16 August 2005 Volume 143 Issue 4 Articles

gfedc plus after or the Acute Coronary Syndrome: Meta-Analysis with Estimates of Risk and Benefit Michael B. Rothberg, Carmel Celestin, Louis D. Fiore, Elizabeth Lawler, and James R. Cook After acute coronary syndromes, warfarin plus aspirin was associated with fewer myocardial infarctions, ischemic , and revascularization procedures. Warfarin was associated with an increase in major bleeding, but mortality did not differ. For patients who are at low or intermediate risk for bleeding, the cardiovascular benefits of warfarin outweigh the bleeding risks.(241-264)

gfedc Clinical Outcomes and Cost-Effectiveness of Strategies for Managing People at High Risk for David M. Eddy, Leonard Schlessinger, and Richard Kahn The authors used a novel decision model to estimate the long-term outcomes and costs of care when patients with impaired glucose tolerance use metformin or enroll in a lifestyle modification program. Although the program used in the Diabetes Prevention Program prevented diabetes in some patients and delayed it in others, the authors found that it was not cost-effective for health plans to implement.(265-273)

gfedc A Prognostic Index for Systemic AIDS-Related Non-Hodgkin Lymphoma Treated in the Era of Highly Active Antiretroviral Mark Bower, Brian Gazzard, Sundhiya Mandalia, Tom Newsom-Davis, Christina Thirlwell, Tony Dhillon, Anne Marie Young, Tom Powles, Andrew Gaya, Mark Nelson, and Justin Stebbing The International Prognostic Index predicts death in non-Hodgkin lymphoma, including AIDS-related lymphoma before highly active antiretroviral therapy (HAART). Since the advent of HAART, the prognosis has improved for AIDS-related non-Hodgkin lymphoma. A combination of the International Prognostic Index and the CD4 cell count defines 4 strata of risk for death in AIDS-related lymphoma in the HAART era.(274-281) Improving Patient Care

gfedc Quality of Care Is Associated with Survival in Vulnerable Older Patients Takahiro Higashi, Paul G. Shekelle, John L. Adams, Caren J. Kamberg, Carol P. Roth, David H. Solomon, David B. Reuben, Lillian Chiang, Catherine H. MacLean, John T. Chang, Roy T. Young, Debra M. Saliba, and Neil S. Wenger The authors studied the effect of the quality of care received by 372 community-dwelling vulnerable older patients on their survival over the next 3 years. Three-year survival improved steadily as the quality score improved. Better performance on process quality measures is strongly associated with better survival in vulnerable older adults. (282-292)

Reviews gfedc Evaluation and Management of the Patient with Pulmonary Arterial Hypertension Lewis J. Rubin and David B. Badesch Pulmonary arterial hypertension raises several diagnostic possibilities: idiopathic or familial cause, systemic diseases (connective tissue disease, HIV , or chronic liver disease), or drugs (fenfluramine anorexigens, amphetamines, or cocaine). In this review, the authors present approaches to the diagnosis and management of pulmonary arterial hypertension, using a typical case to highlight the key management points. (293-300) Perspectives gfedc Promoting Informed Choice: Transforming Health Care To Dispense Knowledge for Decision Making Steven H. Woolf, Evelyn C.Y. Chan, Russell Harris, Stacey L. Sheridan, Clarence H. Braddock, III, Robert M. Kaplan, Alex Krist, Annette M. O'Connor, and Sean Tunis This article is about providing patients with the information they need to participate in difficult decisions about their health care. The authors look to health systems to meet these information needs. Their proposed solutions include expanding information resources for decision support and linking the information to decision counseling. (301-302) Editorials gfedc Trying To Predict the Future for People with Diabetes: A Tough but Important Task Michael M. Engelgau The article by Eddy and colleagues describes a novel approach to modeling the costs and outcomes of health interventions. As applied to programs to alter lifestyle to prevent diabetes in patients with impaired glucose tolerance, the model's predictions are similar in most respects to previous work but differ about cost-effectiveness. This difference leads the authors to diametrically opposed conclusions about whether health plans should offer programs to modify lifestyle. (303-304) gfedc Modeling Complex Medical Decision Problems with the Archimedes Model Margaret L. Brandeau In an ideal world, low-cost, powerful, and ethical clinical trials would decide which treatments work best. In the real world, clinical trials are often too time-consuming, too expensive, unethical, or even impossible to perform. How, then, can we obtain answers to inform patient care? We need a structured framework that uses the best evidence and captures relevant complexities. Decision analysis meets these requirements but raises a new question: "How do we decide if we can trust the predictions of a decision model?" (305-306) gfedc Improving Patient Care Can Set Your Brain on Fire Sankey V. Williams In this issue, Higashi and colleagues conclude that "better performance on process quality measures is strongly associated with better survival among community-dwelling vulnerable older adults." This conclusion, if true, is important because it appears to validate a fundamental assumption underlying most quality improvement efforts: A better process of care will lead to better patient outcomes. (307-308)

On Being a Doctor gfedc Lessons from a Patient William Rifkin Whenever a patient shares with me her tale of misery, I experience the following, perhaps common to most doctors: first, horror at how cruel and harsh the world can be, then appreciation of the relatively insignificant hurts I have experienced in comparison, followed by compassion and a compunction to "make it all better." Finally, once reality sets in, I realize that what I have to offer is very unlikely to help in the end. I often feel that I am simply not giving enough. (309-316)

Letters Screening for Abdominal Aortic Aneurysms Jack L. Cronenwett Ned Calonge and Diana Petitti—RESPONSE(I-14) Target Blood Pressure and Chester B. Good Yujiro Kida Mark J. Sarnak, Andrew S. Levey, and Tom Greene—RESPONSE(I-22) High-Output Associated with Anagrelide Therapy for Essential Thrombocytosis Peter J. Engel, Heide Johnson, Robert P. Baughman, and Arthur I. Richards(I-28) Subungual Splinter Hemorrhages: A Clinical Window to Inhibition of Vascular Endothelial Growth Factor Receptors? Caroline Robert, Sandrine Faivre, Eric Raymond, Jean-Pierre Armand, and Bernard(I-33) Escudier Disseminated Aspergillosis Mimicking Hepatic Veno-Occlusive Disease George L. Daikos, Vassiliki Syriopoulou, George Aperis, Christos Toubanakis, George Petrikkos, and Maria Demonakos Correction: Recommendations for the Diagnosis and Treatment of the Acute Porphyrias Ancillary Content Medical Notices Summaries for Patients Benefits and Harms of Warfarin plus Aspirin after Acute Coronary Events The Outcomes and Costs of Diabetes Prevention with a Diet and Program or Metformin: A Computer Model Estimating Outcome in Patients with HIV-Related Lymphoma Association of Quality of Care with Survival of Elderly Managed Care Patients

Annals of Internal Medicine Article Warfarin plus Aspirin after Myocardial Infarction or the Acute Coronary Syndrome: Meta-Analysis with Estimates of Risk and Benefit Michael B. Rothberg, MD, MPH; Carmel Celestin, MD; Louis D. Fiore, MD, MPH; Elizabeth Lawler, MPH; and James R. Cook, MD, MPH

Background: After the acute coronary syndrome, adding warfa- Data Synthesis: Ten trials involving a total of 5938 patients rin to standard aspirin therapy decreases myocardial infarction and (11 334 patient-years) met the study criteria. Compared with as- but increases major bleeding. pirin alone, warfarin plus aspirin was associated with a decrease in the annual rate of myocardial infarction (0.022 vs. 0.041; rate Purpose: To quantify the risks and benefits of warfarin therapy ratio, 0.56 [95% CI, 0.46 to 0.69]), ischemic stroke (0.004 vs. after the acute coronary syndrome. 0.008; rate ratio, 0.46 [CI, 0.27 to 0.77]), and revascularization Data Sources: MEDLINE from 1990 to October 2004. Additional (0.115 vs. 0.135; rate ratio, 0.80 [CI, 0.67 to 0.95]). Warfarin was data were obtained from study authors. Clinical risk factors were associated with an increase in major bleeding (0.015 vs. 0.006; used to classify hypothetical patients into cardiovascular and rate ratio, 2.5 [CI, 1.7 to 3.7]). Mortality did not differ. bleeding risk groups on the basis of published data. Limitations: Two large studies provided most of the data. Stud- Study Selection: Randomized trials comparing intensive warfa- ies did not include coronary stenting, and results should not be rin therapy (international normalized ratio > 2.0) plus aspirin with applied to patients with stents. Relative risk reductions may not aspirin alone after the acute coronary syndrome. be consistent across risk groups. Data Extraction: Two reviewers independently selected studies Conclusions: For patients with the acute coronary syndrome and extracted data on study design; quality; and clinical outcomes, who are at low or intermediate risk for bleeding, the cardiovas- including myocardial infarction, stroke, revascularization, death, cular benefits of warfarin outweigh the bleeding risks. and major and minor bleeding. Rate ratios for outcomes were calculated and pooled by using the method of DerSimonian and Ann Intern Med. 2005;143:241-250. www.annals.org Laird. For author affiliations, see end of text.

yocardial infarction is a leading cause of illness and should derive the most benefit, which may be offset in Mdeath in the United States (1). Patients with a his- those with increased bleeding risk. To better quantify the tory of myocardial infarction are at increased risk for re- risks and benefits of warfarin therapy for individuals, we current infarction, stroke, and death (2, 3). Several inter- conducted a meta-analysis of randomized trials and then ventions have proven beneficial in the secondary calculated the expected benefit for groups at varying risk prevention of myocardial infarction, including ␤-blockers for cardiovascular disease and bleeding. (4), angiotensin-converting inhibitors (5), lipid- lowering therapy (6), and aspirin (7). In addition, after METHODS acute cardiac events, a marked generation per- Study Selection sists for months after clinical stabilization (8), suggesting a We searched MEDLINE using the following Medical role for anticoagulation beyond the initial use of low-mo- Subject Heading (MeSH) terms and text words: anticoag- lecular-weight (9). Although some studies have ulant or anticoagulation or warfarin and aspirin, combined shown that addition of warfarin to aspirin decreases subse- with myocardial infarction, unstable , coronary arterio- quent risk for cardiovascular events (10–14), other studies sclerosis,orcoronary disease. We limited our search to have not (15, 16). In a meta-analysis, Anand and Yusuf original, English-language articles published between 1 (17) suggested that only moderate- to high-intensity anti- January 1990 and 1 October 2004; earlier trials were not , with a target international normalized ratio (INR) more than 2.0, decreased cardiovascular events more than aspirin alone. A large randomized trial conducted in a primary care setting showed that adding warfarin to aspirin See also: decreased combined cardiovascular end points of myocar- Print dial infarction, stroke, and death by 29% (12). Despite Editors’ Notes ...... 242 these findings, warfarin use has not been widely adopted, Summary for Patients...... I-14 perhaps because of concern that the increased bleeding risk may counter the cardiovascular benefit (18) or that the Web-Only benefits are too small to justify the inconvenience (19). Appendix Tables The risks and benefits that are probably associated Appendix Figures with warfarin therapy are not equally distributed. Patients Conversion of figures and tables into slides at the highest risk for recurrent cardiovascular events www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 241 Article Meta-Analysis of Warfarin after Myocardial Infarction

ignated a subset analysis of countries with good compliance Context before data collection was complete. We included only the Continued thrombin generation persists for several months compliant countries in our analysis. after acute cardiac events. We contacted the OASIS authors for select data on Contribution compliant countries from OASIS (13) and individual end points from the OASIS pilot study (21), which were pub- Data from 10 randomized trials involving 5938 patients lished only as combined end points. We contacted the with the acute coronary syndrome who were not stented showed that, compared with aspirin alone, warfarin plus authors of Huynh and colleagues’ study (22), because the aspirin decreased annual rates of myocardial infarction, published results for all end points were incorrect. ischemic stroke, and revascularization and increased major Statistical Analysis bleeding rates. In patients with low or average bleeding risks, numbers of cardiovascular events prevented by war- We used Stata, version 8.2 (Stata Corp., College Sta- farin plus aspirin exceeded numbers of major bleeding epi- tion, Texas), for all analyses. We calculated the rate ratios sodes caused by it. and 95% CIs for recurrent myocardial infarction, stroke, and bleeding (major or minor) for each study. We used a Implications half-integer correction if no events occurred in 1 group of Benefits of warfarin plus aspirin may exceed harms in pa- the study. We pooled rate ratios by using random-effects tients with the acute coronary syndrome who are not models that used weighting based on the inverse-variance stented and do not have high bleeding risks. model according to DerSimonian and Laird (23). We used the Mantel–Haenszel test to evaluate heterogeneity among –The Editors trial outcomes. We considered statistical significance at a P value less than 0.05. We performed an influence analysis in which we computed the summary rate ratio, omitting likely to reflect current standards of care for secondary pre- ␤ the largest trial (Warfarin, Aspirin, Reinfarction Study vention, including lipid therapy and -blockers. [WARIS II] [12]), to assess for any single study dominance Two reviewers scanned the titles independently for rel- in the analysis. We used the Begg and Mazumdar adjusted evant randomized trials. We checked abstracts and then rank correlation test (24) and the Egger regression asym- manuscripts of all potentially appropriate references. We metry test (25) to evaluate publication bias. included studies if they were randomized, controlled trials of warfarin and aspirin in patients with an acute coronary Weighing Risks and Benefits syndrome. We excluded studies of percutaneous coronary To demonstrate the relative benefits of warfarin plus stenting (for which warfarin is generally not considered aspirin for individual patients at different degrees of risk, appropriate therapy) (20) and low-intensity warfarin ther- Ͻ we calculated the number of myocardial infarctions, apy (target INR 2.0). We assessed studies for proper strokes, and major bleeding episodes expected to result randomization, blinding of patients and investigators to from aspirin therapy, with or without warfarin, for 9 com- treatment allocation, and completeness of follow-up. binations of cardiovascular and bleeding risk. We calcu- Data Extraction lated events in each risk group taking warfarin plus aspirin The 2 reviewers extracted the following data indepen- by multiplying the rate for the corresponding group of dently: study size and duration; aspirin dose; target INR; patients taking aspirin by the rate ratio derived from the and demographic characteristics of trial participants, in- meta-analysis. The absolute differences between these rates cluding age, sex, number of patients with diabetes, smok- represent the number of myocardial infarctions and strokes ing status, systolic blood pressure, and whether the patient prevented by adding warfarin to aspirin. Because the tim- had a Q-wave infarction or received . We ing of events may be important in treatment decisions, we evaluated 7 end points: myocardial infarction, ischemic calculated outcomes at 3 months and 1 year. On the basis stroke, revascularization, death, minor bleeding, major of 5059 patients in the Combination Hemotherapy and bleeding, and intracranial hemorrhage. Each study’s au- Mortality Prevention (CHAMP) study (16), we estimated thors defined major and minor bleeding, and most studies that 51% of myocardial infarctions, 54% of strokes, and reported all 7 end points. When results were presented 50% of bleeding episodes in the first year after acute myo- only as combined end points or were ambiguous, we con- cardial infarction would occur in the first 3 months. Event tacted the study authors to obtain primary data. rates for all 3 outcomes continue to decrease over time. In the Organization to Assess Strategies for Ischemic Thus, considering events in the first 2 years after an acute Syndromes (OASIS) study (13), a multicenter, interna- coronary syndrome, approximately 70% of all outcomes tional trial, investigators noted 35 days into the trial that occur in the first year. To directly compare event rates compliance as measured by achievement of target INR was across studies of differing duration, we converted all results better in some countries than in other countries and des- into first-year rates by using the proportions just described.

242 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Meta-Analysis of Warfarin after Myocardial Infarction Article

Cardiovascular Risk dio della Sopravvivenza nell’Infarto miocardico (GISSI) Many published instruments are available for deter- (28), Secondary Prevention Reinfarction Israeli Nifedipine mining short- and long-term risk for reinfarction, stroke, Trial (SPRINT) (29), or Global Registry of Acute Coro- and mortality. To demonstrate the clinical usefulness of nary Events (GRACE) (30) studies, produce similar ranges estimating both cardiovascular and bleeding risk before of low-, medium-, and high-risk patients. A detailed calcu- prescribing warfarin, we chose clinical examples of low-, lator based on the GRACE model is available at www medium-, and high-risk patients. We derived our estimates .outcomes-umassmed.org/grace. of the rates for recurrent myocardial infarction and stroke for patients receiving aspirin by using the computed hazard Bleeding Risk ratios from a large population-based study of 2700 patients Several instruments are also available for predicting surviving hospitalization after a first myocardial infarction bleeding risk for patients taking warfarin (31–34). We and followed for an average of 3.4 years (26). We chose chose examples of low, medium, and high risk for bleeding this study for its large size, long follow-up period, emphasis by using the Outpatient Bleeding Risk Index devised by on easily measured clinical predictors, and inclusion of Landefeld and Goldman (35), which is the only prediction stroke as an outcome. The following variables were associ- tool prospectively validated in clinical practice (36, 37). ated with increased risk for recurrent myocardial infarction Landefeld and Goldman identified 5 independent risk fac- and stroke: age, treated diabetes, chronic congestive heart tors for major bleeding in outpatients: age 65 years or failure, angina, and serum creatinine level greater than older, history of stroke, history of gastrointestinal bleeding, 123.76 ␮mol/L (Ͼ1.4 mg/dL). We created examples of low-, medium-, and high-risk patients by applying the haz- a specific comorbid condition (renal insufficiency, recent ard ratios for specific risk variables to a baseline risk of 49 myocardial infarction, or severe anemia), and atrial fibril- per 1000 patient-years of observation. Infarction rates per lation. Incidence of major bleeding in 2 prospective vali- 1000 patient-years were 40 for a low-risk patient (no risk dation sets in the first 12 months was 1% for low-risk factors), 82 for a medium-risk patient (for example, diabe- patients (no risk factors), 7% for medium-risk patients (1 Ն tes and serum creatinine level Ͼ123.76 ␮mol/L [Ͼ1.4 or 2 risk factors), and 30% for high-risk patients ( 3 risk mg/dL]), and 188 for a high-risk patient (for example, factors) (36, 37). Approximately half of all bleeding events diabetes and congestive heart failure). Stroke rates for the occurred in the first 3 months of therapy. We derived pre- same patients were 13, 32, and 79, respectively. Other risk dicted rates of bleeding with aspirin alone by dividing the stratification tools, based on the Thrombolysis In Myocar- bleeding rates with warfarin and aspirin by the relative risk dial Infarction (TIMI II) (27), Gruppo Italiano per lo Stu- for warfarin plus aspirin versus aspirin. The difference be-

Figure 1. Forest plot showing rate ratios of myocardial infarction for warfarin plus aspirin compared with aspirin alone.

APRICOT-2 ϭ in the Prevention of Reocclusion In Coronary Thrombolysis-2; ASPECT-2 ϭ Antithrombotics in the Secondary Prevention of Events in Coronary -2; ATACS ϭ Therapy in Acute Coronary Syndromes; OASIS ϭ Organization to Assess Strategies for Ischemic Syndromes; WARIS II ϭ Warfarin, Aspirin, Reinfarction Study. www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 243 Article Meta-Analysis of Warfarin after Myocardial Infarction

Table 1. Baseline Events and Rates for 6 Outcomes in the Aspirin and Combination Aspirin and Warfarin Groups*

Trial Myocardial Infarction Thrombotic Stroke Death

Aspirin Combination Therapy Aspirin Combination Therapy Aspirin Combination Therapy ATACS main, 1994 (38) 9 (0.36) 6 (0.25) – – 2 (0.08) 2 (0.08) ATACS pilot, 1990 (39) 3 (0.14) 0 (0) – – 0 (0) 0 (0) APRICOT-2, 2002 (11) 11 (0.32) 3 (0.09) – – 0 (0) 1 (0.03) ASPECT-2, 2002 (10) 14 (0.05) 10 (0.03) 5 (0.02) 0 (0) 15 (0.05) 9 (0.03) Huynh et al., 2001 (22) 1 (0.03) 2 (0.05) 1 (0.03) 0 (0) 0 (0) 2 (0.05) OASIS main, 2001 (13) 52 (0.14) 30 (0.08) 6 (0.02) 3 (0.01) 35 (0.09) 39 (0.10) OASIS pilot, 1998 (21) 10 (0.40) 5 (0.20) 2 (0.08) 0 (0) – – WARIS II, 2002 (12) 117 (0.03) 69 (0.01) 32 (0.01) 17 (0) 92 (0.02) 95 (0.02) Williams et al., 1997 (40) 5 (0.93) 1 (0.18) 0 (0) 0 (0) 1 (0.19) 0 (0) Zibaeenezhad et al., 2004 (41) 6 (0.09) 4 (0.06) 0 (0) 0 (0) 6 (0.09) 2 (0.03)

* Values are events (events per patient-year). APRICOT-2 ϭ Antithrombotics in the Prevention of Reocclusion In Coronary Thrombolysis-2; ASPECT-2 ϭ Antithrombotics in the Secondary Prevention of Events in -2; ATACS ϭ Antithrombotic Therapy in Acute Coronary Syndromes; OASIS ϭ Organization to Assess Strategies for Ischemic Syndromes; WARIS II ϭ Warfarin, Aspirin, Reinfarction Study. tween these rates represents the absolute number of bleed- tients. Follow-up ranged from 3 months to 4 years; most ing episodes attributable to the addition of warfarin. studies lasted 12 months or less. Aspirin dose varied from 80 to 325 mg daily. Target INR for most studies was 2.0 RESULTS to 2.5. All studies were intention-to-treat analyses, and Study Selection compliance measured by target INR achieved ranged from 2.0 to 3.0. All studies used proper randomization tech- The initial search yielded 656 articles (Appendix Fig- niques and had end points adjudicated by investigators ure 1, available at www.annals.org). Ten studies met the blinded to treatment allocation. Because of the difficulty in inclusion criteria, involving a total of 5938 patients maintaining placebo warfarin therapy, only 2 studies were (11 334 patient-years). All studies reported rates for myo- double-blind. Seven studies had greater than 99% clinical cardial infarction and major and minor bleeding. Nine follow-up, and all studies had follow-up of at least 90%. studies reported rates for intracerebral hemorrhage and death, 8 studies reported revascularization rates, and 7 Myocardial Infarction studies reported rates of ischemic stroke. The annualized rate of myocardial infarction in the Study Design aspirin group ranged from 0.03 to 0.93 (Table 1). Nine of Appendix Table 1 (available at www.annals.org) 10 studies found a risk reduction attributable to warfarin, shows the study designs. There was a mix of large and but only 2 studies were sufficiently powered for the reduc- small studies, with 3 studies including more than 600 pa- tion to reach statistical significance. Reductions in relative

Figure 2. Forest plot showing rate ratios of ischemic stroke for warfarin plus aspirin compared with aspirin alone.

