SYSTEMATIC REVIEW

Hispanic Mortality Paradox: A Systematic Review and Meta-Analysis of the Longitudinal Literature

To investigate the possi- John M. Ruiz, PhD, Patrick Steffen, PhD, and Timothy B. Smith, PhD bility of a mortality advantage, we conducted DESPITE A SIGNIFICANTLY a systematic review and which race and ethnicity are improved on previous reviews by meta-analysis of the pub- more disadvantaged risk factor assessed at study entry and par- using meta-analytic procedures fi lished longitudinal literature pro le, in the United ticipants are followed longitudi- that took into account the differ- reporting Hispanic individ- States often experience similar or nally to mortality. This literature ences in available studies re- uals’ mortality from any better health outcomes across has added a wealth of data for and garding sample size, participant cause compared with any a range of health and disease against a Hispanic mortality ad- characteristics, selection criteria, other race/ethnicity. We contexts compared with non- vantage, but has failed to clarify and outcomes. searched MEDLINE, PubMed, Hispanic Whites (NHWs), an epi- the overall relationship. A number EMBASE, HealthSTAR, and demiological phenomenon com- of factors impede consensus, in- METHODS PsycINFO for published lit- monly referred to as the “Hispanic cluding differences in sample size, erature from January 1990 paradox.” Among the most salient selection criteria, methodologies, Studies were identified through to July 2010. Across 58 studies (4 615 747 features of this advantage is evi- follow-up time, statistical report- 2 techniques. First, we conducted participants), Hispanic popu- dence that Hispanics appear to live ing, and outcomes (i.e., morbidity, extensive electronic database 1--- 3 fi lations had a 17.5% lower longer than NHWs. These speci c-cause mortality, all-cause searches from January 1990 to risk of mortality compared findings are largely based on na- mortality). In addition, at least 5 July 2010, using MEDLINE, with other racial groups tional cohort data, with mortality narrative literature reviews of the PubMed, EMBASE, HealthSTAR, (odds ratio = 0.825; P < .001; data from the US Vital Statistics associated data7---11 were published and PsycINFO. January 1990 was 95% confidence interval = System used in the numerator and in the last decade, asserting the used as the beginning search date 0.75, 0.91). The difference population counts from the US level of interest but failing to pro- because of methodological in mortality risk was greater Census used in the denominator, vide an empirical test (e.g., meta- changes in the use of the terms among older populations yielding a death rate statistic. The analysis) to clarify the discrepancy. such as Hispanic in race and eth- and varied by preexisting classic explanations for these par- Hence, the current status of the nicity data collection and publica- health conditions, with ef- fi 12,13 fects apparent for initially adoxical ndings suggest that ei- Hispanic mortality paradox can tion efforts. To capture the fi healthy samples and those ther the denominator is arti cially best be described as one of great broadest possible sample of rele- fi with cardiovascular diseases. low because of Hispanics return- interest with signi cant logistical vant articles, 3 search term cate- The results also differed by ing to their countries of origin confusion. gories were used: (1) Hispanic racial group: Hispanics had before death (the “salmon bias We systematically reviewed the (Hispanic, Latino, Mexican, lower overall risk of mortality hypotheses”) or that the numera- longitudinal literature, comparing Puerto Rican, Cuban), (2) mortal- than did non-Hispanic Whites tor is not representative due to Hispanic mortality rates with those ity (mortality, death, longevity, and non-Hispanic Blacks, but the healthiest Hispanics migrating of other racial/ethnic groups and survival, life span), and (3) design overallhigherriskofmortality to the (the “healthy conducted a meta-analysis of the (prospective, longitudinal). Sec- than did Asian Americans. migrant hypothesis”). These available data as a definitive test ond, we manually examined the These findings provided hypotheses have been largely of whether there is a relative reference sections of past reviews strong evidence of a His- 4 panic mortality advantage, refuted. The contemporary Hispanic mortality advantage. and of studies meeting the inclu- with implications for con- overarching concern is that the Resolving the validity of the phe- sion criteria to locate articles not fi ceptualizingandaddressing statistical estimation approach nomenon would facilitate future identi ed in the database racial/ethnic health dispar- remains flawed because of un- research efforts to identify con- searches. ities. (Am J Public Health. derreporting of ethnicity on death tributing resilience factors that Published online ahead of certificates. Despite recent data might lead to targeted interven- Inclusion Criteria print January 17, 2013: suggesting that the associated tions. In the present study, we We included only published e1–e9. doi:10.2105/AJPH. error is negligible,5,6 the validity focused on all-cause mortality studies meeting the following cri- 2012.301103) of the paradox remains in ques- (death from any cause) as the teria in the meta-analysis: (1) tion due to its strong ties to this primary dependent variable and written in English or Spanish, (2) methodology. evaluated mortality within specific used a longitudinal design, and (3) One solution to these issues is to disease contexts to the extent that provided quantitative data re- examine longitudinal studies in sufficient data were available. We garding Hispanic mortality at the

Published online ahead of print January 17, 2013 | American Journal of Public Health Ruiz et al. | Peer Reviewed | Systematic Review | e1 SYSTEMATIC REVIEW

