immigrant, % 62 46 <.001 chronic conditions had higher rates of report- Education, % None–8th grade 9 14 <.001 ing poor or fair health than did those without 9th–11th grade 25 24 .92 a chronic condition (38% vs 14%, P<.001). High school graduate 46 47 .99 Characteristics of the neighborhoods of res- College graduate 20 15 .001 idence of the study participants are presented Income, $ in Table 2. Participants were divided into very 01027<.001 deprived, deprived, and not deprived census 1–20000 31 28 .13 tracts. Across these 3 groups, we present the 20001–40000 26 19 <.001 median SES index score and the 5 census >40000 33 26 <.001 variables that comprise the SES index. Health-related characteristics In multivariate analyses, having a chronic Mean body mass index, kg/m2 (SD) 25 (5) 29 (7) <.001 condition and the interaction terms between Any chronic condition, % having a chronic condition and living in a Hypertension . . . 56 . . . deprived or very deprived census tract were Arthritis . . . 34 . . . Diabetes . . . 22 . . . all associated with lower self-rated health Asthma or chronic obstructive pulmonary disease . . . 24 . . . (Table 3). Other characteristics associated Smoking status, % with lower health ratings were male gender, Current smoker 15 17 .21 older age, immigrant status, less education, Past smoker 21 30 <.001 lower income, body mass index more than Never smoked 64 53 <.001 30 kg/m2, current tobacco use, binge drink- Binge drinking in past 30 days, % 7 10 .05 ing, and visiting a health care provider in the a Depression, % 8.8 20.6 <.001 prior year. Health service use, % Table 4 shows the adjusted percentages of Doctor visit in past 12 mo 70 84 <.001 participants expected to report good, very Hospitalization in past 12 mo 13 25 <.001 good, or excellent health, categorized by Unadjusted self-rated health, % chronic condition and census tract status. The Poor 1 9 <.001 reference group for all comparisons was those Fair 13 29 <.001 Good 31 32 .77 without a chronic condition who resided in Very good 28 19 <.001 nondeprived tracts. Among people without a Excellent 27 11 <.001 chronic condition, the adjusted predicted per- Census tract characteristics centage of people who reported their health Median socioeconomic index score (95% CI) –0.58 (–0.63, –0.49) –0.60 (–0.65, –0.50) .04 as good, very good, or excellent was 92.4% Census tract SES index, % for those residing in nondeprived census Not deprived 26 22 .18 tracts, 80.2% for those in deprived tracts, and Deprived 24 26 .13 73.2% for those in very deprived tracts. Peo- Very deprived 50 52 .87 ple with 1 or more chronic conditions had ≥ Residence in census tract 5 years, % 36 49 <.001 lower predicted ratings: good, very good, or Note. CI=confidence interval; SES=socioeconomic status. excellent health was reported by 88.4% of aOf participants (n=1838) screened with the Composite International Diagnostic Interview.40 those in nondeprived tracts, 63.3% of those in deprived tracts, and 52.4% of those in

928 | Research and Practice | Peer Reviewed | Brown et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 2—Characteristics of Census Tracts (From 2000 US Census) of Study Participants in was greater among those living in low-SES the Los Angeles Family and Neighborhoods Survey: Los Angeles, Calif, 2000–2001 census tracts than among those in high-SES census tracts. Not only were the presence of a Census Tract Status chronic condition and lower individual SES Very Deprived Deprived Not Deprived (e.g., income and education) associated with No. of study participants 1339 652 632 lower health status, but having a chronic con- No. of census block groups 128 52 46 dition was associated with substantially No. of census tracts 37 15 13 poorer self-rated health among people in a Residence in neighborhood 37 40 49 deprived area than among those in a more ≥5 years, % advantaged area. Participants with chronic Median socioeconomic index —0.85 (—1.72–0.37) —0.094 (—1.44–0.42) 0.84 (—0.81–2.42) conditions had substantially lower individual (range) income and were less educated than were the Adults 25 years or older without 62.2 31.9 8.5 other study participants. Although a lower a high school degree, % neighborhood SES index was associated with Median family income, $ (range) 27516 (13750–61917) 41293 (16619–67561) 78016 (23615–145361) poorer health status after adjustment for indi- Median home value, $ (range) 147900 (74300–220000) 178000 (96600–242100) 274100 (175700–891700) vidual income and education, it is important Employed residents with blue 62.2 43.0 20.8 to consider that their higher level of socioeco- collar occupations, % nomic disadvantage may have made partici- Unemployed residents, % 13.6 8.1 5.8 pants with chronic conditions particularly Residence in census tract 50.7 50.9 59.6 vulnerable to neighborhood deprivation. >5 years, % Chronic stress (related to crime, poor hous- Chronic conditions, % ing quality, and infrastructure deprivation) None 67 66 72 and lower or poorer availability of resources Any 33 34 28 (such as access to health care, food availabil- Hypertension 20 20 16 ity, and transportation) may contribute to the Osteoarthritis 12 12 10 larger negative association between neighbor- Diabetes mellitus 8 8 5 hood socioeconomic environment and health Asthma/chronic obstructive 9 10 10 for adults with chronic conditions. Among pulmonary disease adults with diabetes, for example, chronic No. of chronic conditions stress has been associated with poorer glyce- 1212420mic control through 2 mechanisms: health be- 2886haviors, such as lower rates of adherence to 3322medication or less physical activity, and neu- 48 4 0.9 0.6 0.5 rohormonal pathways. Mean body mass index, kg/m2 28 27 25 An important mediator of the relationship Depression,a %151710 between neighborhood deprivation and lower self-rated health may be functional status. A a 40 Of participants (n=1838) screened with the Composite International Diagnostic Interview. recent study, for example, indicates that the impact of functional limitation on quality of life is more that 4 times that of the chronic condi- very deprived tracts. The differences between between the 2. Similar results were also ob- tion by itself.49 Inadequate neighborhood re- the adjusted means for persons with a chronic served when we evaluated the number of sources, including access to health care, safe condition (compared with those without one) chronic conditions as opposed to just the places to exercise, healthy foods, and trans- were significantly higher for residents in de- presence of a chronic condition and for analy- portation, may directly contribute to functional prived or very deprived areas than for those ses restricted to people who had lived in the impairment. These factors may pose a greater of nondeprived areas. With the exception of census tract for 5 or more years. barrier for adults with chronic conditions, par- asthma, similar relationships were observed ticularly those with existing disability, and lead for the each of the chronic conditions. DISCUSSION to reduced physical activity, poorer dietary pat- Sensitivity analyses conducted with multi- terns, and lower rates of visits to health provid- level ordered logistic regression models found In this analysis of a population-based study ers. Allostatic load, the cumulative burden as- the same associations between self-rated of adults in Los Angeles County, for both sociated with the body’s adaptation to chronic health and area of residence, the presence of people with a chronic condition and those stress, may be another important mechanism a chronic condition, and the interaction term without, the difference in self-rated health through which neighborhood characteristics

May 2007, Vol 97, No. 5 | American Journal of Public Health Brown et al. | Peer Reviewed | Research and Practice | 929  RESEARCH AND PRACTICE 

TABLE 3—Weighted Odds Ratios for Reporting Good, Very Good, or Excellent Health: Los TABLE 4—Adjusted Percentage of Angeles Family and Neighborhoods Survey, Los Angeles, Calif, 2000–2001 Respondents Expected to Report Good, Very Good, or Excellent Health, by Full Model Full Model Without Depression, With Depression,a Chronic Condition and Neighborhood Odds Ratio (95% CI) P Odds Ratio (95% CI) P Socioeconomic Status Index: Los Angeles Family and Neighborhood Age, y Survey, Los Angeles, Calif, 2000–2001 <30 (reference) 1.00 1.00 30–39 0.78 (0.63, 0.96) .02 0.78 (0.63, 0.96) .02 % of Respondents by Census Tract Status 40–49 0.65 (0.52, 0.82) <.001 0.65 (0.51, 0.82) <.001 >50 0.64 (0.50, 0.83) .001 0.63 (0.49, 0.81) <.001 Not Very Deprived Deprived Deprived Female 1.23 (1.04, 1.44) .01 1.21 (1.03, 1.43) .02 Race/ethnicity No chronic condition 92.4 80.2 73.2 White (reference) 1.00 1.00 Any chronic condition 88.4 63.3a 52.4b Latino 0.81 (0.63, 1.04) .09 0.80 (0.62, 1.03) .10 Hypertension 83.2 63.8a 55.3b African American 0.91 (0.67, 1.23) .54 0.90 (0.67, 1.22) .52 Arthritis 81.2 61.5a 52.8b Asian/Pacific Islander 1.07 (0.76, 1.50) .74 1.02(0.73, 1.43) .91 Diabetes 79.0 65.4 51.2b Immigrant 0.66 (0.49, 0.74) <.001 0.59 (0.48, 0.73) <.001 Asthma/COPD 63.9 52.7 47.4 Education Note. COPD=Chronic obstructive pulmonary disease. College (reference) 1.00 1.00 Adjusted for age, gender, race/ethnicity, income, Some college 0.61 (0.48, 0.76) <.001 0.61 (0.48, 0.76) <.001 education, and immigrant status. aFor this chronic condition, difference between those High school graduate 0.38 (0.29, 0.51) <.001 0.39 (0.29, 0.52) <.001 living in nondeprived versus deprived tracts is greater Less than high school 0.26 (0.18, 0.39) <.001 0.26 (0.18, 0.38) <.001 for people with the condition than for those without Income, $ the condition (P<.05). bFor this chronic condition, difference between those 40000 (reference) 1.00 1.00 living in nondeprived versus very deprived tracts is 20001–40000 0.73 (0.59, 0.90) .07 0.73 (0.58, 0.90) .004 greater for people with the condition than for those 1–20000 0.54 (0.43, 0.68) <.001 0.55 (0.44, 0.69) <.001 without the condition (P<.05). 0 0.41 (0.32, 0.54) <.001 0.42 (0.32, 0.55) <.001 Body mass index, kg/m2 <25 (reference) 1.00 1.00 25–30 0.87 (0.72, 1.05) .14 0.88 (0.73, 1.05) .16 influence health outcomes for adults with 50 >30 0.73 (0.60, 0.90) .005 0.74 (0.60, 0.90) .003 chronic conditions. We observed different relationships for Smoker some of the chronic conditions, most notably Never (reference) 1.00 chronic lung conditions. We cannot be certain Current 0.78 (0.63, 0.97) .02 0.81 (0.65, 1.01) .06 why, for conditions such as asthma, chronic Binge drinker 0.81 (0.69, 0.95) .002 0.81 (0.69, 0.96) .01 obstructive pulmonary disease, and chronic Visit to a primary care provider in 0.78 (0.65, 0.94) .008 0.80 (0.67, 0.97) .02 bronchitis, there is no difference in self-rated the past year health status according to the neighborhood Any chronic conditionb 0.50 (0.38, 0.63) <.001 0.52 (0.41, 0.67) <.001 SES index; other factors may be involved. For Neighborhood socioeconomic index example, residence near a freeway or other de- Nondeprived tracts (reference) 1.00 1.00 terminants of local air quality may be more Deprived tracts 0.88 (0.67, 1.16) .37 0.91 (0.69, 1.21) .52 important predictors of self-rated health among Very deprived tracts 0.90 (0.68, 1.19) .47 0.91 (0.69, 1.20) .50 people with chronic lung problems.51 Another Interaction between neighborhood deprivation explanation for these findings may be the and presence of a chronic condition smaller sample size in this group; still another Chronic condition × nondeprived 1.00 1.00 could be that asthma, COPD, and chronic Chronic condition × deprived 0.68 (0.47, 0.98) .04 0.66 (0.46, 0.97) .03 bronchitis are heterogenous conditions, yet dif- Chronic condition × very deprived 0.59 (0.43, 0.81) .001 0.60 (0.44, 0.82) .002 ferences in their individual health ratings are Depression . . . 0.53 (0.42, 0.66) <.001 masked by grouping them into 1 category. aLimited to those screened using the Composite International Diagnostic Interview.40 Among the limitations of these analyses are bChronic conditions are hypertension, arthritis, diabetes, and asthma/chronic obstructive pulmonary disease. that we used cross-sectional data and have no information on change in health status or

930 | Research and Practice | Peer Reviewed | Brown et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

other health outcomes. Additionally, people more deprived census tracts with undiagnosed University of California, Los Angeles. Anne R. Pebley is in worse health may be more likely to move disease, however, may have led to an overesti- with the School of Public Health, University of California, Los Angeles, and the Rand Corporation, Santa Monica, to or remain in more deprived neighbor- mation of the magnitude of the difference in Calif. hoods. However, restricting the analyses to self-rated health between those with and those Requests for reprints should be sent to Arleen F. people who had lived in the same area for 5 without a chronic condition in deprived and Brown, MD, PhD, UCLA Division of General Internal Medicine and Health Services Research, 911 Broxton or more years and controlling for duration of very deprived tracts. Plaza, Los Angeles, CA 90024 (e-mail: abrown@ residence in these census tracts produced no Our findings of a differential in health sta- mednet.ucla.edu). significant change in the results. Another limi- tus for those with and those without a This article was accepted June 28, 2006. tation is that although we tested for alternate chronic condition by level of neighborhood definitions of neighborhoods using both tract deprivation suggest a need to identify the Contributors and block group data, census-derived charac- specific characteristics of the built and social A.F. Brown originated and supervised the study, conducted analyses, and led the writing of the article. terizations of neighborhoods may not reflect environment that are associated with substan- A. Ang completed the analyses and assisted with the aspects of the social and physical environ- tially lower health ratings among persons writing. A.R. Pebley led the original data collection and ment that influence the health and behaviors with chronic conditions in the most deprived assisted with the study and the writing. All authors helped to conceptualize ideas, interpret findings, and of individuals. However, prior work from neighborhoods. An important step is for re- review drafts of the article. LAFANS suggests that census-tract definitions searchers and health care organizations to of neighborhood size are highly correlated systematically collect information on the envi- Acknowledgments with respondents’ reports of the size and ronmental and social factors that may serve A.F. Brown received support from the Center for boundaries of their neighborhoods.52 Still, as barriers to those with chronic conditions. Health Improvement in Minority Elders/Resource Cen- ters for Minority Aging Research at the University of further work is needed to better understand An understanding of how neighborhood in- California, Los Angeles (UCLA), from the National Insti- the characteristics of areas that influence the fluences health can enhance efforts to im- tutes of Health (NIH), National Institute of Aging (grant health-related experiences of residents. prove health through urban planning, hous- AG–02–004), the UCLA/Drew Project EXPORT/NIH National Center on Minority Health and Health Dispari- Another limitation is that the chronic con- ing policies, and modifying the food resource ties (grant P20 MD00148), and the Paul D. Beeson Ca- ditions were self-reported, with no indepen- environment. reer Development Award (grant AG 26748). dent verification of the results. Data on the In the clinical setting, enhanced awareness We thank Hope Watkins for her assistance with the preparation of the article. validity of self-reported data on chronic con- of the association between neighborhood fac- ditions suggest that accuracy of self-report tors and health for people with chronic con- (with either the medical record or physical ex- ditions may help health care providers iden- Human Participant Protection These analyses were approved by the UCLA institu- amination used as the gold standard) varies tify adults who are most likely to be affected tional review board. by condition. Several validation studies have by disadvantaged neighborhood environ- compared methods of identifying people with ments; they can then tell them where and References the chronic conditions included in our analy- how to obtain important services, such as 1. Haan M, Kaplan GA, Camacho T. Poverty and ses. Although self-reported diabetes shows better food, improved transportation, and health. Prospective evidence from the Alameda the strongest agreement with the medical rec- low-cost exercise facilities. Further work is County Study. Am J Epidemiol. 19 87;125(6): 989–998. ord, there is lower, but generally good, agree- also needed to clarify the specific socio- 2. Yen IH, Kaplan GA. Neighborhood social environment between self-report and medical record economic forces and structural characteristics ment and risk of death: multilevel evidence from the 53 diagnoses of hypertension. In contrast, there that influence health outcomes in general Alameda County Study. Am J Epidemiol. 1999;149: is substantial underreporting of asthma54 and and those that contribute to the differential 898–907. osteoarthritis.55–57 Nonetheless, we found observed for people with chronic conditions. 3. Winkleby MA, Cubbin C. Influence of individual and neighbourhood socioeconomic status on mortality similar patterns for the association between Previous studies have shown that for adults among black, Mexican-American, and white women neighborhood deprivation and health ratings with diabetes, intensive clinical and behav- and men in the United States. J Epidemiol Community for hypertension, arthritis, and diabetes. ioral interventions benefit those who are less Health. 2003;57:444–452. Another potential limitation is that certain educated59 or have low literacy levels60 more 4. Balfour JL, Kaplan GA. Neighborhood environ- groups, including poorer or less educated peo- than those who are more educated or liter- ment and loss of physical function in older adults: evidence from the Alameda County Study. Am J Epi- ple, may be less likely to report a chronic con- ate. We have yet to determine whether inten- demiol. 2002;155(6):506–515. dition owing to lower rates of diagnosis or a sive clinical, behavioral, or policy interven- 5. Patel KV, Eschbach K, Rudkin LL, Peek MK, lack of awareness of the condition. A recent tions can modify neighborhood effects on Markides KS. Neighborhood context and self-rated study suggests that less education and more co- people with chronic conditions. health in older Mexican-Americans. Ann Epidemiol. 2003;13:620–628. morbid conditions are associated with underreporting of conditions including hypertension 6. Stafford M, Marmot M. Neighbourhood deprivation and health: does it affect us all equally? Int J Epi- 58 and diabetes, but the absolute differences ob- About the Authors demiol. 2003;32:357–366. served for different levels of education were Arleen F. Brown and Alfonso Ang are with the Division of 7. Harburg E, Erfurt JC, Chape C, Hauenstein LS, relatively small. Misclassification of people in General Internal Medicine and Health Services Research, Schull WJ, Schork MA. Socioecological stressor areas