Studies with excluded rate ratios had no events in either group. ASPECT-2 ϭ Antithrombotics in the Secondary Prevention of Events in Coronary Thrombosis-2; OASIS ϭ Organization to Assess Strategies for Ischemic Syndromes; WARIS II ϭ Warfarin, Aspirin, Reinfarction Study.

244 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Meta-Analysis of Warfarin after Myocardial Infarction Article

Table 1—Continued

Major Bleeding Minor Bleeding Revascularization

Aspirin Combination Therapy Aspirin Combination Therapy Aspirin Combination Therapy 0 (0) 3 (0.12) 3 (0.12) 7 (0.30) 16 (0.64) 12 (0.50) 0 (0) 0 (0) 3 (0.41) 3 (0.35) 12 (1.6) 22 (2.58) 2 (0.06) 2 (0.06) 2 (0.06) 5 (0.15) 43 (1.2) 17 (0.50) 3 (0.01) 7 (0.02) 16 (0.06) 50 (0.17) 39 (0.14) 32 (0.11) 0 (0) 2 (0.05) 2 (0.05) 9 (0.24) 3 (0.08) 5 (0.13) 16 (0.04) 35 (0.09) 12 (0.03) 21 (0.06) 167 (0.45) 136 (0.36) 1 (0.04) 2 (0.08) 12 (0.48) 28 (1.14) – – 8 (0.00) 28 (0.01) 39 (0.01) 133 (0.03) 454 (0.10) 430 (0.09) 0 (0) 1 (0.18) 0 (0) 2 (0.36) 3 (0.56) 4 (0.72) 2 (0.03) 5 (0.07) 2 (0.03) 13 (0.19) – –

risk ranged from 29% to 100%, with an overall risk reduc- 73%). Removing the largest study improved the risk reduc- tion of 44% (Figure 1). Excluding the 2 largest studies had tion to 82% (CI, Ϫ1% to 97%). little effect on the risk reduction, which increased to 45% Hemorrhagic strokes were counted separately and in- (95% CI, 18% to 63%). Event-free survival curves contin- cluded in the estimates of major bleeding. Overall, 4 hem- ued to diverge for at least 5.5 years of therapy. Pairwise orrhagic strokes occurred in the warfarin group and 1 oc- correlation analysis suggested that studies showing greater curred in the aspirin group, translating to 1 additional risk reductions with warfarin therapy tended to have more intracranial hemorrhage per 1800 patient-years of com- ϭ Ͻ bleeding complications as well (r 0.7; P 0.036). bined anticoagulation. Stroke The annualized risk for ischemic stroke in the aspirin Revascularization group ranged from 0.0 to 0.08, with a weighted average of The annualized risk for revascularization ranged from 0.008. All 5 studies in which at least 1 stroke was reported 0.076 to 1.3. Five of 7 studies showed decreased rates of found a risk reduction for the warfarin plus aspirin group, percutaneous transluminal coronary angioplasty or coro- but only 1 risk reduction was statistically significant (Fig- nary artery bypass grafting for patients receiving warfarin ure 2). Reductions in relative risk ranged from 50% to therapy, but only 1 rate reached statistical significance (Ap- 100%, with an overall risk reduction of 54% (CI, 23% to pendix Figure 2, available at www.annals.org). Hazard ra-

Figure 3. Forest plot showing rate ratios of death for warfarin plus aspirin compared with aspirin alone.

The study with an excluded rate ratio had no events in either group. APRICOT-2 ϭ Antithrombotics in the Prevention of Reocclusion In Coronary Thrombolysis-2; ASPECT-2 ϭ Antithrombotics in the Secondary Prevention of Events in Coronary Thrombosis-2; ATACS ϭ Antithrombotic Therapy in Acute Coronary Syndromes; OASIS ϭ Organization to Assess Strategies for Ischemic Syndromes; WARIS II ϭ Warfarin, Aspirin, Reinfarction Study. www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 245 Article Meta-Analysis of Warfarin after Myocardial Infarction

Figure 4. Forest plot showing rate ratios of major bleeding for warfarin plus aspirin compared with aspirin alone.

The study with an excluded rate ratio had no events in either group. APRICOT-2 ϭ Antithrombotics in the Prevention of Reocclusion In Coronary Thrombolysis-2; ASPECT-2 ϭ Antithrombotics in the Secondary Prevention of Events in Coronary Thrombosis-2; ATACS ϭ Antithrombotic Therapy in Acute Coronary Syndromes; OASIS ϭ Organization to Assess Strategies for Ischemic Syndromes; WARIS II ϭ Warfarin, Aspirin, Reinfarction Study. tios ranged from 0.51 to 1.70, with an overall relative risk Timing and Risk Groups reduction of 20% (CI, 5% to 33%). As expected, studies lasting 3 months had the highest rates of myocardial infarction, stroke, and bleeding. Rates for Death all 3 outcomes decreased with increasing study duration. Pro- No study showed a statistical difference in mortality jected first-year rates of myocardial infarction and major (Figure 3). The combined studies showed a 4% decrease in bleeding episodes varied considerably among the studies, but overall mortality in the warfarin group, but this did not relative risk reductions from warfarin were the same for all reach statistical significance. outcomes, regardless of study duration or indication (Table 2). Figure 5 shows the cardiovascular events averted and ex- Bleeding cess bleeding episodes caused by adding warfarin to aspirin for Nine studies showed an increased risk for major bleed- 1 year for different combinations of cardiovascular and bleed- ing associated with warfarin therapy (Figure 4). The annu- ing risk. For most studies, the benefit of warfarin clearly out- alized risk for major bleeding in the warfarin group ranged weighs the risk. Individuals at the highest risk for cardiovas- from 0.6% to 18.0%, with an overall risk of 1.5%. The cular events (for example, a patient with diabetes and relative risk for major bleeding with warfarin compared congestive heart failure) could avert 83 myocardial infarctions with aspirin was 2.5 (CI, 1.7 to 3.7). The relative risk for and 43 strokes per 1000 patient-years of warfarin therapy, minor bleeding was 2.6 (CI, 2.0 to 3.3) (Appendix Figure while those at the lowest cardiovascular risk would avoid 18 3, available at www.annals.org). myocardial infarctions and 7 strokes. Regardless of cardiovas- cular risk, adding warfarin to aspirin will cause an excess of 6 Study Diagnostics to 180 major bleeding episodes per 1000 patient-years, de- The Mantel–Haenszel test showed no statistically sig- pending on the individual’s bleeding risk. Because rates for all nificant heterogeneity for any end point among the studies outcomes decrease over time, approximately half of the benefit (Appendix Table 2, available at www.annals.org). An in- and risk in the first year of therapy accrues in the first 3 fluence analysis excluding 1 study at a time and finally the months. Conversely, both benefits and risks are reduced by 2 largest studies simultaneously (WARIS II [12] and OA- 60% in the second year of therapy. SIS [13]) had no substantial effect on the summary rate ratio. No evidence suggested publication bias for any end point, with the exception of stroke, and only the Egger DISCUSSION regression asymmetry test suggested some bias. Stroke was Our meta-analysis bolsters the finding that after an not a primary end point in any trial. acute coronary syndrome, warfarin plus aspirin decreases

246 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Meta-Analysis of Warfarin after Myocardial Infarction Article the rate of myocardial infarction or stroke more than aspi- have bleeding episodes because they received inadequate rin alone. Although the probability of major bleeding is anticoagulation or had stopped taking warfarin altogether. also increased, the benefits far outweigh the risks for many An annual bleeding rate of 1.5% is probably more repre- patients. For individuals at high cardiovascular risk but at sentative of adherent patients. Similarly, adherent patients low risk for bleeding (for example, a 58-year-old man with could expect fewer myocardial infarctions and strokes than diabetes and congestive heart failure), adding warfarin to were seen in the intention-to-treat analysis, as was found in standard aspirin therapy could avert 83 myocardial infarc- both the OASIS trial (13) and the Thrombosis Prevention tions and 43 strokes per 1000 patient-years of therapy at a Trial (44) when each was analyzed according to adherence cost of just 6 major bleeding episodes. The number needed to warfarin. to treat for 3 months to prevent 1 major cardiovascular Our study has several limitations. Although the de- event is 16, while the number needed to harm for the same signs of the individual studies were similar, they were not period is 333. Even when it is not possible to fully separate identical. All meta-analyses contain some degree of heter- bleeding risk from cardiovascular risk, warfarin still seems ogeneity, as do multicenter randomized trials. Despite mi- beneficial, as long as patients at the highest risk for bleed- nor variations in eligibility criteria, study duration, and ing are excluded from therapy. aspirin dosage, the studies selected were remarkably simi- We found no difference in relative risk reduction over lar. Although 2 large trials accounted for most patients, time, but the absolute rates of all end points decrease steadily. As a result, the greatest benefits and risks occur in without these, the relative risk reduction for myocardial the first 3 months of therapy. Still, the curves for the com- infarction was almost identical, emphasizing the homoge- bined end points continue to diverge for at least 5 years. neity of the trials. For patients at high cardiovascular risk, the gains beyond 1 We chose to include only the compliant countries year are likely to be substantial, whereas those at low risk from the OASIS trial (13) because including countries may choose to stop therapy after 3 months. where fewer than 70% of the patients took the therapy Despite an almost 3-fold increase in major bleeding, would bias our results heavily in favor of the null hypoth- overall bleeding rates among patients taking warfarin and esis. Although this violates the strictest interpretation of aspirin during 5500 patient-years of observation were low, intention to treat, the designation of compliant countries similar to the rate seen in patients with atrial fibrillation or was made only 35 days into the study and without exam- deep (42, 43). The low rate in WARIS ining clinical outcomes. Because entire countries were ex- II (0.6% per year) may have resulted from the intention- cluded, patient-specific bias could not occur. It is also en- to-treat design (12). Some patients presumably did not couraging that the effect size seen in compliant countries

Table 2. Projected 1-Year Rates for Myocardial Infarction and Bleeding, by Indication*

Study, Year (Reference) Indication MI Major Bleeding

Aspirin Warfarin Relative Aspirin Warfarin Relative Risk Risk Primarily including patients with MI ASPECT-2, 2002 (10) Acute MI (87%), unstable angina 0.05 0.03 0.69 0.01 0.02 2.26 (13%) APRICOT-2, 2002 (11) Acute MI after thrombolysis 0.16 0.04 0.28 0.03 0.03 1.03 WARIS II, 2002 (12) Acute MI 0.05 0.03 0.56 0.00 0.01 3.32 Zibaeenezhad et al., 2004 (41) Acute MI 0.09 0.06 0.67 0.03 0.07 2.50 Williams et al., 1997 (40) STEMI (63%), unstable angina, or 0.35 0.07 0.19 0.00 0.07 2.91 non–Q-wave MI (37%) Overall rate† 0.03 0.02 0.54 0.003 0.008 2.73

Primarily including patients with unstable angina ATACS main, 1994 (38) Unstable angina (68%), 0.16 0.11 0.69 0.00 0.06 7.26 non–Q-wave MI (32%) ATACS pilot, 1990 (39) Unstable angina (66%), 0.06 0.00 0.22 0.00 0.00 non–Q-wave MI (34%) Huynh et al., 2001 (22) Unstable angina, non–Q-wave MI 0.02 0.04 2.07 0.01 0.03 5.16 OASIS pilot, 1998 (21) Unstable angina (80%), 0.20 0.10 0.51 0.02 0.04 2.02 non–Q-wave MI (20%) OASIS main, 2001 (13) Unstable angina (86%), 0.08 0.05 0.58 0.02 0.05 2.20 non–Q-wave MI (14%) Overall rate† 0.16 0.04 0.58 0.04 0.09 2.49

* To compare rates, we assumed that 25% of events occur in the first 3 months, 50% occur in the first year, 70% occur in the first 2 years, and 100% occur by 4 years. APRICOT-2 ϭ Antithrombotics in the Prevention of Reocclusion In Coronary Thrombolysis-2; ASPECT-2 ϭ Antithrombotics in the Secondary Prevention of Events in Coronary Thrombosis-2; ATACS ϭ Antithrombotic Therapy in Acute Coronary Syndromes; MI ϭ myocardial infarction; OASIS ϭ Organization to Assess Strategies for Ischemic Syndromes; STEMI ϭ ST-segment elevation myocardial infarction; WARIS II ϭ Warfarin, Aspirin, Reinfarction Study. † Overall rates are raw rates per person-year, not annualized. www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 247 Article Meta-Analysis of Warfarin after Myocardial Infarction

Figure 5. Predicted myocardial infarctions (MIs) and thrombotic strokes averted and excess bleeding episodes caused in 1000 patients as a result of adding warfarin to aspirin for 1 year, stratified by bleeding and MI risk.

Numbers of cardiovascular events are the differences between the MI rates (aspirin rate Ϫ aspirin rate ϫ risk ratio) plus the difference in stroke rates (baseline rates not shown) multiplied by 1000 patient-years. The number of bleeding episodes represents the difference in bleeding rates (combination rate Ϫ combination rate/risk ratio for major bleeding episode) multiplied by 1000. Letters and corresponding dots represent actual combinations of bleeding and cardiovascular risk from Table 2. The shaded area in the parallelogram represents unfavorable risk or benefit. APRICOT-2 ϭ Antithrombotics in the Prevention of Reocclusion In Coronary Thrombolysis-2; ASPECT-2 ϭ Antithrombotics in the Secondary Prevention of Events in Coronary Thrombosis-2; ATACS ϭ Antithrombotic Therapy in Acute Coronary Syndromes; OASIS ϭ Organization to Assess Strategies for Ischemic Syndromes; WARIS II ϭ Warfarin, Aspirin, Reinfarction Study. from OASIS (13) was almost identical to that seen in farin to prevent reinfarction and stroke. The most recent WARIS II (12). American College of Cardiology/American Heart Associa- All studies were conducted in the 1990s before coro- tion guidelines (53) for treatment of myocardial infarction nary artery stenting became widespread. As a result of sev- were published before the results of WARIS II were avail- eral comparative trials, warfarin is not considered a stan- able. Three general objections have been raised to warfarin: dard of care in stented patients (45–47). However, outside The increased bleeding risk offsets any potential cardiovas- of large metropolitan areas in the United States, most pa- cular benefit (18), the cardiovascular gains are too small to tients with myocardial infarction still do not undergo justify the inconvenience of warfarin therapy (19), and stenting. Moreover, none of these studies examined long- studies have not demonstrated a mortality benefit (54). term anticoagulation after the initial stenting period. On the basis of our analysis, the benefits of warfarin Finally, we assumed that the decrease in relative risk should outweigh the harms for most patients. Even those for myocardial infarction and stroke would be consistent for whom the absolute numbers of major bleeding episodes across risk groups. Although this assumption has been and cardiovascular events are similar may benefit because found to be true with (48), simvastatin (49), cardiovascular events are usually more serious than bleed- and aspirin (50), additional trials of moderate- to high-risk ing events. Approximately 25% of reinfarctions are fatal, patients, including those with diabetes and heart failure, and among patients who survive, risk for congestive heart would be helpful, especially because warfarin may not ben- failure and disability increases. Ischemic stroke is even efit diabetic patients (51) and aspirin may be harmful in more likely to result in permanent disability. An atrial fi- patients with heart failure (52). brillation study showed that warfarin therapy reduces not Despite the clear benefits demonstrated in WARIS II only the frequency but also the severity of strokes (55). (12), no recommendations have been made for use of war- Apart from intracranial hemorrhage, bleeding episodes