individual level compared with reported within studies were of the article until consensus was infection, 7% with , 5% that of other racial/ethnic groups. transformed to the natural log ORs obtained. with renal disease, and the We excluded studies in which for analyses and then transformed Aggregate effect sizes were cal- remaining 20% with a variety of the outcome was not explicitly back to ORs for interpretation. culated using random effects conditions, including liver disease stated as mortality (e.g., combined When effect size data were models following confirmation of and dementia. Research reports outcomes of morbidity and mor- reported in any metric other than heterogeneity. A random effects most often (91%) considered all- tality), studies of , ORs or the natural log ORs, we approach yields results that gen- cause mortality, but some re- single-case designs, and reports transformed those values using eralize beyond the sample of stricted evaluations to mortality with exclusively aggregated data statistical software programs and studies actually reviewed.15 We associated with CVD (5%) or (e.g., census-level statistics). We macros (Comprehensive Meta- assumed that the results would other specific causes (4%). Only 8 included all other types of quanti- Analysis14). In many cases, we differ as a function of participant studies (14%) involved a medical tative research designs that were calculated effect sizes from fre- characteristics (i.e., age, gender) intervention21,24,25,33,35,45,62,73; longitudinal and yielded a statisti- quency data in matrixes of mor- and study design (i.e., length of most merely tracked participants’ cal estimate of the risk of mortality tality status by ethnicity. In cases follow-up). Random effects models mortality over time. Participants among Hispanic populations com- when frequency data were not take this between-studies variation were followed for an average of pared with that of other racial/ reported, we recovered the cell into account, whereas fixed effects 6.9 years (SD = 5.9; range =1 ethnic groups. There were no age probabilities from the reported models do not.16 month to 33 years). Preferred limitations other than those re- risk ratio and marginal probabili- Reporting Items for Systematic lated to studies of infant mortality. ties. Across studies, we assigned RESULTS Reviews and Meta-Analyses and However, the published literature OR values less than 1.00 to data Meta-analysis of Observational on mortality was largely skewed indicative of decreased mortality Figure 1 shows the study selec- Studies in guide- toward older ages, as reflected among Hispanics and OR values tion process. Statistically non- lines were adhered to in the design here. greater than 1.00 to data indica- redundant effect sizes were and reporting of this study.75,76 tive of increased mortality among extracted from 58 studies (Table Data Abstraction Hispanics relative to the compari- 1).17---74 Data were reported from Omnibus Analysis Articles were independently son group(s). 4 615 747 total participants, with Across the 58 studies, the ran- coded by 2 teams with 2 members When multiple effect sizes were an average composition of 26% dom effects weighted average each. A third independent mem- reported within a study at the Hispanic participants within stud- effect size was OR = 0.825 ber then compared the 2 ratings, same time, we averaged the values ies. The mean ages of participants (P < .001; 95% confidence inter- resolving discrepancies through (weighted by SE) to avoid violating at initial evaluation were 54.6 val [CI] = 0.75, 0.91). Consistent joint review with the teams. the assumption of independent years (SD =11.6) for Hispanics with the hypothesis, Hispanic Coders extracted several objec- samples. When a study contained and 56.1 years (SD =11.7) for ethnicity was associated with a tively verifiable characteristics of multiple effect sizes across time, comparison groups. Hispanic 17.5% mortality advantage. the studies: (1) the number of we extracted the data from the participants consisted of 44% As shown in Figure 2, ORs participants and their composition longest follow-up period. If a study women, and comparison groups ranged from 0.39 to 2.75, with by age, ethnicity, gender, and pre- used statistical controls in calcu- included 45% women. a very large degree of heteroge- existing health conditions (if any), lating an effect size, we extracted Research reports typically failed neity across studies (I2 = 96%; fi s2 as well as the cause of mortality; the data from the model utilizing the to describe the speci c ethnic Q(57) = 1564; P < .001; = (2) length of follow-up; and (3) fewest statistical controls. We coded heritage of the Hispanic partici- 0.12), suggesting that systematic research design. Given the sub- theresearchdesignusedratherthan pants (80% omitted this informa- effect size variability was unac- stantial heterogeneity among His- the estimate risk of individual study tion), but 8 studies (15%) were counted for. Thus, it was likely panic peoples exemplified by dif- bias. The coding protocol is avail- specific to Mexican Ameri- that factors associated with the ferences in culture, traditions, and able from the authors. cans,20,29,33,42,47,52,67,72 1 study studies themselves (e.g., publica- importantly, health outcomes, we Information obtained from the was specific to Puerto Rican tion status), participant character- further sought to code by country studies was extracted directly from Americans,48 and 5 studies istics (e.g., age, health status), and, of origin or nativity when such the reports. As a result, the inter- (9%) involved participants or the research design (e.g., length data were available. rater agreement was high for cat- from a variety of ethnic back- of follow-up) might have moder- Data within studies were often egorical variables (mean Cohen’s grounds.24,25,31,36,51 Several stud- ated the overall results. We reported in terms of odds ratios j = 0.97; SD = 0.02) and for con- ies (22%) involved initially therefore conducted additional (ORs), the likelihood of mortality tinuous variables (mean intraclass healthy participants, but 24% of analyses to determine the extent to contrasted by ethnic group. Be- correlation = 0.93; SD = 0.14). studies involved patients with which the variability in the effect cause OR values cannot be mean- Discrepancies across coders were cardiovascular disease (CVD), sizes was moderated by these ingfully aggregated, all effect sizes resolved through further scrutiny 12% with , 10% with HIV variables.

e2 | Systematic Review | Peer Reviewed | Ruiz et al. American Journal of Public Health | Published online ahead of print January 17, 2013 SYSTEMATIC REVIEW

5877 citations identified through database searches 4 citations identified through search of reference (January 1990–July 2010) lists

5618 excluded based on redundancy, review of title, review of abstract

263 potential citations identified for further 92 additional citations excluded review

113 excluded 22 lacked mortality data 31 lacked mortality data by race 171 full-text articles assessed for eligibility 14 had insufficient data to calculate ORs 5 were Hispanic only samples 7 lacked sufficient Hispanic sample size 5 lacked Hispanic mortality data 6 combined morbidity and mortality data 2 did not report race 2 reported combined minority data 19 duplicate studies

58 articles included in meta-analysis

Note. OR = odds ratio. FIGURE 1—Selection of articles for meta-analysis: 1990 to 2010.

Evaluation for Egger’s regression test79 and the region of nonsignificance if publica- Based on these 4 analyses, we Publication Bias alternative to that test recommen- tion bias was present. In this case, concluded that the data did not To assess the possibility of ded by Peters et al.,80 which is the data underrepresented studies reflect publication bias per se, publication bias,77 we conducted better suited to data in OR format. with relatively fewer participants but that they might represent 4 analyses. First, we calculated The results of these 2 analyses that demonstrated lower mortality aconservativeestimateofrisk Orwin’s fail-safe N,78 the theoret- failed to reach statistical significance rates among Hispanics. Finally, we for mortality among Hispanic ical number of unpublished stud- (P > .05). Third, we generated employed the “trim and fill” meth- populations. ies with effect sizes averaging zero a “funnel plot”81 of the studies’ log odology described by Duval and (no effect) that would need to be ORs by the SEs. The data obtained Tweedie.82,83 This analysis indi- Moderation by Participant and located to reduce the overall from this meta-analysis were not cated that when 14 estimated Study Characteristics magnitude of the results obtained symmetrically distributed around “missing” studies were included in To investigate whether the to a trivial estimate of 1.0 > OR > the grand mean; there appeared to the analysis, the overall effect size lower risk of mortality among 0.95. Based on this calculation, at be multiple studies “missing” from was calculated to be OR = 0.70 Hispanic populations varied as least 367 additional studies aver- the bottom left corner of the distri- (95% CI = 0.64, 0.77), indicating a function of participant charac- aging OR = 1.0 would need to be bution.However,thesestudieswere that Hispanic participants were teristics within studies, we con- found to render the results of the in the opposite corner from what 30% less likely to die than were ducted analyses involving partici- present meta-analysis as negligi- would have been expected. Typi- comparison group members over pants’ age, gender, and preexisting ble. Second, we utilized both cally, “missing” studies were in the thesametimeperiod. diagnoses. We also investigated