May 2007, Vol 97, No. 5 | American Journal of Public Health Brown et al. | Peer Reviewed | Research and Practice | 931  RESEARCH AND PRACTICE 

and black–white blood pressure: Detroit. J Chronic Dis. 27. Ellen IG, Mijanovich T, Dillman K. Neighborhood 45. Ford ES, Moriarty DG, Zack MM, Mokdad AH, 1973;26(9):595–611. effects on health: exploring the links and assessing the Chapman DP. Self-reported body mass index and health- 8. Devesa SS, Diamond EL. Socioeconomic and ra- evidence. J Urban Aff. 2001;23:391–408. related quality of life: findings from the Behavioral Risk cial differences in lung cancer incidence. Am J Epi- 28.Branch LG, Lu L. Transition probabilities to de- Factor Surveillance System. Obes Res. 2001;9(1):21–31. demiol. 19 83;118(6):818–831. pendency, institutionalization and death among the el- 46. Zhang J, Yu KF. What’s the relative risk? JAMA. 9. Jenkins CD. Social environment and cancer mor- derly over a decade. J Aging Health. 19 8 9;1:370–408. 1998;290:1690–1691. tality in men. N Engl J Med. 19 83;308(7):395–398. 29. Mor V, Wilcox V, Rakowski W, Hiris J. Functional 47.Carpenter J, Bithell J. Bootstrap confidence inter- 10.Crombie IK, Kenicer MB, Smith WC, Tunstall- transitions among the elderly: patterns, predictors, and vals: when, which, what? A practical guide for medical Pedoe HD. Unemployment, socioenvironmental factors, related hospital use. Am J Public Health. 1994;84(8): statisticians. Stat Med. 2000;19:1141–1164. and coronary heart disease in Scotland. Br Heart J. 1274–1280. 48. Barglow P, Hatcher R, Edidin DV, et al. Stress 1989;61( 2):172–177. 30. Mossey JM, Shapiro E. Self-rated health: a predic- and metabolic control in diabetes: psychosomatic evi- 11.Adler NE, Boyce WT, Chesney MA, Folkman S, tor of mortality among the elderly. Am J Public Health. dence and evaluation of methods. Psychosom Med. Syme L. Socioeconomic inequalities in health: no easy 19 82;72(8):800–808. 1984;46:127–144. solution. JAMA. 19 93;269(4):3140–3145. 31.Idler EL, Kasl SV, Lemke JH. Self-evaluated 49. Netuveli G, Wiggins RD, Hildon Z, Montgomery 12. Diez-Roux AV, Nieto FJ, Muntaner C, et al. Neigh- health and mortality among the elderly in New Haven, SM, Blane D. Functional limitation in long standing borhood environments and coronary heart disease: a Connecticut, and Iowa and Washington counties, Iowa, illness and quality of life: evidence from a national sur- multilevel analysis. Am J Epidemiol. 19 97;146:48–63. 19 82–1986. Am J Epidemiol. 1990;131(1):91–103. vey. BMJ. 2005;331(7529):1382–1383. 13.Roberts EM. Socioeconomic position and health: 32. Cott CA, Gignac MA, Badley EM. Determinants of 50. Seeman TE, Crimmins E. Social environment ef- the independent contribution of community socioeco- self rated health for Canadians with chronic disease fects on health and aging: integrating epidemiologic nomic context. Ann Rev Sociol. 19 9 9;25:489–516. and disability. J Epidemiol Community Health. 1999; and demographic approaches and perspectives. Ann N 14 .Yen IH, Syme SL. The social environment and 53(11):731–736. Y Acad Sci. 2001;954:88–117. health: a discussion of the epidemiological literature. 33. Molarius A, Janson S. Self-rated health, chronic dis- 51.Oyana TJ, Rogerson P, Lwebuga-Mukasa JS. Geo- Annu Rev Public Health. 1999;20:287–308. eases, and symptoms among middle-aged and elderly graphic clustering of adult asthma hospitalization and 15. Diez Roux AV, Merkin SS, Arnett D, et al. Neigh- men and women. J Clin Epidemiol. 2002;55(4):364–370. residential exposure to pollution at a United States– borhood or residence and incidence of coronary heart 34.Froom P, Melamed S, Triber I, Ratson NZ, Canada border crossing. Am J Public Health. 2004; disease. N Engl J Med. 2001;345:99–106. Hermoni D. Predicting self-reported health: the 94(7):1250–1257. 16. Morenoff J, Lynch JW. What makes a place CORDIS study. Prev Med. 2004;39(2):419–423. 52. Sastry N, Pebley AR, Zonta M. Neighborhood def- healthy? Neighborhood influences on racial/ethnic dis- 35. Sastry N, Ghosh-Dastidar B, Adams J, Pebley AR. initions and the spatial dimensions of daily life in Los parities in health over the life course. In: Anderson NB, The design of a multilevel longitudinal survey of chil- Angeles. Rand Working Paper DRU-2400/8-LAFANS. Bulatao RA, Cohen B, eds. Critical Perspectives on Ra- dren, families, and communities: the Los Angeles Fam- Available at: http://www.rand.org/pubs/drafts/ cial and Ethnic Differences in Health in Late Life. Wash- ily and Neighborhood Survey. Soc Sci Res. 2006;35(4): DRU2400.8/index.html. Accessed January 25, 2007. ington, DC: National Academy Press; 2004:406–449. 1000–1024. 53. Bush TL, Miller SR, Golden AL, Hale WE. Self- 17. Macintyre S, Maciver S, Sooman A. Area, class 36. Sastry N, Pebley AR. Non-response in the Los Ange- report and medical record report agreement of selected and health: should we be focusing on places or people? les Family and Neighborhood Survey. Rand Corp, Santa medical conditions in the elderly. Am J Public Health. Int Soc Policy. 19 93;22:213–234. Monica, Calif, 2003. Available at: http://www.rand.org/ 1989;79:1554–1556. 18. Sampson R, Raudenbush S, Earls F. Neighbor- pubs/drafts/DRU2400.7. Accessed March 5, 2007. 54. Mohangoo AD, van der Linden MW, Schellevis hoods and violent crime: a multilevel study of collec- 37.2000 US Census, Summary File 3. Available at: FG, Raat H. Prevalence estimates of asthma or COPD tive efficacy. Science. 19 97;277:918–924. http://www.census.gov/Press-Release/www/2002/ from a health interview survey and from general prac- 19. Diez Roux AV. Bringing context back into epide- sumfile3.html. Accessed January 25, 2007. titioner registration: what’s the difference? Eur J Public miology: variables and fallacies in multilevel analysis. 38.Diez Roux AV, Kiefe CI, Jacobs DR Jr, et al. Area Health. 2006;16:101–105. Am J Public Health. 19 98;88:216–222. characteristics and individual-level socioeconomic posi- 55. Simpson CF, Boyd CM, Carlson MC, Griswold ME, 20.Pickett KE, Pearl M. Multilevel analyses of neigh- tion indicators in three population-based epidemiologic Guralnik JM, Fried LP. Agreement between self-report bourhood socioeconomic context and health outcomes: studies. Ann Epidemiol. 2001;11(6):395–405. of disease diagnoses and medical record validation in a critical review. J Epidemiol Community Health. 2001; 39.Ruo B, Rumsfeld JS, Hlatky MA, Liu H, Browner disabled older women: factors that modify agreement. 55(2):111–122. WS, Whooley MA. Depressive symptoms and health- J Am Geriatr Soc. 2004;52(1):123–127. 21. Evans GW, Kantrowitz E. Socioeconomic status related quality of life: the Heart and Soul Study. JAMA. 56. Harlow SD, Linet MS. Agreement between ques- and health: the potential role of environmental risk ex- 2003;290(2):215–221. tionnaire data and medical records: the evidence for ac- posure. Annu Rev Public Health. 2002;23:303–331. 40. Knäuper B, Cannell CF, Schwarz N, Bruce ML, curacy of recall. Am J Epidemiol. 1989;129:233–248. 22. Steptoe A, Feldman P. Neighborhood problems as Kessler RC. Improving accuracy of major depression 57. Bowlin SJ, Morrill BD, Nafziger AN, Jenkins PL, sources of chronic stress: development of a measure age-of-onset reports in the US National Comorbidity Lewis C, Pearson TA. Validity of cardiovascular disease of neighborhood problems, and associations with socio- Survey. Int J Methods Psychiatr Res. 1999;8(1):39–48. risk factors assessed by telephone survey: the Behav- economic status and health. Ann Behav Med. 2001;23: 41.Idler EL, Russell LB, Davis D. Survival, functional ioral Risk Factor Survey. J Clin Epidemiol. 19 93;46(6): 177–185. limitations, and self-rated health in the NHANES I epi- 561–571. 23. Bodenheimer T, Wagner EH, Grumbach K. Im- demiologic follow-up study, 1992. Am J Epidemiol. 58. Okura Y, Urban LH, Mahoney DW, Jacobsen SJ, proving primary care for patients with chronic illness. 2000;152:874–883. Rodeheffer RJ. Agreement between self-report ques- N Engl J Med. 2002;288:1775–1779. 42. Holtbrugge W, Schumacher M. A comparison of tionnaires and medical record data was substantial for 24.Wagner EH, Austin BT, Davis C, Hindmarsh M, regression models for the analysis of ordered categori- diabetes, hypertension, myocardial infarction and stroke Bonomi A. Improving chronic illness care: translating cal data. Appl Stat. 19 91;40:249–259. but not for heart failure. J Clin Epidemiol. 2004;57(10): evidence into action. Health Aff. 2001;20:64–78. 43. Ananth CV, Kleinbaum DG. Regression models 1096–1103. 25. Glasgow RE, Wagner EH, Kaplan RM, Vinicor F, for ordinal responses: a review of methods and applica- 59. Goldman DP, Smith JP. Can patient self-management Smith L, Norman J. If diabetes is a public health prob- tions. Int J Epidemiol. 19 97;26:1323–1333. help explain the SES health gradient? Proc Natl Acad Sci lem, why not treat it as one? A population-based ap- 44. Okosun IS, Choi S, Matamoros T, Dever GE. Obe- U S A. 2002;99(16):10929–10934. proach to chronic illness. Ann Behav Med. 1999; 21(2): sity is associated with reduced self-rated general health 60.Rothman RL, DeWalt DA, Malone R, et al. Influ- 159–170. status: evidence from a representative sample of white, ence of patient literacy on the effectiveness of a pri- 26.Verbrugge LM, Jette AM. The disablement pro- black, and Hispanic Americans. Prev Med. 2001;32(5): mary care-based diabetes disease management process. Soc Sci Med. 19 9 4;38(1):1–14. 429–436. gram. JAMA. 2004;292(14):1711–1716.

932 | Research and Practice | Peer Reviewed | Brown et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Relationships Between Self-Reported Unfair Treatment and Prescription Medication Use, Illicit Drug Use, and Alcohol Dependence Among Filipino Americans

| Gilbert C. Gee, PhD, Jorge Delva, PhD, and David T. Takeuchi, PhD

Stressors may lead to illness and behavior Objectives. We examined associations between self-reported unfair treatment change. Stressors can promote drug, alcohol, and prescription medication use, illicit drug use, and alcohol dependence. and medication use via 2 different pathways. Methods. We used data from the Filipino American Community Epidemiological First, they can cause illness and subsequent Survey, a cross-sectional investigation involving 2217 Filipino Americans inter- use of medications. Acute and chronic stress- viewed in 1998–1999. Multinomial logistic and negative binomial regression ors have been associated with mortality and analyses were used in assessing associations between unfair treatment and the with conditions ranging from the common substance use categories. cold to depression and heart disease.1–3 Indi- Results. Reports of unfair treatment were associated with prescription drug viduals may be prescribed medications to use, illicit drug use, and alcohol dependence after control for age, gender, location treat their illness and reduce their stress lev- of residence, employment status, educational level, ethnic identity level, nativity, language spoken, marital status, and several health conditions. els. Second, individuals may use substances Conclusions. Unfair treatment may contribute to illness and subsequent use of to cope with stressors, a theme that appears prescription medications. Furthermore, some individuals may use illicit drugs and in several frameworks, including the relapse alcohol to cope with the stress associated with such treatment. Addressing the an- 4 prevention model, the tension reduction tecedents of unfair treatment may be a potential intervention route. (Am J Public 5,6 7 model, the self-medication hypothesis, Health. 2006;96:933–940. doi:10.2105/AJPH.2005.075739) and the stress–coping model of addiction.8,9 Yet, despite decades of research, questions remain as to the types of stressors that might including racial discrimination and internal- METHODS contribute to substance use.10 ized oppression.33–35 We focused on 1 type of stressor, unfair We sought to extend previous research by Sample treatment, defined as discriminatory behavior examining the associations of perceptions of We analyzed data from the 1998–1999 on the part of institutions and individuals unfair treatment with alcohol dependence Filipino American Community Epidemiologi- directed toward individuals with less power and use of prescription and illicit drugs. Be- cal Study. A detailed description of this study and the groups to which they belong. Unfair cause unfair treatment may lead to illness and can be found elsewhere.45 To be included in treatment is an important stressor linked to subsequent use of medications, we hypothe- the study, individuals had to be of Filipino illnesses such as depression, chronic disease, sized that unfair treatment would be associ- heritage, had to be 18 years or older, and and hypertension.11–18 It may structure one’s ated with greater use of prescription medica- had to reside in San Francisco or Honolulu. life circumstances, pose direct threats to one’s tions. Furthermore, given that individuals One eligible person within a household was safety, and erode one’s sense of self.19–22 may misuse substances to cope with stressors randomly selected and administered an in- In these ways, unfair treatment is a socially associated with unfair treatment, we hypothe- person interview in English, Tagalog, or Ilo- derived stressor that operates at multiple sized that unfair treatment would be associ- cano. The response rate among eligible indi- levels.12,23,24 ated with potential misuse of medications, use viduals was 78%, resulting in a sample of Unfair treatment may be related to sub- of illicit drugs, and alcohol dependence. 2285. After exclusion of 68 respondents with stance use. It has been shown to be associ- We used a large, community-based sample missing data, a final sample of 2217 was ated with alcohol use among bus drivers25,26 of Filipino Americans, the second largest Asian available for the analyses. and African American adults,27 with cigarette American ethnic group, to examine these hy- smoking among African American youths28,29 potheses.36,37 Historically, Filipino Americans Dependent Variables and adults,30 and with substance use among have experienced considerable unfair treat- We examined 4 outcomes: alcohol depen- American Indian children31 and African ment,38–41 and they continue to experience dence, illicit drug use, and prescription and American parents and children.32 These rela- such treatment today.42–44 Recent research has nonprescription use of medications. We also tionships may be partially explained by use shown that unfair treatment is associated with assessed several distinct drug subcategories. of substances as a way to cope with the chronic health conditions45 and with depres- Alcohol dependence. The University of stressors associated with minority status, sive symptoms46 among this population. Michigan Composite International Diagnostic