248 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Meta-Analysis of Warfarin after Myocardial Infarction Article rarely result in long-term sequelae. In our meta-analysis, Chestnut Street, Springfield, MA 01199; e-mail, Michael.Rothberg warfarin resulted in 1 excess nonfatal intracranial hemor- @bhs.org. rhage per 1800 patient-years of therapy. Current author addresses and author contributions are available at www While the absolute benefit to participants in WARIS .annals.org. II was small because of their low overall risk (3% per year for myocardial infarction and 1% per year for stroke), pa- References tients were mostly young and nondiabetic and had Q-wave 1. American Heart Association. Heart Disease and Stroke Statistics—2004 Up- infarctions with preserved left ventricular function. Our date. Dallas, TX: American Heart Assoc; 2004. Accessed at www.americanheart. study demonstrated that warfarin produces the same rela- org on 22 April 2004. 2. Moss AJ, Benhorin J. Prognosis and management after a first myocardial tive risk reduction in high-risk patients, such as those in infarction. N Engl J Med. 1990;322:743-53. [PMID: 2407955] the Antithrombotic Therapy in Acute Coronary Syn- 3. Kornowski R, Goldbourt U, Zion M, Mandelzweig L, Kaplinsky E, Levo Y, dromes (ATACS) (38, 39) and Antithrombotics in the Pre- et al. Predictors and long-term prognostic significance of recurrent infarction in the year after a first myocardial infarction. SPRINT Study Group. Am J Cardiol. vention of Reocclusion In Coronary Thrombolysis-2 1993;72:883-8. [PMID: 8213543] (APRICOT-2) (11) studies, yielding considerable absolute 4. Yusuf S, Peto R, Lewis J, Collins R, Sleight P. Beta blockade during and after benefit. Whether this benefit is worth the inconvenience of myocardial infarction: an overview of the randomized trials. Prog Cardiovasc Dis. 1985;27:335-71. [PMID: 2858114] frequent blood draws and dietary restrictions is a decision 5. Yusuf S, Sleight P, Pogue J, Bosch J, Davies R, Dagenais G. Effects of an that is best left to individual patients and their physicians. angiotensin-converting-, ramipril, on cardiovascular events in Despite decreasing myocardial infarctions and strokes, high-risk patients. The Heart Outcomes Prevention Evaluation Study Investiga- warfarin had no demonstrable mortality benefit. However, tors. N Engl J Med. 2000;342:145-53. [PMID: 10639539] 6. Randomised trial of cholesterol lowering in 4444 patients with coronary heart mortality in all groups was relatively low, making it diffi- disease: the Scandinavian Simvastatin Survival Study (4S). Lancet. 1994;344: cult to detect a difference. Moreover, study durations were 1383-9. [PMID: 7968073] short and mortality due to congestive heart failure may be 7. Wallentin LC. Aspirin (75 mg/day) after an episode of unstable coronary artery disease: long-term effects on the risk for myocardial infarction, occurrence delayed several years. A similar reduction in nonfatal myo- of severe angina and the need for revascularization. Research Group on Instability cardial infarction was shown in the Clopidogrel in Unsta- in Coronary Artery Disease in Southeast Sweden. J Am Coll Cardiol. 1991;18: ble Angina to Prevent Recurrent Events (CURE) trial (56), 1587-93. [PMID: 1960301] 8. Merlini PA, Bauer KA, Oltrona L, Ardissino D, Cattaneo M, Belli C, et al. which also did not demonstrate a mortality benefit among Persistent activation of coagulation mechanism in unstable angina and myocar- 12 500 patients who were followed for up to 1 year. dial infarction. Circulation. 1994;90:61-8. [PMID: 8026047] Like insulin, warfarin engenders resistance in primary 9. Yusuf S, Mehta SR, Xie C, Ahmed RJ, Xavier D, Pais P, et al. Effects of reviparin, a low-molecular-weight heparin, on mortality, reinfarction, and strokes care physicians and patients alike. Despite ample evidence in patients with acute myocardial infarction presenting with ST-segment eleva- for more than a decade that warfarin reduces stroke in tion. JAMA. 2005;293:427-35. [PMID: 15671427] atrial fibrillation, as well as clear guidelines from the Amer- 10. van Es RF, Jonker JJ, Verheugt FW, Deckers JW, Grobbee DE. Aspirin and coumadin after acute coronary syndromes (the ASPECT-2 study): a randomised ican College of Cardiology/American Heart Association, controlled trial. Lancet. 2002;360:109-13. [PMID: 12126819] only one half of eligible patients receive warfarin therapy 11. Brouwer MA, van den Bergh PJ, Aengevaeren WR, Veen G, Luijten HE, (57), an increase from one third in 1993 (58). In the face Hertzberger DP, et al. Aspirin plus versus aspirin alone in the preven- tion of reocclusion after fibrinolysis for acute myocardial infarction: results of the of similar resistance, acceptance of warfarin in patients Antithrombotics in the Prevention of Reocclusion In Coronary Thrombolysis with the acute coronary syndrome may take years, by (APRICOT)-2 Trial. Circulation. 2002;106:659-65. [PMID: 12163424] which time less cumbersome, albeit more expensive, alter- 12. Hurlen M, Abdelnoor M, Smith P, Erikssen J, Arnesen H. Warfarin, aspi- rin, or both after myocardial infarction. N Engl J Med. 2002;347:969-74. natives should be available. Routine use of clopidogrel for [PMID: 12324552] at least 9 months is quickly becoming the standard of care, 13. Effects of long-term, moderate-intensity oral anticoagulation in addition to despite studies that found the practice to be non–cost- aspirin in unstable angina. The Organization to Assess Strategies for Ischemic effective (59, 60). Moreover, the 23% reduction in myo- Syndromes (OASIS) Investigators. J Am Coll Cardiol. 2001;37:475-84. [PMID: 11216966] cardial infarction shown with clopidogrel is considerably 14. Thrombosis prevention trial: randomised trial of low-intensity oral anticoag- smaller than the 44% reduction shown with warfarin (56). ulation with warfarin and low-dose aspirin in the primary prevention of ischaemic Although newer oral that do not require heart disease in men at increased risk. The Medical Research Council’s General Practice Research Framework. Lancet. 1998;351:233-41. [PMID: 9457092] monitoring may someday replace warfarin, the benefits of 15. Randomised double-blind trial of fixed low-dose warfarin with aspirin after warfarin therapy are available to patients today. myocardial infarction. Coumadin Aspirin Reinfarction Study (CARS) Investiga- tors. Lancet. 1997;350:389-96. [PMID: 9259652] From Baystate Medical Center, Springfield, Massachusetts, and Tufts 16. Fiore LD, Ezekowitz MD, Brophy MT, Lu D, Sacco J, Peduzzi P, et al. Department of Veterans Affairs Cooperative Studies Program com- University School of Medicine, Massachusetts Veterans Epidemiology paring combined warfarin and aspirin with aspirin alone in survivors of acute Research and Information Center, Veterans Affairs Boston Healthcare myocardial infarction: primary results of the CHAMP study. Circulation. 2002; System, Boston, Massachusetts. 105:557-63. [PMID: 11827919] 17. Anand SS, Yusuf S. Oral therapy in patients with coronary artery disease: a meta-analysis. JAMA. 1999;282:2058-67. [PMID: 10591389] Potential Financial Conflicts of Interest: None disclosed. 18. Topol EJ. Warfarin alone or with aspirin was superior to aspirin alone after acute myocardial infarction but increased bleeding. ACP J Club. 2003;138:34. [PMID: 12614119] Requests for Single Reprints: Michael B. Rothberg, MD, MPH, Divi- 19. Pullicino P, Thompson JL. Warfarin, aspirin, or both after myocardial in- sion of General Medicine and Geriatrics, Baystate Medical Center, 759 farction [Letter]. N Engl J Med. 2003;348:256-7; author reply 256-7. [PMID: www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 249 Article Meta-Analysis of Warfarin after Myocardial Infarction

12529470] culprit lesion in unstable coronary artery disease with warfarin and aspirin versus 20. Leon MB, Baim DS, Popma JJ, Gordon PC, Cutlip DE, Ho KK, et al. A aspirin alone: preliminary study. J Am Coll Cardiol. 1997;30:364-9. [PMID: clinical trial comparing three antithrombotic-drug regimens after coronary-artery 9247506] stenting. Stent Anticoagulation Restenosis Study Investigators. N Engl J Med. 41. Zibaeenezhad MJ, Mowla A, Sorbi MH. Warfarin and aspirin versus aspirin 1998;339:1665-71. [PMID: 9834303] alone in patients with acute myocardial infarction: a pilot study. Angiology. 2004; 21. Anand SS, Yusuf S, Pogue J, Weitz JI, Flather M. Long-term oral antico- 55:17-20. [PMID: 14759085] agulant therapy in patients with unstable angina or suspected non-Q-wave myo- 42. Risk factors for stroke and efficacy of antithrombotic therapy in atrial fibril- cardial infarction: organization to assess strategies for ischemic syndromes (OA- lation. Analysis of pooled data from five randomized controlled trials. Arch Intern SIS) pilot study results. Circulation. 1998;98:1064-70. [PMID: 9736592] Med. 1994;154:1449-57. [PMID: 8018000] 22. Huynh T, Théroux P, Bogaty P, Nasmith J, Solymoss S. Aspirin, warfarin, 43. Palareti G, Leali N, Coccheri S, Poggi M, Manotti C, D’Angelo A, et al. or the combination for secondary prevention of coronary events in patients with Bleeding complications of oral anticoagulant treatment: an inception-cohort, pro- acute coronary syndromes and prior coronary artery bypass . Circulation. spective collaborative study (ISCOAT). Italian Study on Complications of Oral 2001;103:3069-74. [PMID: 11425770] Anticoagulant Therapy. Lancet. 1996;348:423-8. [PMID: 8709780] 23. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 44. Rudnicka AR, Ashby D, Brennan P, Meade T. Thrombosis prevention trial: 1986;7:177-88. [PMID: 3802833] compliance with warfarin treatment and investigation of a retained effect. Arch 24. Begg CB, Mazumdar M. Operating characteristics of a rank correlation test Intern Med. 2003;163:1454-60. [PMID: 12824095] for publication bias. Biometrics. 1994;50:1088-101. [PMID: 7786990] 45. Schömig A, Neumann FJ, Kastrati A, Schühlen H, Blasini R, Hadamitzky 25. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis M, et al. A randomized comparison of antiplatelet and anticoagulant therapy detected by a simple, graphical test. BMJ. 1997;315:629-34. [PMID: 9310563] after the placement of coronary-artery stents. N Engl J Med. 1996;334:1084-9. 26. Kaplan RC, Heckbert SR, Furberg CD, Psaty BM. Predictors of subsequent [PMID: 8598866] coronary events, stroke, and death among survivors of first hospitalized myocar- 46. Bertrand ME, Legrand V, Boland J, Fleck E, Bonnier J, Emmanuelson H, dial infarction. J Clin Epidemiol. 2002;55:654-64. [PMID: 12160913] et al. Randomized multicenter comparison of conventional anticoagulation versus 27. Wiviott SD, Morrow DA, Frederick PD, Giugliano RP, Gibson CM, antiplatelet therapy in unplanned and elective coronary stenting. The full antico- McCabe CH, et al. Performance of the thrombolysis in myocardial infarction agulation versus aspirin and (fantastic) study. Circulation. 1998;98: risk index in the National Registry of Myocardial Infarction-3 and -4: a simple 1597-603. [PMID: 9778323] index that predicts mortality in ST-segment elevation myocardial infarction. J 47. Leon MB, Baim DS, Popma JJ, Gordon PC, Cutlip DE, Ho KK, et al. A Am Coll Cardiol. 2004;44:783-9. [PMID: 15312859] clinical trial comparing three antithrombotic-drug regimens after coronary-artery 28. Marchioli R, Avanzini F, Barzi F, Chieffo C, Di Castelnuovo A, Franzosi stenting. Stent Anticoagulation Restenosis Study Investigators. N Engl J Med. MG, et al. Assessment of absolute risk of death after myocardial infarction by use 1998;339:1665-71. [PMID: 9834303] of multiple-risk-factor assessment equations: GISSI-Prevenzione mortality risk 48. Budaj A, Yusuf S, Mehta SR, Fox KA, Tognoni G, Zhao F, et al. Benefit of chart. Eur Heart J. 2001;22:2085-103. [PMID: 11686666] clopidogrel in patients with acute coronary syndromes without ST-segment ele- 29. Kornowski R, Goldbourt U, Boyko V, Behar S. Clinical predictors of re- vation in various risk groups. Circulation. 2002;106:1622-6. [PMID: 12270853] infarction among men and women after a first myocardial infarction. SPRINT 49. MRC/BHF Heart Protection Study of cholesterol lowering with simvastatin Study Group. Secondary Prevention Reinfarction Israeli Nifedipine Trial. Cardi- in 20,536 high-risk individuals: a randomised placebo-controlled trial. Lancet. ology. 1995;86:163-8. [PMID: 7728808] 2002;360:7-22. [PMID: 12114036] 30. Eagle KA, Lim MJ, Dabbous OH, Pieper KS, Goldberg RJ, Van de Werf 50. Collaborative meta-analysis of randomised trials of antiplatelet therapy for F, et al. A validated prediction model for all forms of acute coronary syndrome: prevention of death, myocardial infarction, and stroke in high risk patients. BMJ. estimating the risk of 6-month postdischarge death in an international registry. 2002;324:71-86. [PMID: 11786451] JAMA. 2004;291:2727-33. [PMID: 15187054] 51. Smith P, Hurlen M, Abdelnoor M, Arnesen H. No benefit by warfarin in 31. Kuijer PM, Hutten BA, Prins MH, Büller HR. Prediction of the risk of post-MI patients with diabetes mellitus [Abstract]. Pathophysiol Haemost bleeding during anticoagulant treatment for venous thromboembolism. Arch In- Thromb. 2002;32:S76. tern Med. 1999;159:457-60. [PMID: 10074953] 52. Cleland JG, Findlay I, Jafri S, Sutton G, Falk R, Bulpitt C, et al. The 32. Gitter MJ, Jaeger TM, Petterson TM, Gersh BJ, Silverstein MD. Bleeding Warfarin/Aspirin Study in Heart failure (WASH): a randomized trial comparing and thromboembolism during anticoagulant therapy: a population-based study in antithrombotic strategies for patients with heart failure. Am Heart J. 2004;148: Rochester, Minnesota. Mayo Clin Proc. 1995;70:725-33. [PMID: 7630209] 157-64. [PMID: 15215806] 33. van der Meer FJ, Rosendaal FR, Vandenbroucke JP, Briët E. Assessment of 53. Ryan TJ, Antman EM, Brooks NH, Califf RM, Hillis LD, Hiratzka LF, et a bleeding risk index in two cohorts of patients treated with oral anticoagulants. al. 1999 update: ACC/AHA guidelines for the management of patients with Thromb Haemost. 1996;76:12-6. [PMID: 8819244] acute myocardial infarction: executive summary and recommendations: a report 34. McMahan DA, Smith DM, Carey MA, Zhou XH. Risk of major hemor- of the American College of Cardiology/American Heart Association Task Force rhage for outpatients treated with warfarin. J Gen Intern Med. 1998;13:311-6. on Practice Guidelines (Committee on Management of Acute Myocardial Infarc- [PMID: 9613886] tion). Circulation. 1999;100:1016-30. [PMID: 10468535] 35. Landefeld CS, Goldman L. Major bleeding in outpatients treated with war- 54. Hoey J. Warfarin, acetylsalicylic acid or both? CMAJ. 2002;167:1036. farin: incidence and prediction by factors known at the start of outpatient ther- [PMID: 12403748] apy. Am J Med. 1989;87:144-52. [PMID: 2787958] 55. Hylek EM, Go AS, Chang Y, Jensvold NG, Henault LE, Selby JV, et al. 36. Wells PS, Forgie MA, Simms M, Greene A, Touchie D, Lewis G, et al. The Effect of intensity of oral anticoagulation on stroke severity and mortality in atrial outpatient bleeding risk index: validation of a tool for predicting bleeding rates in fibrillation. N Engl J Med. 2003;349:1019-26. [PMID: 12968085] patients treated for deep venous thrombosis and pulmonary . Arch 56. Yusuf S, Zhao F, Mehta SR, Chrolavicius S, Tognoni G, Fox KK, et al. Intern Med. 2003;163:917-20. [PMID: 12719200] Effects of clopidogrel in addition to aspirin in patients with acute coronary syn- 37. Beyth RJ, Quinn LM, Landefeld CS. Prospective evaluation of an index for dromes without ST-segment elevation. N Engl J Med. 2001;345:494-502. predicting the risk of major bleeding in outpatients treated with warfarin. Am J [PMID: 11519503] Med. 1998;105:91-9. [PMID: 9727814] 57. Go AS, Hylek EM, Borowsky LH, Phillips KA, Selby JV, Singer DE. 38. Cohen M, Adams PC, Parry G, Xiong J, Chamberlain D, Wieczorek I, et Warfarin use among ambulatory patients with nonvalvular atrial fibrillation: the al. Combination antithrombotic therapy in unstable rest angina and non-Q-wave anticoagulation and risk factors in atrial fibrillation (ATRIA) study. Ann Intern infarction in nonprior aspirin users. Primary end points analysis from the ATACS Med. 1999;131:927-34. [PMID: 10610643] trial. Antithrombotic Therapy in Acute Coronary Syndromes Research Group. 58. Stafford RS, Singer DE. National patterns of warfarin use in atrial fibrilla- Circulation. 1994;89:81-8. [PMID: 8281698] tion. Arch Intern Med. 1996;156:2537-41. [PMID: 8951296] 39. Cohen M, Adams PC, Hawkins L, Bach M, Fuster V. Usefulness of anti- 59. Gaspoz JM, Coxson PG, Goldman PA, Williams LW, Kuntz KM, Hunink thrombotic therapy in resting angina pectoris or non-Q-wave myocardial infarc- MG,etal.Cost effectiveness of aspirin, clopidogrel, or both for secondary pre- tion in preventing death and myocardial infarction (a pilot study from the Anti- vention of coronary heart disease. N Engl J Med. 2002;346:1800-6. [PMID: thrombotic Therapy in Acute Coronary Syndromes Study Group). Am J Cardiol. 12050341] 1990;66:1287-92. [PMID: 2244556] 60. Marshall T. Coronary heart disease prevention: insights from modelling in- 40. Williams MJ, Morison IM, Parker JH, Stewart RA. Progression of the cremental cost effectiveness. BMJ. 2003;327:1264. [PMID: 14644970]

250 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Annals of Internal Medicine

Current Author Addresses: Dr. Rothberg: Division of General Medi- cine and Geriatrics, Baystate Medical Center, 759 Chestnut Street, Springfield, MA 01199. Dr. Celestin: Baystate Medical Center, 759 Chestnut Street, Springfield, MA 01199. Dr. Fiore and Ms. Lawler: Massachusetts Veterans Epidemiology Re- search and Information Center, 150 South Huntington Avenue, Boston, MA 02130. Dr. Cook: Cardiac Services, Baystate Medical Center, 759 Chestnut Street, Springfield, MA 01199.

Author Contributions: Conception and design: M.B. Rothberg, C. Celestin. Analysis and interpretation of the data: M.B. Rothberg, L.D. Fiore, E. Lawler, J.R. Cook. Drafting of the article: M.B. Rothberg, J.R. Cook. Critical revision of the article for important intellectual content: M.B. Rothberg, C. Celestin, L.D. Fiore, E. Lawler, J.R. Cook. Final approval of the article: M.B. Rothberg, C. Celestin, L.D. Fiore, E. Lawler, J.R. Cook. Provision of study materials or patients: L.D. Fiore, E. Lawler. Statistical expertise: J.R. Cook. Collection and assembly of data: M.B. Rothberg, C. Celestin, L.D. Fiore.

www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 W-49 Appendix Table 1. Design of Trials Included in the Meta-Analysis*

Study, Year Population Patients, Mean Men, Diabetes, Target Aspirin Duration of Randomization Double- Patients (Reference) Studied n Age, y % % INR Dose, mg Therapy and Proper Blind Lost to Generation of Follow-up, Allocation n/n† Sequence ATACS pilot, 1990 Unstable angina, 69 62.5 57 38 3.0–4.5 325/80‡ 3 mo Not stated No 5/93 (39) non–Q-wave MI ATACS main, 1994 Unstable angina, 214 61.5 68 8 2.0–3.0 162.5 3 mo Not stated No 0/214 (38) non–Q-wave MI Williams et al., Acute MI, unstable 57 57 86 – 2.0–2.5 150 3 mo Yes Yes 0/57 1997 (40) angina OASIS pilot, 1998 Unstable angina, 197 64 67 – 2.5 325 3 mo Not stated No 0/197 (21) non–Q-wave MI OASIS main, 2001 Unstable angina, 3712 64 61 20 2.5 80–325 5 mo Telephone No 15/3712 (13) non–Q-wave MI randomization service Huynh et al., 2001 CABG patients with 90 67 77 21 2.0–2.5 80 12 mo Not stated Yes 2/90 (22) unstable angina or non–Q-wave MI APRICOT-2, 2002 Acute MI after 274 58 82 6 2.0–3.0 80 3 mo Telephone No 0/274 (11) thrombolytics randomization service ASPECT-2, 2002 Acute MI, unstable 668 61 78 10 2.0–2.5 80 26 mo Central No 2/669 (10) angina telephone service WARIS II, 2002 Acute MI 2414 61 76 9 2.0–2.5 160/75‡ 4 y Centrally by No 14/3630 (12) permuted blocks Zibaeenezhad et Acute MI 140 61 76 – 2.0–3.0 100 12 mo Not stated No 0/140 al., 2004 (41)

* APRICOT-2 ϭ Antithrombotics in the Prevention of Reocclusion In Coronary Thrombolysis-2; ASPECT-2 ϭ Antithrombotics in the Secondary Prevention of Events in Coronary Thrombosis-2; ATACS ϭ Antithrombotic Therapy in Acute Coronary Syndromes; CABG ϭ coronary artery bypass grafting; INR ϭ international normalized ratio; MI ϭ myocardial infarction; OASIS ϭ Organization to Assess Strategies for Ischemic Syndromes; WARIS II ϭ Warfarin, Aspirin, Reinfarction Study. † Values are patients lost to follow-up/total patients. ‡ Lower dose is for patients assigned to warfarin therapy.

Appendix Table 2. Tests of Heterogeneity*

Variable Heterogeneity P Value I2† ␶2 Statistic Ischemic stroke 1.91 0.86 0.0% 0.0000 Death 3.73 0.81 0.0% 0.0000 Intracranial bleeding 0.41 0.81 0.0% 0.0000 Myocardial infarction 4.70 0.86 0.0% 0.0000 Major bleeding 4.12 0.85 0.0% 0.0000 Minor bleeding 10.43 0.32 13.7% 0.0251 PTCA or CABG 7.37 0.50 0.0% 0.0000

* CABG ϭ coronary artery bypass grafting; PTCA ϭ percutaneous transluminal coronary angioplasty. † I2 is the variation in rate ratio attributable to heterogeneity.

W-50 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Appendix Figure 1. Process of study selection.

Appendix Figure 2. Forest plot showing rate ratios of percutaneous transluminal coronary angioplasty or coronary artery bypass grafting for warfarin plus aspirin compared with aspirin alone.

The study with an excluded rate ratio had no events in either group. APRICOT-2 ϭ Antithrombotics in the Prevention of Reocclusion In Coronary Thrombolysis-2; ASPECT-2 ϭ Antithrombotics in the Secondary Prevention of Events in Coronary Thrombosis-2; ATACS ϭ Antithrombotic Therapy in Acute Coronary Syndromes; OASIS ϭ Organization to Assess Strategies for Ischemic Syndromes; WARIS II ϭ Warfarin, Aspirin, Reinfarction Study. www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 W-51 Appendix Figure 3. Forest plot showing rate ratios of minor bleeding for warfarin plus aspirin compared with aspirin alone.