Published online ahead of print January 17, 2013 | American Journal of Public Health Ruiz et al. | Peer Reviewed | Systematic Review | e3 SYSTEMATIC REVIEW

any differences across studies that TABLE 1—Characteristics of Included Studies: 1990–2010 may be due to length of follow-up, Hispanic, Mean Age, Follow-Up, Health Status at type of research design, and cause Source Total, No. No. (%) Female, % Years Years Study Entry Analysis Category of mortality. To establish whether the aver- Alexander et al.17 90 316 9835 (11) 55 69 1 CVD CVD age age of the sample accounted Assassi et al.18 250 71 (28) 87 47 6 Scleroderma Other for significant between-studies Brogan et al.19 1027 31 (3) 35 35 < 1 Respiratory failure Other variance, the effect sizes from the Brown et al.20 327 125 (38) 38 37 5 HIV/AIDS HIV/AIDS 53 studies that reported partici- Bush et al.21 2486 92 (4) 40 65 5 CVD CVD pants’ average age at intake were Chen et al.22 281 100 (36) 19 59 3 Cancer Cancer correlated with the corresponding Cohen et al.23 15 610 2600 (17) 17 36 3 HIV/AIDS HIV/AIDS effect size for that study. The Cohen et al.24 27 788 734 (3) 26 59 < 1 CVD CVD resulting random effects weighted Cooper-Dehoff et al.25 22 576 8045 (36) 61 66 3 CVD CVD correlation was –0.28 (P = .03), Cromwell et al.26 692 574 9868 (1) NA > 65 1 CVD CVD indicating that studies with older Cunningham et al.27 200 36 (18) 5 38 6 HIV/AIDS HIV/AIDS populations tended to demon- Echols et al.28 7007 344 (5) 38 63 1 CVD CVD strate lower risk of mortality Eden et al.29 107 64 (60) 73 62 7 Other among Hispanic participants rela- Feinglass et al.30 25 568 3628 (14) 44 72 5 Extremity bypass Other tive to comparison groups. As Fernandez et al.31 396 220 (56) 86 35 10 Autoimmune Other a first step to verify that this Frankenfield et al.32 7723 994 (13) 46 59 1 Kidney disease Other association was specific to chro- Freedman et al.33 15 329 970 (6) 55 44 12 Cancer Cancer nological age, we investigated the Gomez et al.34 41 901 2061 (5) 50 > 65 7 Cancer Cancer possible confounding association Gortmaker et al.35 1028 358 (35) 50 7 4 HIV/AIDS HIV/AIDS with trends over time. However, Hartmann et al.36 980 483 (41) 50 66 5 Stroke Other when we correlated the effect sizes Harzke et al.37 1 238 317 311 082 (25) 0 28 5 None apparent None/community with the year of initial data col- Havranek et al.38 7495 1789 (24) 49 56 < 1 CVD CVD lection and with a variable created Helzner et al.39 323 179 (55) 70 87 4 Dementia Other by subtracting the average age of Henderson et al.40 71 798 41 665 (58) 52 63 6 None apparent None/community participants at the start of the Jokela et al.41 8544 1736 (20) 50 20 25 None apparent None/community study from the year of initial data Lee et al.42 446 312 (70) 61 > 60 8 None apparent None/community collection (an estimate of the av- Liao et al.43 696 697 52 725 (8) 53 38 9 None apparent None/community erage year of participant birth), the Lin et al.44 553 307 33 954 (6) 54 > 25 11 None apparent None/community resulting values of r = –0.08 and Mak et al.45 15 376 1613 (10) 34 64 3 CVD CVD r = 0.22 did not approach statisti- Manoharan et al.46 400 67 (17) 33 67 14 Cancer Cancer cal significance (P > .1). Thus, the Medina et al.47 584 236 (40) 60 62 4 Diabetes Other findings within studies did not Mendenhall et al.48 428 63 (15) 0 49 5 Liver Disease Other consistently change over time. Murthy et al.49 100 618 10 393 (10) 47 59 2 Kidney disease Other Because older populations were Ostir et al.50 506 153 (30) 51 81 5 None apparent None/community more likely to receive treatment Palmas et al.51 1178 451 (38) 55 72 7 Diabetes Other than were younger populations, Patel et al.52 66 397 1114 (2) 56 73 8 None apparent None/community we conducted a second analysis to Peralta et al.53 39 550 12 076 (31) 59 62 4 Kidney disease Other verify the association observed Perez E et al.54 312 91 (29) 46 58 20 Cancer Cancer with participant age by simulta- Perez M et al.55 44 171 2625 (6) 9 54 8 CVD CVD neously regressing participant age Plurad et al.56 3998 2495 (62) 18 33 7 Sepsis Other and the type of research study Robinson et al.57 6677 673 (10) 45 57 5 Kidney disease Other (intervention vs observation) on Sacco et al.58 394 82 (21) 51 63 1 Stroke Other study effect size. In this model, the Sacco et al.59 2670 1443 (54) 63 66 9 None apparent None/community average age of participants Schupf et al.60 2247 876 (39) 66 76 3 None apparent None/community remained statistically significant Segev et al.61 79 034 9846 (12) 59 39 6 None apparent None/community (b = –0.28, P = .04), but the type Serna et al.62 5122 413 (8) 41 NA 5 Cancer Cancer of research study (intervention Continued vs observation) did not. The dif- ferences observed in risk for

e4 | Systematic Review | Peer Reviewed | Ruiz et al. American Journal of Public Health | Published online ahead of print January 17, 2013 SYSTEMATIC REVIEW

When considered along with Continued TABLE 1— the consistent state and national Shaw et al.63 346 075 7823 (2) 47 61 < 1 CVD CVD vital statistics evidence, including Silverberg et al.64 4787 661 (14) 10 37 9 HIV/AIDS HIV/AIDS the recent Centers for Disease Smyth et al.65 581 323 (56) 0 25 33 Heroin addiction Other Control and Prevention report Stefanidis et al.66 408 296 (73) 44 54 16 Cancer Cancer clearly stating a Hispanic ethnicity 3 Steffen-Batey et al.67 406 196 (48) 41 59 7 CVD CVD mortality advantage, it might be Sudano et al.68 8400 723 (9) 52 56 6 None apparent None/community time to move beyond the question Swenson et al.69 1862 921 (49) 57 52 11 Diabetes Other of the existence of the Hispanic Tedaldi et al.70 1301 225 (17) 20 38 5 HIV/AIDS HIV/AIDS mortality paradox and onto inves- Waring et al.71 956 37 (4) 73 72 13 Dementia Other tigations into the causes of such Wei et al.72 3735 2630 (70) 59 43 8 None apparent None/community resilience. An important concep- Wolf et al.73 9303 979 (11) 44 61 1 Kidney disease Other tual consideration was that the Young et al.74 337 870 26 544 (8) 1 64 2 Diabetes Other observed mortality advantage, as well as the broader health out- Note. CVD = cardiovascular disease; NA = not available. come advantages evident in the , may reflect resilience at several points in the mortality appeared to be moder- advantage among individuals with non-Hispanics, a rate that was course of disease. Hispanics might ated by participant age. CVD and an estimated 16% ad- highly comparable to the 20% be less susceptible than some Similar random effects vantage among persons with a va- advantage reported by Arias et al.5 other races to illness in general or weighted correlations with the riety of other preexisting health using the alternative death statistic to specific conditions with high gender composition of each sam- conditions. However, Hispanics estimation strategy. The omnibus mortality rates, such as CVD. It ple (using percentage who were diagnosed with HIV/AIDS or finding in the present study was was also possible that the rate of female; r = –0.23) and the length cancer had a risk of mortality that moderated by age, such that the disease progression might be of time participants were followed did not significantly differ from effect became stronger among slower among Hispanics, resulting (r = 0.07) did not reach statistical non-Hispanics. older participants, a finding similar in lower morbidity and greater significance (P > .05). Further- Because studies compared His- to that which was recently longevity. Finally, the mortality more, no differences in the aver- panic participants with different reported using the estimation ap- advantage might reflect an advan- age effect sizes were found be- ethnic groups, we conducted proach.85 However, the date of tage in survival and recovery from tween studies using prospective a random effects weighted analysis data collection did not moderate acute clinical events (e.g., myocar- versus retrospective designs of variance across the several the effect, suggesting that the tra- dial infarction, stroke). Hence,