May 2007, Vol 97, No. 5 | American Journal of Public Health Gee et al. | Peer Reviewed | Research and Practice | 933  RESEARCH AND PRACTICE 

Interview (UM-CIDI), a modified version of the drug misuse was a sum of the illicit drugs Control Variables World Health Organization’s Composite Inter- used and the prescription drugs misused Our analyses controlled for age, gender, national Diagnostic Interview, was used to (range: 0–9). nativity (US vs foreign born), language (Eng- measure alcohol dependence. The UM-CIDI is lish, Tagalog/Ilocano, or both), employment a standardized survey designed to be adminis- Independent Variables status (currently vs not currently employed), tered by trained interviewers, and it is struc- We assessed 2 distinct measures of unfair place of residence (Honolulu or San Fran- tured to allow for clinical diagnoses according treatment. Everyday unfair treatment was a cisco), years of education, and marital status to the criteria of the Diagnostic and Statistical 9-item scale assessing frequency of routine (married vs other). In addition, we included Manual of Mental Disorders (revised third edi- experiences of unfair treatment within the a 9-item measure of ethnic identity derived tion; DSM-III-R) and the International Classifi- preceding 30 days. This scale, derived from from the Multigroup Ethnic Identity Measure cation of Diseases, 10th Revision.47–50 A stan- qualitative research,52,53 was first used in the and adapted for Filipino Americans.57,58 It is dard protocol was used to classify respondents Detroit Area Study.54,55 The scale assessed scored from 1 to 4, with higher scores indi- as dependent or not dependent on alcohol.51 how often (1=never, 5=very often) respon- cating a stronger sense of affiliation with Fil- Illicit drug use. Questions about lifetime dents experienced the following types of un- ipino Americans (Cronbach α=0.74). use of illicit drugs were adapted from the fair treatment: encountering prejudice and Because unfair treatment has been associ- UM-CIDI. Respondents were asked whether discrimination from others; being treated with ated with depression46 and chronic health they had ever used inhalants, marijuana, co- less courtesy and less respect than others; re- conditions45 among Filipino Americans and caine, hallucinogens, or heroin. Respondents ceiving poorer service at restaurants or stores; because these conditions predict medication were assigned a score of 1 for each drug used people acting as if they are “afraid of you,” as use, we also controlled for previous health and a score of 0 otherwise. if “you are dishonest,” or as if they are “better conditions. We included 3 types of mental Prescription drug use and misuse. Questions than you are”; being called names or insulted; health problems for which data were available about medications also were derived from the and being threatened or harassed. from our survey: lifetime depression, manic epi- UM-CIDI. Respondents were asked about the Higher scores on the scale indicated higher sodes, and dysthymia. Data on all 3 measures following 4 categories of substances: seda- frequencies of everyday unfair treatment. The were obtained from the UM-CIDI short form. tives (e.g., barbiturates, “downers”), tranquiliz- scale’s Cronbach α coefficient was 0.88. It Clinical case diagnoses, according to DSM-III-R ers (e.g., benzodiazepines, “nerve pills”), stim- has been shown that scores on this scale pre- criteria, were derived from a computerized ulants (e.g., amphetamines, “uppers”), and dict self-rated health among African Ameri- scoring algorithm.47,59 Respondents were as- analgesics (e.g., nonsteroidal anti-inflamma- cans55,56 and chronic health conditions and signed a score of 1 for each diagnosed disor- tory drugs, “painkillers”). Respondents were depression among Filipino Americans.45,46 der and were assigned a score of 0 otherwise. given examples of both lay and pharmaceuti- Unfair events was a count of respondents’ Also, we used a checklist adapted from the cal names for the substances in each category endorsements to being “treated unfairly or Medical Outcomes Study60–62 to examine cur- and then asked about their history of use of badly” during the preceding 12 months be- rent physical health conditions. The conditions these substances. cause of their race or ethnicity, because they assessed were diabetes, hypertension, arthri- Responses were coded into 1 of 3 mutually spoke a different language, or because they tis, physical disability (e.g., loss of a limb, birth exclusive categories: prescription use, misuse, spoke with an accent. Scores ranged from 0 defect), trouble breathing because of emphy- and never used. Prescription use meant that to 3, with higher scores indicating more un- sema or lung disease, cancer, neurological the drug was used as prescribed by a doctor. fair events. Scores on this measure have been conditions (e.g., epilepsy, Parkinson’s disease), Misuse indicated that the drug was used with- shown to predict chronic health conditions stroke, major paralysis, heart failure, angina, out a physician’s prescription, used for non- among Filipino Americans.45 other heart disease, back problems, stomach prescription purposes, or used more often The unfair events measure differed from ulcers, chronic inflamed bowel, enteritis or than prescribed or that respondents believed the everyday unfair treatment scale in 3 colitis, thyroid disease, kidney failure, trouble that they were so dependent on the medica- ways. First, the events measure explicitly seeing, and migraine headaches. The disor- tion they could not stop using it. Never used focused on race/ethnicity, whereas in the ders were summed; the range within our referred to no lifetime use of the drug. everyday measure race/ethnicity issues were sample was 0 to 14. In addition, we created 4 indexes. Total implied rather than explicit. Second, the illicit drug use was a count of the number of events instrument examined experiences oc- Data Analysis illicit drugs used (range: 0–5). Total prescrip- curring during the previous 12 months rather We first explored the data using bivariate tion drug use was a count of the number of than routine (everyday) experiences. Third, analyses and then conducted multivariate drugs used strictly for prescription purposes the 2 measures were scored differently, in analyses to adjust for controls. We performed (range: 0–4). Total prescription drug misuse that unfair events was a count rather than a analyses of variance to assess differences in was a count of the drugs used for nonpre- scale. The correlation between the measures unfair treatment between individuals who used scription purposes, as just described (range: was low (0.37), suggesting that they tapped prescription medications, individuals who mis- 0–4). Total illicit drug use and prescription different aspects of the construct. used them, and individuals who had never

934 | Research and Practice | Peer Reviewed | Gee et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 1—Selected Characteristics of 41 years old, employed, married, and foreign suggested group differences in reporting of un- Participants in the 1998–1999 Filipino born; spoke both English and Tagalog/ fair treatment for tranquilizers, no pairwise American Community Epidemiological Ilocano; had a moderately high ethnic iden- comparisons were significant, possibly as a re- Study tity; and had 11 years of education. Most of sult of the small sample of misusers. the respondents were healthy, with low rates Table 3 displays the results of the multi- Sample of depression, manic episodes, and dysthymia. variate multinomial logistic regression analy- (n=2217) On average, respondents reported 1 health ses. In these analyses, “never used” was the Female, % 50.70 condition and low frequencies of everyday reference category. Model 1 assessed the as- Marital status, % unfair treatment and unfair events. sociation between unfair treatment and drug Single 25.70 Table 2 displays mean levels of both mea- use with control for age, gender, location of Widowed/separated/divorced 15.41 sures of unfair treatment according to sub- residence, employment status, educational Married 58.89 stance use category. There were significantly level, ethnic identity, nativity, language, and Foreign born, % 79.56 more reports of everyday unfair treatment marital status. Reporting of everyday unfair Language spoken in household, % among individuals classified as alcohol de- treatment (relative risk ratio=1.38; 95% con- Tagalog/Ilocano 30.62 pendent than among individuals not classi- fidence interval [CI]=1.05, 1.82) and unfair Tagalog/Ilocano and English 48.98 fied as such (means of 2.11 and 1.37, respec- events (relative risk ratio=1.24; 95% CI= English 20.39 tively; P <.001). However, no significant 1.00, 1.54) predicted use of prescribed tran- Mean age, y (SD) 41.57 (13.30) differences were observed between these 2 quilizers. A similar pattern emerged for use of Mean ethnic identity score (SD) 3.55 (0.47) groups in mean levels of unfair events. Re- prescribed analgesics. There was also a mar- Mean educational level, y (SD) 11.5 (5.21) ports of everyday unfair treatment and unfair ginally significant association between unfair Employed, % 74.86 events were more common among individu- events and misuse of stimulants (relative risk Mean everyday unfair treatment 1.38 (0.56) score (SD) als who had used illicit drugs in their lifetime ratio=1.31; 95% CI=1.01, 1.69). Mean unfair events score (SD) 0.17 (0.60) than among those who had never used these Model 2 added controls for lifetime depres- Mean no. of health conditions (SD) 1.03 (1.46) drugs (the exception being no difference in sion, manic episodes, dysthymia, and chronic Major depression, % 3.50 unfair events among inhalant users). For ex- health conditions. The results were similar to Dysthymia, % 2.79 ample, those who had used any illicit drug those of model 1 with the exception that the Manic episodes, % 0.20 reported a higher level of everyday unfair associations between unfair events and pre- treatment than those who had not used illicit scription use of tranquilizers and misuse of drugs (means of 1.65 and 1.31, respectively; stimulants were no longer significant. used them. In addition, we conducted post hoc P <.001). We also conducted multivariate logistic re- analyses using Sidak multiple comparison tests Several interesting findings emerged when gression analyses focusing on the individual for pairwise comparisons. We used logistic re- individuals who used prescription drugs, mis- illicit drugs and alcohol dependence (table gression to examine the association of each used prescription drugs, or had never used available on request). Everyday unfair treat- measure of unfair treatment with alcohol de- these drugs were compared on the 2 unfair ment was associated with use of inhalants pendence and with each of the illicit drugs, treatment measures. Individuals who had mis- (odds ratio [OR] = 1.90; 95% CI = 1.25, followed by multinomial logistic regression to used prescription drugs generally reported 2.88), marijuana (OR = 1.31; 95% CI = 1.03, analyze the nominal outcomes of prescription more experiences of everyday unfair treat- 1.67), heroin (OR = 2.69; 95% CI = 1.55, drug use and misuse. We initially used Poisson ment and unfair events than individuals in 4.65), and alcohol dependence (OR = 2.22; regression analyses for count measures (total the other groups, especially those who had 95% CI = 1.36, 3.62). However, everyday illicit drug use, total prescription drug use, total never used prescription drugs. unfair treatment was not significantly associ- prescription drug misuse, and total illicit drug No clear pattern emerged between prescrip- ated with use of cocaine or hallucinogens. use and prescription drug misuse) but switched tion drug users and nonusers. Reports of both Unfair events were not associated with use to negative binomial regression after detecting measures of unfair treatment were more fre- of any of the illicit drugs or with alcohol overdispersion (i.e., violation of the assumption quent among prescription users than nonusers dependence. of Poisson regression that means and variances in the case of analgesics; differences were non- Because we were interested in the general are equal). Weights were applied in the analy- significant for sedatives and tranquilizers. Un- relationship between unfair treatment and ses to account for differential probabilities of expectedly, stimulant users reported less every- substance use and because of our concerns selection within a household. day unfair treatment and fewer unfair events about the low prevalence rates in several than those who had never used stimulants. drug categories, we summed the drug cate- RESULTS However, the associations between discrimina- gories to create the indexes described earlier tion and stimulant use disappeared in multi- (total illicit drug use, total prescription drug Table 1 summarizes the characteristics of variate analyses (described subsequently). use, total prescription drug misuse, total illicit the study sample. The typical respondent was Finally, although the analysis of variance drug use and prescription drug misuse). In the

May 2007, Vol 97, No. 5 | American Journal of Public Health Gee et al. | Peer Reviewed | Research and Practice | 935  RESEARCH AND PRACTICE 

TABLE 2—Unadjusted Mean Scores on Measures of Unfair Treatment, by Type of Substance Used: Filipino American Community Epidemiological Study, 1998–1999

Ever Used Misused Never Used Dependent Not Dependent No. Mean (SD) No. Mean (SD) No. Mean (SD) No. Mean (SD) No. Mean (SD)

Alcohol 26 2221 Everyday unfair treatmenta 2.11 (1.02) 1.37 (0.55)† Unfair events 0.17 (0.60) 0.35 (0.69) Illicit drugs Inhalants 38 2242 Everyday unfair treatment 2.07 (0.77) 1.37 (0.55) Unfair events 0.24 (0.71) 0.17 (0.60) Marijuana 466 1815 Everyday unfair treatment 1.64 (0.69) 1.32 (0.50) Unfair events 0.26 (0.69) 0.15 (0.58) Cocaine 134 2112 Everyday unfair treatment 1.75 (0.75) 1.36 (0.54) Unfair events 0.38 (0.86) 0.16 (0.58) Hallucinogens 83 2198 Everyday unfair treatment 1.93 (0.80) 1.36 (0.54) Unfair events 0.39 (0.82) 0.16 (0.59) Heroin 12 2269 Everyday unfair treatment 2.37 (0.91) 1.38 (0.55) Unfair events 0.50 (0.90) 0.17 (0.60) Any illicit drugs 478 1800 Everyday unfair treatment 1.65 (0.03) 1.31 (0.01) Unfair events 0.26 (0.03) 0.15 (0.01) Prescription drugs Sedatives 243 80 1959 Everyday unfair treatmentb,c 1.34 (0.57) 1.60 (0.61) 1.38 (0.56) Unfair events 0.20 (0.66) 0.29 (0.73) 0.16 (0.59) Tranquilizers 241 48 1993 Everyday unfair treatment 1.45 (0.63) 1.57 (0.60) 1.37 (0.55) Unfair events 0.25 (0.74) 0.25 (0.76) 0.16 (0.58) Stimulants 103 112 2066 Everyday unfair treatmentb,c,d 1.22 (0.46) 1.74 (0.73) 1.37 (0.55) Unfair eventsb,c,d 0.12 (0.47) 0.39 (0.81) 0.16 (0.59) Analgesics 648 222 1411 Everyday unfair treatmentd 1.48 (0.58) 1.41 (0.56) 1.33 (0.55) Unfair eventsd 0.26 (0.74) 0.18 (0.62) 0.13 (0.52) Any medication 655 353 1273 Everyday unfair treatmentc,d 1.45 (0.58) 1.50 (0.62) 1.32 (0.53) Unfair eventsb,d 0.26 (0.75) 0.23 (0.67) 0.11 (0.48)

aSignificant difference between alcohol dependent and not dependent (Sidak multiple comparisons test). bSignificant difference between ever used and misused (Sidak multiple comparisons test). cSignificant difference between misused and never used (Sidak multiple comparisons test). dSignificant difference between ever used and never used (Sidak multiple comparisons test).

first model (Table 4), negative binomial re- marital status, location of residence, employ- The second model showed that everyday gression was used to examine the association ment status, educational level, ethnic identity, unfair treatment was associated with increased between total illicit drug use and the 2 mea- nativity, and language had been taken into reports of prescription drug use (b=0.14, P≤ sures of unfair treatment. Everyday unfair account (b=0.18, P≤.05). However, unfair .05). Unfair events were also associated with treatment was associated with increasing events were not associated with illicit drug prescription drug use (b=0.15, P≤.001). In counts of illicit drug use after age, gender, use (b=0.07, P >.05). the prescription drug use analysis, we took a