APRICOT-2 ϭ Antithrombotics in the Prevention of Reocclusion In Coronary Thrombolysis-2; ASPECT-2 ϭ Antithrombotics in the Secondary Prevention of Events in Coronary Thrombosis-2; ATACS ϭ Antithrombotic Therapy in Acute Coronary Syndromes; OASIS ϭ Organization to Assess Strategies for Ischemic Syndromes; WARIS II ϭ Warfarin, Aspirin, Reinfarction Study.

W-52 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Article A Prognostic Index for Systemic AIDS-Related Non-Hodgkin Lymphoma Treated in the Era of Highly Active Antiretroviral Therapy Mark Bower, MA, PhD; Brian Gazzard, MD; Sundhiya Mandalia, PhD; Tom Newsom-Davis, MB BS; Christina Thirlwell, MB BS; Tony Dhillon, MB BS; Anne Marie Young, MB BS; Tom Powles, MD; Andrew Gaya, MB BS; Mark Nelson, MD; and Justin Stebbing, MA, PhD

Background: The established International Prognostic Index for Prognostic Index factors of age, tumor stage, lactate dehydrogen- lymphomas has not included patients with systemic AIDS-related ase level, Eastern Cooperative Oncology Group performance sta- non-Hodgkin lymphoma. tus, and number of extranodal sites were confirmed to be signif- icant variables. Regression modeling for patients in whom disease Objective: To establish the most appropriate prognostic index was diagnosed after 1996 revealed only 2 independent predictors for use in patients with systemic AIDS-related non-Hodgkin lym- of death: International Prognostic Index risk group and CD4 cell phoma. count. These predictors yielded 4 internally validated risk strata Design: A prospective study involving univariate and multivari- with predicted 1-year survival rates of 82%, 47%, 20%, and 15% able analyses of patients with AIDS-related non-Hodgkin lym- (P < 0.001). Prognostic risk scores in the highest quartile yielded phoma whose data were used to examine standard and new a likelihood ratio for death of 7.90 (hazard ratio, 1.0), whereas a criteria for survival after diagnosis. prognostic score less than 1.0 yielded a likelihood ratio of 0.23 (hazard ratio, 0.15 [95% CI, 0.06 to 0.33]). Setting: The Chelsea and Westminster cohort of HIV-1–infected persons. Limitations: The sample was small, and different HAART regi- mens were used. Patients: 9621 HIV-positive patients, 111 in whom AIDS-related non-Hodgkin lymphoma was treated after 1996, in the era of Conclusions: For patients with AIDS-related non-Hodgkin lym- highly active antiretroviral therapy (HAART). phoma that was diagnosed in the era of HAART, application of the International Prognostic Index remains useful. The addition of Intervention: Cox proportional hazards regression analysis to CD4 cell count provides further independent prognostic informa- determine the prognostic significance of multiple clinicopathologic tion. Patients who present with AIDS-related non-Hodgkin lym- variables. phoma and a low CD4 cell count have a poor prognosis; this information can be used to guide therapeutic options. Results: Survival of patients with AIDS-related non-Hodgkin lym- phoma has increased in the HAART era (log-rank chi-square, 9.23; Ann Intern Med. 2005;143:265-273. www.annals.org .Univariate analyses using the established International For author affiliations, see end of text .(0.002 ؍ P

he lymphomas are a diverse group of malignant disor- in the post-HAART era, and phase II studies describe me- Tders that vary in their molecular features, genetics, clin- dian duration of survival of 15 to 34 months (8–15), an ical presentation, treatment, and outcome. Major advances interval similar to that observed among all patients with in our understanding of the biology of these diseases have advanced-stage, high-grade non-Hodgkin lymphoma. It is been made, leading to new and classifications. hypothesized that these improvements are associated with a Although combination chemotherapy cures intermediate- change in prognostic factors. We sought to develop a new or high-grade aggressive non-Hodgkin lymphomas in prognostic model for HIV-associated non-Hodgkin lym- many patients, approximately 50% of patients die of the phoma in the era of HAART, a treatment that has been disease (1). Because Ann Arbor disease staging does not available in established market economies since 1996. predict outcome (2, 3), the International Prognostic Index Small prognostic studies of AIDS-related non- was introduced in 1993 to segregate aggressive lymphomas Hodgkin lymphoma in patients who presented in the pre- in terms of survival (4). From 2031 patients studied, 4 risk HAART era suggest that application of the International groups were derived on the basis of age, tumor stage, serum Prognostic Index may be useful in HIV-infected patients lactate dehydrogenase level, performance status, and num- (16, 17). We therefore aimed to confirm the validity of the ber of extranodal disease sites. International Prognostic Index, to identify additional prog- As we enter the third decade of the AIDS epidemic, it is apparent that many are more common in people infected with HIV. Non-Hodgkin lymphoma remains the See also: second most common tumor in such patients (after Kaposi Print sarcoma), and the rate of death from systemic AIDS-re- Editors’ Notes ...... 266 lated non-Hodgkin lymphoma remains high (5, 6). The Summary for Patients...... I-28 median duration of survival reported with chemotherapy before the availability of highly active antiretroviral therapy Web-Only (HAART) was 2 to 13 months (7). The outcome of AIDS- Conversion of figures and tables into slides related non-Hodgkin lymphoma appears to have improved

© 2005 American College of Physicians 265 Article Prognostic Index for Systemic AIDS-Related Lymphoma

marrow, paranasal, or paraspinal involvement received a Context further 5 doses of intrathecal chemotherapy. Between 1996 The International Prognostic Index (IPI) predicts survival in and 1998, the patients with AIDS-related non-Hodgkin patients with lymphoma, but its applicability to AIDS- lymphoma diagnosed in the era of HAART received che- related lymphomas in the era of highly active antiretroviral motherapy with bleomycin, etoposide, vincristine, metho- therapy has not been evaluated. The IPI stratifies patients trexate, prednisolone/cyclophosphamide, and doxorubicin into risk groups on the basis of age, tumor stage, serum (18 patients) or cyclophosphamide, hydroxydoxorubicin, lactate dehydrogenase levels, performance status, and vincristine, and prednisone (3 patients). A further 21 pa- number of extranodal sites. tients received chemotherapy with cisplatin, vinblastine, Contribution and bleomycin (17 patients); radiotherapy alone (3 pa- Among 111 patients with AIDS-related lymphoma diag- tients); or best supportive care (1 patient). Since 1999, 59 nosed since 1996, the IPI and CD4 cell count separated patients have been treated with infused cyclophosphamide, patients into 4 strata with 1-year survival rates of 82%, doxorubicin, and etoposide chemotherapy (13, 19); 2 have 47%, 20%, and 15%. received cyclophosphamide, hydroxydoxorubicin, vincris- tine, and prednisone; and 8 (including 3 in whom disease Implications was diagnosed at autopsy) received best supportive care only. The IPI and CD4 cell count can help physicians predict the The CD4 cell subset analysis was performed by using prognosis of patients with AIDS-related lymphoma. whole blood stained with murine antihuman monoclonal antibodies to CD4 (TetraOne [Beckman Coulter, High –The Editors Wycombe, United Kingdom]) on an Epics XL-MCL mul- tiparametric flow cytometer (Beckman Coulter). nostic factors for patients with AIDS-related non-Hodgkin Statistical Analysis lymphoma in the era of HAART, and to devise a new Variables were compared between groups by using the prognostic model for these patients. chi-square test for nominal variables and the Mann– Whitney U test for nonparametric variables. Survival was METHODS calculated from the day of diagnosis of AIDS-related non- Patients Hodgkin lymphoma until death or the date of last follow- The Chelsea and Westminster HIV cohort is one of up. Curves for overall duration of survival were plotted the largest in Europe. Clinical information on 9621 HIV-1 according to the method of Kaplan and Meier (20). The seropositive patients has been accumulated since 1986. All log-rank method was used to test for the significance of patients in whom lymphoma was diagnosed were identified differences in survival distributions (21) and univariate prospectively; these included 215 patients with AIDS-re- Cox proportional hazards regression analysis was used to lated non-Hodgkin lymphoma, 60 with primary central determine the prognostic significance of clinicopathologic nervous system lymphomas, and 26 with Hodgkin disease. variables at presentation with AIDS-related non-Hodgkin We estimated prognostic factors for AIDS-related lymphoma. Cox multivariable modeling was used to deter- non-Hodgkin lymphoma in the HAART era in patients mine independent variables predictive of survival by enter- receiving HAART. Patients with Hodgkin disease and pri- ing all variables that were significant in univariate analysis mary central nervous system lymphomas were excluded. (at a level of P Ͻ 0.15). A prognostic model was then The HAART era is defined as commencing on 1 January constructed from these data by dividing each ␤ coefficient 1996, when this treatment became routinely available at in the final multivariable models with significant predictors our institution and many others. One hundred eleven pa- by the lowest ␤ to 2 decimal places. Using these point tients with AIDS-related non-Hodgkin lymphoma received values, a risk score was assigned to each patient by sum- a diagnosis after this date. Highly active antiretroviral ther- ming the values for each present. The prognostic apy is defined as a combination of at least 3 antiretroviral score derived was then grouped into quartiles so that ap- agents, including a analogue backbone com- proximately equal numbers of patients were included in bined with a protease inhibitor or a nonnucleoside reverse each of these categories. The chosen cutoff values for the transcriptase inhibitor or both classes of drug, according to prognostic risk scores were further investigated by using generally accepted definitions (18). receiver-operating characteristic methods. All patients had histologically confirmed diagnoses of Because the performance of prognostic models may be AIDS-related non-Hodgkin lymphoma, and more than optimistically overestimated when they are determined on 95% had aggressive B-cell disease. All patients had full the basis of a small sample, a higher apparent performance staging at diagnosis, including examination of bone mar- than that observed in an independent sample of patients row and cerebrospinal fluid. All patients received a single not considered in the modeling process may result (22, dose of intrathecal chemoprophylaxis with their staging 23). To formally confirm the validity of the prognostic lumbar puncture. Patients with Burkitt lymphoma or bone index based on a small sample, we used the internal em-

266 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Prognostic Index for Systemic AIDS-Related Lymphoma Article

Figure 1. Kaplan–Meier overall survival curve for 215 patients with systemic AIDS-related non-Hodgkin lymphoma (NHL) diagnosed in the era before (104 patients) and after (111 patients) highly active antiretroviral therapy (HAART).

All causes of death are included (log-rank chi-square, 9.23; P ϭ 0.002). pirical distribution function, placing equal probabilities on did patients whose disease was diagnosed before the every original data value, as described elsewhere (24). Non- HAART era, the former patients were older (P Ͻ 0.001) parametric bootstrapping was used to draw a sample by and had a higher serum lactate dehydrogenase level (P Ͻ selecting independent bootstrap values (25–28). Each of 0.001) (data not shown). Univariate Cox proportional haz- these consisted of 111 data points drawn with replacement ards regression analysis identified many prognostic factors where each sample unit was replaced in the data set, such for survival after diagnosis of AIDS-related non-Hodgkin that they could be chosen subsequently in random selec- lymphoma in the HAART era, including low CD4 cell tion. Resampled data were used to generate bootstrap esti- count, stage III or IV disease, B-class symptoms, lower mates of the hazard ratio, based on the multivariable model Eastern Cooperative Oncology Group performance status, presented for the HAART era. These were determined by bone marrow and meningeal involvement, and more than 2000 iterations of such resampling. 1 extranodal site at presentation (Table 1). The originally described International Prognostic In- RESULTS dex criteria (age, disease stage, serum lactate dehydrogen- The overall duration of survival was significantly ase, performance status, and extent of extranodal disease) greater for patients in whom non-Hodgkin lymphoma was were analyzed separately to assess their prognostic signifi- diagnosed in the HAART era compared with those in cance. All factors except age were found to be of significant whom the disease was diagnosed in the pre-HAART era prognostic value; however, only 6 patients were older than (log-rank chi-square, 9.23; P ϭ 0.002) (Figure 1). Among 60 years of age. These criteria were grouped to create 4 risk the 215 patients with AIDS-related non-Hodgkin lym- categories: low, low-intermediate, high-intermediate, and phoma, the actuarial overall survival rate, including all high. International Prognostic Index risk group was a sig- causes of death, was 32% at 2 years (95% CI, 25% to nificant prognostic factor in both patients whose disease 39%) and 26% at 5 years (CI, 19% to 33%). was diagnosed in the HAART era (P Ͻ 0.001) and those Although patients whose disease was diagnosed in the in whom disease was diagnosed before HAART (P ϭ HAART era had significantly higher CD4 cell counts at 0.011). The International Prognostic Index was found to presentation (median value, 144 ϫ 106 cells/L vs. be significantly associated with overall survival for the 45 ϫ 106 cells/L; P Ͻ 0.001), better Eastern Cooperative whole cohort combined (P Ͻ 0.001). Oncology Group performance status (P ϭ 0.003), and Multivariable modeling was performed by including fewer previous AIDS-defining illnesses (P Ͻ 0.001) than the International Prognostic Index scores and excluding www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 267 Article Prognostic Index for Systemic AIDS-Related Lymphoma

Table 1. Clinicopathologic Characteristics of 111 Patients with AIDS-Related Non-Hodgkin Lymphoma Diagnosed in the Era of Highly Active Antiretroviral Therapy, and Likelihood of Death after Diagnosis, by Univariate Cox Proportional Hazards Regression

Variable in Univariate Model Patients Who Died Hazard Ratio (95% CI) †*(%) n ,(62 ؍ n) Sex Female 6 (67) 1.01 (0.98–1.03) Male 56 (55) 1.211 (0.52–2.81) CD4 cell count, by interquartile range Ͻ22 ϫ 106 cells/L 15 (82) 3.75 (1.75–8.04) 23–93 ϫ 106 cells/L 14 (70) 2.25 (1.06–4.79) 94–231 ϫ 106 cells/L 20 (54) 1.43 (0.71–2.88) Ͼ231 ϫ 106 cells/L 13 (36) 1 CD4 cell count Ͻ100 ϫ 106 cells/L 30 (76) 2.17 (1.31–3.59) Ն100 ϫ 106 cells/L 32 (44) 1 Previous AIDS diagnosis Yes 19 (64) 1.23 (0.72–2.11) No 43 (53) 1 Stage IorII 7 (27) 0.30 (0.14–0.66) III or IV 55 (65) 1 Class of symptoms A 8 (31) 0.35 (0.17–0.73) B 54 (64) 1 Eastern Cooperative Oncology Group performance status Ͻ1 14 (25) 0.20 (0.10–0.37) Ͼ1 48 (77) 1 Bone marrow involvement Yes 17 (77) 2.41 (1.36–4.26) No 45 (51) 1 Meningeal disease at diagnosis Yes 14 (82) 2.26 (1.24–4.12) No 48 (51) 1 Liver involved at presentation Yes 23 (64) 1.32 (0.79–2.20) No 39 (52) 1 >1 extranodal site at presentation Yes 32 (68) 1.76 (1.07–2.90) No 30 (47) 1 Increased serum lactate dehydrogenase level Yes 46 (64) 1.76 (0.99–3.10) No 16 (41) 1 Burkitt lymphoma Yes 11 (79) 2.15 (1.11–4.16) No 51 (53) 1 Hepatitis C diagnosed before AIDS-related lymphoma Not tested 38 (51) 0.82 (0.49–1.30) Positive 2 (50) 0.77 (0.18–3.26) Negative 22 (67) 1 Most recent hepatitis C status Not tested 50 (64) 2.30 (1.19–4.43) Positive 1 (25) 0.52 (0.07–4.05) Negative 11 (37) 1 International Prognostic Index risk group High 25 (96) 7.66 (3.11–18.85) High-intermediate 18 (60) 3.42 (1.35–8.63) Low-intermediate 13 (42) 1.90 (0.72–5.01) Low 6 (25) 1

* Percentages are derived from the full cohort of 111 patients. For example, 56 of the 62 patients who died were male, and these patients made up 55% of men with AIDS-related lymphoma diagnosed in the era of highly active antiretroviral therapy (that is, 102 of 111 patients were male). † The mean age of the patients who died was 44.2 years, SD 10.2.

268 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Prognostic Index for Systemic AIDS-Related Lymphoma Article

Table 2. Multivariable Cox Proportional Hazards Regression Model Showing Significant Independent Predictors of Death after Diagnosis of AIDS-Related Non-Hodgkin Lymphoma in the Era of Highly Active Antiretroviral Therapy*

Variable Hazard Ratio (95% CI)† P Value Hazard Ratio Bootstrap Estimate (Efron 95th-Percentile Confidence Limits) CD4 cell count Ͻ100 ϫ 106 cells/L 2.08 (1.20–3.60) 0.009 2.39 (2.32–2.47) Ն100 ϫ 106 cells/L 11

International Prognostic Index risk group High 4.88 (1.544–15.43) 0.007 6.67 (6.02–7.33) High-intermediate 2.74 (0.94–8.04) 0.066 3.40 (3.15–3.65) Low-intermediate 1.73 (0.57–5.21) Ͼ0.2 2.16 (1.20–2.32) Low 1 1

* The model includes the International Prognostic Index risk group and excludes the list of variables from which this was derived that were significant in the univariate model. The validity of the prognostic score was calculated internally on the 111 patients in whom AIDS-related non-Hodgkin lymphoma was diagnosed in the era of highly active antiretroviral therapy; average bootstrap-estimated hazard ratios are shown, with corresponding Efron 95th-percentile confidence limits (24–27). † Adjusted for age, sex, type B symptoms, bone marrow and meningeal disease at diagnosis, Burkitt lymphoma, and other variables in the model (see text). the variables from which this score is derived (age, disease these samples are presented to demonstrate the validity of stage, extranodal disease, serum lactate dehydrogenase the prognostic index. The average bootstrap-estimated haz- level, and performance status) to identify more variables ard ratios shown in Table 2 for both CD4 cell count and that would add to this model. Multivariable Cox propor- the International Prognostic Index risk groups all overlap tional hazards regression modeling for patients in whom with the predicted 95% CIs calculated by using Cox pro- disease was diagnosed since 1996 revealed only 2 indepen- portional hazards regression based on the study sample. In dent predictors of death: International Prognostic Index addition, the Efron 95th-percentile confidence limits fall risk group and CD4 cell count at presentation (Table 2). A within this interval, indicating internal validity between prognostic weighting was derived for each variable, and a predicted hazard ratios and bootstrap-estimated hazard ra- total prognostic risk score was calculated by addition of tios (25–28). Validity of predicted hazard ratios were also these weightings for both variables (Table 3). The highest confirmed in a separate analysis on the patients who re- prognostic weightings were high International Prognostic ceived a diagnosis before the era of HAART. Index score (2.9), high-intermediate International Prog- In the HAART era, a prognostic risk score greater than nostic Index score (1.84), and CD4 cell count less than 2.90 yielded a likelihood ratio for death of 7.90 (hazard 100 ϫ 106 cells/L (1.34). As with the International Prog- ratio, 1.0), whereas a prognostic risk score less than 1.0 nostic Index, the prognostic risk scores were divided into yielded a likelihood ratio of 0.23 (hazard ratio, 0.15 [95% quartiles: less than 1.0, 1.0 to 1.83, 1.84 to 2.90, and CI, 0.06 to 0.33]). A receiver-operating characteristic curve greater than 2.90. Figure 2 shows Kaplan–Meier survival analysis of this model demonstrated the sensitivity and curves for each quartile group, and Table 4 shows the specificity of the cutoff values of the prognostic risk scores likelihood ratio for death. derived from Cox proportional hazards regression coeffi- Bootstrap resampling was used to estimate the hazard cient (Figure 3). ratio of the final multivariable model (24). Robust param- eter estimates and standard errors were calculated from these models. Six hundred bootstrap samples were gener- DISCUSSION ated by using resampling with replacement, and averages of The aim of this prospective cohort study was to assess the validity of the International Prognostic Index in iden- tifying specific risk groups with AIDS-related non- Table 3. Prognostic Weightings Used To Yield the Total Hodgkin lymphoma in the HAART era and to propose Prognostic Risk Score* additional prognostic markers in these patients. The Inter- Variable Weighting national Prognostic Index (4) remains clinically useful in CD4 cell count patients with AIDS-related non-Hodgkin lymphoma that Ͻ100 ϫ 106 cells/L 1.34 was diagnosed during the HAART era. In addition, the Ͼ100 ϫ 106 cells/L 0 CD4 cell count remains an independent significant prog- ϭ International Prognostic Index risk group nostic variable (P 0.009). High 2.90 When all 215 patients with AIDS-related non- High-intermediate 1.84 Hodgkin lymphoma were combined as a single cohort, we Low-intermediate 1 Low 0 found that 3 other variables other than established Inter- national Prognostic Index risk group (P Ͻ 0.001 for a high * See Table 4 for results of application of the total prognostic risk score. International Prognostic Index score vs. other score catego- www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 269 Article Prognostic Index for Systemic AIDS-Related Lymphoma

Figure 2. Product-limit survival plot for 111 patients with systemic AIDS-related non-Hodgkin lymphoma (NHL) diagnosed in the era of highly active antiretroviral therapy.