(Q1,57 = 0.1; P > .05). Studies comparisons conducted within jectory of this mortality effect did further research is needed to as- evaluating all-cause mortality had studies (such that each study con- not change (i.e., weaken) over certain whether the observed effect sizes of equivalent magni- tributed as many effect sizes as it time. The Hispanic mortality ad- Hispanic mortality advantage re- tude to those from the studies in had unique comparisons with dif- vantage varied as a function of flects advantages at specific points which a specific cause of death ferent ethnic groups84). As shown preexisting health status at study in the disease course and whether was evaluated (i.e., cancer; Q= in Table 3, there was a significant entry. Specifically, Hispanics dis- such time-point differences vary 0.3; P > .05). Thus, the omnibus difference across ethnicity (Q= played a significant mortality ad- by disease context. results presented earlier were not 6.5; P < .05). Hispanic participants vantage among studies of initially Several risk and resilience fac- moderated by these variables. were less likely to die over time healthy samples and in the context tors might contribute to these ef- As shown in Table 2, statisti- compared with both NHWs and of CVD and other health condi- fects, including potential biological cally significant differences were non-Hispanic Blacks (NHBs), but tions, such as renal disease. With (e.g., genetics, immune function- found across participants’ type they were more likely to die than respect to studies of persons with ing), behavioral (e.g., diet, smok- of health condition at the point were Asian Americans during the cancer and HIV/AIDS, Hispanics ing), psychological (e.g., stress, of initial evaluation (Q=11.5; same follow-up period. and non-Hispanics experienced personality), and social (e.g., ac- P = .02). Community samples of equivalent mortality risk. Findings culturation, social cohesion) dif- Hispanics with no identified health DISCUSSION also indicated that although His- ferences.86 Although not assessed impairment had the greatest mor- panics had a significant overall in the present study, lower so- tality advantage (estimated 30%) Results of this meta-analysis mortality advantage relative to cioeconomic status (SES) is a ro- relative to non-Hispanics. Hispa- showed that Hispanic ethnicity NHWs and NHBs, they were bust predictor of worse health nic ethnicity was also associated was associated with a 17.5% marginally disadvantaged relative outcomes.87 However, the pre- with a 25% reduced mortality lower mortality rate relative to to Asian Americans. sent findings challenged the

Published online ahead of print January 17, 2013 | American Journal of Public Health Ruiz et al. | Peer Reviewed | Systematic Review | e5 SYSTEMATIC REVIEW

Study Name Study Statistics OR (95% CI) the relationship between SES and Lower Upper disease risk among Hispanics, such OR z P limit limit -Value -Value that there is a buffering effect of Alexander et al.17 0.811 0.774 0.850 -8.708 < .001 Assassi et al.18 1.252 0.652 2.405 0.676 .499 SES associated with low levels of Brogan et al.19 2.282 1.040 5.008 2.057 .04 and a more tradi- Brown et al.20 1.000 0.622 1.607 0.000 > .999 Bush et al.21 0.992 0.617 1.594 -0.033 .974 tional SES gradient effect at higher Chen et al.22 0.834 0.489 1.421 -0.669 .503 acculturation levels.88 Accultura- Cohen et al.23 0.615 0.524 0.722 -5.927 < .001 Cohen et al.24 0.920 0.713 1.187 -0.638 .523 tion might be a proxy for social Cooper-Dehoff et al.25 0.629 0.564 0.702 -8.268 < .001 behaviors and cultural values that Cromwell et al.26 0.845 0.807 0.883 -7.348 < .001 Cunningham et al.27 2.670 1.103 6.462 2.177 .029 buffer against the stress of eco- Echols et al.28 0.743 0.478 1.155 -1.320 .187 nomic and environmental disad- Eden et al.29 0.849 0.374 1.926 -0.392 .695 Feinglass et al.30 0.880 0.730 1.060 -1.347 .178 vantages. It was also possible that Fernandez et al.31 0.961 0.443 2.084 -0.101 .919 the relative impact of traditional Frankenfield et al.32 0.951 0.804 1.126 -0.581 .561 Freedman et al.33 2.477 1.950 3.146 7.434 < .001 risk factors, such as diabetes and Gomez et al.34 0.997 0.902 1.102 -0.059 .953 lipids, differ by ethnicity and con- Gortmaker et al.35 0.706 0.439 1.137 -1.432 .152 Hartmann et al.36 0.564 0.418 0.762 -3.739 < .001 tribute to the observed paradox. Harzke et al.37 1.462 1.255 1.704 4.872 < .001 More research is needed to iden- Havranek et al.38 0.609 0.526 0.705 -6.613 < .001 Helzner et al.39 0.521 0.333 0.815 -2.860 .004 tify risk and resilience mechanisms Henderson et al.40 0.426 0.400 0.454 -26.656 < .001 as well as to understand poten- Jokela et al.41 0.930 0.622 1.389 -0.356 .722 Lee et al.42 0.763 0.478 1.219 -1.130 .259 tially complex interaction patterns Liao et al.43 0.548 0.518 0.579 -21.500 < .001 that may explain the observed Lin et al.44 0.418 0.391 0.446 -25.676 < .001 Mak et al.45 1.214 0.969 1.521 1.687 .092 effects. Manoharan et al.46 1.169 0.691 1.976 0.582 .561 The present study is a reminder Medina et al.47 1.131 0.766 1.670 0.618 .537 Mendenhall et al.48 2.751 1.534 4.934 3.396 .001 to physicians and researchers Murthy et al.49 0.915 0.869 0.963 -3.423 .001 about the heterogeneity in racial/ Ostir et al.50 1.039 0.699 1.543 0.188 .851 Palmas et al.51 0.630 0.473 0.839 -3.164 .002 ethnic minority health. Despite Patel et al.52 0.658 0.584 0.742 -6.852 < .001 similar risk factor profiles, His- Peralta et al.53 0.393 0.364 0.423 -24.605 < .001 54 panics had significantly lower all- Perez, E et al. 0.506 0.289 0.886 -2.385 .017 Perez, M et al.55 0.602 0.512 0.707 -6.195 < .001 cause mortality relative to NHBs. Plurad et al.56 0.786 0.655 0.943 -2.591 .010 fi 57 Such ndings support a need for Robinson et al. 0.965 0.810 1.148 -0.404 .686 Sacco et al.58 0.511 0.257 1.0171.912 - .056 Hispanic-specific comparative re- Sacco et al.59 0.428 0.334 0.548 -6.738 < .001 60 search to determine where such Schupf et al. 0.852 0.659 1.102 -1.221 .222 Segev et al.61 0.989 0.422 2.320 -0.025 .98 differences occur in specific dis- Serna et al.62 1.225 1.001 1.499 1.971 .049 63 ease courses and outcomes and to Shaw et al. 0.450 0.413 0.490 -18.410 < .001 Silverberg et al.64 0.590 0.430 0.811 -3.253 .001 investigate potential racial and Smyth et al.65 1.048 0.756 1.454 0.281 .778 66 ethnic differences in the relative Stefanidis et al. 1.779 1.147 2.759 2.571 .01 Steffen-Batey et al.67 0.822 0.528 1.280 -0.867 .386 weight or influence of identified 68 Sudano et al. 0.756 0.549 1.040 -1.718 .086 risk factors for disease. Given Swenson et al.69 0.862 0.675 1.102 -1.184 .236 Tedaldi et al.70 1.259 0.886 1.788 1.285 .199 evidence of Hispanic heteroge- 71 Waring et al. 0.691 0.342 1.393 -1.034 .301 neityinhealthoutcomes,sub- Wei et al.72 0.887 0.609 1.292 -0.625 .532 Wolf et al.73 0.854 0.703 1.037 -1.596 .11 group comparative research is 74 Young et al. 0.824 0.780 0.871 -6.893 < .001 also warranted. Combined 0.825 0.746 0.912 -3.767 < .001 0.1 0.2 0.5 1 2 5 10 Decreased Mortality Increased Mortality Limitations We could not entirely rule out the possibility of selection bias as Note. CI = confidence interval; OR = odds ratio. an alternative explanation for the findings. Although we made sig- FIGURE 2—Meta-analysis of Hispanic ethnicity and all-cause mortality: 1990–2010. nificant efforts to identify all rele- vant published studies, and data generalizability of this relationship possible that SES either does not moderated by some third variable. checks indicated no significant given the typically lower SES of contribute to risk among His- For example, emerging data sug- violations of publication distribu- Hispanics relative to NHWs. It is panics or confers risk only as gested that acculturation moderates tion, our results might yet reflect