936 | Research and Practice | Peer Reviewed | Gee et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 3—Use of Prescription Drugs and Measures of Unfair Treatment: Filipino American TABLE 4—Measures of Unfair Treatment Community Epidemiological Study, 1998–1999 and Categories of Drug Use: Filipino American Community Epidemiological Model 1a Model 2b Study, 1998–1999 Prescription Prescription Prescription Prescription Drug Use vs Drug Misuse vs Drug Use vs Drug Misuse vs b (SE) Nonuse, Relative Nonuse, Relative Nonuse, Relative Nonuse, Relative RR (95% CI) RR (95% CI) RR (95% CI) RR (95% CI) Total illicit drug use Everyday unfair treatment 0.18** (0.07) Sedatives Unfair events 0.07 (0.68) Everyday unfair treatment 0.93 (0.65, 1.32) 1.01 (0.69, 1.47) 0.88 (0.62, 1.26) 0.94 (0.63, 1.39) Total prescription drug use Unfair events 1.20 (0.94, 1.52) 1.13 (0.83, 1.54) 1.18 (0.93, 1.49) 1.09 (0.78, 1.52) Everyday unfair treatment 0.14* (0.07) Tranquilizers Unfair events 0.15*** (0.04) Everyday unfair treatment 1.38 (1.05, 1.82) 0.83 (0.48, 1.42) 1.33 (1.01, 1.75) 0.80 (0.44, 1.43) Total prescription drug misuse Unfair events 1.24 (1.00, 1.54) 0.94 (0.61, 1.46) 1.22 (0.98, 1.52) 0.94 (0.60, 1.49) Everyday unfair treatment 0.01 (0.10) Stimulants Unfair events 0.08 (0.73) Everyday unfair treatment 0.72 (0.39, 1.36) 1.13 (0.80, 1.60) 0.71 (0.38, 1.35) 1.02 (0.73, 1.45) Total illicit drug or prescription Unfair events 0.84 (0.54, 1.30) 1.31 (1.01, 1.69) 0.84 (0.54, 1.29) 1.25 (0.97, 1.61) drug misuse Analgesics Everyday unfair treatment 0.16* (0.07) Everyday unfair treatment 1.32 (1.06, 1.64) 0.95 (0.69, 1.32) 1.30 (1.04, 1.62) 0.93 (0.67, 1.38) Unfair events 0.07 (0.06) Unfair events 1.36 (1.11, 1.65) 1.12 (0.86, 1.45) 1.35 (1.11, 1.64) 1.10 (0.85, 1.44) Note. Values were derived from multivariate tests of Note.RR=risk ratio; CI=confidence interval.Values were derived from multinomial logistic regression analyses. Everyday associations between measures of unfair treatment unfair treatment and unfair events were modeled separately. and types of drugs used via negative binomial a This model controlled for age, gender, location of residence, employment status, educational level, ethnic identity, nativity, regression.All analyses controlled for age, gender, language, and marital status. location of residence, employment status, educational b This model controlled for the factors controlled in model 1 along with lifetime depression, manic episodes, dysthymia, and level, ethnic identity, nativity, language, and marital chronic health conditions. status. Prescription drug use also controlled for lifetime depression, manic episodes, dysthymia, and chronic health conditions. Everyday unfair treatment and unfair events were modeled separately. Dependent variables more conservative approach by controlling for the association between everyday unfair treat- were counts of the number of drugs used. For illicit lifetime depression, manic episodes, dys- ment and total prescription drug use. Reports drug use, scores ranged from 0 to 5. For prescription thymia, and chronic health conditions in addi- of unfair events were similarly moderated by drug use and prescription drug misuse, scores ranged from 0 to 4. For any use of illicit drugs or prescription tion to the covariates described earlier. nativity and ethnic identity. drug misuse, scores ranged from 0 to 9. The third model indicated that total pre- *P≤.05; **P≤.01; ***P≤.001. scription drug misuse was not associated with DISCUSSION everyday unfair treatment or unfair events. Finally, the fourth model showed that total il- Our data suggest that reports of unfair dependence among Filipino Americans. licit drug and prescription misuse was signifi- treatment are associated with substance A1-unit increase in reports of everyday un- cantly associated with increased reports of use, echoing similar findings in the litera- fair treatment was associated with 2-fold everyday unfair treatment (b=0.16, P≤.05) ture.25,28,29,31,63 These studies suggest that greater odds of being classified as alcohol de- but not significantly associated with reports substance use may serve as a way of coping pendent, although caution must be exercised of unfair events (b=0.07, P>.05). with psychosocial stressors and that unfair in interpreting this result because of the small Supplemental analyses revealed no consis- treatment may be a particularly relevant numbers of individuals classified as such. tent moderating effects according to age, gen- stressor for minority populations.33,34 Consis- Next, consider prescription medications. der, or health status, but there were signs of tent with the stress perspective, research has To our knowledge, the present study is the moderation according to nativity and ethnic also shown that unfair treatment is associated first to report that unfair treatment may be identity. Specifically, US-born status was asso- with a variety of health outcomes13 ,18 , 2 0 and associated with use of prescription medica- ciated with more reports of illicit drug use that coping resources may buffer the effects tions, although a recent study suggests that than foreign-born status, as well as increased of unfair treatment.16,45,46,64 unfair treatment may be associated with de- reports of everyday unfair treatment and un- First, consider alcohol use. Some studies lays in filling prescriptions.65 Two different fair events (table available on request). How- have reported associations between percep- measures of unfair treatment, one focusing ever, the association between everyday unfair tions of unfair treatment and alcohol use on everyday experiences and another focus- treatment and illicit drug use was reduced among African Americans27,32 and bus driving on past year events, were associated among individuals born in the United States. ers.26,63 We built on earlier work by showing with prescription drug use. If unfair treat- Likewise, ethnic identity appeared to diminish that unfair treatment is associated with alcohol ment increases the risk of illness, then the

May 2007, Vol 97, No. 5 | American Journal of Public Health Gee et al. | Peer Reviewed | Research and Practice | 937  RESEARCH AND PRACTICE 

association between unfair treatment and communication, June 2005). It might be that context of several limitations. First, we were prescription medication use is self-evident. unfair treatment is associated with a series not able to establish temporal relationships In the analyses involving prescription med- of symptoms that are more likely to lead cli- between study variables because of the cross- ications, we controlled for depression and nicians to prescribe tranquilizers than seda- sectional design. Thus, although theory sug- chronic health conditions, 2 outcomes previ- tives. Information on specific drugs used and gests that unfair treatment may lead to sub- ously shown to be associated with unfair specific prescribing practices was not avail- stance use, it is also possible that substance treatment among Asian Americans,62 as well able in the present data set, so we were un- use causes individuals to experience and re- as dysthymia, manic episodes, and other so- able to investigate this potential explanation. port stigma and unfair treatment.69 Second, ciodemographic characteristics.16,45,46 Obvi- A focused examination of the types of symp- we relied on self-reported data, which may ously, these conditions represent only a se- toms associated with unfair treatment, as have introduced response effects such as re- lect sample of the universe of potential well as the types of medications prescribed, call bias and socially desirable reporting.70 health conditions one could include in such awaits future study. Future work should include biomarkers of analyses; the association between unfair Finally, consider illicit drugs and misused substance use and alternative measures of treatment and health probably would have medications. Consistent with the stress per- unfair treatment. been eliminated had we controlled for all spective is the finding that reports of every- Third, we examined lifetime prescription possible health outcomes. day unfair treatment were also associated and nonprescription use of medications. Al- Alternatively, one could argue that our in- with use of inhalants, marijuana, and heroin. though it would have been desirable to exam- clusion of both physical and mental health However, misuse of prescription drugs was ine current drug use, we did not do so be- conditions “overcontrolled” for the association not associated with unfair treatment. Bivariate cause of the low rates of current use within between medication use and unfair treatment analyses indicated a trend in which prescrip- our sample. Fourth, it is unclear how our re- (and, hence, that our test was conservative), tion drug misuse was associated with in- sults might generalize to other populations. given that it has been theorized that prescrip- creased reports of unfair treatment. Given However, we overcame one limitation of pre- tion drug use and health outcomes fall along that misuse was relatively rare, it is possible vious research by focusing on a specific Asian the same etiological pathway. That stated, that there was inadequate power to fully un- subgroup rather than examining an aggregate both perspectives involve the same sugges- cover such an association. Alternatively, this of Asian Americans.37,71,72 tion, namely that unfair treatment may be result may indicate that there is indeed no Granted these caveats, our study provides an important factor that contributes to dimin- association between prescription drug misuse novel findings on an increasingly recognized ished well-being. and perceptions of unfair treatment. health risk factor, unfair treatment. We have In terms of specific prescription drugs, it Everyday unfair treatment was more con- presented evidence of associations between appears that unfair treatment was associated sistently associated with substance use than unfair treatment and substance use among with use of analgesics and tranquilizers, were unfair events. This suggests that rou- Filipino Americans, even after control for a classes of drugs often used to treat anxiety, tine experiences of unfair treatment may number of potential confounders. Although pain, and stress disorders. These results, con- take a greater toll on well-being than more we examined the stress process, it is impor- sonant with the perspective of unfair treat- acute experiences.18 , 5 2 However, it would be tant to examine the upstream production of ment as a stressor that may cause mental ill- premature to rule out competing explana- stressors.18 ,73–76 Should the present findings ness and distress,22,66,67 suggest that future tions. The unfair events measure was more be replicable, enduring, and causal, they sug- studies might consider potential associations explicitly racialized and included fewer items gest that a “war on discrimination” may aid between unfair treatment and pain, tension, than the everyday unfair treatment measure. the purported “war on drugs.” More broadly, and related problems. However, an anom- Thus, differences in the content of the mea- policies that unravel the interlocking systems alous finding in the present study was the sures as well as their psychometric proper- that maintain and promote oppression may lack of an association between use of seda- ties may have accounted for the divergence foster not only a civil and just society but a tives and unfair treatment. This result was in findings. Future investigations would ben- healthy one as well. unexpected given that sedatives and tran- efit from improved instruments, and these quilizers have similar effects and are used to studies should evaluate how measures of un- treat similar symptoms. One explanation fair treatment that focus on specific charac- About the Authors may be that sedatives are often prescribed teristics (e.g., race) compare with measures Gilbert C. Gee is with the School of Public Health, Univer- for sleep disturbances, and sleep problems that do not focus on these characteristics.68 sity of Michigan, Ann Arbor. Jorge Delva is with the School might have no relationship with unfair That stated, it is remarkable that associations of Social Work, University of Michigan. David T. Takeuchi is with the Department of Sociology and School of Social treatment. with substance use were found with both Work, University of Washington, Seattle. In addition, tranquilizers are often pre- measures. Requests for reprints should be sent to Gilbert C. Gee, scribed for sleep disorders, especially Although this study provides insights into PhD, University of Michigan, 1420 Washington Heights, Room M5224, Ann Arbor, MI 48109-2029 (e-mail: when these disorders co-occur with anxiety the roles of unfair treatment and substance [email protected]). and other symptoms (J. A. Himley, oral use, our findings should be considered in the This article was accepted September 29, 2005.

938 | Research and Practice | Peer Reviewed | Gee et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Contributors 16. Noh S, Kaspar V. Perceived discrimination and 33. Caetano R, Clark CL, Tam T. Alcohol consump- G.C. Gee originated the study and led the writing and depression: moderating effects of coping, acculturation tion among racial/ethnic minorities: theory and re- analyses. J. Delva and D.T. Takeuchi assisted with the and ethnic support. Am J Public Health. 2003;93: search. Alcohol Health Res World. 1998;22:233–241. writing and the analysis. 232–238. 34. Gil AG, Wagner EF, Vega WA. Acculturation, 17.Pavalko E, Mossakowski KN, Hamilton V. Does familism and alcohol use among Latino adolescent Acknowledgments perceived discrimination affect health? Longitudinal re- males: longitudinal relations. J Community Psychol. This study was supported by a grant from the National lationships between work discrimination and women’s 2000;28:443–458. Institute on Alcohol Abuse and Alcoholism to David T. physical and emotional health. J Health Soc Behav. 35. D’Avanzo CE, Frye B, Froman R. Culture, stress Takeuchi (R0109633). 2003;44:18–34. and substance use in Cambodian refugee women. J Stud We thank Joseph Himley for helpful comments on 18.Williams DR, Neighbors H, Jackson JS. Racial/ethnic Alcohol. 1994;55:420–426. the article and Juan Chen for help in data preparation. discrimination and health: findings from community 36. Lai E, Arguelles D, eds. The New Face of Asian studies. Am J Public Health. 2003;93:200–208. Pacific America: Numbers, Diversity and Change in the Human Participant Protection 19.Williams DR. Racism and health: a research 21st Century. San Francisco, Calif: AsianWeek; 2004. This study was approved by the institutional review agenda. Ethn Dis. 1996;6:1–6. 37.Ro M. Moving forward: addressing the health of board of the University of Washington. Participants 20. Krieger N. Embodying inequality: a review of con- Asian American and Pacific Islander women. Am J Pub- provided informed consent to take part in the study. cepts, measures, and methods for studying health con- lic Health. 2002;92:516–519. sequences of discrimination. Int J Health Serv. 1999;29: 38. Abel EK. “Only the best class of immigration”: References 295–352. public health policy toward Mexicans and Filipinos in 1. McEwen BS. Protective and damaging effects of 21. Jones CP. Invited commentary: “race,” racism, and Los Angeles, 1910–1940. Am J Public Health. 2003; stress mediators. N Engl J Med. 19 9 8;338:171–179. the practice of epidemiology. Am J Epidemiol. 2001; 93:932–939. 2. Cohen S, Tyrrell DA, Smith AP. Psychological 154:299–304. 39. Chan S. Asian Americans: An Interpretive History. stress and susceptibility to the common cold. N Engl 22. Clark R, Anderson A, Clark VR, Williams DR. Boston, Mass: Twayne Publishers; 1991. J Med. 19 91;325:606–612. Racism as a stressor for African Americans. Am 40. Jung M-K. Interracialism: the ideological transfor- 3. Kelly S, Hertzman C, Daniels M. Searching for the Psychol. 1999;54:805–816. mation of Hawaii’s working class. Am Sociol Rev. 2003; biological pathways between stress and health. Annu 23. Carmichael S, Hamilton CV. Black Power: The 68:373–400. Rev Public Health. 19 97;18:437–462. Politics of Liberation in America. New York, NY: Vintage 41. San Juan E Jr. Configuring the Filipino diaspora in 4. Marlatt GA, Gordon JR. Relapse Reprevention: Books; 1967. the United States. Diaspora. 1994;3:117–133. Maintenance Strategies in the Treatment of Addictive Be- 24. Jones CP. Levels of racism: a theoretical frame- 42. Agbayani-Siewert P, Revila L. Filipino Americans. haviors. New York, NY: Guilford Press; 1985. work and gardener’s tale. Am J Public Health. 2000;90: Newbury Park, Calif: Sage Publications; 1995. 1212–1215. 5. Cogner JJ. Reinforcement theory and the dynam- 43. San Juan E Jr. The predicament of Filipinos in the ics of alcoholism. J Stud Alcohol. 1956;17:296–305. 25. Yen IH, Ragland DR, Greiner BA, Fisher J. Work- United States. In: Aguilar SK, ed. The State of Asian 6. Sher KJ, Levenson RW. Risk for alcoholism and place discrimination and alcohol consumption: findings America: Activism and Resistance in the 1990s. Boston, individual differences in the stress-response-dampening from the San Francisco Muni Health and Safety Study. Mass: South End Press; 1994:205–218. Ethn Dis. 1999;9:70–80. effect of alcohol. J Abnorm Psychol. 19 82;100: 44.Yamane L. Native-born Filipina/o Americans and 427–448. 26.Yen IH, Ragland DR, Greiner BA, Fisher JM. labor market discrimination. Feminist Economics. 2002; 7. Khantzian EJ. The self-medication hypothesis of Racial discrimination and alcohol-related behavior in 8:125–144. urban transit operators: findings from the San Fran- addictive disorders: focus on heroin and cocaine de- 45. Gee GC, Chen J, Spencer M, et al. Social support pendence. Am J Psychiatry. 19 85;142:1259–1264. cisco Muni Health and Safety Study. Public Health Rep. 1999;114:448–458. as a buffer for perceived unfair treatment among Fil- 8. Shiffman S. Relapse following smoking cessation: ipino Americans: differences between Honolulu and a situational analysis. J Consult Clin Psychol. 1982;50: 27. Martin JK, Tuch SA, Roman PM. Problem drink- San Francisco. Am J Public Health. 2006;96:677–684. 71–86. ing patterns among African Americans: the impacts of reports of discrimination, perceptions of prejudice, and 46. Mossakowski KN. Coping with perceived dis- 9. Wills TA, Shiffman S. Coping and Substance Abuse: “risky” coping strategies. J Health Soc Behav. 2003;44: crimination: does ethnic identity protect mental A Conceptual Framework. Orlando, Fla: Academic Press 408–425. health? J Health Soc Behav. 2003;44:318–331. Inc; 1985. 28. Bennett GG, Wolin KY, Robinson EL, Fowler S, 47.Kessler RC, McGonagle KA, Zhao S, et al. Life- 10.Sinha R. How does stress increase risk of drug Edwards CL. Perceived racial/ethnic harassment and time and 12-month prevalence of DSM-III–R psychiat- abuse and relapse? Psychopharmacology. 2001;158: tobacco use among African American young adults. ric disorders in the United States: results from the Na- 343–359. Am J Public Health. 2005;95:238–240. tional Comorbidity Study. Arch Gen Psychiatry. 1994; 51:8–19. 11. Caughy MO, O’Campo PJ, Muntaner C. Experi- 29. Guthrie BJ, Young AM, Williams DR, Boyd CJ, ences of racism among African American parents and Kintner EK. African American girls’ smoking habits 48. Composite International Diagnostic Interview. Ge- the mental health of their preschool-aged children. Am and day-to-day experiences with racial discrimination. neva, Switzerland: World Health Organization; 1990. J Public Health. 2004;94:2118–2124. Nurs Res. 2002;51:183–190. 49. Diagnostic and Statistical Manual of Mental Disor- 12. Gee GC. A multilevel analysis of the relationship 30. Landrine H, Klonoff EA. The Schedule of Racist ders, Revised Third Edition. Washington, DC: American between institutional racial discrimination and health Events: a measure of racial discrimination and a study Psychiatric Association; 1987. status. Am J Public Health. 2002;92:615–623. of its negative physical and mental health conse- 50. International Classification of Diseases, 10th Revi- 13. Harrell JP, Hall S, Taliaferro J. Physiological re- quences. J Black Psychol. 1996;22:144–168. sion. Geneva, Switzerland: World Health Organization; sponses to racism and discrimination: an assessment of 31. Whitbeck LB, Hoyt DR, McMorris BJ, Chen X, 19 92. the evidence. Am J Public Health. 2003;93:243–248. Stubben JD. Perceived discrimination and early sub- 51. Cottler LB, Grant BF, Balise B. Concordance of 14 . Krieger N, Sidney S. Racial discrimination and stance abuse among American Indian children. J Health DSM-IV alcohol and drug use criteria and diagnoses as blood pressure: the CARDIA study of young black and Soc Behav. 2001;42:405–424. measured by AUDADIS-ADR, CIDI and SCAN. Drug white adults. Am J Public Health. 1996;86:1370–1378. 32. Gibbons FX, Gerrard M, Cleveland MJ, Wills TA, Alcohol Depend. 1997;47:185–205. 15. LaVeist TA, Rolley NC, Daila C. Prevalence and Brody G. Perceived discrimination and substance use 52. Essed P. Understanding Everyday Racism: An Inter- patterns of discrimination among US health care con- in African American parents and their children: a disciplinary Theory. Newbury Park, Calif: Sage Publica- sumers. Int J Health Serv. 2003;33:331–344. panel study. J Pers Soc Psychol. 2004;86:517–529. tions; 1991.