Patients are separated into whole-cohort quartiles of prognostic risk score. All causes of death are included. P Ͻ 0.001 (log-rank chi-square test). ries) were significant independent predictors of death: pre- further analysis using this therapeutic distinction does not vious AIDS-defining illnesses (P ϭ 0.016), presence of change our results. Burkitt lymphoma (P ϭ 0.026), and CD4 cell count at The duration of survival differed significantly in the diagnosis (P Ͻ 0.001). The disappearance of previous pre- and post-HAART eras among patients with AIDS- AIDS-defining illnesses and reduced incidence of Burkitt related non-Hodgkin lymphoma (P ϭ 0.002) (Figure 1). lymphoma in the era of HAART has probably resulted After the introduction of HAART, the incidence of Kaposi from improvement in the immune status of these patients sarcoma decreased within a few years, but a concomitant as measured by the increase in CD4 cell count at presen- decrease in AIDS-related non-Hodgkin lymphoma inci- tation over these periods (1986 to 1995 vs. 1996 to the dence was not observed over a longer period, and some present; P Ͻ 0.001). Although our HAART-era data may studies reported no decrease at all (29, 30). In a prospective be confounded by the different therapies used, we have European study, the incidence of AIDS-defining illnesses previously shown no difference in the prevention of AIDS- decreased from 30.7 per 100 patient-years in 1994 to 2.5 related non-Hodgkin lymphoma (6) according to the type per 100 patient-years in 1998 (31). Only the incidence of of HAART regimen (protease inhibitor–based or non- non-Hodgkin lymphoma increased as an AIDS-defining nucleoside reverse transcriptase inhibitor–based) used, and illnesses during this time (4% in 1994 compared with 16%

Table 4. Likelihood Ratio for Death from AIDS-Related Non-Hodgkin Lymphoma in the Era of Highly Active Antiretroviral Therapy, and Sensitivity and Specificity Associated with Interquartile Range Cutoff Values

Prognostic Risk Score Patients Who Survived Patients Who Died Likelihood Ratio Hazard Ratio Sensitivity Specificity (n (%) for Death* (95% CI) (95% CI) (95% CI ,(62 ؍ n (%) (n ,(49 ؍ n) By quartile Ͻ1.00 28 (78) 8 (22) 0.23 0.15 (0.06–0.33) 1.00–1.83 14 (52) 13 (48) 0.74 0.44 (0.22–0.88) 1.84–2.90 5 (19) 21 (81) 3.32 1.17 (0.64–2.17) Ͼ2.90 2 (9) 20 (91) 7.90 1

By interquartile range cutoff values Ͻ1.0 28 8 0.87 (0.76–0.94) 0.57 (0.42–0.72) Ն1.0 21 54 Յ1.83 42 21 0.66 (0.53–0.78) 0.86 (0.73–0.94) Ͼ1.83 7 41 Յ2.90 47 42 0.32 (0.21–0.45) 0.96 (0.86–1.00) Ͼ2.90 2 20

* A likelihood ratio Ͼ 1 indicates greater risk for death.

270 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Prognostic Index for Systemic AIDS-Related Lymphoma Article

Figure 3. Receiver-operating characteristic curve showing sensitivity and false-positive error rate of mortality using the quartile cutoff values for the prognostic risk score, derived from the Cox proportional hazards regression coefficient.

Data from 111 patients are included. The sensitivity gives the degree of certainty that patients who fall in a particular prognostic risk score group will not die. The diagonal line displays ties, and the points on the curve refer to the sensitivity and 1 Ϫ specificity (the risk for false-positive results). 1 Ϫ specificity may also be referred to as the type I error rate. This demonstrates that the cutoffs established for risk scores are predictive of mortality. in 1998). In a meta-analysis of 23 prospective cohort stud- less than 1.0 (the lowest quartile) yields a likelihood mor- ies involving 47 936 patients living in established market tality ratio of 0.23 and a hazard ratio of 0.15 (CI, 0.06 to economies, a decrease in the incidence of non-Hodgkin 0.33). These values correspond to a likelihood ratio for lymphoma was reported from 6.2 cases per 1000 person- death of 7.90 and a corresponding hazard ratio of 1.0 (Ta- years in 1992 to 3.6 cases per 1000 patient-years in 1999 ble 4) among patients in the highest score quartile. In our (29). The epidemiologic features of Kaposi sarcoma and cohort, 70 patients in whom AIDS-related non-Hodgkin non-Hodgkin lymphoma may differ because improvement lymphoma was diagnosed in the HAART era had CD4 cell in immune function due to HAART prevents Kaposi sar- counts greater than 100 ϫ 106 cells/L at the time of diag- coma during short-term therapy, but perhaps a longer du- nosis. Of these patients, 14 were in the low International ration of therapy is required to prevent non-Hodgkin lym- Prognostic Index risk group, 21 were in the low-interme- phoma. In support of this hypothesis, we observed that diate risk group, 20 were in the high-intermediate risk increased B-cell counts protected against Kaposi sarcoma group, and 15 were in the high-risk group. The 2-year (32), and Kaposi sarcoma lesions may resolve with survival rates for each group were 83% (CI, 60% to HAART alone (33–35); neither of these scenarios has been 100%), 70% (CI, 49% to 90%), 42% (CI, 20% to 65%), the case with non-Hodgkin lymphoma. These data suggest and 7% (CI, 0% to 20%), respectively. In a historical that as well as the reported decrease in incidence of non- comparison, the first 3 values do not significantly differ Hodgkin lymphoma, patients also have improved survival, from the values of 88%, 74%, and 62% derived from use most likely because of better overall immune status. of the original International Prognostic Index. In contrast, The International Prognostic Index was found to be however, HIV-positive patients in the high-risk group have useful in 69 patients with AIDS-related non-Hodgkin lym- a significantly worse 2-year overall survival rate than their phoma (17). Like that study, ours is limited by a small HIV-negative comparators (20% vs. 7%). sample; however, we can propose a new prognostic scoring Previous prognostic work has clarified that available scheme designed for patients in whom AIDS-related non- clinical information can provide a foundation for long- Hodgkin lymphoma was diagnosed in the HAART era. term survival estimates, particularly when combined with a This scoring method is based on International Prognostic physician’s clinical estimate (36). Use of the International Index risk group and CD4 cell count only (Table 2) and Prognostic Index with CD4 cell count provides useful in- was internally validated by bootstrap resampling that formation for therapy and suggests that patients with a shows tight and overlapping confidence limits around the poor prognosis for AIDS-related non-Hodgkin lymphoma resampled and original hazard ratios. A prognostic score should be considered for additional treatment. This as- www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 271 Article Prognostic Index for Systemic AIDS-Related Lymphoma sumes increased importance in light of the data suggesting sone and highly active antiretroviral therapy for patients with acquired immuno- that high-dose chemotherapy plus autologous hematopoi- deficiency syndrome-related Burkitt lymphoma/leukemia. . 2002; 94:1492-9. [PMID: 11920506] etic stem-cell transplantation is feasible in patients with 10. Ratner L, Lee J, Tang S, Redden D, Hamzeh F, Herndier B, et al. Che- AIDS-related non-Hodgkin lymphoma in terms of harvest- motherapy for human immunodeficiency virus-associated non-Hodgkin’s lym- ing, engraftment, adverse events, and control of HIV (37). phoma in combination with highly active antiretroviral therapy. J Clin Oncol. The advent of HAART has dramatically reduced morbidity 2001;19:2171-8. [PMID: 11304769] 11. Thirlwell C, Stebbing J, Nelson M, Gazzard B, Bower M. CDE chemo- and mortality due to HIV infection (38). Because HIV- therapy plus HAART for AIDS related non-Hodgkin’s lymphoma [Abstract]. In: infected patients are living longer, development of im- 6th International Congress on Drug Therapy in HIV Infection, 17-21 November proved therapies for cancer and lymphoma that are based 2002, Glasgow, United Kingdom. London: Mediscript; 2002:94. on accurate prognostic information has assumed greater 12. Navarro JT, Ribera JM, Oriol A, Vaquero M, Romeu J, Batlle M, et al. Influence of highly active anti-retroviral therapy on response to treatment and clinical importance. Patients with AIDS-related non- survival in patients with acquired immunodeficiency syndrome-related non- Hodgkin lymphoma who are in a poor prognostic category Hodgkin’s lymphoma treated with cyclophosphamide, hydroxydoxorubicin, vin- should be considered for a clinical trial of newer ap- cristine and prednisone. Br J Haematol. 2001;112:909-15. [PMID: 11298585] proaches. Since the advent of HAART, being in a good 13. Sparano JA, Lee S, Chen MG, Nazeer T, Einzig A, Ambinder RF, et al. prognostic quartile appropriately defines patients that can Phase II trial of infusional cyclophosphamide, doxorubicin, and etoposide in patients with HIV-associated non-Hodgkin’s lymphoma: an Eastern Cooperative benefit from standard curative-intent therapy. Oncology Group Trial (E1494). J Clin Oncol. 2004;22:1491-500. [PMID: 15084622] From The Chelsea and Westminster , London, United King- 14. Little R, Pearson D, Steinberg S, Elwood P, Yarchoan R, Wilson W. dom. Dose-adjusted EPOCH chemotherapy in previously untreated HIV-associated non-Hodgkin’s lymphoma. Proc Am Soc Clin Oncol. 1999;18:10a. 15. Little RF, Pittaluga S, Grant N, Steinberg SM, Kavlick MF, Mitsuya H, et al. Highly effective treatment of acquired immunodeficiency syndrome-related Potential Financial Conflicts of Interest: None disclosed. lymphoma with dose-adjusted EPOCH: impact of antiretroviral therapy suspen- sion and tumor biology. Blood. 2003;101:4653-9. [PMID: 12609827] 16. Navarro JT, Ribera JM, Oriol A, Vaquero M, Romeu J, Batlle M, et al. Requests for Single Reprints: Mark Bower, MA, PhD, or Justin Steb- International prognostic index is the best prognostic factor for survival in patients bing, MA, PhD, The Chelsea and Westminster Hospital, 369 Fulham with AIDS-related non-Hodgkin’s lymphoma treated with CHOP. A multivari- Road, London SW10 9NH, United Kingdom; e-mail, m.bower@ ate study of 46 patients. Haematologica. 1998;83:508-13. [PMID: 9676023] imperial.ac.uk or [email protected]. 17. Rossi G, Donisi A, Casari S, Re A, Cadeo G, Carosi G. The International Prognostic Index can be used as a guide to treatment decisions regarding patients Current author addresses and author contributions are available at www with human immunodeficiency virus-related systemic non-Hodgkin lymphoma. Cancer. 1999;86:2391-7. [PMID: 10590382] .annals.org. 18. Yeni PG, Hammer SM, Hirsch MS, Saag MS, Schechter M, Carpenter CC, et al. Treatment for adult HIV infection: 2004 recommendations of the International AIDS Society-USA Panel. JAMA. 2004;292:251-65. [PMID: 15249575] 19. Bower M, McCall-Peat N, Ryan N, Davies L, Young AM, Gupta S, et al. References Protease inhibitors potentiate chemotherapy-induced neutropenia. Blood. 2004; 104:2943-6. [PMID: 15238428] 1. Lossos IS, Czerwinski DK, Alizadeh AA, Wechser MA, Tibshirani R, Bot- 20. Kaplan E, Meier P. Nonparametric estimation from incomplete observa- stein D, et al. Prediction of survival in diffuse large-B-cell lymphoma based on tions. J Am Stat Assoc. 1958;53:457-81. the expression of six genes. N Engl J Med. 2004;350:1828-37. [PMID: 21. Peto R, Pike MC, Armitage P, Breslow NE, Cox DR, Howard SV, et al. 15115829] Design and analysis of randomized clinical trials requiring prolonged observation 2. Benzel JE. International conference on leukemia-lymphoma. Ann Arbor, of each patient. II. Analysis and examples. Br J Cancer. 1977;35:1-39. [PMID: Michigan, October, 1967. Del Med J. 1968;40:78-81. [PMID: 5644827] 831755] 3. Rosenberg SA. Validity of the Ann Arbor staging classification for the non- 22. Harrell FE Jr, Lee KL, Califf RM, Pryor DB, Rosati RA. Regression mod- Hodgkin’s lymphomas. Cancer Treat Rep. 1977;61:1023-7. [PMID: 902260] elling strategies for improved prognostic prediction. Stat Med. 1984;3:143-52. 4. A predictive model for aggressive non-Hodgkin’s lymphoma. The Interna- tional Non-Hodgkin’s Lymphoma Prognostic Factors Project. N Engl J Med. [PMID: 6463451] 1993;329:987-94. [PMID: 8141877] 23. Steyerberg EW, Eijkemans MJ, Harrell FE Jr, Habbema JD. Prognostic 5. Stebbing J, Bower M. What can oncologists learn from HIV? Lancet Oncol. modeling with logistic regression analysis: in search of a sensible strategy in small 2003;4:438-45. [PMID: 12850195] data sets. Med Decis Making. 2001;21:45-56. [PMID: 11206946] 6. Stebbing J, Gazzard B, Mandalia S, Teague A, Waterston A, Marvin V, et al. 24. Steyerberg EW, Harrell FE Jr, Borsboom GJ, Eijkemans MJ, Vergouwe Y, Antiretroviral treatment regimens and immune parameters in the prevention of Habbema JD. Internal validation of predictive models: efficiency of some proce- systemic AIDS-related non-Hodgkin’s lymphoma. J Clin Oncol. 2004;22:2177- dures for logistic regression analysis. J Clin Epidemiol. 2001;54:774-81. [PMID: 83. [PMID: 15169806] 11470385] 7. Stebbing J, Marvin V, Bower M. The evidence-based treatment of AIDS- 25. Efron B, Gong G. A leisurely look at the bootstrap, the jack-knife, and related non-Hodgkin’s lymphoma. Cancer Treat Rev. 2004;30:249-53. [PMID: cross-validation. The American Statistician. 1983;37:36-48. 15059648] 26. Efron B, Tibshirani R. Bootstrap methods for standard errors, confidence 8. Vaccher E, Spina M, di Gennaro G, Talamini R, Nasti G, Schioppa O, et al. intervals, and other measures of statistical accuracy. Statistical Science. 1986;1:54- Concomitant cyclophosphamide, doxorubicin, vincristine, and prednisone che- 75. motherapy plus highly active antiretroviral therapy in patients with human im- 27. Efron B, Tibshirani R. An Introduction to the Bootstrap. Number 57 in munodeficiency virus-related, non-Hodgkin lymphoma. Cancer. 2001;91:155- Monographs on Statistics and Applied Probability New York: Chapman and 63. [PMID: 11148572] Hall; 1993. 9. Cortes J, Thomas D, Rios A, Koller C, O’Brien S, Jeha S, et al. Hyper- 28. Efron B, Halloran E, Holmes S. Bootstrap confidence levels for phylogenetic fractionated cyclophosphamide, vincristine, doxorubicin, and dexametha- trees. Proc Natl Acad SciUSA.1996;93:13429-34. [PMID: 8917608]

272 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Prognostic Index for Systemic AIDS-Related Lymphoma Article

29. International Collaboration on HIV and Cancer. Highly active antiretroviral resolution of Kaposi’s sarcoma? J Antimicrob Chemother. 2003;51:1095-8. therapy and incidence of cancer in human immunodeficiency virus-infected [PMID: 12668573] adults. J Natl Cancer Inst. 2000;92:1823-30. [PMID: 11078759] 35. Murdaca G, Campelli A, Setti M, Indiveri F, Puppo F. Complete remission 30. Matthews GV, Bower M, Mandalia S, Powles T, Nelson MR, Gazzard BG. of AIDS/Kaposi’s sarcoma after treatment with a combination of two nucleoside Changes in acquired immunodeficiency syndrome-related lymphoma since the reverse transcriptase inhibitors and one non-nucleoside reverse transcriptase in- introduction of highly active antiretroviral therapy. Blood. 2000;96:2730-4. hibitor [Letter]. AIDS. 2002;16:304-5. [PMID: 11807324] [PMID: 11023505] 36. Knaus WA, Harrell FE Jr, Lynn J, Goldman L, Phillips RS, Connors AF Jr, 31. Mocroft A, Katlama C, Johnson AM, Pradier C, Antunes F, Mulcahy F, et et al. The SUPPORT prognostic model. Objective estimates of survival for seri- al. AIDS across Europe, 1994-98: the EuroSIDA study. Lancet. 2000;356:291-6. ously ill hospitalized adults. Study to understand prognoses and preferences for [PMID: 11071184] outcomes and risks of treatments. Ann Intern Med. 1995;122:191-203. [PMID: 32. Stebbing J, Gazzard B, Newsom-Davis T, Nelson M, Patterson S, Gotch F, 7810938] et al. Nadir counts are significantly correlated with the risk of Kaposi’s 37. Gabarre J, Marcelin AG, Azar N, Choquet S, Levy V, Levy Y, et al. High- sarcoma. Int J Cancer. 2004;108:473-4. [PMID: 14648716] dose therapy plus autologous transplantation for human 33. Jacobson LP, Yamashita TE, Detels R, Margolick JB, Chmiel JS, Kingsley immunodeficiency virus (HIV)-related lymphoma: results and impact on HIV LA, et al. Impact of potent antiretroviral therapy on the incidence of Kaposi’s disease. Haematologica. 2004;89:1100-8. [PMID: 15377471] sarcoma and non-Hodgkin’s lymphomas among HIV-1-infected individuals. 38. Palella FJ Jr, Delaney KM, Moorman AC, Loveless MO, Fuhrer J, Satten Multicenter AIDS Cohort Study. J Acquir Immune Defic Syndr. 1999;21(Suppl GA, et al. Declining morbidity and mortality among patients with advanced 1):S34-41. [PMID: 10430217] human immunodeficiency virus infection. HIV Outpatient Study Investigators. 34. Stebbing J, Portsmouth S, Gazzard B. How does HAART lead to the N Engl J Med. 1998;338:853-60. [PMID: 9516219]

www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 273 Current Author Addresses: Drs. Bower, Gazzard, Mandalia, and Nel- Analysis and interpretation of the data: M. Bower, B. Gazzard, T. New- son: The Chelsea and Westminster Hospital, 369 Fulham Road, London som-Davis, C. Thirwell, T. Dhillon, J. Stebbing. SW10 9NH, United Kingdom. Drafting of the article: M. Bower, A. Gaya, J. Stebbing. Drs. Newsom-Davis, Dhillon, Young, and Powles: Charing Cross and Critical revision of the article for important intellectual content: Hammersmith NHS Trust, London W6 8RF, United King- M. Bower, A.M. Young, T. Powles, A. Gaya, J. Stebbing. dom. Final approval of the article: M. Bower, B. Gazzard, C. Thirwell, Dr. Thirlwell: Cancer Research UK, Lincoln’s Inn Fields, London T. Dhillon, A.M. Young, A. Gaya, J. Stebbing. WC2A 3PX, United Kingdom. Provision of study materials or patients: M. Bower, A.M. Young, Dr. Gaya: Northwick Park Hospital, Watford Road, London HA1 3UJ, J. Stebbing. United Kingdom. Statistical expertise: M. Bower, S. Mandalia, J. Stebbing. Dr. Stebbing: St. Bartholomew’s Hospital, Bodley Scott Chemotherapy Obtaining of funding: M. Bower, B. Gazzard, J. Stebbing. Unit, East Wing, West Smithfield, London EC1A 7BE, United King- Administrative, technical, or logistic support: M. Bower, B. Gazzard, dom. A.M. Young, J. Stebbing. Author Contributions: Conception and design: M. Bower, B. Gazzard, Collection and assembly of data: M. Bower, B. Gazzard, T. Dhillon, S. Mandalia, M. Nelson, J. Stebbing. A.M. Young, J. Stebbing.

www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 W-69 Improving Patient Care Quality of Care Is Associated with Survival in Vulnerable Older Patients Takahiro Higashi, MD, PhD; Paul G. Shekelle, MD, PhD; John L. Adams, PhD; Caren J. Kamberg, MSPH; Carol P. Roth, RN, MPH; David H. Solomon, MD; David B. Reuben, MD; Lillian Chiang, MD; Catherine H. MacLean, MD, PhD; John T. Chang, MD, MPH; Roy T. Young, MD; Debra M. Saliba, MD, MPH; and Neil S. Wenger, MD, MPH

Background: Although assessment of the quality of medical average, 53% of the care processes prescribed in quality indica- care often relies on measures of process of care, the linkage tors (range, 27% to 88%). Eighty-six (23%) persons died during between performance of these process measures during usual the 3-year follow-up. There was a graded positive relationship clinical care and subsequent patient outcomes is unclear. between quality score and 3-year survival. After adjustment for sex, health status, and health service use, quality score was not Objective: To examine the link between the quality of care that associated with mortality for the first 500 days, but a higher patients received and their survival. quality score was associated with lower mortality after 500 days Design: Observational cohort study. (hazard ratio, 0.64 [95% CI, 0.49 to 0.84] for a 10% higher quality score). Setting: Two managed care organizations. Limitations: The observational design limits causal inference Patients: Community-dwelling high-risk patients 65 years of age regarding the effect of quality of care on survival. or older who were continuously enrolled in the managed care organizations from 1 July 1998 to 31 July 1999. Conclusions: Better performance on process quality measures is strongly associated with better survival among community-dwell- Measurements: Quality of care received by patients (as mea- ing vulnerable older adults. sured by a set of quality indicators covering 22 clinical conditions) and their survival over the following 3 years.