e6 | Systematic Review | Peer Reviewed | Ruiz et al. American Journal of Public Health | Published online ahead of print January 17, 2013 SYSTEMATIC REVIEW

specific comparative studies in- respective institutional review boards, TABLE 2—Analyses of Weighted Average Effect Sizes Across volving Hispanics as opposed to approval was not sought given the nature of the study and its use of published, Type of Preexisting Health Condition: 1990–2010 fi generalizing ndings of Black- de-identified data. Type of Health Conditiona Studies, No. OR (95% CI) White differences. A next challenge is to identify factors that promote References None apparent (community samples) 13 0.70 (0.58, 0.85) resilience across the life span, and 1. Markides KS, Coreil J. The health of Cardiovascular disease 11 0.75 (0.61, 0.91) in turn, have the potential for Hispanics in the southwestern United States: an epidemiologic paradox. Public Cancer 7 1.21 (0.92, 1.59) j informing interventions for all. Health Rep. 1986;101(3):253---265. HIV/AIDS 6 0.86 (0.64, 1.17) 2. Sorlie PD, Backlund E, Johnson NJ, Other conditions 21 0.84 (0.72, 0.99) Rogot E. Mortality by Hispanic status in the United States. JAMA. 1993;270 Note. CI = confidence interval; OR = odds ratio, transformed from random effects weighted About the Authors John M. Ruiz is with the Department of (20):2464---2468. natural log OR; Qb = Q-value for variance between groups. a Psychology, University of North Texas, 3. Arias E. United States Life Tables by Qb = 115; P = .02. Denton. Patrick Steffen is with the Hispanic Origin. Vol 152. Hyattsville, Department of Psychology, Brigham Young MD: National Center for Health Statistics; University, Provo, UT. Timothy B. Smith is 2010. with the Department of Counseling 4. Abraído-Lanza AF, Dohrenwend BP, some degree of bias. For example, SES and health behaviors, which Psychology, Brigham Young University. Ng-Mak DS, Turner JB. The Latino mor- fl Correspondence should be sent to John M. limiting inclusion to only those were shown to in uence out- tality paradox: a test of the “salmon bias” 89 Ruiz, PhD, Department of Psychology, studies in English or Spanish comes. To these points, we and healthy migrant hypotheses. Am J University of North Texas, 1155 Union Public Health. 1999;89(10):1543---1548. might have resulted in a language would note that we did not exam- Circle #311280, Denton, TX 76203-5017 bias. The number of available ine unpublished manuscripts that (e-mail: [email protected]). Reprints can be 5. Arias E, Eschbach K, Schauman WS, Backlund EL, Sorlie PD. The Hispanic studies also limited our ability to could also affect findings. Finally, ordered at http://www.ajph.org by clicking the “Reprints” link. mortality advantage and ethnic misclassi- fi examine mortality in speci c con- the analyzed sample was predom- This article was accepted October 16, fication on US death certificates. Am J texts, including diabetes, autoim- inantly Mexican American, which 2012. Public Health. 2010;100(Suppl 1):S171--- S177. mune conditions, injury, neuro- likely limited generalizability Contributors 6. Smith DP, Bradshaw BS. Rethinking logic disorders, and others, as well across Hispanic subgroups, partic- the Hispanic paradox: death rates and life J. M. Ruiz had the idea for the project. as test effects of acculturation or ularly given evidence of significant expectancy for US non-Hispanic White J. M. Ruiz, P. Steffen, and T. B. Smith and Hispanic populations. Am J Public generational status. We were also heterogeneity in Hispanic sub- conceptualized the study design. J. M. Health. 2006;96(9):1686---1692. unable to address questions re- group mortality outcomes.90,91 Ruiz and P. Steffen oversaw the literature 7. Franzini L, Ribble JC, Keddie AM. garding whether the observed ef- review and article procurement. P. Understanding the Hispanic paradox. Conclusions Steffen and T. B. Smith oversaw data fect was constant or decreased extraction. T. B. Smith conducted the Ethn Dis. 2001;11(3):496---518. over time. Study availability might These findings should serve as statistical analyses. J. M. Ruiz, P. Steffen, 8. Markides KS, Eschbach K. Aging, also have limited our ability to a cornerstone to document a com- and T. B. Smith drafted the article. J. M. migration, and mortality: current status of Ruiz, P. Steffen, and T. B. Smith had full detect subtle effects, as in the parative Hispanic mortality ad- research on the Hispanic paradox. J Ger- access to all the data in the meta-analysis ontol B Psychol Sci Soc Sci. 2005;60(Spec context of cancer and HIV, where vantage in the context of a disad- and take responsibility for the integrity of No 2):S68---S75. observed effects might have been vantaged risk factor profile and to the data and the accuracy of the data analysis. 9. Lerman-Garber I, Villa AR, Caballero significant with a larger number demonstrate important heteroge- E. Diabetes and cardiovascular disease. Is there a true Hispanic paradox? Rev of studies. Lack of reporting also neity in racial/ethnic minority Acknowledgments fi Invest Clin. 2004;56(3):282---296. limited our ability to examine health. Furthermore, these nd- This research was supported by the 10. Morales LS, Lara M, Kington RS, Departments of Psychology at the Uni- several key moderators, including ings highlighted the need for Valdez RO, Escarce JJ. Socioeconomic, versity of North Texas and Brigham cultural, and behavioral factors affecting Young University. Hispanic health outcomes. J Health Care We would like to thank Erin Kauff- Poor Underserved. 2002;13(4):477---503. man, Courtney C. Prather, Lauren Smith TABLE 3—Odds of Survival by Race Compared With Hispanics: (University of North Texas) and Jameson 11. Palloni A, Morenoff JD. Interpreting VanDyke and Randy Gilliland (Brigham the paradoxical in the Hispanic paradox: 1990–2010 Young University) for their assistance in demographic and epidemiologic ap- proaches. Ann N Y Acad Sci. 2001;954: a the literature search and coding. We Race Studies, No. OR (95% CI) would also like to thank Linda Gallo (San 140---174. Diego State University), Karen Matthews Non-Hispanic Black 40 0.87 (0.76, 0.99) 12. Greico EM, Cassidy R. C. Overview of (University of Pittsburgh), and Bert Race and Hispanic Origin. Washington, Asian 9 1.19 (0.90, 1.56) Uchino (University of Utah) for their use- DC: US Census Bureau; 2001. ful comments on this article. Non-Hispanic White 53 0.81 (0.73, 0.91) 13. Gibson C, Jung K. Historical Census Statistics on Population Totals by Race, Note. CI = confidence interval; OR = odds ratio, transformed from random effects weighted Human Participant Protection 1790 to 1990, and by Hispanic Origin, natural log OR; Qb = Q-value for variance between groups. 1970 to 1990, for the United States Re- a P No protocol approval was necessary be- Qb = 6.5; = .04. cause data were obtained from secondary gions, Divisions, and States. Washington, sources. After consultation with our DC: US Census Bureau; 2002.