May 2007, Vol 97, No. 5 | American Journal of Public Health Gee et al. | Peer Reviewed | Research and Practice | 939  RESEARCH AND PRACTICE 

53. Feagin JR. The continuing significance of race: an- health research on racism and health. Soc Sci Med. tiblack discrimination in public places. Am Sociol Rev. 2005;61:1576–1596. Behind the Mask 19 91;56:101–116. 69. Link BG, Phelen JC, Bresnahan M, Stueve A, How the World Survived SARS, 54. Williams DR, Yu Y, Jackson JS, Anderson NB. Ra- Pescosolido BA. Public conceptions of mental illness: the First Epidemic of the cial differences in physical and mental health: socioeco- labels, causes, dangerousness, and social distance. Am Twenty–First Century nomic status, stress and discrimination. J Health Psychol. J Public Health. 1999;89:1328–1333. 19 97;2:335–351. By Timothy J. Brookes, in collabora- 70. Meyer IH. Prejudice as stress: conceptual and 55. Williams DR, Spencer MC, Jackson JS. Race, measurement problems. Am J Public Health. 2003;93: tion with Omar A. Khan, MD, MHS. stress, and physical health: the role of group identity. 262–265. In: Contrada RJ, Ashmore RD, eds. Self, Social Identity, Foreword by David L. Heymann, MD, and Physical Health. New York, NY: Oxford University 71. Ghosh C. Healthy People 2010 and Asian Americans/ World Health Organization Press Inc; 1999:71–100. Pacific Islanders: defining a baseline of information. Am J Public Health. 2003;93:2093–2098. 56. Schulz A, Williams D, Israel B, et al. Unfair treatment, neighborhood effects, and mental health in the 72. Srinivasan S, Guillermo T. Toward improved Detroit metropolitan area. J Health Soc Behav. 2000; health: disaggregating Asian American and Native 41:314–332. Hawaiian/Pacific Islander data. Am J Public Health. 2000;90:1731–1734. 57. Phinney J. The Multigroup Ethnic Identity Mea- sure: a new scale for use with adolescents and young 73. LaVeist TA, Wallace JM. Health risk and inequita- adults from diverse groups. J Adolesc Res. 1992;7: ble distribution of liquor stores in African American 156–176. neighborhoods. Soc Sci Med. 2000;51:613–617. 58. Gong F, Gage SL, Tacata L. Helpseeking behavior 74. Link BG, Phelan J. Social conditions as fundamen- among Filipino Americans: a cultural analysis of face tal causes of disease. J Health Soc Behav. 1995;(extra and language. J Community Psychol. 2003;31: issue):80–94. 469–488. 75. Lillie-Blanton M, Anthony JC, Schuster CR. Prob- 59. Gong F, Takeuchi DT, Agbayani-Siewert P, ing the meaning of racial/ethnic group comparisons in Tacata L. Acculturation, psychological distress, and al- crack cocaine smoking. JAMA. 1993;269:993–997. cohol use: investigating the effects of ethnic identity and religiosity. In: Chun K, Organista PB, Marin G, eds. 76.Pearlin LI. The sociological study of stress. J Health Acculturation: Advances in Theory, Measurement, and Soc Behav. 1989;30:241–256. his exciting new book tells the story of Applied Research. Washington, DC: American Psycho- logical Association; 2003:189–206. the recent outbreak of SARS—Severe Acute Respiratory Syndrome, the first 60. Alonso J, Ferrer M, Bandek B, et al. Health-related T true epidemic of the 21st century. It follows quality of life associated with chronic conditions in the beginnings of this epidemic from the eight countries: results from the International Quality of Life Assessment (IQOLA) Project. Qual Life Res. point of view of the virus–hunters as well as 2004;13:283–289. of those most affected. ISBN 0-87553-046-X • Softcover • 2004 • 262 pages 61. Stewart AL, Greenfield S, Hays RD, et al. Func- $21.00 APHA Members • $25.00 Nonmembers tional status and well-being of patients with chronic conditions: results from the Medical Outcomes Study. American Public Health JAMA. 1989;262:907–913. Association 62. Ware JE. The status of health assessment. Annu 800 I Street, NW, Rev Public Health. 1994;16:327–354. Washington, DC 20001 www.apha.org 63. Yen IH, Ragland DR, Greiner BA, Fisher JM. Ra- cial discrimination and alcohol consumption in urban transit operators: findings from the San Francisco Muni ORDER TODAY! Health and Safety Study. Public Health Rep. 1999;114: American Public Health Association 448–458. Publication Sales 64. Noh S, Beiser M, Kaspar V, Hou F, Rummens J. Web: www.apha.org Perceived racial discrimination, depression, and coping: E-mail: [email protected] a study of Southeast Asian refugees in Canada. J Health Tel: 888-320-APHA Soc Behav. 1999;40:193–207. FAX: 888-361-APHA 65. Van Houtven CH, Voils CI, Oddone EZ, et al. Per- ceived discrimination and reported delay of pharmacy prescriptions and medical tests. J Gen Intern Med. 2005;20:578–583. 66.Kessler RC, Michelson KD, Williams DR. The prevalence, distribution and mental health correlates of perceived discrimination in the United States. J Health Soc Behav. 1999;40:208–230. 67.Williams DR, Willams-Morris R. Racism and mental health: the African American experience. Ethn Health. 2000;5:243–268. 68. Krieger N, Smith K, Naishadham D, Hartman C, Barbeau EM. Experiences of discrimination: validity and reliability of a self-report measure for population

940 | Research and Practice | Peer Reviewed | Gee et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

Differences in Frequency of Violence and Reported Injury Between Relationships With Reciprocal and Nonreciprocal Intimate Partner Violence

| Daniel J. Whitaker, PhD, Tadesse Haileyesus, MS, Monica Swahn, PhD, and Linda S. Saltzman, PhD

Prevention of violence between intimate part- Objectives. We sought to examine the prevalence of reciprocal (i.e., perpe- ners is an important public health goal. Na- trated by both partners) and nonreciprocal intimate partner violence and to de- tional estimates indicate that approximately termine whether reciprocity is related to violence frequency and injury. 25% of women report being victims of a part- Methods. We analyzed data on young US adults aged 18 to 28 years from the ner’s physical or sexual violence at some 2001 National Longitudinal Study of Adolescent Health, which contained infor- point in their life, and approximately 1.5 mil- mation about partner violence and injury reported by 11 370 respondents on lion women and 835000 men are physically 18761 heterosexual relationships. assaulted or raped by intimate partners in the Results. Almost 24% of all relationships had some violence, and half (49.7%) United States annually.1 Intimate partner vio- of those were reciprocally violent. In nonreciprocally violent relationships, women lence (IPV) is associated with a number of were the perpetrators in more than 70% of the cases. Reciprocity was associated negative psychological and physical health with more frequent violence among women (adjusted odds ratio [AOR]=2.3; 95% confidence interval [CI]=1.9, 2.8), but not men (AOR=1.26; 95% CI=0.9, 1.7). Re- consequences including posttraumatic stress garding injury, men were more likely to inflict injury than were women (AOR=1.3; disorder, depression, physical injury, repro- 95% CI=1.1, 1.5), and reciprocal intimate partner violence was associated with ductive health problems, irritable bowel syn- greater injury than was nonreciprocal intimate partner violence regardless of the 2–4 drome, and chronic pain. IPV costs approx- gender of the perpetrator (AOR=4.4; 95% CI=3.6, 5.5). imately $5.8 billion per year, which includes Conclusions. The context of the violence (reciprocal vs nonreciprocal) is a strong only direct medical and mental health costs predictor of reported injury. Prevention approaches that address the escalation and work productivity losses to victims.5 of partner violence may be needed to address reciprocal violence. (Am J Public The women’s movement brought initial at- Health. 2007;97:941–947. doi:10.2105/AJPH.2005.079020) tention to the problem of partner violence directed at women and to the need for funding et al. found that in about half of the cases, vi- was initiated by each partner at least 40% of to address that problem.6 Much of the initial olence was reciprocal.10 Similar results were the time.10 Additionally, studies of community research on IPV was conducted with severely found in the National Survey of Families and samples found that a relatively low percent- abused women and supported the assumption Households.8 Studies reviewed by Gray and age of women endorsed self-defense as a pri- that IPV is primarily perpetrated by men Foshee11 found that among violent adolescent mary motive for violence.13 ,14 These data sug- against women. Data is mounting, however, relationships, the percentage of relationships gest that self-defense cannot fully explain the that suggests that IPV is often perpetrated by in which there was reciprocal partner violence reciprocal violence phenomenon. both men and women against their partner.7,8,9 ranged from 45% to 72%. A recent meta- Little is known about reciprocal violence It is also becoming recognized that perpetra- analysis found that a woman’s perpetration of with regard to its context or severity. We tion of IPV by both partners within a relation- violence was the strongest predictor of her sought to examine the prevalence of recipro- ship is fairly common. This phenomenon has being a victim of partner violence.12 cal and nonreciprocal IPV in a large, nation- been described with terms such as mutual vi- Reciprocal partner violence does not ap- ally representative sample of young adults. olence, symmetrical violence, or reciprocal vio- pear to be only comprised of self-defensive We also sought to examine the seriousness of lence. Here we use the terms reciprocal and acts of violence. Several studies have found IPV in relationships with reciprocal versus nonreciprocal to indicate IPV that is perpe- that men and women initiate violence against nonreciprocal IPV using 2 indices: violence trated by both partners (reciprocal) or 1 part- an intimate partner at approximately the same frequency and injury occurrence. Family con- ner only (nonreciprocal) in a given relationship. rate. For example, Gray and Foshee11 specifi- flict theory,15 which asserts that IPV occurs Reciprocity of IPV does not necessarily mean cally asked adolescents about their initiation as a result of escalating conflicts, would pre- that the frequency or the severity of the vio- of violence and found that among the violent dict that reciprocal IPV should be more seri- lence is equal or similar between partners. relationships studied, 66% were characterized ous than nonreciprocal IPV because recipro- Several studies have found that much of by both partners initiating violence at least cal IPV would indicate that both partners partner violence is reciprocal. For example, in once. In the National Family Violence Survey, are engaging in the escalation of conflict. We their national studies of family violence, Straus both men and women reported that violence also examined gender as a predictor of the

May 2007, Vol 97, No. 5 | American Journal of Public Health Whitaker et al. | Peer Reviewed | Research and Practice | 941  RESEARCH AND PRACTICE 

seriousness of the violence. Gender is at the TABLE 1—Weighted Estimates of Sample Characteristics of Individuals and Relationships: Young forefront of feminist theories of partner vio- Adults Aged 18–28 Years, National Longitudinal Study of Adolescent Health, United States, 2001 lence16 and it has been consistently found that male perpetrators are more likely to in- Total Men Women Variable N=11370 n=5219 n=6151 flict injury than female perpetrators.7 Thus, we examined the gender main effect on the Sample Characteristics N=11370 seriousness of violence and the interaction Mean age, years (SD) 22.0 (1.7) 22.1 (1.7) 21.9 (1.7) between reciprocity and gender to under- Race/ethnicity,a no. (%) stand whether the reciprocity effect differed White 6370 (69.9) 2924 (69.6) 3446 (70.1) for men and women. Black 2283 (15.1) 971 (15.0) 1312 (15.3) Hispanic 1653 (10.7) 805 (11.0) 848 (10.3) METHODS Other 889 (4.3) 433 (4.4) 456 (4.3) Education,a no. (%) Participants Less than high school 1377 (14.4) 714 (16.1) 663 (12.8) All participants were part of the National High-school graduate 3649 (32.3) 1781 (34.4) 1868 (30.2) Longitudinal Study of Adolescent Health (Add Some college 4513 (38.8) 2011 (37.0) 2502 (40.5) Health), and participated in the third wave of College graduate 1825 (14.5) 708 (12.4) 1117 (16.5) data collection during 2001. Add Health used Mean number of relationships in 3.08 (2.7) 3.05 (2.7) 3.10 (2.6) a multistage stratified cluster design to iden- past 5 years (SD) tify a nationally representative sample of ado- Ever victimized by IPV, no. (%) 3046 (26.8) 1269 (24.8) 1777 (28.8) lescents (complete details regarding Add Ever perpetrated IPV, no. (%) 3121 (26.5) 884 (17.3) 2237 (35.5) Health are found elsewhere17 ). In 1995, Relationship Characteristics N=18761 18 924 adolescents in middle and high school Relationship lasted ≥3 months 16608 (89.3) 7233 (85.9) 9375 (92.5) (aged 12 to 21 years) participated in Wave I in duration,a no. (%) of Add Health’s in-home interview. Six years Sex in relationship,a no. (%) 16755 (89.8) 7522 (88.9) 9233 (90.6) later, 14322 participants, 77.4% of those Cohabitation type,a no. (%) who completed the Wave I survey (aged 18 to Never married or living together 11630 (62.7) 5535 (65.8) 6095 (59.7) 28 years at Wave III), completed the in-home Lived together but not married 4317 (24.5) 1874 (23.7) 2443 (25.4) survey of Wave III of the Add Health study. Married 2392 (12.8) 912 (10.5) 1480 (14.9) In other analyses, the Add Health study team Notes. SD=standard deviation; IPV=intimate partner violence. Percentages reflect weighted estimates of the distribution of determined that participant nonresponse for the variables for the US young adult population. Wave III had minimal impact on the sample’s aBecause of small amounts of missing data, the numbers do not sum to the full sample size. representativeness.18 Our analyses involve only the Wave III data and focus on the questions on intimate Among the 14322 participants, 2952 and relationships included in the current relationships. In 1 section of the Wave III in- were excluded either because they reported analyses. terview, participants were asked to report an no relationships (n = 2584) or only same-sex “inventory” of all their sexual or romantic relationships (n = 368), which left a subset of Measures relationships during the past 5 years (sexual 11 370 participants. These 11370 reported All relationship-level questions were asked and romantic relationships were not further on 18761 relationships that included part- separately for each relationship (e.g., respon- defined). Participants were asked a short se- ner violence data (4085 participants re- dents with 2 partners were asked each set ries of questions about each relationship (e.g., ported 1 relationship, 7182 reported 2, and of questions twice, once for each partner). To partner age and gender, relationship length, 103 reported 3 or more). In most cases (all assess perpetration of physical violence within marital status, sexual contact), and then spe- but 97) violence questions were asked of intimate relationships, respondents answered cific types of relationships (primarily impor- “important” relationships, with importance 2 questions (“How often in the past year have tant ones) were selected and more detailed defined by a preset algorithm that consid- you threatened your partner with violence, questions were developed to gather more in- ered factors such as marital status, recency, pushed or shoved him/her, or thrown some- formation. In all, the 14322 participants with and duration of relationship. (Additional thing at him/her that could hurt,” and “How sample weights for Wave III reported 38894 and detailed information on the relationship often in the past year have you slapped, hit, relationships. We analyzed the subset of these selection can be obtained from the Add or kicked your partner”) on the following relationships that were heterosexual relation- Health study team [http://www.cpc.unc.edu/ scale: 0=never, 1=once, 2=twice, 3=3–5 ships and that had data on violence toward projects/addhealth].) Table 1 shows descrip- times, 4=6–10 times, 5=11–20 times, and from the partner. tive information on the sample of participants 6=more than 20 times, 7=did not happen