Results: The 372 vulnerable older patients were eligible for a Ann Intern Med. 2005;143:274-281. www.annals.org mean of 21 quality indicators (range, 8 to 54) and received, on For author affiliations, see end of text.

s clinicians, the public, and health systems become patients has hindered the acceptance of quality indicators Amore aware that many Americans do not receive nec- as a way to measure and improve health outcomes (8). essary care, the importance of measuring and improving The Assessing Care of Vulnerable Elders (ACOVE) quality of care has gained increasing attention (1–3). Al- project developed a set of process quality criteria that were though quality of care can theoretically be measured by judged by clinical experts to improve patient outcomes on outcomes (what happens to patients), process (what pro- the basis of clinical evidence and professional opinion (9– viders do) is often preferred (3–5) because process is under 11). Combined with mortality information available relatively greater control of providers, needs a shorter time through the National Death Index, our study evaluated the frame, can directly inform improvement, and may not re- process–outcome relationship. While the development quire statistical adjustment for severity of illness (6, 7). method conferred content validity on the process measures, Typically, process measures evaluate the proportion of eli- we aimed to assess the predictive validity of the quality gible patients who receive care as recommended (for exam- measurement system by examining the relationship be- ple, the proportion of patients Ն 65 years of age receiving tween the quality of care received by sampled participants pneumococcal vaccine). and their subsequent survival. To be a meaningful measure of quality, a process of care must be related to improved patient outcomes. For See also: many quality indicators, this relationship is based on evi- dence of efficacy from randomized, controlled trials, usu- Print ally among a select patient population. However, the rela- Editors’ Notes ...... 275 tionship between performance on process of care quality Editorial comment...... 305 indicators and better health outcomes remains a largely Summary for Patients...... I-33 untested assumption for general populations of patients Web-Only receiving care in community settings. The lack of a dem- Appendix Table onstrated relationship between performance on process Conversion of figures and tables into slides quality measures and outcome advantage in a cohort of

Improving Patient Care is a special section within Annals supported in part by the U.S. Department of Health and Human Services (HHS) Agency for Healthcare Research and Quality (AHRQ). The opinions expressed in this article are those of the authors and do not represent the position or endorsement of AHRQ or HHS.

274 © 2005 American College of Physicians Quality of Care–Survival Relationship in Vulnerable Older Patients Improving Patient Care

METHODS The ACOVE Project Context The ACOVE project developed and implemented a Quality-of-care evaluation often focuses on how often set of quality indicators that focuses on process of care for patients receive certain tests or treatments. Theoretically, clinical conditions important in the care of vulnerable the content of care should predict patient survival, but the older patients. Details of the methods of selecting condi- evidence is inconclusive. tions and developing quality indicators have been de- Contribution scribed in previous reports (9, 10, 12). We selected quality This study used 207 criteria to assess good care in 372 indicators by using systematic reviews of the medical liter- vulnerable elderly patients. When care did not meet these ature followed by deliberations by several panels of clinical standards, patients were more likely to die during the 3 experts using formal consensus methods to assess the valid- years of follow-up. ity of quality indicators. This process resulted in 236 qual- ity indicators covering 22 clinical areas (continuity of care, Implications dementia, depression, diabetes mellitus, end-of-life care, In vulnerable older patients, the content of care is associ- falls and mobility problems, hearing loss, heart failure, hos- ated with mortality. This finding supports the use of pro- pital care, hypertension, ischemic heart disease, malnutri- cess measures in the evaluation of quality of care and tion, management, osteoarthritis, osteoporosis, shows that good care may prolong life. pain management, pneumonia, pressure ulcer, screening and prevention, stroke and atrial fibrillation, urinary in- –The Editors continence, and vision care) across the continuum of care, patients as vulnerable. Among them, 420 (88%) patients including prevention, diagnosis, treatment, and follow-up. consented to participate in the study and 372 (78%) pa- Each quality indicator contains an “if” clause that defines tients had medical records for the 13-month period from 1 the patient who is eligible to receive it and a “then” clause July 1998 to 31 July 1999 that were able to be abstracted. that describes what care is recommended (for example: “If We collected all participants’ medical records, includ- a vulnerable elder has had a myocardial infarction, then he ing those for inpatient care, outpatient care, nursing home or she should be offered a ␤-blocker”). If the medical care, home care, and mental health care. Trained nurses record describes a contraindication to the recommended abstracted charts to apply quality indicators. A senior nurse care, the patient is not eligible for the quality indicator. reviewer assessed completed abstractions, and physician Furthermore, we explicitly defined certain indicators as be- overreaders reviewed them for clinical assessment. We eval- ing not applicable, and therefore not included, when as- uated inter-rater reliability by reabstraction of 10% of the sessing the care of patients with advanced dementia or poor medical records, which contained 698 quality indicators. prognosis (13). Agreement was 97% for quality indicator eligibility and We applied the ACOVE quality indicators to a sample 95% for overall quality score. We collected patient charac- of vulnerable older patients in 2 large managed care orga- teristics, including age, sex, cognitive function measured by nizations, 1 in the northeastern United States and the the Blessed Orientation–Memory–Concentration test (15), other in the southwestern United States (11). Each man- and mental health score derived from Medical Outcomes aged care plan had more than 20 000 senior members and Study Short Form-36 items (16), at the time of the recruit- contracted with a network of providers for delivery of care. ment telephone interview. The RAND institutional review Eligibility criteria included continuous enrollment in the board approved the study protocol. managed care organization with no out-of-network care Among the 236 ACOVE quality indicators, 207 could during the 13-month study period and no active treatment be implemented in the field trial either by medical record for malignant conditions except for nonmelanoma skin (183 indicators) or patient interview (24 indicators). Be- cancer. We identified vulnerable older persons by tele- cause some patients died before the interview was con- phone interview using the Vulnerable Elders Survey-13 ducted, we used only quality indicators for which informa- (VES-13) (14). The VES-13 is a 13-item questionnaire tion was available in medical records. Among these, 160 that produces a vulnerability score ranging from 0 to 10 quality indicators had at least 1 eligible patient; 43 focused based on age, self-reported health, and function. Patients on prevention, 42 on diagnosis, 47 on treatment, and 28 with scores of 3 or higher are at 4 times the risk for death on follow-up care. These 160 quality indicators covered all or functional decline over the next 2 years and are therefore 22 conditions. The Appendix Table (available at www defined as vulnerable. We excluded non–English-language .annals.org) contains the list of quality indicators used in speakers because interviews were available only in English. our report, the number of eligible patients, and the pass Among 3207 community-dwelling patients 65 years of age rate for each indicator. and older who were randomly selected from the 2 managed Statistical Analysis care plans, we conducted screening interviews with 2278 We calculated quality scores for each patient on the patients (9% through proxies) and identified 475 (21%) basis of the percentage of ACOVE quality indicators for www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 275 Improving Patient Care Quality of Care–Survival Relationship in Vulnerable Older Patients which an eligible patient received recommended care. We quality score and patient sickness level, represented by pa- obtained death, date, and cause-of-death data for ACOVE tient age and VES-13 score. For this examination, we used participants from the National Death Index during 3 years correlation coefficients, as well as a comparison of mean after the quality measurement period (from August 1999 quality scores between younger (Ͻ85 years of age) versus to September 2002). older (Ն85 years of age) patients and between healthier We used both unadjusted and adjusted analyses to ex- (VES-13 score Ͻ 7) versus sicker (VES-13 score Ն 7) pa- amine the link between patient survival and quality score. tients. Furthermore, we compared quality for the 39 pa- For the unadjusted analysis, we first divided the sample in tients identified as having advanced dementia, documented half on the basis of quality score (that is, Ն median and poor prognosis, or preferences not to receive aggressive care Ͻ median) and examined the difference in survival curves (13) versus the remaining 333 patients in the sample. between patients with higher quality and patients with Sensitivity Analyses lower quality by using the log-rank test. Second, we calcu- Since the main analysis defined quality score as a sim- lated survival for 10 equal intervals of quality score from ple percentage of the recommended care received, we con- the lowest quality score to the highest quality score in the ducted 2 sensitivity analyses. The first sensitivity analysis sample and graphically assessed the graded relationship be- repeated the main analysis by using weights proportionate tween quality score and survival. to the number of quality indicators for which patients were We used the Cox proportional hazards survival model eligible. These weights reflect the stability of quality scores in adjusted analyses. Because the proportional hazards as- by placing greater emphasis on scores calculated from more sumption for the multivariate survival analysis did not hold care processes and reducing the effect of unstable quality for the entire observation period, we used a piecewise scores, making it less likely to find a relationship by model that allowed the coefficients for quality to vary be- chance. The second sensitivity analysis aimed to adjust for tween 500 days or less and more than 500 days, as sug- differences in the level of difficulty satisfying individual gested by the Kaplan–Meier survival curve in the unad- quality indicators by creating an alternative quality score by justed analysis. Covariates included sex, VES-13 score subtracting from each person’s score the population mean (including age), mental health, number of hospitalizations score for the set of quality indicators for which the patient and office visits during the quality measurement period, was eligible. This alternative score represents the quality and number of conditions that patients had during the above or below the average score of the population eligible quality measurement period among 13 comorbid condi- for the set of quality indicators for which the patient was tions (dementia, depression, diabetes mellitus, heart failure, eligible. hypertension, ischemic heart disease, osteoarthritis, osteo- porosis, pressure ulcer, atrial fibrillation, urinary inconti- Role of a Potential Omitted Confounder nence, chronic obstructive pulmonary disease, and chronic We performed an additional sensitivity analysis by us- renal failure). The mental health score ranged from 1 to 6, ing a simulation technique to assess the potential effects of and we created 3 categories on the basis of the score (Ͻ2, an omitted confounder variable. We assumed the omitted very good; 2 to 3, good; and Ͼ3, fair). An indicator vari- variable to be binary and generated it to correlate with able designated patients who were not interviewed for both death and quality of care (similar results are obtained mental health items because of cognitive impairment. for positive correlations with death and negative correla- To further examine a plausible mechanism for the tions with quality). We selected values to illustrate the quality–survival link, we examined the relationship be- magnitude of the correlations required to eliminate the tween survival and high-prevalence individual quality indi- quality effect on survival. We conducted statistical analyses cators. For the individual quality indicators, we calculated by using Stata, version 8.2 (Stata Corp., College Station, the relative risk for death over the 3 years for patients who Texas). were eligible and received recommended care (that is, pass) Role of the Funding Source in comparison with patients who were eligible but did not This study was supported by a contract with Pfizer receive the recommended care (that is, fail). Because qual- Inc. The funding source had no role in the design, analysis, ity indicators with few eligible patients cannot produce or interpretation of the study or in the decision to submit reliable estimates of this ratio, we evaluated only those the manuscript for publication. quality indicators for which at least 50 patients received the recommended care and at least 50 patients did not receive the care. In addition, we compared cause of death between RESULTS patients who received high-quality care and those who re- Sample Characteristics ceived low-quality care. Among 420 vulnerable older patients who consented We also performed analyses to evaluate whether pa- to participate in the study, 372 had available medical tients who were sicker received lower-quality care, perhaps records for quality-of-care measurement (11). They had a because they were perceived to be on an immutable trajec- mean age of 81 years; 64% were women, and the mean tory toward death, by studying the relationship between vulnerability score was 5.3. During the 3-year period from

276 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Quality of Care–Survival Relationship in Vulnerable Older Patients Improving Patient Care quality score of 62%, SD 7% (range, 52% to 88%) and *(372 ؍ Table 1. Description of the Study Sample (n participants in the lower half had a mean quality score of Characteristic Value 44%, SD 6% (range, 22% to 52%). Figure 1 shows the Demographic Kaplan–Meier survival curves for the upper and lower Mean age, y 80.6, SD 6.8 Women, % 64 quality half samples. Participants receiving higher-quality High school graduate, % 59 care had significantly lower mortality (18%) than patients receiving lower-quality care (28%) (log-rank test; P ϭ Clinical Mean VES-13 score 5.3, SD 2.3 0.02). Furthermore, when patients were grouped into 10 Mean self-reported health (5-point scale) 2.6 equal intervals of quality score, there was a graded relation- Mean activities of daily living disabilities 0.49 ship between quality score and 3-year survival (Figure 2). (6-point scale) Mean instrumental activities of daily living 1.2 The unadjusted piecewise Cox model showed that the haz- disabilities (6-point scale) ard ratio associated with a 10% quality score increase was Mental health, %† Very good 77 1.16 (95% CI, 0.86 to 1.56) for up to 500 days and 0.68 Good 16 (CI, 0.54 to 0.87) after 500 days. Fair 7 After adjustment for sex, vulnerability score (including Cognitive impairment, %‡37 Mean hospitalizations, n§ 0.28, SD 0.72 age and functional status), mental health, number of hos- Mean office visits, n§ 8.7, SD 5.7 pitalizations and outpatient visits, and number of condi- tions by using the piecewise Cox proportional hazards Comorbid conditions, % Dementia 8 model, higher quality was not associated with mortality Depression 16 within 500 days after the quality measurement period (haz- Diabetes mellitus 24 ard ratio, 1.19 [CI, 0.86 to 1.64] for a 10% higher quality Heart failure 15 Hypertension 61 score) but was significantly associated with lower mortality Ischemic heart disease 31 after 500 days (hazard ratio, 0.64 [CI, 0.49 to 0.84] for a Osteoarthritis 24 10% higher quality score). Osteoporosis 12 Pressure ulcer 2 Analyses of Possible Mechanisms of the Atrial fibrillation 13 Quality–Survival Association Urinary incontinence 9 COPD or related disorders 25 Nine quality indicators had at least 50 patients who Chronic renal failure 7 passed and 50 patients who failed. For 8 of these 9 quality indicators, patients who received recommended care were * COPD ϭ chronic obstructive pulmonary disease; VES-13 ϭ Vulnerable Elders Survey-13. less likely to die than those who did not receive such care † Information was available for 285 patients. Mental health score on a scale of 1 to (Table 2). The National Death Index included the cause of 6 points: very good, Ͻ2 points; good, 2–3 points; or fair, Ͼ3 points. For 87 patients, this information was not available because proxies answered the screening death for 78 patients. Twenty-one (27%) patients died of interview questions. ‡Cognitive impairment was defined as either proxy respondent needed for inter- view or patient scored 17 points or less on the Blessed Orientation–Memory– Concentration test. Figure 1. Kaplan–Meier survival curves for patients grouped § During 13-mo quality measurement period. into the upper and lower half of quality.

August 1999 to September 2002, 86 (23%) patients died. Overall, the 372 participants had a mean quality score of 53%, SD 11% (range, 22% to 88%), indicating that they received, on average, 53% of the care recommended in the ACOVE quality indicators. Each participant was, on aver- age, eligible for 21 quality indicators (9 prevention, 3 di- agnosis, 6 treatment, and 2 follow-up [values do not add up to 21 because of rounding]). Table 1 summarizes other patient characteristics. Among 48 participants for whom medical records could not be used for quality-of-care evaluation, 28 partic- ipants received no care during the 13-month observation period and 20 patients had incomplete or illegible medical records. Overall, the 3-year mortality rate for these 48 par- ticipants was 23%. Quality–Survival Association Patients in the upper half of quality received a mean quality score of When we split the sample in half on the basis of qual- 62%, and patients in the lower half had a mean quality score of 44%. ity score, participants in the upper half received a mean Survival curves differed by the log-rank test (P ϭ 0.02). www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 277 Improving Patient Care Quality of Care–Survival Relationship in Vulnerable Older Patients

Figure 2. Three-year survival for 10 equal intervals of quality Role of a Potential Omitted Confounder score. Table 3 presents the revised estimates and 95% CIs for the effect of quality on survival more than 500 days after the observation period when we incorporated a hypo- thetical confounder into the proportional hazards model. We selected these particular values to illustrate the magni- tude of the correlations that would be required to eliminate the quality effect. Substantial correlations of a potential confounder with quality of care and mortality would be required to eliminate the finding of a reduction in mortal- ity associated with quality of care. Assessment of Alternative Hypothesis We recognized that the relationship between higher quality score and survival could be explained by an alter- native hypothesis: Physicians provide sicker patients with less care because they presume that such patients are likely to die anyway. This would result in lower-quality care be- Relationship of survival to quality captured in 10 equal intervals (r ϭ ing related to higher mortality. To investigate this possibil- 0.77). ity, we studied the relationship between quality score and patient sickness level, represented by patient age and VES-13 score, which was significantly related to death over cardiovascular diseases; 16 (21%) died of respiratory dis- 3 years in the multivariate analysis (relative risk for death eases; 9 (12%) died of malignant ; and 9 (12%) for each point of the scale was 1.18; P Ͻ 0.001). We found died of neurologic disorders, including dementia. The no relationship between quality score and patient age or cause of death did not differ between patients in the upper VES-13 score: Pearson correlation coefficients were Ϫ0.03 half of quality of care and those in the lower half. and Ϫ0.01, respectively, and graphical assessment showed Sensitivity Analyses no relationship (Figure 3). The mean quality score did not Both sensitivity analyses produced results similar to significantly differ between older participants and younger the main analysis. The analysis using weights to reduce the participants (54% for those 65 to 84 years of age vs. 53% effect of unstable quality scores yielded a hazard ratio of for those Ն85 years of age; P Ͼ 0.2) or between healthier 1.15 (CI, 0.79 to 1.68) within 500 days and 0.61 (CI, 0.45 patients and sicker patients (that is, 54% for those with a to 0.81) after 500 days for a 10% increase in quality score. VES-13 score Ͻ 7 vs. 53% for those with a VES-13 The analysis in which quality scores were adjusted for the score Ͼ 7; P Ͼ 0.2). Mean quality score for the 39 patients difficulty of passing each patient’s set of quality indicators who had advanced dementia or poor prognosis or who provided a hazard ratio of 1.23 (CI, 0.85 to 1.76) within declined aggressive care was similar to that for the remain- 500 days and 0.64 (CI, 0.49 to 0.84) after 500 days. ing 333 patients (55% vs. 53%, respectively; P Ͼ 0.2).

Table 2. Relative Risk for Death When Receiving Recommended Care versus Not Receiving Recommended Care by Individual Quality Indicators

Quality Indicator Care Eligible Mortality if Received Mortality if Did Not Receive Relative Risk for Process Patients, Recommended Care, % Recommended Care, % Death Pass/Fail n (95% CI)* Influenza vaccine 372 21 27 0.78 (0.54–1.14) Ask about falls annually 372 26 22 1.18 (0.78–1.77) Pneumococcal vaccine 370 12 27 0.46 (0.26–0.79) Annual evaluation of 363 21 25 0.87 (0.57–1.31) urinary incontinence Weight measurement at 355 19 25 0.74 (0.49–1.11) every visit Follow-up of 189 23 26 0.89 (0.53–1.50) in outpatient setting Document response to 180 20 32 0.62 (0.37–1.04) drug therapy Cognitive screen at new 130 24 31 0.77 (0.43–1.35) evaluation Targeted physical 123 16 18 0.87 (0.40–1.90) examination for pain

* If providing recommended care is associated with reduced mortality, relative risk is Ͻ 1.0.