Published online ahead of print January 17, 2013 | American Journal of Public Health Ruiz et al. | Peer Reviewed | Systematic Review | e7 SYSTEMATIC REVIEW

14. Borenstein M, Hedges L, Rothstein 26. Cromwell J, McCall NT, Burton J, 37. Harzke AJ, Baillargeon J, Paar DP, 49. Murthy BV, Molony DA, Stack AG. H. Comprehensive Meta-analysis. Version Urato C. Race/ethnic disparities in utili- Pulvino J, Murray OJ. Chronic liver dis- Survival advantage of Hispanic patients 2. Englewood, NJ: Biostat; 2005. zation of lifesaving technologies by ease mortality among male prison inmates initiating dialysis in the United States is fi fi 15. Hedges LV, Vevea JL. Fixed- and Medicare ischemic heart disease bene - in Texas, 1989-2003. Am J Gastroenterol. modi ed by race. J Am Soc Nephrol. random-effects models in meta-analysis. ciaries. Med Care. 2005;43(4):330---337. 2009;104(6):1412---1419. 2005;16(3):782---790. Psychol Methods. 1998;3(4):486---504. 27. Cunningham WE, Mosen DM, 38. Havranek EP, Froshaug DB, 50. Ostir GV, Goodwin JS. High anxiety 16. Mosteller F, Colditz GA. Under- Morales LS, Andersen RM, Shapiro MF, Emserman CD, et al. Left ventricular is associated with an increased risk of standing research synthesis (meta-analy- Hays RD. Ethnic and racial differences in hypertrophy and cardiovascular mortality death in an older tri-ethnic population. sis). Annu Rev Public Health. 1996;17: long-term survival from hospitalization by race and ethnicity. Am J Med. 2008; J Clin Epidemiol. 2006;59(5):534---540. 1---23. for HIV infection. J Health Care Poor 121(10):870---875. 51. Palmas W, Pickering TG, Teresi J, Underserved. 2000;11(2):163---178. 17. Alexander M, Grumbach K, Remy L, 39. Helzner EP, Scarmeas N, Cosentino et al. Ambulatory blood pressure moni- Rowell R, Massie BM. Congestive heart 28. Echols MR, Mahaffey KW, Banerjee S, Tang MX, Schupf N, Stern Y. Survival toring and all-cause mortality in elderly failure hospitalizations and survival in A, et al. Racial differences among high- in Alzheimer disease: a multiethnic, people with diabetes mellitus. Hyperten- California: patterns according to race/ risk patients presenting with non--- population-based study of incident cases. sion. 2009;53(2):120---127. Neurology. 2008;71(19):1489---1495. ethnicity. Am Heart J. 1999;137(5): ST-segment elevation acute coronary 52. Patel KV, Eschbach K, Ray LA, 919---927. syndromes (results from the SYNERGY 40. Henderson SO, Bretsky P, Henderson Markides KS. Evaluation of mortality data trial). Am J Cardiol. 2007;99(3):315--- 18. Assassi S, Del Junco D, Sutter K, et al. BE, Stram DO. Risk factors for cardiovas- for older : implica- 321. Clinical and genetic factors predictive of cular and cerebrovascular death among tions for the Hispanic paradox. Am J mortality in early systemic sclerosis. Ar- 29. Eden SV, Meurer WJ, Sanchez BN, African Americans and Hispanics in Los Epidemiol. 2004;159(7):707---715. Angeles, California. Acad Emerg Med. thritis Rheum. 2009;61(10):1403---1411. et al. Gender and ethnic differences in 53. Peralta CA, Shlipak MG, Fan D, et al. subarachnoid hemorrhage. Neurology. 2001;8(12):1163---1172. 19. Brogan TV, Thiagarajan RR, Rycus Risks for end-stage renal disease, cardio- 2008;71(10):731---735. PT, Bartlett RH, Bratton SL. Extracorpo- 41. Jokela M, Elovainio M, Singh- vascular events, and death in Hispanic real membrane oxygenation in adults 30. Feinglass J, Sohn MW, Rodriguez H, Manoux A, Kivimaki MIQ. Socioeconomic versus non-Hispanic white adults with with severe respiratory failure: a multi- Martin GJ, Pearce WH. Perioperative status, and early death: the US National chronic kidney disease. J Am Soc Nephrol. center database. Intensive Care Med. outcomes and amputation-free survival Longitudinal Survey of Youth. Psychosom 2006;17(10):2892---2899. Med. 2009;71(3):322---328. 2009;35(12):2105---2114. after lower extremity bypass surgery in 54. Perez EA, Gutierrez JC, Moffat FL Jr, California hospitals, 1996-1999, with 20. Brown LS Jr, Siddiqui NS, Chu AF. 42. Lee DJ, Markides KS. Activity and et al. Retroperitoneal and truncal sarco- follow-up through 2004. J Vasc Surg. Natural history of HIV-1 infection and mortality among aged persons over an mas: prognosis depends upon type not 2009;50(4):776---783. predictors of survival in a cohort of HIV-1 eight-year period. J Gerontol. 1990;45(1): location. Ann Surg Oncol. 2007;14(3): seropositive injecting drug users. J Natl 31. Fernández M, Alarcon GS, McGwin S39---S42. 1114---1122. Med Assoc. 1996;88(1):37---42. G Jr, et al. Using the Short Form 6D, as an 43. Liao Y, McGee DL, Cooper RS. 55. Perez MV, Yaw TS, Myers J, overall measure of health, to predict 21. Bush D, Martin LW, Leman R, Mortality among US adult Asians and Froelicher VF. Prognostic value of the damage accrual and mortality in patients fi fi Chandler M, Haywood LJ. Atrial fibrilla- Paci c Islanders: ndings from the Na- computerized ECG in Hispanics. Clin with systemic lupus erythematosus: tion among African Americans, Hispanics tional Health Interview Surveys and the Cardiol. 