942 | Research and Practice | Peer Reviewed | Whitaker et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

in the past year, but happened prior to that. or not (yes = codes 1–7; no = code 0) and Research Triangle Park, NC) to accommodate Two parallel questions assessed the partner’s conducted binary logistic regression. the complex sampling design and to provide perpetration of violence toward the respon- Each logistic regression model included rec- accurate standard errors for analyses. dent. Responses to the questions were highly iprocity (nonreciprocal vs reciprocal) and per- correlated (respondent’s perpetration, petrator gender (men vs women) as predictors, RESULTS r =0.65; partner’s perpetration, r =0.78) along with several control variables: respon- and were thus averaged to create indices of dent gender (men vs women), respondent Table 2 shows the proportion of all rela- IPV perpetration by the respondent and IPV race/ethnicity (White, Black, Hispanic, other), tionships with any IPV, the proportion of perpetration by the partner. Injuries from education (less than high school, high-school violent relationships with reciprocal and non- partner violence were assessed with a single graduate, some college, college graduate), rela- reciprocal IPV, and the proportion of relation- question for the perpetration of injuries upon tionship length (less than 3 months vs greater ships with nonreciprocal IPV with perpetra- the partner (“How often has partner had an than 3 months), and relationship type (ever tors who were men versus those who were injury, such as a sprain, bruise, or cut because married, ever lived together but not married, women. Proportions were reported for the of a fight with you”), and a parallel question never lived together nor married). Finally, to overall sample and by respondent gender. assessed the partner’s perpetration of injuries properly analyze the data, we configured data Overall, IPV was reported in 23.9% of rela- to the respondent. Analyses were conducted so that each potential perpetrator in a relation- tionships, with women reporting a greater at the relationship level with respondents pro- ship (i.e., the respondent and the partner) was proportion of violent relationships than men viding data about their own perpetration and considered a separate case. This was neces- (28.4% vs 19.3%; P<.01). Among violent their partners perpetration (data was not di- sary because comparisons of reciprocal IPV relationships, nearly half (49.7%) were char- rectly collected from partners and was there- with respect to violence frequency and injury acterized as reciprocally violent. Women re- fore not available). occurrence would be within-subject compar- ported a significantly greater proportion of vi- isons (i.e., they would be on the same line of olent relationships that were reciprocal versus Analytic Plan data), whereas comparisons of nonreciprocal nonreciprocal than did men (women=51.5%; To examine the prevalence of nonreciprocal IPV would be between-subject comparisons. men=46.9%; P<.03). Among relationships and reciprocal IPV, we first classified each re- Additionally, all analyses were weighted to with nonreciprocal violence, women were relationship as having either no IPV (neither the provide national estimates.19 Weights were ported to be the perpetrator in a majority of respondent nor the partner perpetrated vio- assigned to each participant on the basis of cases (70.7%), as reported by both women lence against the other) or any IPV (either the grade of education, gender, and race, and ac- (67.7%) and men (74.9%). To look at the respondent or the partner perpetrated violence cording to the sampling frame, which over- data another way, women reported both against the other). We classified relationships sampled specific groups of adolescents. Analy- greater victimization and perpetration of vio- with IPV as having either reciprocal IPV (both ses were conducted with SAS version 9.1 lence than did men (victimization=19.3% respondent and partner perpetrated violence (SAS Institute Inc, Cary, NC) and SUDAAN vs 16.4%, respectively; perpetration=24.8% against the other) or nonreciprocal IPV (either version 9 (Research Triangle Institute, vs 11.4%, respectively). In fact, women’s the respondent or the partner perpetrated against the other, but not both). Finally, we divided the relationships with nonreciprocal IPV TABLE 2—Weighted Estimates of Violence Occurrence for Reciprocally and Nonreciprocally into those that were perpetrated by men ver- Violent Relationships, Overall and by Gender: Young Adults Aged 18–28 Years, National sus those perpetrated by women. Longitudinal Study of Adolescent Health, United States, 2001 To examine the seriousness of IPV by rec- Overall, no. (%) Men, no. (%) Women, no. (%) iprocity (nonreciprocal vs reciprocal), we re- Variable N=18761 n=8531 n=10230 stricted the analyses to only those relationships with IPV and used logistic regression All relationships to model reports of violence frequency and Nonviolent 14152 (76.1) 6897 (80.7) 7255 (71.6) injury occurrence. For violence frequency, Violent 4609 (23.9) 1634 (19.3) 2975 (28.4) because responses were nonnormally Among violent relationships distributed and the response options were Reciprocal IPV 2270 (49.7) 738 (46.9) 1532 (51.5) not evenly spaced, we collapsed response Nonreciprocal IPV 2339 (50.3) 896 (53.1) 1443 (48.5) codes 1–6 into 3 ordinal categories of vio- Among cases with nonreciprocal IPV lence frequency (low = responses 1 or 2; Perpetrated by men 670 (29.3) 232 (25.1) 438 (32.3) medium = response 3; high = responses 4–6) Perpetrated by women 1669 (70.7) 664 (74.9) 1005 (67.7) and conducted ordinal logistic regression. Notes. IPV=intimate partner violence. Percentages reflect weighted estimates of the distribution of the variables for the US For injury occurrence, we coded whether vi- young adult population. olence perpetration had resulted in an injury

May 2007, Vol 97, No. 5 | American Journal of Public Health Whitaker et al. | Peer Reviewed | Research and Practice | 943  RESEARCH AND PRACTICE 

TABLE 3—Weighted Estimates of Violence Frequency and Injury Occurrence by Reciprocity adults, with about half of violent relation- Status and Perpetrator Gender: Young Adults Aged 18–28 Years, National Longitudinal ships being characterized by reciprocal vio- Study of Adolescent Health, United States, 2001 lence. More importantly, we found that violence was perpetrated more frequently Violence Frequency, no. (%) (by women only) and was more likely to re- Low Medium High Injury occurrence, sult in injury when it was reciprocal as op- Variable (n=4447) (n=1549) (n=760) no. (%) posed to nonreciprocal. Reciprocity Our findings that half of relationships with Nonreciprocal 1721 (73.3) 413 (18.5) 167 (8.2) 266 (11.6) violence could be characterized as reciprocally 8,9,11 Reciprocal 2726 (60.6) 1136 (25.6) 593 (13.8) 1271 (28.4) violent are consistent with prior studies. Perpetrator gender We were surprised to find, however, that Men against women 1883 (65.1) 623 (21.1) 368 (13.8) 850 (28.8) among relationships with nonreciprocal vio- Women against men 2564 (64.6) 926 (24.9) 392 (10.5) 687 (18.1) lence, women were the perpetrators in a ma- Gender by reciprocity jority of cases, regardless of participant gender. Men against women: nonreciprocal 457 (69.0) 119 (17.6) 76 (13.4) 136 (20.0) One possible explanation for this, assuming Men against women: reciprocal 1426 (64.0) 504 (22.1) 292 (13.9) 714 (31.4) that men and women are equally likely to initi- 20 Women against men: nonreciprocal 1264 (75.0) 294 (18.9) 91 (6.1) 130 (8.1) ate physical violence, is that men, who are Women against men: reciprocal 1300 (57.2) 632 (29.1) 301 (13.7) 557 (25.3) typically larger and stronger, are less likely to retaliate if struck first by their partner. Thus, Note. Percentages reflect weighted estimates of the distribution of the variables for the US young adult population. some men may be following the norm that “men shouldn’t hit women” when struck first by their partner. A different explanation is that greater perpetration of violence was reported (adjusted odds ratio [AOR]=1.19; P=.17), men are simply less willing to report hitting by both women (female perpetrators=24.8%, which indicated that the frequency of vio- their partner than are women.21 male perpetrators=19.2%) and by men lence perpetrated by men did not vary by rec- This explanation cannot account for the (female perpetrators=16.4%, male perpetra- iprocity. For perpetrators who were women, data, however, as both men and women re- tors=11.2%). IPVwas more frequent when perpetrated in ported a larger proportion of nonreciprocal Next we restricted the analyses to only vio- the context of reciprocal IPV versus nonrecip- violence perpetrated by women than by men. lent relationships and examined violence fre- rocal IPV (AOR=2.23; P<.001). In other One might be tempted to think that men who quency and reported injury occurrence as a words, women perpetrated IPV more fre- perpetrate violence in nonreciprocal relation- function of reciprocity and perpetrator gen- quently in the context of reciprocal violence ships are the traditional male “batterer.” How- der. Table 3 shows the percentages for vio- than in nonreciprocal violence. As can be ever, the data were not consistent with this lence frequency and injury occurrence by rec- seen Table 3, a greater percentage of women representation; women who were victims of iprocity and perpetrator gender. To analyze in reciprocally violent relationships perpe- nonreciprocal violence experienced less vio- the frequencies in Table 3, we conducted lo- trated medium and high levels of violence lence and a lower likelihood of injury than did gistic regression to examine the relationship (29.1% and 13.7%, respectively), than did women who were victims of violence in recip- between reciprocity and perpetrator gender women perpetrators in nonreciprocally violent rocally violent relationships. Some have sug- and the 2 indices of the seriousness of the vi- relationships (18.9% and 6.1%, respectively). gested that survey studies, such as this one, olence. For each dependent variable, the ini- For injury occurrence, both perpetrator gen- likely exclude the more severely abused tial model included the main effects of reci- der and reciprocity were significant predictors, women typically studied in clinical settings.22 procity and perpetrator gender along with the but the interaction was not significant. Injury Thus, our findings may represent 1 form of reciprocity by perpetrator gender interaction. was more likely when violence was perpe- partner violence—what Johnson23 has called If the interaction was not significant, it was trated by men than by women (men=28.8% common couple violence or situational violence— dropped from the model. If the interaction vs women=18.8%; AOR=1.30), and in that is likely to be found in broader popula- was significant, we computed the reciprocity relationships for which IPV was reciprocal ver- tion samples rather than in clinical samples. effect separately for perpetrators who were sus nonreciprocal (reciprocal=28.4% vs non- In analyses of reports of violence frequency men and those who were women. reciprocal=11.6 %; AOR=4.41). and injury occurrence, 2 clear findings Table 4 shows results of the logistic regres- emerged. First, perpetrators who were men sion models. For violence frequency, the main DISCUSSION were more likely to inflict an injury on a part- effects of perpetrator gender and reciprocity ner than were those who were women, regard- were both significant and there was a signifi- Our findings show that reciprocal vio- less of reciprocity status. This replicates find- cant interaction. For perpetrators who were lence was about as common as nonrecipro- ings in the literature at large that women are men, the reciprocity effect was nonsignificant cal violence in this national sample of young more likely to be injured by partner violence

944 | Research and Practice | Peer Reviewed | Whitaker et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

TABLE 4—Ordinal Logistic Regression Results for Violence Frequency and Binary Logistic cannot test this hypothesis using this study’s Regression Results for Injury Occurrence: Young Adults Aged 18–28 Years, National data. An escalation explanation is supported Longitudinal Study of Adolescent Health, United States, 2001 by longitudinal studies that show that violence between relationship partners tends to esca- Violence Frequency Injury Occurrence late over time from verbal abuse to physical Variable AORa (95% CI) AOR (95% CI) abuse26–28 and that victimization from vio- Reciprocityb lence is a strong predictor of perpetration of Reciprocal violence NAb 4.41 (3.56, 5.47)* violence.12 , 2 9 The escalation of negative, coer- Nonreciprocal violence NAb 1.0 cive interactions has been central to, and Perpetrator strongly supported in, Patterson’s30 work, Men against women NAb 1.30 (1.14, 1.48)* which describes family processes that support Women against men NAb 1.0 the development of aggression, and has been Perpetrator gender×reciprocity interaction P<.001 P=.13c suggested to play a role in dating violence.25 Men against women perpetration In such cases, it may be important to work Reciprocal violence 1.26 (0.92, 1.72) with both relationship partners to help them Nonreciprocal violence 1.0 understand when and how conflict escalates to Women against men perpetration violence and how to interrupt that process. In- Reciprocal violence 2.30 (1.88, 2.82)* tervention with violent couples has been ex- Nonreciprocal violence 1.0 tremely controversial but has recently been recognized as viable in some cases, such as when Notes. AOR=adjusted odds ratio; CI=confidence interval; NA=not applicable. Percentages reflect weighted estimates of the distribution of the variables for the US young adult population. there is low-to-moderate violence, when both a All odds ratios are adjusted for respondent gender, respondent race, respondent education, relationship length, and partners agree to counseling and wish to re- relationship type. main an intact couple, when violence is recipro- b Because of a significant reciprocity×perpetrator gender interaction, results are presented separately for perpetrators who were men and those who were women. cal, and when there are low levels of intimida- c The reciprocity×gender interaction was dropped from the model because it was not significant. tion, fear, and control.31–33 Couples counseling *P<.001. would not be appropriate for patterns of partner violence in which there is severe abuse, high levels of fear on the part of the victim, and than are men.1,7 Second, relationships with re- that it is critical to begin to study some of the control of one partner by the other. ciprocal violence resulted in more frequent relationship processes that contribute to recip- violence (by women only) and a greater likeli- rocal partner violence as those are most likely Limitations hood of injury caused by both male and fe- to result in injury. There are several limitations of this work. male perpetrators. Reciprocal violence was The first set centers around the measures of more dangerous for the victim, both men and Implications for Prevention and partner violence. All measures were assessed women, than was nonreciprocal violence. In Intervention using only participant reports about their own fact, men in relationships with reciprocal vio- The finding that IPV is more frequently perpetration of violence and that of their part- lence were reportedly injured more often perpetrated by women and is more likely to ners. The data are thus subject to all the bi- (25.2%) than were women in relationships result in injury when perpetrated in the con- ases and limitations inherent to this form of with nonreciprocal violence (20.0%); this is text of reciprocal IPV can best be understood data collection, such as recall bias, social de- important as violence perpetrated by women is in the context of a conflict-based theoretical sirability bias, and reporting bias. Regarding often seen as not serious.10 An important model, which suggests that conflict leads to reporting biases, there has been much discus- caveat to these findings is that we do not know increasingly coercive interactions that may spi- sion of whether there are differences in re- the extent or severity of the injuries reported, ral into violence.15 , 2 4,25 For example, suppose ported IPV by the gender of the reporter. A only that they were reported to have occurred. partner A shoves partner B and that partner B meta-analysis of the reliability of the conflict These findings highlight the importance of does not retaliate but instead storms out of the tactics scale concluded that there is evidence considering relationship violence in the con- house; the violence may end as nonreciprocal of underreporting by both genders, and that text of the relationship. Many authors have violence with no injury. If Partner B retaliates underreporting may be greater for men,34 for noted that research and prevention should by slapping or punching partner A, the vio- more severe acts of IPV.21 It would have been begin to shift away from the sole focus on lence then becomes reciprocal and injury be- ideal to collect violence data from both part- violence by men against women given the comes more likely with each escalating blow. ners, but those data were not collected from accumulation of data indicating that partner This pattern suggests that retaliation may be a the full Add Health sample. violence is perpetrated by both men and primary mechanism for the increased injury A second measurement issue pertains to the women.10 , 2 0 The data presented here suggest associated with reciprocal violence, though we scope of violence measures. The 3 questions