278 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Quality of Care–Survival Relationship in Vulnerable Older Patients Improving Patient Care

DISCUSSION Table 3. Correlation of the Potential Omitted Confounder with We found that better quality of care provided to com- Death and Quality That Would Be Required To Eliminate the munity-dwelling vulnerable older persons was associated Observed Quality–Survival Relationship with higher 3-year survival. The relationship between qual- Correlation of Omitted Correlation of Omitted Hazard Ratio for ity of care and survival was robust to analysis in several Variable with Death Variable with Quality Death (95% different ways. The process–outcome link remained after CI)* weighting for quality score stability and after adjustment Ϫ0.8 0.2 0.99 (0.74–1.33) for quality indicator difficulty. The alternative explanation, Ϫ0.8 0.3 1.13 (0.81–1.57) Ϫ0.6 0.2 0.82 (0.61–1.11) that physicians elect to provide less care to persons on a Ϫ0.6 0.3 1.11 (0.82–1.51) downward trajectory, is less likely because we did not ob- Ϫ0.4 0.4 0.89 (0.67–1.20) serve a relationship between care quality and sickness or Ϫ0.4 0.5 1.00 (0.74–1.36) Ϫ0.2 0.6 0.82 (0.60–1.13) age. Although our study is observational, our results satisfy most of the factors explicated by Hill (17) for making * Estimated hazard ratios are associated with a 10% higher quality score for the causal inference in observational studies, the most impor- period of 500 d after the quality measurement period. tant of which are strength of association, temporality of the cause and effect, the presence of a dose–response gradient, effect that is consistent with current knowledge. Only con- plausibility of causal mechanisms, coherence with current sistency and specificity of the association are not satisfied knowledge, consistency with other studies, and specificity by our results. Consistency cannot be tested by only 1 of the association. Our finding of a moderately strong as- study, and specificity cannot be tested because our only sociation between process and outcome has no temporal outcome measure is mortality. Furthermore, the formal ambiguity between process and outcome, with evidence of sensitivity analysis for a potential unmeasured omitted con- a dose–response relationship, and a plausible mechanism of founder revealed that this confounder must be very strongly related to both quality and survival to explain the quality–survival relationship beyond the adjustment for Figure 3. Relationship between quality score and age (top) and other covariates. Therefore, we believe that the most plau- relationship between quality score and Vulnerable Elders sible interpretation of our results is that the receipt of bet- Survey-13 (VES-13) score (bottom). ter-quality care was causally linked with improvement in 3-year mortality in our sample of community-dwelling vul- nerable older adults. Although many studies have measured process quality on the basis of indicators that have content validity derived from their link to outcomes in the medical literature, our study is, to our knowledge, the first to show the predictive validity of a broad-based, process-of-care quality measure- ment system using patient survival among community- dwelling older persons. Previous evaluation of the process– outcome link has focused on a narrower range of conditions and care domains. In 1 study, Kahn and col- leagues (18) used explicit criteria to examine the quality of care for older patients hospitalized with congestive heart failure, myocardial infarction, pneumonia, cerebrovascular accident, and . Better processes of care in 4 of these 5 conditions were related to lower 30-day mortality. In another study (19), the quality of medication prescrib- ing was statistically significantly related to preserved basic self-care function, although not survival, among commu- nity-dwelling older persons. Several other studies (20–22) investigated the process–outcome link to examine the va- lidity of annual hospital mortality statistics released by the Health Care Financing Administration (now the Centers for Medicare & Medicaid Services) (23). These studies found a weak relationship or no relationship between pro- cess quality and hospital mortality. We speculate that our study could detect an association between process quality and mortality because ACOVE quality measurement was both broad in coverage and specifically constructed to as- www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 279 Improving Patient Care Quality of Care–Survival Relationship in Vulnerable Older Patients sess those aspects of health care judged most likely to pre- tion between quality and survival more conservative be- vent death and loss of function. Furthermore, we assessed cause the assumption of equal weights biases the results care in a particularly vulnerable patient population with toward the null value. Fourth, we calculated quality scores 23% mortality over 3 years, which enabled us to detect this on the basis of individual sets of quality indicators because association even in a relatively small group over a relatively individuals are eligible for different care processes. There- short time period. Among the frequently occurring indica- fore, the same quality score can signify different care across tors that were commonly both passed and failed, receiving patients. We tried to adjust for patient sickness with gen- recommended care was associated with improved survival. eral measures, such as the VES-13, the number of comor- Some of these indicators have randomized, controlled trial bid conditions, and health service use, and these analyses evidence of a direct relationship with reduced mortality upheld our main result. Finally, our study has limited gen- (for example, influenza vaccine). Others may begin a path- eralizability because our sample comes from noninstitu- way to processes that may reduce mortality (such as the tionalized members of 2 managed care organizations. Rep- regular measurement of weight), and still others may not lications in a fee-for-service Medicare population and other lead directly or indirectly to reduced mortality but may be patient groups are needed. markers for careful medical care (which is then related to In conclusion, our study reports that better quality of reduced mortality). care, as measured by a broad set of quality indicators, is asso- Our findings show that performance on a comprehen- ciated with better survival among community-dwelling vul- sive measure of process of care is linked to survival. This nerable older persons. Although resource requirements may should motivate policies that promote such measurement pose an obstacle to implementing such a broad-based quality for older patients. Process-of-care measurement is valuable measurement system, advances in information technology in many ways: The results identify targets for improve- may substantially ameliorate the data collection burden in the ment; it does not require complicated case-mix adjustment; near future. Since only about half of recommended care was and the effect of a quality intervention can be measured in received, poor care may be responsible for unnecessary deaths a more timely manner compared with most outcome mea- among vulnerable older patients. The process measures in sures. Yet process measures have disadvantages as well, such ACOVE can be implemented by health care providers. An as the need to frequently update quality criteria to keep up important next step is to evaluate whether interventions can with advances in medical knowledge and technology, lack be implemented that improve the delivery of these processes of face validity for patients, and the cost of measurement. to vulnerable older patients and whether these improvements The cost of medical record abstraction can be particularly lead, as our results suggest, to improvement in mortality. high when evaluating a broad range of care. Recent ad- vances in information technology, however, will increase From RAND Health, Santa Monica, California, and Washington, DC, the practicality of process-of-care measurement (1). Our and the University of California, Los Angeles, and the Greater Los An- study supports the investment of resources to expedite im- geles Veterans Affairs Healthcare System, Los Angeles, California. plementation of broad measurement strategies integrated into routine clinical documentation (for example, through Acknowledgments: The authors thank Robert Brook, MD, ScD, for electronic medical records) as a critical step toward improv- inspiration and guidance; Robin P. Hertz, PhD, senior director of out- ing outcomes for vulnerable populations. comes research and population studies at Pfizer Inc, for providing valu- able support; and Patricia Smith and Victor Gonzalez for their technical Our study has several limitations. First, as mentioned assistance. earlier, this is an observational study, meaning that causal inference is guarded because the possibility of confounding by unmeasured variables cannot be ruled out. However, Grant Support: Supported by a contract from Pfizer Inc. Dr. Higashi is supported by a St. Luke’s Life Science Institute Fellowship Award. Dr. observational studies will probably be the predominant Shekelle was a Senior Research Associate of the Veterans Affairs Health study design to assess the relationship between quality and Services Research & Development Service. Dr. Chiang is supported by a survival, since it would be unethical to deliberately ran- Bureau of Health Professionals Geriatrics Research Faculty Training Pro- domly assign persons to lower-quality care. Furthermore, gram. Drs. MacLean and Saliba are Research Associates of the Veterans our results meet most of the factors proposed by Hill (17) Affairs Health Services Research & Development Service. Dr. Chang is when considering causation from observational studies. supported by a National Research Service Award (PE-19001) and the The most important unmet factor is consistency. Our University of California, Los Angeles, Specialty Training and Advanced Research (STAR) Program. study needs replication in other populations. Second, this study focused only on survival. Functional capability and Potential Financial Conflicts of Interest: Stock ownership or options quality of life are important to vulnerable older patients (other than mutual funds): R.T. Young (Pfizer Inc). and should be additional targets in future studies. Third, in calculation of the quality score, we treated individual qual- Requests for Single Reprints: Neil S. Wenger, MD, MPH, RAND, ity indicators as equal. Providing recommended care for 1700 Main Street, Santa Monica, CA 90407. some quality indicators undoubtedly has a greater effect on survival than others. This makes our estimates of associa- Current author addresses are available at www.annals.org.

280 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Quality of Care–Survival Relationship in Vulnerable Older Patients Improving Patient Care

References RK, et al. Appropriateness of quality indicators for older patients with advanced 1. Institute of Medicine. Crossing the Quality Chasm. Washington DC: Na- dementia and poor prognosis. J Am Geriatr Soc. 2003;51:902-7. [PMID: tional Acad Pr; 2001. 12834508] 2. Schuster MA, McGlynn EA, Brook RH. How good is the quality of health 14. Saliba D, Elliott M, Rubenstein LZ, Solomon DH, Young RT, Kamberg care in the United States? Milbank Q. 1998;76:517-63, 509. [PMID: 9879302] CJ, et al. The Vulnerable Elders Survey: a tool for identifying vulnerable older 3. McGlynn EA, Asch SM, Adams J, Keesey J, Hicks J, DeCristofaro A, et al. people in the community. J Am Geriatr Soc. 2001;49:1691-9. [PMID: The quality of health care delivered to adults in the United States. N Engl J Med. 11844005] 2003;348:2635-45. [PMID: 12826639] 15. Katzman R, Brown T, Fuld P, Peck A, Schechter R, Schimmel H. Valida- 4. National Committee on Quality Assurance. HEDIS 2004 Summary Table of tion of a short Orientation-Memory-Concentration Test of cognitive impair- Measures and Product Lines. Washington, DC: National Committee on ment. Am J Psychiatry. 1983;140:734-9. [PMID: 6846631] Quality Assurance; 2004. Accessed at www.ncqa.org/Programs/HEDIS 16. McHorney CA, Ware JE Jr, Lu JF, Sherbourne CD. The MOS 36-item /Hedis%202004%20Summary%20Table.pdf on 22 June 2004. Short-Form Health Survey (SF-36): III. Tests of data quality, scaling assump- 5. Jencks SF, Cuerdon T, Burwen DR, Fleming B, Houck PM, Kussmaul AE, tions, and reliability across diverse patient groups. Med Care. 1994;32:40-66. et al. Quality of medical care delivered to Medicare beneficiaries: a profile at state [PMID: 8277801] and national levels. JAMA. 2000;284:1670-6. [PMID: 11015797] 17. Hill AB. The environment and disease: association or causation? Proc R Soc 6. Lilford R, Mohammed MA, Spiegelhalter D, Thomson R. Use and misuse of Med. 1965;58:295-300. [PMID: 14283879] process and outcome data in managing performance of acute medical care: avoid- 18. Kahn KL, Rubenstein LV, Draper D, Kosecoff J, Rogers WH, Keeler EB, ing institutional stigma. Lancet. 2004;363:1147-54. [PMID: 15064036] et al. The effects of the DRG-based prospective payment system on quality of 7. Mant J, Hicks N. Detecting differences in quality of care: the sensitivity of care for hospitalized Medicare patients. An introduction to the series. JAMA. measures of process and outcome in treating acute myocardial infarction. BMJ. 1990;264:1953-5. [PMID: 2120473] 1995;311:793-6. [PMID: 7580444] 19. Hanlon JT, Fillenbaum GG, Kuchibhatla M, Artz MB, Boult C, Gross CR, 8. Brook RH, McGlynn EA, Cleary PD. Quality of health care. Part 2: measur- et al. Impact of inappropriate drug use on mortality and functional status in ing quality of care [Editorial]. N Engl J Med. 1996;335:966-70. [PMID: representative community dwelling elders. Med Care. 2002;40:166-76. [PMID: 8782507] 11802089] 9. Wenger NS, Shekelle PG. Assessing care of vulnerable elders: ACOVE project 20. Thomas JW, Holloway JJ, Guire KE. Validating risk-adjusted mortality as overview. Ann Intern Med. 2001;135:642-6. [PMID: 11601946] an indicator for quality of care. Inquiry. 1993;30:6-22. [PMID: 8454316] 10. Shekelle PG, MacLean CH, Morton SC, Wenger NS. ACOVE quality 21. Jencks SF, Daley J, Draper D, Thomas N, Lenhart G, Walker J. Interpret- indicators. Ann Intern Med. 2001;135:653-67. [PMID: 11601948] ing hospital mortality data. The role of clinical risk adjustment. JAMA. 1988; 11. Wenger NS, Solomon DH, Roth CP, MacLean CH, Saliba D, Kamberg 260:3611-6. [PMID: 3057250] CJ, et al. The quality of medical care provided to vulnerable community-dwelling 22. Park RE, Brook RH, Kosecoff J, Keesey J, Rubenstein L, Keeler E, et al. older patients. Ann Intern Med. 2003;139:740-7. [PMID: 14597458] Explaining variations in hospital death rates. Randomness, severity of illness, 12. Sloss EM, Solomon DH, Shekelle PG, Young RT, Saliba D, MacLean CH, quality of care. JAMA. 1990;264:484-90. [PMID: 2195173] et al. Selecting target conditions for quality of care improvement in vulnerable 23. Health Care Financing Administration. Medicare Hospital Mortality Infor- older adults. J Am Geriatr Soc. 2000;48:363-9. [PMID: 10798460] mation, 1986. HCFA publication 01-002. Washington, DC: U.S. Department 13. Solomon DH, Wenger NS, Saliba D, Young RT, Adelman AM, Besdine of Health and Human Services; 1987.

www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 281 Current Author Addresses: Dr. Higashi: Department of Epidemiology Drs. Solomon, Adams, and Wenger and Ms. Roth: RAND, 1700 Main and Healthcare Research, Kyoto University, Yoshida-konoe-cho, Sakyo- Street, M-26, Santa Monica, CA 90407-2138. ku, Kyoto 606-8501, Japan. Ms. Kamberg: RAND, 1200 South Hayes Street, Arlington, VA 22202. Dr. Chang: Division of General Internal Medicine, University of Cali- Dr. Young: Division of General Internal Medicine, University of Cali- fornia, Los Angeles, 911 Broxton Plaza, Los Angeles, CA 90095-1736. fornia, Los Angeles, 200 Medical Plaza, Los Angeles, CA 90095-1736. Drs. Shekelle, MacLean, and Saliba: Greater Los Angeles Veterans Affairs Drs. Reuben and Chiang: Division of Geriatrics, University of Califor- Healthcare System, 11301 Wilshire Boulevard, Los Angeles, CA 90073. nia, Los Angeles, 200 Medical Plaza, Los Angeles, CA 90095-1736.

W-70 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Appendix Table. Quality Indicators Using Medical Records as the Information Source, Eligible Patients, and Pass Rates*

Text of Quality Indicators Patients Pass Triggered, Rates, n† %† Continuity and coordination of care IF an outpatient vulnerable elder is started on a new prescription 189 66 medication, and he or she has a follow-up visit with the prescribing physician, THEN the medical record at the follow-up visit should document 1 of the following: 1) The medication is being taken, 2) the physician asked about the medication (e.g., side effects or adherence or availability), or 3) the medication was not started because it was not needed or because it was changed. IF a vulnerable elder is discharged from a hospital to home or to a nursing 62 90 home, and the hospital medical record specifies a follow-up appointment for a physician visit or a treatment (e.g., physical therapy or radiation oncology), THEN the medical record should document that the visit or treatment took place or that it was postponed or was not needed. IF a vulnerable elder is discharged from a hospital to home or to a nursing 52 41 home, THEN there should be a discharge summary in the outpatient physician or nursing home medical record within 6 weeks. IF a vulnerable elder is discharged from a hospital to home or to a nursing 14 71 home, and the transfer form or discharge summary indicates that a test result is pending, THEN the outpatient or nursing home medical record should include the test result within 6 weeks of hospital discharge. IF a vulnerable elder is discharged from a hospital to home, and he or she 11 55 received a new prescription medication or a change in medication before discharge, THEN the outpatient medical record should document or acknowledge the medication change within 6 weeks of discharge. IF a vulnerable elder is under the outpatient care of Ն2 physicians, and 1 642 physician prescribed a new prescription medication or a change in medication, THEN subsequent medical record entries by the nonprescribing physician should acknowledge the medication change.

Dementia IF a vulnerable elder is admitted to a hospital or is new to a physician 130 52 practice, THEN there should be documentation of a multidimensional assessment of cognitive ability. IF a vulnerable elder is admitted to a hospital or is new to a physician 130 18 practice, THEN there should be an assessment of functional status. IF a vulnerable elder with dementia has a caregiver (and, if capable, the 28 26 patient assents), THEN the physician should discuss or refer the patient and caregiver for discussion about patient safety, provide education on how to deal with conflicts at home, and inform them about community resources for dementia. IF a vulnerable elder receives a new diagnosis of dementia, THEN the 650 diagnosing physician should advise the patient not to drive a motor vehicle, request that the Department of Motor Vehicles (or equivalent) retests the patient’s ability to drive, or refer the patient to a drivers’ safety or education course that includes assessment of driving ability consistent with state laws. IF a vulnerable elder has dementia, THEN he or she should be screened for 560 depression during the initial evaluation period. IF a vulnerable elder receives a new diagnosis of dementia, THEN a serum 520

vitamin B12 and TSH test should be performed. IF a vulnerable elder with dementia has cerebrovascular disease, THEN he or 2 100 she should be offered appropriate stroke prophylaxis.

Depression IF a vulnerable elder presents with new onset of 1 of the following 34 26 symptoms: sad mood, feeling down, insomnia or difficulties with sleep, apathy or loss of interest in pleasurable activities, reports of memory loss, unexplained weight loss greater than 5% in the past month or 10% over 1 year, or unexplained fatigue or low energy; THEN the patient should be asked about or treated for depression or should be referred to a mental health professional within 2 weeks of presentation. IF a vulnerable elder receives a diagnosis of a new depression episode, 13 0 THEN the medical record should document at least 3 of the 9 DSM-IV target symptoms for major depression within the first month of diagnosis. IF a vulnerable elder receives a diagnosis of a new depression episode, 13 0 THEN the medical record should document on the day of diagnosis the presence or absence of suicidal ideation and psychosis (consisting of, at a minimum, auditory hallucinations or delusions).

Continued on following page

www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 W-71 Appendix Table—Continued

Text of Quality Indicators Patients Pass Triggered, Rates, n† %† IF a vulnerable elder receives a diagnosis of depression, THEN antidepressant 13 69 treatment, psychotherapy, or electroconvulsive therapy should be offered within 2 weeks after diagnosis unless there is documentation within that period that the patient has improved or unless the patient has substance abuse or dependence, in which case treatment may wait until 8 weeks after the patient is in a drug- or alcohol-free state. IF a vulnerable elder is started on an antidepressant medication, THEN the 10 90 following medications should not be used as first- or second-line therapy: tertiary amine tricyclics (amitriptyline, imipramine, doxepin, clomipramine, or trimipramine), monoamine oxidase inhibitors (unless atypical depression is present), benzodiazepines, or stimulants (except methylphenidate). IF a vulnerable elder has no meaningful symptom response after 6 weeks of 922 treatment, THEN 1 of the following treatment options should be initiated by the 8th week of treatment: Medication dose should be optimized or the patient should be referred to a psychiatrist (if initial treatment was medication) or medication should be initiated or referral to a psychiatrist should be offered (if initial treatment was psychotherapy alone). IF a vulnerable elder responds only partially after 12 weeks of treatment, 825 THEN 1 of the following treatment options should be instituted by the 16th week of treatment: Switch to a different medication class or add a second medication to the first (if initial treatment includes medication), add psychotherapy (if the initial treatment was medication), try medication (if initial treatment was psychotherapy without medication), consider electroconvulsive therapy, or refer to a psychiatrist. IF a vulnerable elder with a history of cardiac disease is started on a tricyclic 10 antidepressant, THEN baseline electrocardiography should be performed before initiation of or within 3 months before treatment.

Diabetes mellitus IF a vulnerable elder has diabetes, THEN his or her blood pressure should be 85 59 checked at each outpatient visit. IF a vulnerable elder has diabetes, THEN his or her glycated hemoglobin 84 80 level should be measured at least every 12 months. IF a diabetic vulnerable elder is not blind, THEN he or she should receive an 84 48 annual dilated eye examination performed by an ophthalmologist, optometrist, or diabetes specialist. ALL diabetic vulnerable elders should be offered daily aspirin therapy. 59 41 IF a diabetic vulnerable elder has elevated blood pressure, THEN he or she 44 79 should be offered a therapeutic intervention to lower blood pressure within 3 months if blood pressure is 150–160/90–100 mm Hg and within 1 month if blood pressure is greater than 160/100 mm Hg. IF a diabetic vulnerable elder does not have established renal disease and is 43 19 not receiving an ACE inhibitor or ACE blocker, THEN he or she should receive an annual test for proteinuria. IF a diabetic vulnerable elder has a fasting total cholesterol level Ն6.2 12 92 mmol/L (Ն240 g/dL), THEN he or she should be offered an intervention to lower cholesterol. IF a vulnerable elder has an elevated glycated hemoglobin level, THEN he or 961 she should be offered a therapeutic intervention aimed at improving glycemic control within 3 months if the glycated hemoglobin level is 9.0% to 10.9%, and within 1 month if the glycated hemoglobin level is Ն11%. IF a diabetic vulnerable elder has proteinuria, THEN he or she should be 520 offered therapy with an ACE inhibitor or ACE receptor blocker.