2007;30(4):189---194. XLVII, results from a multiethnic US co- and Caucasians: clinical features and out- National Longitudinal Mortality Study. hort. Arthritis Rheum. 2007;57(6):986--- 56. Plurad DS, Lustenberger T, Kilday P, comes from the AFFIRM trial. J Natl Med Ethn Dis. 1999;9(3):423---433. 992. et al. The association of race and survival Assoc. 2006;98(3):330---339. 44. Lin CC, Rogot E, Johnson NJ, Sorlie from sepsis after injury. Am Surg.2010; 32. Frankenfield DL, Rocco MV, Roman 22. Chen LM, Li G, Reitzel LR, et al. PD, Arias E. A further study of life 76(1):43---47. SH, McClellan WM. Survival advantage Matched-pair analysis of race or ethnicity expectancy by socioeconomic factors in for adult Hispanic hemodialysis patients? 57. Robinson BM, Joffe MM, Pisoni RL, in outcomes of head and neck cancer the National Longitudinal Mortality Findings from the end-stage renal disease Port FK, Feldman HI. Revisiting survival patients receiving similar multidisciplin- Study. Ethn Dis. 2003;13(2):240---247. clinical performance measures project. J differences by race and ethnicity among ary care. Cancer Prev Res (Phila). 2009;2 45. Mak KH, Bhatt DL, Shao M, et al. Am Soc Nephrol. 2003;14(1):180---186. hemodialysis patients: the Dialysis Out- (9):782---791. Ethnic variation in adverse cardiovascu- comes and Practice Patterns Study. JAm 33. Freedman DM, Looker AC, Chang 23. Cohen CJ, Iwane MK, Palensky JB, lar outcomes and bleeding complications Soc Nephrol. 2006;17(10):2910---2918. SC, Graubard BI. Prospective study of in the Clopidogrel for High Athero- et al. A national HIV community cohort: 58. Sacco RL, Hauser WA, Mohr JP, serum vitamin D and cancer mortality in thrombotic Risk and Ischemic Stabiliza- design, baseline, and follow-up of the Foulkes MA. One-year outcome after the United States. J Natl Cancer Inst. tion, Management, and Avoidance (CHA- AmFAR Observational Database. Ameri- cerebral infarction in whites, blacks, and 2007;99(21):1594---1602. RISMA) study. Am Heart J. 2009;157 can Foundation for AIDS Research Hispanics. Stroke. 1991;22(3):305---311. Community-Based Clinical Trials Net- 34. Gomez SL, O’Malley CD, Stroup A, (4):658---665. 59. Sacco RL, Khatri M, Rundek T, et al. work. J Clin Epidemiol. 1998;51(9): Shema SJ, Satariano WA. Longitudinal, 46. Manoharan M, Ayyathurai R, de Los Improving global vascular risk prediction 779---793. population-based study of racial/ethnic Santos R, Nieder AM, Soloway MS. Pre- differences in colorectal cancer survival: with behavioral and anthropometric fac- 24. Cohen MG, Granger CB, Ohman EM, sentation and outcome following radical impact of neighborhood socioeconomic tors. The multiethnic NOMAS (Northern et al. Outcome of Hispanic patients cystectomy in Hispanics with bladder status, treatment and comorbidity. BMC Manhattan Cohort Study). J Am Coll Car- treated with thrombolytic therapy for cancer. Int Braz J Urol. 2008;34(6):691--- Cancer diol. 2009;54(24):2303---2311. acute myocardial infarction: results from . 2007;7:193. 698, discussion 698. 60. Schupf N, Costa R, Luchsinger J, the GUSTO-I and III trials. Global Utili- 35. Gortmaker SL, Hughes M, Cervia J, 47. Medina RA, Pugh JA, Monterrosa A, zation of Streptokinase and TPA for Oc- et al. Effect of combination therapy in- Cornell J. Minority advantage in diabetic Tang MX, Lee JH, Mayeux R. Relationship cluded Coronary Arteries. J Am Coll Car- cluding protease inhibitors on mortality end-stage renal disease survival on he- between plasma lipids and all-cause mor- diol. 1999;34(6):1729---1737. among children and adolescents infected modialysis: due to different proportions of tality in nondemented elderly. J Am Ger- iatr Soc. 2005;53(2):219---226. 25. Cooper-DeHoff RM, Aranda JM Jr, with HIV-1. N Engl J Med. 2001;345 diabetic type? Am J Kidney Dis. 1996;28 Gaxiola E, et al. Blood pressure control (21):1522---1528. (2):226---234. 61. Segev DL, Muzaale AD, Caffo BS, and cardiovascular outcomes in high-risk 36. Hartmann A, Rundek T, Mast H, 48. Mendenhall CL, Gartside PS, Roselle et al. Perioperative mortality and long- Hispanic patients---findings from the In- et al. Mortality and causes of death after GA, Grossman CJ, Weesner RE, Chedid A. term survival following live kidney dona- ternational Verapamil SR/Trandolapril first ischemic stroke: the Northern Man- Longevity among ethnic groups in alco- tion. JAMA. 2010;303(10):959---966. Study (INVEST). Am Heart J. 2006;151 hattan Stroke Study. Neurology. 2001;57 holic liver disease. Alcohol Alcohol. 1989; 62. Serna DS, Lee SJ, Zhang MJ, et al. (5):1072---1079. (11):2000---2005. 24(1):11---19. Trends in survival rates after allogeneic