May 2007, Vol 97, No. 5 | American Journal of Public Health Whitaker et al. | Peer Reviewed | Research and Practice | 945  RESEARCH AND PRACTICE 

included in the Add Health study do not cap- of violence severity. Research, and prevention the National Violence Against Women Survey. Washing- ture all forms of violence that occur between and treatment approaches should begin to ex- ton, DC: US Dept of Justice; 1998. relationship partners, including many of the amine the specific context of partner violence 2. Campbell JC. Health consequences of intimate partner violence. Lancet. 2002;359:1331–1336. more severe forms of partner violence on the to improve prevention efforts. This includes 3. Coker AL, Davis KE, Arias I, et al. Physical and Conflict Tactics Scale (e.g., used a knife or understanding the distal and immediate mental health effects of intimate partner violence for gun, choked, or burned). Questions about causes and motives that lead to partner vio- men and women. Am J Prev Med. 2002;23:260–268. emotional, verbal, psychological, or sexual ag- lence. Many authors have noted that there 4. Plichta SB. Intimate partner violence and physical gression were also not included. Similarly, are many forms of partner violence22 and dif- health consequences: policy and practice implications. J Interpers Violence. 2004;11:1296–1323. only a single item assessed injury to victims ferent types of perpetrators who are violent and it focused on injury frequency and ex- for different reasons.35,36 Research is needed 5. National Center for Injury Prevention and Con- trol. Cost of Intimate Partner Violence Against Women in cluded injury severity and whether medical that uses both representative samples and the United States. Atlanta, Ga: Centers for Disease Con- attention was needed or sought. Thus, it is samples of victims and perpetrators from clin- trol and Prevention; 2003. unclear whether the data presented here ical settings to fully understand the range and 6. Dobash RE, Dobash RP. Violence Against Wives. would be similar had the violence and injury scope of partner violence. New York, NY: Free Press; 1979. assessment been more thorough or if differ- 7. Archer J. Sex differences in aggression between heterosexual partners: a meta-analytic review. Psychol ent forms of violence had been measured and About the Authors Bull. 2000;126:651–680. analyzed separately. Perhaps more important At the time of this study, Daniel J. Whitaker and Linda S. 8. Brush LD. Violent acts and injurious outcomes in than the limited measures of violence and in- Saltzman were with the Division of Violence Prevention, married couples: methodological issues in the National jury is the fact that no data were collected National Center for Injury Prevention and Control, Centers Survey of Families and Households. Gender Society. for Disease Control and Prevention, Atlanta, Ga. Tadesse 1990;4:56–67. about the causes or function of violence. Such Haileyesus is with the Office of Statistics and Program- 9. Straus MA, Gelles RJ. How violent are American data are needed to understand why relation- ming, National Center for Injury Prevention and Control. families? Estimates from the National Family Violence ships with reciprocal violence are more vio- Monica Swahn is with the Office on Smoking and Health, Resurvey and other studies. In: Straus MA, Gelles RJ, Centers for Disease Control and Prevention. lent and more likely to result in injury. We eds. Physical Violence in American Families: Risk Factors Requests for reprints should be sent to Daniel Whitaker, and Adaptations to Violence in 8,145 Families. New speculated that retaliation may lead to esca- Centers for Disease Control and Prevention, 4770 Buford Brunswick, NJ: Transaction Publishers; 1995:95–112. lating violence and injury, but data are Highway, NE, MS K-60, Atlanta, GA 30341 (e-mail: [email protected]). 10. Straus MA. Women’s violence toward men is a needed to examine this hypothesis. Future This article was accepted May 25, 2006. serious social problem. In: Gelles RJ, Loseke DR, eds. studies should focus on the causes and con- Note. The findings and conclusions in this report are Current Controversies on Family Violence. 2nd ed. New- bury Park, Calif: Sage; 2004:55–77. text of reciprocal and nonreciprocal IPV. those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention. 11.Gray HM, Foshee V. Adolescent dating violence: Another limitation is that the Add Health differences between one-sided and mutually violent study obtained partner violence data prima- Contributors profiles. J Interpers Violence. 19 97;12:126–141. rily about relationships considered to be im- D. J. Whitaker designed the research, conducted 12. Stith SM, Smith DB, Penn CE, Ward DB, Tritt D. portant as defined by the Add research team. analyses, and was responsible for writing the article. Intimate partner physical abuse perpetration and victimization risk factors: a meta-analytic review. Aggress Thus, it is not clear how this selection bias T. Haileyesus was responsible for conducting the analyses. All authors contributed to the design of the re- Violent Behav. 2004;10:65–98. may have impacted the findings—that is, search and to the writing of the article. 13.DeKeseredy WS, Schwartz MD. Woman Abuse on whether the findings would be the same with Campus: Results From the Canadian National Survey. a fuller sample of relationships. However, our Acknowledgments Thousand Oaks, Calif: Sage; 1998. findings are consistent with previous research This research uses data from the National Longitudinal 14 .Follingstad DR, Wright S, Lloyd S, Sebastian JA. on other samples that have shown reciprocal Study of Adolescent Health, a program project designed Sex differences in motivations and effects in dating by J. Richard Udry, Peter S. Bearman, and Kathleen Mul- violence. Fam Relat. 19 91;40:51–57. partner violence is fairly common with ado- lan Harris, and funded by grant P01-HD31921 from the 15. Straus MA, Gelles RJ. Violence in American fami- 11 8,9 lescents and with broader populations. National Institute of Child Health and Human Develop- lies: how much is there and why does it occur? In: Finally, as noted, the data collected were part ment, with cooperative funding from 17 other agencies. Nunnally EW, Chilman CS, Fox FM, eds. Troubled Rela- This article was completed after the unexpected tionship. Newbury Park, Calif: Sage; 1988;141–162. of a nationally representative sample selected death of Linda Saltzman in 2005. 16. Yllo KA. Through a feminist lens: gender, diver- when participants were in middle and high The authors would like to thank the following in- sity, and violence: extending the feminist framework. dividuals for their helpful comments on this article: school. The use of a nationally representative In: Loseke DR, Gelles RJ, Cavanaugh MM, eds. Current Kathleen Basile, Linda Dahlberg, Ileana Arias, Lee Controversies on Family Violence. 2nd ed. Thousand sample greatly increases the generalizability Annest, John R. Lutzker, and Brenda Le. of the findings, but this particular sample is of Oaks, Calif: Sage; 2004:19–34. limited range in age (18–28 years) and likely 17. Harris KM, Florey F, Tabor J, Bearman PS, Jones J, Human Participant Protection Udry JR. The National Longitudinal Study of Adolescent does not include the most severely abused This research was reviewed and approved by the insti- Health: Research Design. Available at: http://www.cpc. victims who are subjected to extreme control tutional review board of the Centers for Disease Con- unc.edu/projects/addhealth/design. Accessed Septem- trol and Prevention. by their partners and may be unable or un- ber 15, 2005. willing to participate in research.22 18. Chantala K, Kalsbeek WD, Andraca E. Non-response References in Wave III of the Add Health Study. Chapel Hill, North This study indicates that reciprocity of 1. Tjaden P, Thoennes N. Prevalence, Incidence, and Carolina: Carolina Population Center, University of partner violence is an important correlate Consequences of Violence Against Women: Findings From North Carolina; 2004. Available at: http://www.cpc.

946 | Research and Practice | Peer Reviewed | Whitaker et al. American Journal of Public Health | May 2007, Vol 97, No. 5  RESEARCH AND PRACTICE 

unc.edu/projects/addhealth/files/W3nonres.pdf. 35. Gottman J, Jacobson N. When Men Batter Women: Accessed March 22, 2007. New Insights Into Ending Abusive Relationships. New York, NY: Simon and Schuster; 1998. 19. Chantala K, Tabor J. Strategies to Perform a Design- Based Analysis Using the Add Health Data. Chapel Hill, 36. Holtzworth-Munroe A, Stuart GL. Typologies of North Carolina: Carolina Population Center, University male batterers: three subtypes and the differences of North Carolina; 1999. Available at: http://www.cpc. among them. Psychol Bull. 1994;116:476–497. unc.edu/projects/addhealth/files/weight1.pdf. Accessed September 15, 2005. 20. Dutton DG, Nicholls TL. The gender paradigm in domestic violence research and theory: Part 1—The conflict of theory and data. Aggress Violent Behav. 2005;10:680–714. 21.Archer J. Sex differences in physically aggressive acts between heterosexual partners: a meta-analytic review. Aggress Violent Behav. 2002;7:313–351. 22. Johnson MP. Patriarchal terrorism and common couple violence: two forms of violence against women. J Marriage Fam. 19 95;57:283–294. 23. Johnson MP, Leone JM. The differential effects of Public Health and intimate terrorism and situational couple violence: findings from the National Violence Against Women Sur- vey. J Fam Issues. 2005;26:322–349. Podiatric Medicine 24. Riggs DS, O’Leary KD. Aggression between heterosexual dating partners: an examination of a causal model of courtship aggression. J Interpers Violence. Principles and Practice, 2nd Edition 1996;11:519–540. Arthur E. Helfand, DPM, Ed. 25. Wekerle C, Wolfe DA. Dating violence in mid- adolescence: theory, significance, and emerging preven- This new 2nd edition has been developed as a text for practitioners, tion initiatives. Clin Psychol Rev. 19 99;19:435–456. students, and individuals in both the podiatric and public health profes- 26. Murphy CM, O’Leary KD. Psychological aggres- sions. It is intended to serve as a model for the future, so that the public sion predicts physical aggression in early marriage. can be assured of their foot health and that as society ages, patients can J Consult Clin Psychol. 1989;57:579–582. retain their mobility, minimize complications related to chronic diseases, 27. O’Leary KD, Malone J, Tyree A. Physical aggres- and retain their health and vigor late in life. This presentation has been sion in early marriage: prerelationship and relationship authored by a group of professionals with a long history of involvement effects. J Consult Clin Psychol. 1994;62:594–602. in the professions of podiatric medicine and in public health. The health 28. Schumacher JA, Leonard KE. Husbands’ and care delivery system is ever changing. There is a need to focus concern wives’ marital adjustment, verbal aggression, and physi- in the future quality of podiatric medical care, the delivery of care and cal aggression as longitudinal predictors of physical availability of podiatric care for generations to come. aggression in early marriage. J Consult Clin Psychol. 2005;73:28–37. ORDER TODAY! American Public Health Association 29. Bookwala J, Frieze IH, Smith C, Ryan K. Predic- PUBLICATION SALES tors of dating violence: a multivariate analysis. Violence IBSN 0-87553-071-0 WEB: www.apha.org Vict. 19 92;7:297–311. 608 pages • softcover • 2006 E-MAIL: [email protected] 30.Patterson GR. The early development of coercive $39.19 APHA Members (plus s&h) TEL: 888-320-APHA family process. In: Reid JB, Patterson GR, eds. Antiso- $55.95 Nonmembers (plus s&h) cial Behavior in Children and Adolescents: A Develop- FAX: 888-361-APHA mental Analysis and Model for Intervention. Washington, DC: American Psychological Association; 2002: 25–44. 31. Bograd M, Mederos F. Battering and couples therapy: universal screening and selection of treatment modality. J Marital Fam Ther. 1999;25:291–312. 32. Stith SM, Rosen KH, McCollum EE, Thomsen CJ. Treating intimate partner violence within intact couple relationships: outcomes of multi-couple versus individual couple therapy. J Marital Fam Ther. 2004;30: 305–318. 33. O’Leary KD. Conjoint therapy for partners who engage in physically aggressive behavior: rationale and research. J Aggression Maltreat Trauma. 2002;5: 145–164. 34.Archer JA. Assessment of the reliability of the Conflict Tactics Scales: a meta-analytic review. J Inter- pers Violence. 19 99;14:1263–1289.

May 2007, Vol 97, No. 5 | American Journal of Public Health Whitaker et al. | Peer Reviewed | Research and Practice | 947 MARKETPLACE Annual Meeting and Exposition Dates and Sites:

2007 • November 3–7 • Washington, DC 2008 • October 25–29 • San Diego, CA 2009 • November 7–11 • Philadelphia, PA 2010 • November 6-10 • Denver, CO

Disability in Local and NEW IN PAPERBACK Global Worlds Disease and Democracy BENEDICTE INGSTAD AND The Industrialized World Faces AIDS SUSAN REYNOLDS WHYTE, EDITORS PETER BALDWIN Leading scholars explore global changes “A historical masterpiece!” in disability awareness, technology, and —Lawrence O. Gostin, author of policy from the viewpoint of disabled Public Health Law people and their families in a wide range California/Milbank Books on Health and the Public of local contexts. $24.95 paperback $55.00 hardcover, $21.95 paperback Medicare Matters What Geriatric Medicine Can Teach When Bodies American Health Care CHRISTINE K. CASSEL Remember The Strange Case of UPDATED EDITION WITH A NEW PREFACE Experiences and Politics of AIDS the Broad Street Pump “A concise and thoughtful discussion of in South Africa Medicare with much to offer policy makers, John Snow and clinicians, and other citizens.” DIDIER FASSIN the Mystery of Cholera “This is a remarkable book. As Fassin —Journal of the American Medical Association SANDRA HEMPEL California/Milbank Books on Health and the Public dissects the deadly powers of today, “An excellent account both of the $16.95 paperback he also unrelentingly looks for human disease and of the man whose courage alternatives to turn the AIDS tragedy and commitment belatedly put an end around.” —JoãoBiehl, author of Vita Congratulations to Alexandra Minna Stern, Disability in Local and Global Worlds California Series in Public Anthropology to one of the recurrent nightmares of author of Eugenic Nation: Faults and $55.00 hardcover, $21.95 paperback the mid-19th century.” —The Spectator Frontiers of Better Breeding in Modern $24.95 hardcover America, winner of the Viseltear Book Award, American Public Health Association. At bookstores or order (800) 822-6657 • www.ucpress.edu from Photograph

948 | Marketplace American Journal of Public Health | May 2007, Vol 97, No. 5 The 2007 Summer Program in Applied Biostatistical & Epidemiological Methods July 9th through 20th, 2007 AT THE OHIO STATE UNIVERSITY, COLUMBUS, OH

The 2007 Summer Program in Applied Biostatistical & Epidemiological Methods consists of 12 week long courses in the application of statistical methodology & epidemiology to a wide range of biomedical and public health problems. Our faculty is internationally known for both their biostatistical expertise and their teaching ability. The program brings participants up-to-date on new approaches to the analysis of data from epidemiologic studies as well as the analysis of biologic, clinical trials, and laboratory data. All courses are designed to provide valuable hands-on experience in the analysis and interpretation of real data.

The goal of the Practice-Based Epidemiology Series (PBE) is to provide practicing epidemiologists and public health researchers with general and specific epidemiology training that emphasizes methods and techniques that can be directly applied to better public health performance at the state and local health department levels. The eight courses in the PBE Series aim to apply teaching examples and real-world scenarios to core epidemiology principles and practice techniques.