e8 | Systematic Review | Peer Reviewed | Ruiz et al. American Journal of Public Health | Published online ahead of print January 17, 2013 SYSTEMATIC REVIEW

hematopoietic stem-cell transplantation Am Soc Nephrol. 2008;19(7):1379--- NJ: Lawrence Erlbaum Associates; for acute and chronic leukemia by eth- 1388. 2012:539---568. nicity in the United States and Canada. J 74. Young BA, Maynard C, Boyko EJ. 88. Gallo LC, de Los Monteros KE, Clin Oncol. 2003;21(20):3754---3760. Racial differences in diabetic nephropa- Allison M, et al. Do socioeconomic gradi- 63. Shaw LJ, Shaw RE, Merz CN, et al. thy, cardiovascular disease, and mortality ents in subclinical atherosclerosis vary Impact of ethnicity and gender differences in a national population of veterans. Di- according to acculturation level? Analyses on angiographic coronary artery disease abetes Care. 2003;26(8):2392---2399. of Mexican-Americans in the Multi-Ethnic prevalence and in-hospital mortality in Study of Atherosclerosis. Psychosom Med. 75. Moher D, Liberati A, Tetzlaff J, the American College of Cardiology- 2009;71(7):756---762. Altman DG. Preferred reporting items for National Cardiovascular Data Registry. systematic reviews and meta-analyses: the 89. Blue L, Fenelon A. Explaining low Circulation. 2008;117(14):1787---1801. PRISMA statement. J Clin Epidemiol. mortality among US immigrants relative 64. Silverberg MJ, Leyden W, 2009;62(10):1006---1012. to native-born Americans: the role of Quesenberry CP Jr, Horberg MA. Race/ . Int J Epidemiol. 2011;40 76. Stroup DF, Berlin JA, Morton SC, ethnicity and risk of AIDS and death (3):786---793. et al. Meta-analysis of observational stud- among HIV-infected patients with access ies in epidemiology: a proposal for 90. Heron M, Hoyert DL, Murphy SL, to care. J Gen Intern Med. 2009;24(9): Xu J, Kochanek KD, Tejada-Vera B. reporting. Meta-analysis Of Observational 1065---1072. Deaths: final data for 2006. Natl Vital Stat Studies in Epidemiology (MOOSE) group. Rep. 2009;57(14):1---134. 65. Smyth B, Hoffman V, Fan J, Hser YI. JAMA. 2000;283(15):2008---2012. Years of potential life lost among heroin 91. Borrell LN, Crawford ND. All-cause 77. Rosenthal R. The file drawer prob- addicts 33 years after treatment. Prev mortality among Hispanics in the United lem and tolerance for null results. Psychol Med. 2007;44(4):369---374. States: exploring heterogeneity by nativity Bull. 1979;86(3):638---641. 66. Stefanidis D, Pollock BH, Miranda J, status, country of origin, and race in the et al. Colorectal cancer in Hispanics: 78. Orwin RG. A fail-safe N for effect National Health Interview Survey-linked a population at risk for earlier onset, size in meta-analysis. J Educ Stat. 1983;8 Mortality Files. Ann Epidemiol. 2009;19 advanced disease, and decreased survival. (2):157---159. (5):336---343. Am J Clin Oncol. 2006;29(2):123---126. 79. Egger M, Davey Smith G, Schneider 67. Steffen-Batey L, Nichaman MZ, Goff M, Minder C. Bias in meta-analysis DC Jr, et al. Change in level of physical detected by a simple, graphical test. BMJ. activity and risk of all-cause mortality or 1997;315(7109):629---634. reinfarction: The Corpus Christi Heart 80. Peters JL, Sutton AJ, Jones DR, Project. Circulation. 2000;102(18): Abrams KR, Rushton L. Comparison of 2204---2209. two methods to detect publication bias in 68. Sudano JJ, Baker DW. Explaining US meta-analysis. JAMA. 2006;295(6):676--- racial/ethnic disparities in health declines 680. and mortality in late middle age: the roles 81. Begg CB. Publication bias. In: of , health behaviors, Cooper H, Hedges LV, eds. The Handbook and health insurance. Soc Sci Med. 2006; of Research Synthesis. New York: Russell 62(4):909---922. Sage Foundation; 1994:399---409. 69. Swenson CJ, Trepka MJ, Rewers MJ, 82. Duval S, Tweedie R. Trim and fill: Scarbro S, Hiatt WR, Hamman RF. Car- a simple funnel-plot-based method of diovascular disease mortality in Hispanics testing and adjusting for publication bias and non-Hispanic whites. Am J Epidemiol. in meta-analysis. Biometrics. 2000;56 2002;156(10):919---928. (2):455---463. 70. Tedaldi EM, Absalon J, Thomas AJ, 83. Duval S, Tweedie R. A non- Shlay JC, van den Berg-Wolf M. Ethnicity, parametric “trim and fill” method of ac- race, and gender. Differences in serious counting for publication bias in meta- adverse events among participants in analysis. J Am Stat Assoc. 2000;95 an antiretroviral initiation trial: results (449):89---98. of CPCRA 058 (FIRST Study). J Acquir Immune DeficSyndr. 2008;47(4): 84. Cooper HM, Lindsay JJ. Research 441---448. synthesis and meta-analysis. In: Bickman L, Rog DJ, eds. Handbook of Applied Social 71. Waring SC, Doody RS, Pavlik VN, Research Methods. Thousand Oaks: Sage Massman PJ, Chan W. Survival among Publications, Inc.; 1998:315---337. patients with dementia from a large multi-ethnic population. Alzheimer Dis 85. Borrell LN, Lancet EA. Race/eth- Assoc Disord. 2005;19(4):178---183. nicity and all-cause mortality in US adults: revisiting the Hispanic paradox. Am J 72. Wei M, Valdez RA, Mitchell BD, Public Health. 2012;102(5):836---843. Haffner SM, Stern MP, Hazuda HP. Mi- gration status, socioeconomic status, and 86. Ruiz JM, Steffen P. Latino Health. In: mortality rates in Mexican Americans and Friedman HS, ed. The Oxford Handbook of non-Hispanic whites: the San Antonio Health Psychology. New York: Oxford Heart Study. Ann Epidemiol. 1996;6 University Press; 2011:805---823. (4):307---313. 87. Ruiz JM, Steffen P, Prather CC. 73. Wolf M, Betancourt J, Chang Y, et al. Socioeconomic status and health. In: Impact of activated vitamin D and race on Baum A, Revenson TA, Singer JE, eds. survival among hemodialysis patients. J Handbook of Health Psychology. Mahwah,

Published online ahead of print January 17, 2013 | American Journal of Public Health Ruiz et al. | Peer Reviewed | Systematic Review | e9