2007 Course Offerings

Week 1 (July 9 – 13, 2007) Week 2 (July 16 – 20, 2007)

Applied Logistic Regression Applied Survival Analysis • Stanley Lemeshow • David W. Hosmer, Jr. Basic Biostatistics for Epidemiologists Environmental Epidemiology • J. Jackson Barnette • Harvey Checkoway Current and Emerging Infectious Disease Epidemiology Epidemiology of Maternal and Child Health • Gregory C. Gray • William M. Sappenfield Epidemiology Data Sources, Uses and Analysis Software Intermediate Epidemiology Using ActivEpi • Siran M. Koroukian • David G. Kleinbaum Geographic Information Systems (GIS) Meta-Analysis • Ellen K. Cromley • William R. Shadish Human-Made and Natural Disaster Epidemiology Public Health Field Epidemiology • Eric K. Noji • Richard C. Dicker

*Core courses necessary to be Practice-based Epidemiology certified by The Ohio State University Practice-Based Epidemiology Series

For more information, please email us at [email protected] or visit our web site at www.biostatistics.osu.edu

American Journal of Public Health | May 2007, Vol 100, No. 4 Marketplace | 949 JOURNAL OF PUBLIC HEALTH MANAGEMENT AND PRACTICE ISSUE FOCUS ON: Public Health Finance Guest Editors: Peggy A. Honoré, DHA & Brian W. Amy, MD, MHA, MPH

his March/April 2007 issue of the Journal of Public Health Management and Practice provides a comprehensive T collection of papers that substantially increases the body of knowledge on public health finance. Topics, covering a broad range of issues such as theories and definitions, funding formulas and financing levels, financial indicators and performance measurement, and budgeting are relevant for policymaking, research, educating, and application to practice. Distinguished contributing authors include U.S. Senator Edward M. Kennedy; Garth Graham, MD, MPH, HHS Assistant Deputy Secretary for Minority Health; Kevin U. Stephens, Sr, MD, JD, New Orleans Health Department Health Officer; and many others. The publication was developed with support from The Robert Wood Johnson Foundation that also sponsored a journal release event at the National Press Club and a national public health finance meeting as a means of sharing information from this significant publication. APHA SPECIAL OFFER: Offer 1) Receive a special rate of $79.95 (includes s&h, mention code D7K755AB) to the bi-monthly publication Journal of Public Health Management & Practice (price includes print and online for individual subscribers with paid orders). or Offer 2) Receive just the special Finance issue of Journal of Public Health Management & Practice for $21.95 (includes s&h, mention code D7K755SC)

SIMPLY: Mail your check with this ad to: JPHMP,Lippincott Williams & Wilkins, PO Box 1600, Hagerstown, MD 21741. You can also order by visiting LWW.com or calling 1-800-638-3030. When ordering, please be sure to mention the promotion code above to receive this exclusive member offer.

950 | Marketplace American Journal of Public Health | May 2007, Vol 97, No. 5 CHILDREN’S HOSPITAL COLUMBUS, OHIO Injury Research Faculty Position JOB OPPORTUNITIES Available CENTER FOR INJURY Qualitative Researcher Support: Help for Qualitative Researchers and Doctoral Students RESEARCH AND POLICY in All Disciplines: We help you by coding your he Center for Injury Research and project based upon your research design, or we Policy, located in the Columbus train you to use leading qualitative software. T Children’s Research Institute of Private training by web conference or in our Columbus Children’s Hospital, is seeking California office. Customized for individuals applicants for a tenure track faculty position in injury research. The Center is ex- and small teams. DataSense, LLC, email: panding its research in the areas of trau- [email protected] www.datasense.org, matic brain injury, impact biomechanics, Ph:(661) 821-1909. EOE. motor vehicle occupant injury, and injury surveillance and epidemiology. Faculty members will have an appointment in the Department of Pediatrics, The Ohio State University College of Medicine. Joint appointments with other departments and colleges within the university are easily arranged. Applicants should have a doctoral degree in the medical, public health, or related field, and a track record in research productivity. Applicants will be considered at the Assistant, Associate, or full Professor levels. Applications from mid-career researchers are strongly encouraged. Salary and benefits are very competitive and are based on experience and academic rank. An attractive startup package will be tailored to the faculty member’s needs. The startup package includes sufficient funding to purchase equipment and support research staff, postdoctoral fellows, and/or graduate students. Children’s Hospital and the Ohio State University are affirmative action/equal opportunity employers. Qualified women, minorities, Vietnam-era veterans, disabled veterans, and individuals with disabilities are encouraged to apply.

Individuals interested in applying for a position should forward a letter of application and CV to: Gary A. Smith, MD, DrPH, Director Center for Injury Research and Policy, Columbus Children’s Research Institute, Children’s Hospital 700 Children’s Drive Columbus, Ohio 43205 Telephone (614)722-2400 FAX (614)722-2448 e-mail: [email protected].

May 2007, Vol 97, No. 5 | American Journal of Public Health Job Opportunities | 951 952 | Job Opportunities American Journal of Public Health | May 2007, Vol 97, No. 5 Associate/Full Professor he Department of Epidemiology and Community Health (DECH), emerging School of Public Health, School of Medicine, Virginia Commonwealth University (VCU) invites applications for tenure-track and tenured faculty positions in T any sub-discipline of epidemiology. Candidates whose research is in chronic diseases, infectious diseases, epidemiologic methods, genetic/molecular epidemiology, maternal and child health, nutrition, injuries, or occupational and environmental epidemiology are particularly encouraged to apply. Requirements include a doctoral degree in epidemiology or a related field, demonstrated ability to develop and maintain an independent research program and experience in teaching at the graduate level. Rank and salary will be determined according to the candidate’s credentials. For appointment to full professor, applicants must have a strong record of funded research, maintain an extramurally-supported independent research program, and regularly publish research results in peer-reviewed journals. VCU is developing a School of Public Health within which DECH is a core department. Now is an exciting time to join the Department and to play a role in the evolution of the first School of Public Health in Virginia. More than in many schools, successful candidates can help shape the direction of the developing School of Public Health. The department offers M.P.H., Ph.D. and several dual degree programs. The Department is an integral part of one of the country’s most comprehensive health sciences centers — the Schools of Medicine, Pharmacy, Nursing, Dentistry, and Allied Health Professions, as well as the Massey Cancer Center are all part of the Medical College of Virginia Campus. Located in Richmond, Virginia’s state capital, the Department offers excellent opportunities for interaction and collaboration with academic colleagues from a variety of health- related fields, practitioners, and policy makers. Applications will be reviewed as received, and the positions will remain open until filled. Send curriculum vitae, a letter high- lighting qualifications and interests, and the names and addresses of three references to: Dr. Tilahun Adera, Chair; Department of Epidemiology and Community Health, Virginia Commonwealth University, P. O. Box 980212, Richmond, VA 23298-0212, phone: 804.828.9785, email: [email protected].

Virginia Commonwealth University is an equal opportunity/affirmative action employer. Women, minorities, and persons with disabilities are encouraged to apply.

EOE

May 2007, Vol 97, No. 5 | American Journal of Public Health Job Opportunities | 953 Want to reach your targeted audience INSTANTLY? Banners are a quick way to promote: • Programs • Conferences • Seminars • Workshops • Call for Papers

GO TO: www.apha.org/publications/advertising For details or email [email protected]

Protect, Prevent, Live Well EOE

954 | Job Opportunities American Journal of Public Health | May 2007, Vol 97, No. 5 Assistant Professor of Public Health Assistant or Associate Professor Psychiatry Dept. of Psychiatry, University of Illinois at Chicago Cardiovascular Diseases Epidemiologist is seeking non-tenure Assistant The George Washington University Professor with a background in both public health and nursing to adminis- School of Public Health and Health Services tratively direct the departmental Quality Management Program and School of Medicine and Health Sciences serve as the clinical director of the First Episode Psychosis Program. This position involves a 50% departmental he Department of Epidemiology and Biostatistics of the GWU School of QA administrative appointments and Public Health and Health Services, in collaboration with the Division of 50- administrative/clinical/research T Cardiology, Department of Medicine, GWU School of Medicine and Health appointment within the Psychosis Sciences, is recruiting for a dynamic, full-time faculty member at the rank of Assistant Program. Responsibilities include de- or Associate Professor. Rank, salary and employment in either a tenure or non-tenure veloping and maintaining QA initia- track position will be commensurate with experience. tives for the department while also serving as the liaison to state organiza- The successful applicant will have a joint appointment in the School of Public tions. Additionally, responsibilities in Health and Health Services and the School of Medicine and Health Sciences. This the First Episode Psychosis Program new faculty member will receive salary support from the Department of include administrative oversight, pro- Epidemiology and Biostatistics for research, teaching and mentoring activities; and re- gram development, and ceive salary support and be eligible for research funds from the GWU Division of research. Applicants should have Ph.D. Cardiology Cheney Cardiovascular Institute. degree. The Department of Epidemiology and Biostatistics has expertise in HIV/AIDS, For fullest consideration send resume cancer, behavioral, and aging epidemiology, racial disparities, geographical health and cover letter by May 23, 2007 to: information systems, and biostatistical methods, and a strong affiliation with the GWU Ena Casas, UIC Dept. of Psychiatry, 1601 Biostatistics Center renowned for its leadership in the design and coordination of large W. Taylor (M/C 912), Chicago, IL 60612. multi-center clinical trials in cardiovascular disease, diabetes, and maternal/fetal UIC is an AA/EOE medicine. The Division of Cardiology has participated in a broad range of cardiovascular research including single- and multi-center clinical trials (Women’s Health Initiative, acute and chronic coronary disease, heart failure, and cardiac arrhythmia studies), and has collaborated with the GWU Biostatistics Center on their coordinating center activities.

Basic Qualifications: Applicants must have either a doctoral degree in epidemiology or in medicine with advanced training and experience in epidemiology, strong verbal and written communication skills, and a record of peer-reviewed publications, especially on topics related to cardiovascular disease. Preferred Qualifications: Applicants should have demonstrated success for securing ex- ternally-funded research grants and have evidence of teaching and mentoring experience in graduate-level epidemiology. Responsibilities will include maintaining an active research program in the epidemiology of AJPH WEBSITE cardiovascular diseases by developing collaborative relationships between Public Health and Cardiovascular Institute clinical researchers at GWU; teaching graduate-level courses in epidemiologic methods and in topical areas of expertise; and ad- The online Journal joins the many vising masters and doctoral-level students. Review of applications will begin on June 1, 2007 and will continue until the position is filled. Applications from women and minorities online benefits offered to APHA are strongly encouraged. members at www.apha.org, including: Application Procedure: To be considered, interested applicants should submit the follow- • full issues of The Nation’s Health ing documents electronically: 1) a curriculum vitae; 2) a statement of research interests and accomplishments and plans for maintaining or developing an independently-funded re- •APHA Membership Directory search program; and 3) a statement of teaching and mentoring experience and description and of the courses you are interested in teaching, to: Search Committee, Cardiovascular Diseases Epidemiology • Annual Meeting information c/o Rachel Talbot, Department Manager Department of Epidemiology and Biostatistics School of Public Health and Health Services The George Washington University [email protected] (please submit all materials electronically to this e-mail address) Only complete applications will be considered. Additional information about the SPHHS and SMHS can be found at http://www.gwumc.edu/sphhs/ and Check out the Journal at http://www.gwumc.edu/smhs/. www.ajph.org The George Washington University is an Equal Opportunity/Affirmative Action Employer.

May 2007, Vol 97, No. 5 | American Journal of Public Health Job Opportunities | 955 SCHOOL OF

WEST LAFAYETTE, INDIANA PUBLIC HEALTH The Department of Health Management and Policy at the School of Public Health at Drexel University invites applications for a tenure- Applications are being accepted track position as Associate or Full Professor in Health Management and Policy. for the Fall Semester class Masters Successful candidates must hold a doctoral degree in public health (Ph.D., Dr. P.H.,Sc. D., M. D.), medicine or health-policy relevant field, in Public Health (M.P.H.) and have a history of productive scholarship. Must also show evidence of successful track record in teaching and supervising students CONCENTRATIONS IN: and developing independent funded research. This position will play Community Health & Health Communication a key role in developing, overseeing and teaching in a new doctoral program in Health Policy and Social Justice. We are looking for indi- Visit the Purdue University Graduate School Webpage viduals with expertise and interest in one or more of the following areas: health policy and its relationship to health disparities; social to apply www.gradschool.purdue.edu and economic inequality and access to health care for the poor and Select: Department of Health & Kinesiology underserved; health and social policy analysis; using health services research methods to understand race, class, culture and ethnicity in For information contact: health and illness. Applicants should send a curriculum vitae and cover letter that includes a GC HYNER description of research interests to: [email protected] John A. Rich, MD, MPH Department of Health & Kinesiology Chair, Health Management and Policy Purdue University Drexel University School of Public Health 1505 Race Street, MS 660 West Lafayette, Indiana 47907-2046 Philadelphia, PA 19102-1192

EOE Drexel University is an Equal Opportunity/Affirmative Action Employer.

Assistant/Associate/Full Professor of Pharmacy Administration University of New Mexico College of Pharmacy

The University of New Mexico College of Pharmacy invites applications for a calendar-year, Pharmacy Administration faculty position at the Assistant, Associate or Full Professor rank. esponsibilities of this position include developing or participating in an extramurally funded research program, participating n the continuing development of a self-sustaining and nationally recognized graduate program, and providing progressive Rinstruction in both the professional and graduate program. The nature of this position will require strong evidence for either potential or accomplishment in the pharmacy administration arena (pharmacoeconomics, health policy and health services/health outcomes research). Outstanding collaborative opportunities exist with the UNM Health Sciences Center, the VA Cooperative Studies Program, the Center for Pharmacoeconomic and Outcomes Research at the Lovelace Respiratory Research Institute, and the Lovelace Clinic Foundation. Additional collaborative opportunities exist with co-appointments in the MPH program and the new Robert Wood Johnson Center for Health Policy at the University of New Mexico.

Minimum requirements and desirable qualifications are listed by eligible rank at the following website: http://hsc.unm.edu/hr/. The salary for this position and the benefits package are competitive. Additional information about the University of New Mexico Health Sciences Center College of Pharmacy is available at the following website: http://hsc.unm.edu/. Applications will be accepted until the position is filled. If interested, please submit a signed letter of intent describing career goals, curriculum vitae, and names and addresses of at least three references to: Patricia L. Marshik, Pharm.D., Associate Professor, College of Pharmacy, MSC09 5360, 1 University of New Mexico, Albuquerque, NM 87131-0001; TELEPHONE: (505) 272-0579; E-MAIL: [email protected].

THE UNIVERSITY OF NEW MEXICO IS AN EQUAL OPPORTUNITY AND AFFIRMATIVE ACTION EMPLOYER AND EDUCATOR

956 | Job Opportunities American Journal of Public Health | May 2007, Vol 97, No. 5 ASSISTANT/ASSOCIATE PROFESSORS Stress Medicine, Health and Productivity Research

Join a newly established international team addressing key health challenges in the global economy. The Division of Occupational and Environmental Medicine (DOEM), Department of Family Medicine, within the School of Medicine, Wayne State University is dedicated to the promotion of occupational and environmental health and the prevention of disease through research, education and service. The DOEM has secured multi-year funding to expand its research into the areas of: 1: Stress Medicine including basic sciences as well as individual and organizational interventions; 2: Health and productivity and its link to customer perception and quality, and; 3: Wireless technologies and their impact on health and working life We are looking to fill three positions at the Assistant/Associate Professor level. Tenure track positions may be available for those candidates with appropriate records. Candidates should have a background in occupational and environmental health, physiology, psychology or other relevant area. Commensurate with prior experience, candidates will be expected to secure external funding and to serve as co-investigators on multidisciplinary research teams focusing on DOEM’s prioritized areas. Teaching opportunities will predominantly be at the graduate level. Requirements for all positions include a doctoral degree in a relevant field as well as a track record of publishing original observations in peer-reviewed journals. Candidates should have a proven record and/or a strong promise of obtaining external funding. All candidates should provide a statement of interest, curriculum vitae (including a list of publications and funded grants, teaching and community service experience) and a list containing the names, addresses, phone numbers and e-mail of three references. Positions open until filled. Wayne State University is an equal opportunity/affirmative action employer. Applications should send a letter, curriculum vitae and statement of interest to: Bengt Arnetz, M.D., Ph.D., M.P.H.,M.Sci.Epi., Professor and Director—Division of Occupational and Environmental Medicine, C/O Vickie Muhammad, Wayne State University, Department of Family Medicine, 101 E. Alexandrine, Room 223, Detroit, MI 48201; [email protected]

May 2007, Vol 97, No. 5 | American Journal of Public Health Job Opportunities | 957