End-of-life care ALL vulnerable elders should have in their outpatient charts 1) an advance 370 4 directive indicating the patient’s surrogate decision maker, or 2) documentation of a discussion about who would be a surrogate decision maker or a search for a surrogate, or 3) indication that there is no identified surrogate. IF a vulnerable elder with dementia, coma, or altered mental status is 20 25 admitted to the hospital, THEN within 48 hours of admission, the medical record should 1) contain an advance directive indicating the patient’s surrogate decision maker, 2) document a discussion about who would be a surrogate decision maker or a search for a surrogate, or 3) indicate that there is no identified surrogate.

W-72 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Appendix Table—Continued

Text of Quality Indicators Patients Pass Triggered, Rates, n† %† IF a vulnerable elder with decision-making capacity has orders written in the 10 70 hospital or the nursing home to withhold or withdraw a particular treatment modality (e.g., DNR order or an order not to initiate ), THEN the medical record should document 1) patient participation in the decision or 2) why the patient chose not to participate in the decision. IF a vulnerable elder has an advance directive in the outpatient, inpatient, or 825 nursing home medical record or the patient reports the existence of an advance directive in an interview, and the patient receives care in a second venue, THEN 1) the advance directive should be present in the medical record at the second venue or 2) documentation should acknowledge its existence, its contents, and the reason that it is not in the medical record. IF a vulnerable elder is admitted directly to the intensive care unit (from the 617 outpatient setting or emergency department) and survives 48 hours, THEN within 48 hours of admission, the medical record should document consideration of the patient’s preferences for care or that these could not be elicited or are unknown. IF a vulnerable elder carries a diagnosis of severe dementia, is admitted to 2 100 the hospital, and survives 48 hours, THEN within 48 hours of admission, the medical record should document consideration of the patient’s previous preferences for care or that these could not be elicited or are unknown. IF a vulnerable elder requires mechanical ventilation during a hospitalization 2 100 (except short-term and postoperative mechanical ventilation), THEN the medical record should document within 48 hours of the initiation of mechanical ventilation the goals of care and the patient’s preference for mechanical ventilation or why this information is unavailable.

Fall or illness problem ALL vulnerable elders should have documentation that they were asked at 372 25 least annually about the occurrence of recent falls. IF a vulnerable elder reported Ն 2 falls in the past year or 1 fall with injury 57 49 requiring treatment, THEN there should be documentation of a basic fall history. IF a vulnerable elder reported Ն 2 falls in the past year or 1 fall with injury 57 3 requiring treatment, THEN there should be documentation of a basic fall examination. IF a vulnerable elder reported Ն 2 falls in the past year or 1 fall with injury 57 30 requiring treatment, THEN there should be an examination with documented recommendations. IF a vulnerable elder reports or is found to have new or worsening difficulty 22 23 with ambulation, balance, and/or mobility, THEN there should be documentation that a basic gait, mobility, and balance evaluation was performed within 6 months that resulted in specific diagnostic and therapeutic recommendations. IF a vulnerable elder is found to have problems with gait, strength (e.g., 14 71 Յ4/5 on manual muscle testing or needs arms to rise from a chair), or endurance (e.g., dyspnea on mild exertion), THEN an exercise program should be offered. IF a vulnerable elder demonstrates decreased balance or proprioception or 13 62 increased postural sway, THEN an appropriate exercise program should be offered and an evaluation for an assistive device performed.

Hearing loss IF a vulnerable elder fails a hearing screening, THEN he or she should be 18 94 offered a formal audiologic evaluation within 3 months. IF a vulnerable elder has a hearing problem or fails an audiologic screening, 15 83 THEN he or she should have an ear examination within 3 months. ALL vulnerable elders should have a hearing screening examination as part 40 of the initial evaluation. IF a vulnerable elder is a hearing aid candidate, THEN he or she should be 450 offered hearing rehabilitation.

Heart failure IF a vulnerable elder has heart failure and LV ejection fraction Յ 0.4 (or 23 65 unknown), THEN he or she should be offered an ACE inhibitor or receptor blocker.

Continued on following page

www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 W-73 Appendix Table—Continued

Text of Quality Indicators Patients Pass Triggered, Rates, n† %† IF a vulnerable elder has heart failure, has LV ejection fraction Յ 0.4, and is 21 48 NYHA class I–III, THEN he or she should be offered a ␤-blocker unless a contraindication (e.g, uncompensated heart failure) has been documented. IF a vulnerable elder has heart failure, has LV ejection fraction Յ 0.4, and 9 100 does not have AF, THEN from among the 3 generations of calcium-channel blocker medications, he or she should not be treated with a first- or second-generation calcium-channel blocker. IF a vulnerable elder is hospitalized with heart failure, THEN he or she 8 100 should have serum electrolytes, creatinine, and blood urea nitrogen levels measured within 1 day of hospitalization. IF a vulnerable elder has heart failure and AF, THEN he or she should be 771 offered anticoagulation to achieve an INR of 2.0 to 3.0. IF a vulnerable elder receives a new diagnosis of heart failure, THEN he or 683 she should have a history taken at the time of diagnosis and hospitalization that documents the presence or absence of previous MI, documented coronary artery disease, revascularization, current symptoms of chest pain or angina, history of hypertension, history of diabetes, history of hypercholesterolemia, history of valvular heart disease, history of thyroid disease, , current medications, and a description of functional capacity (e.g., NYHA functional status). IF a vulnerable elder receives a new diagnosis of heart failure, THEN he or 667 she should be offered an evaluation of LV ejection fraction within 1 month. IF a vulnerable elder receives a new diagnosis of heart failure, THEN he or 6 100 she should have the following elements of the physical examination documented at the time of presentation: weight, blood pressure and heart rate, lung examination, cardiac examination, and abdominal or lower-extremity examination. IF a vulnerable elder receives a new diagnosis of heart failure, THEN he or 667 she should undergo the following studies within 1 month of the diagnosis (unless they have already been performed within the previous 3 months): chest radiography, electrocardiography, CBC, serum sodium and potassium levels, serum creatinine level, and TSH level in patients with AF or heart failure with no obvious cause. IF a vulnerable elder has heart failure and AF and he or she has documented 333 contraindications to anticoagulation, THEN he or she should be offered aspirin. IF a vulnerable elder has heart failure and LV ejection fraction Յ 0.4, THEN 1 100 he or she should not be treated with a type I antiarrhythmic agent unless an implantable cardioverter defibrillator is in place.

Hospital care IF a vulnerable elder is admitted to the hospital for any acute or chronic 57 97 illness or any surgical procedure, THEN the evaluation should include within 24 hours: 1) diagnoses and 2) prehospital and current medications. IF a vulnerable elder is admitted to the hospital for any acute or chronic 57 20 illness or any surgical procedure, THEN documentation of cognitive status should be performed within 24 hours. IF a vulnerable elder enters the hospital, THEN discharge planning should 57 67 begin within 48 hours. IF a hospitalized vulnerable elder has peptic stress ulcer risk factors, THEN 10 45

the patient should receive prophylaxis with either an H2-blocker, sucralfate, or a proton-pump inhibitor. IF a hospitalized vulnerable elder has a definite or suspected diagnosis of 10 60 delirium, THEN an evaluation for potentially precipitating factors must be undertaken and identified causes treated. IF a hospitalized vulnerable elder has a definite or suspected diagnosis of 944 delirium, THEN identified potential causes should be treated. IF a hospitalized vulnerable elder is at very high risk for venous thrombosis, 4 100 THEN the patient should have venous thromboembolism prophylaxis.

Hypertension IF a vulnerable elder requires pharmacotherapy for treatment of 59 93 hypertension in the outpatient setting, THEN a once- or twice-daily medication should be used unless there is documentation about the need for agents that require more frequent dosing.

W-74 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Appendix Table—Continued

Text of Quality Indicators Patients Pass Triggered, Rates, n† %† IF a vulnerable elder has hypertension and has renal parenchymal disease 19 63 with a serum creatinine level Ͼ133 ␮mol/L (Ͼ1.5 mg/dL) or Ͼ 1gof protein/24 hours of collected urine, THEN therapy with an ACE inhibitor should be offered. IF a vulnerable elder has hypertension and asthma, THEN ␤-blocker therapy 15 100 for hypertension should not be used. IF a vulnerable elder remains hypertensive after nonpharmacologic 11 64 intervention, THEN pharmacologic antihypertensive treatment should be initiated. IF a vulnerable elder receives a new diagnosis of hypertension, THEN within 633 4 weeks of the diagnosis, electrocardiography should be performed. IF a vulnerable elder receives a new diagnosis of hypertension, THEN there 633 should be documentation about the presence or absence of other cardiovascular risk factors. IF a vulnerable elder receives a new diagnosis of hypertension, THEN 633 nonpharmacologic therapy with lifestyle modification for treatment of hypertension should be recommended, including dietary sodium restriction and weight loss if patient is Ͼ 10% more than ideal body weight. IF a vulnerable elder receives a new diagnosis of hypertension and the blood 333 pressure is below 170/90 mm Hg, THEN there should be evidence that 3 or more blood pressure measures Ն 140/90 mm Hg were obtained before the diagnosis.

Ischemic heart disease IF a vulnerable elder has established CHD and is not receiving warfarin, 73 66 THEN he or she should be offered antiplatelet therapy. IF a vulnerable elder has had a MI, THEN he or she should be offered a 53 53 ␤-blocker. IF a vulnerable elder has established CHD and LDL cholesterol level Ͼ3.36 16 47 mmol/L (Ͼ 130 mg/dL) and a trial of step II diet therapy was not offered or was ineffective, THEN he or she should be offered cholesterol-lowering medication. IF a vulnerable elder with established CHD smokes, THEN he or she should 850 be offered counseling for smoking cessation at least annually and have this documented in the medical record. IF a vulnerable elder has established CAD and his or her cholesterol level is 333 not known, THEN he or she should undergo a fasting cholesterol evaluation, including total LDL and HDL cholesterol. IF a vulnerable elder has an acute MI or unstable angina, did not undergo 30 angiography, and does not have contraindications to revascularization, THEN he or she should be offered noninvasive stress testing 4–21 days after the infarction or anginal event. IF a vulnerable elder is hospitalized with an acute MI, THEN he or she 250 should be offered assessment of LV function before discharge or within 3 days after hospital discharge. IF a vulnerable elder has an acute MI or unstable angina, THEN he or she 20 should be given aspirin therapy within 1 hour of presentation. IF a vulnerable elder has unstable angina or an acute MI, THEN he or she 250 should be offered ␤-blocker therapy within 12 hours of presentation. IF a vulnerable elder has had a recent MI or recent coronary bypass graft 20 surgery, THEN he or she should be offered cardiac rehabilitation. IF a vulnerable elder has clinically significant left main or clinically significant 10 3-vessel coronary artery disease with LV ejection fraction Ͻ 0.5, THEN he or she should be offered coronary artery bypass graft surgery.

Malnutrition ALL vulnerable elders should be weighed at each physician office visit and 355 42 these weights should be documented in the medical record. IF a vulnerable elder is hospitalized, THEN his or her nutritional status should 57 47 be documented during the hospitalization by evaluation of oral intake or serum biochemical testing (e.g., albumin, prealbumin, or cholesterol levels). IF a vulnerable elder has documented involuntary weight loss or 33 52 hypoalbuminemia (Ͻ35 g/L), THEN she or he should receive an evaluation for potentially reversible causes of poor nutritional intake.

Continued on following page

www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 W-75 Appendix Table—Continued

Text of Quality Indicators Patients Pass Triggered, Rates, n† %† IF a vulnerable elder has documented involuntary weight loss or 33 76 hypoalbuminemia (Ͻ35 g/L), THEN he or she should receive an evaluation for potentially relevant comorbid conditions, including medications that might be associated with decreased appetite (e.g., digoxin, fluoxetine, anticholinergics), depressive symptoms, and cognitive impairment. IF a vulnerable elder has involuntary weight loss Ͼ 10% of body weight 13 77 over Յ1 year, THEN weight loss (or a related disorder) should be documented in the medical record as an indication that the physician recognized malnutrition as a potential problem.

Medication use IF a vulnerable elder does not need control of seizures, THEN barbiturates 372 99 should not be used. IF a vulnerable elder requires analgesia, THEN meperidine should not be 369 99 used. ALL vulnerable elders should not be prescribed a medication with strong 366 98 anticholinergic effects if alternatives are available. IF a vulnerable elder is prescribed a new drug, THEN the patient (or, if 259 18 incapable, a caregiver) should receive education about the purpose of the drug, how to take it, and expected side effects or important adverse reactions. IF a vulnerable elder is prescribed a new drug, THEN the prescribed drug 258 98 should have a clearly defined indication documented in the record. EVERY new drug that is prescribed to a vulnerable elder on an ongoing 180 65 basis for a chronic medical condition should have a documentation of response to therapy within 6 months. IF a vulnerable elder is prescribed a thiazide or loop diuretic, THEN he or she 127 80 should have electrolyte levels checked at least yearly. IF a vulnerable elder is prescribed an oral hypoglycemic drug, THEN 89 99 chlorpropamide should not be used. IF a vulnerable elder is prescribed warfarin, THEN an INR should be 44 53 determined at least every 6 weeks. IF a vulnerable elder is newly started on a diuretic, THEN serum potassium 25 34 and creatinine levels should be checked within 1 month of the initiation of therapy. IF a vulnerable elder is newly started on an ACE inhibitor, THEN serum 23 37 potassium and creatinine levels should be checked within 1 month of the initiation of therapy. IF a vulnerable elder is prescribed warfarin, THEN an INR should be 11 45 determined within 4 days after initiation of therapy and at least every 6 weeks.

Osteoarthritis IF a vulnerable elder is treated with COX-2 nonselective NSAIDs, THEN 50 4 there should be evidence that the patient was advised of the risks associated with these drugs. IF a vulnerable elder is older than 75 years of age and/or has a history of 38 11 , gastrointestinal bleeding, or current warfarin use and the patient is being treated with a COX-2 nonselective NSAID, THEN he or she should be offered concomitant treatment with either or a proton-pump inhibitor. IF oral pharmacologic therapy is initiated to treat osteoarthritis, THEN 37 43 acetaminophen should be the first drug used, unless there is a documented contraindication to use. IF an ambulatory vulnerable elder receives a new diagnosis of symptomatic 19 16 osteoarthritis of the knee and has no contraindications to exercise and is physically and mentally able to exercise, THEN a directed or supervised strengthening or aerobic exercise program should be prescribed within 3 months of diagnosis. IF a vulnerable elder with severe symptomatic osteoarthritis of the knee or 10 90 hip has not responded to nonpharmacologic and pharmacologic therapy, THEN the patient should be offered referral to an orthopedic surgeon to be evaluated for total joint replacement within 6 months unless a contraindication to surgery is documented. IF oral pharmacologic therapy for osteoarthritis is changed from 333 acetaminophen to a different oral agent, THEN there should be evidence that the patient has had a trial of maximum-dose acetaminophen (suitable for age and comorbid conditions).

W-76 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Appendix Table—Continued

Text of Quality Indicators Patients Pass Triggered, Rates, n† %†

Osteoporosis ALL female vulnerable elders who smoke should be counseled annually 25 48 about smoking cessation. IF a vulnerable elder has a new diagnosis of osteoporosis, THEN during the 12 42 initial evaluation period, an underlying cause of osteoporosis should be sought by checking medication use and current alcohol use. IF a female vulnerable elder receives a new diagnosis of osteoporosis, 10 60 THEN the patient should be offered treatment with hormone replacement therapy or bisphosphonates or calcitonin within 3 months of diagnosis. IF a vulnerable elder is taking corticosteroids for more than 1 month, THEN 771 the patient should be offered calcium and vitamin D. IF an ambulatory vulnerable elder has an osteoporotic fracture diagnosed, 70 THEN physical therapy or an exercise program should be offered within 3 months.

Pain management IF a vulnerable elder has a newly reported, chronic painful condition, THEN 123 40 a targeted history should be performed within 1 month. IF a vulnerable elder has a newly reported, chronic painful condition, THEN 123 58 a physical examination should be performed within 1 month. IF a vulnerable elder has a newly reported, chronic painful condition, THEN 121 86 treatment should be offered. IF a vulnerable elder is treated for a chronic painful condition, THEN he or 70 66 she should be assessed for a response within 6 months. IF a vulnerable elder has been prescribed a COX-2 nonselective NSAID for 50 10 the treatment of chronic pain, THEN the medical record should indicate whether he or she has a history of peptic ulcer disease and, if a history is present, justification of NSAID use should be documented. IF a vulnerable elder with chronic pain is treated with opioids, THEN he or 46 0 she should be offered a bowel regimen or the medical record should document the potential for constipation or explain why bowel treatment is not needed.

Pneumonia IF a vulnerable elder with no history of allergy to the pneumococcal vaccine 372 29 is not known to have already received a pneumococcal vaccine or if the patient received it more than 5 years ago (if before age 65 years), THEN a pneumococcal vaccine should be offered. IF a vulnerable elder has no history of anaphylactic hypersensitivity to eggs 372 66 or to other components of the influenza vaccine, THEN the patient should be offered an annual influenza vaccination. IF a vulnerable elder is admitted to the hospital with pneumonia, THEN 888 antibiotics should be administered within 8 hours of hospital arrival. IF a vulnerable elder is admitted to the hospital with community-acquired 7 100 pneumonia with , THEN the patient should receive oxygen therapy. IF a vulnerable elder with community-acquired pneumonia is to be 7 100 discharged home, THEN the patient should not be unstable on the day before or the day of discharge. IF a smoker develops pneumonia, THEN the smoker should be advised to 333 quit smoking.

Prevention and screening ALL vulnerable elders newly admitted to a physician practice should 414 receive within 6 months the elements of a comprehensive geriatric assessment. ALL vulnerable elders newly admitted to a physician practice should receive 344 within 6 months recommendations from the comprehensive geriatric assessment. IF the elements of a comprehensive geriatric assessment are performed, 2 100 THEN follow-up should assure the implementation of recommendations. IF a vulnerable elder has valvular or congenital heart disease, intracardiac 1 100 valvular prosthesis, hypertrophic cardiomyopathy, mitral valve prolapse with regurgitation, or previous episode of endocarditis and a high-risk procedure is planned, THEN endocarditis prophylaxis should be given.

Continued on following page www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 W-77 Appendix Table—Continued

Text of Quality Indicators Patients Pass Triggered, Rates, n† %† Pressure ulcer IF a vulnerable elder is admitted to an intensive care unit or a medical or 11 59 surgical unit of a hospital and cannot reposition himself or herself or has limited ability to do so, THEN risk assessment for pressure ulcers should be performed on admission. IF a vulnerable elder is identified as at risk for pressure ulcer development or 90 a pressure ulcer risk assessment score indicates that the person is at risk, THEN preventive intervention must be instituted within 12 hours, addressing repositioning needs and pressure reduction (or management of tissue loads). IF a vulnerable elder presents with a pressure ulcer, THEN the pressure ulcer 933 should be assessed for 1) location, 2) depth and stage, 3) size, and 4) presence of necrotic tissue. IF a vulnerable elder is identified as at risk for pressure ulcer development 683 and has malnutrition (involuntary weight loss Ͼ 10% over 1 year or low albumin or prealbumin levels), THEN nutritional intervention or dietary consultation should be instituted. IF a vulnerable elder presents with a clean full-thickness pressure ulcer and 250 has no improvement at 4 weeks post-treatment, THEN 1) the appropriateness of the treatment plan and 2) the presence of cellulitis or osteomyelitis should be assessed. IF a vulnerable elder with a full-thickness pressure ulcer presents with 10 systemic signs and symptoms of infection, such as elevated temperature, leukocytosis, confusion, and agitation, and these signs and symptoms are not due to another identified cause, THEN the ulcer should be debrided of necrotic tissue within 12 hours. IF a vulnerable elder presents with a partial-thickness pressure ulcer and has 133 no improvement at 2 weeks post-treatment, THEN the appropriateness of the treatment plan should be assessed.

Stroke and AF IF a vulnerable elder has AF for Ͼ 48-hour duration and has any high-risk 18 94 condition (impaired LV function; women age Ͼ 75 years; hypertension or systolic blood pressure Ͼ 160 mm Hg; or previous ischemic stroke, TIA, or systemic embolism), THEN he or she should be offered oral anticoagulation, or antiplatelet therapy if the medical record documents a reason not to give anticoagulant therapy. IF a vulnerable elder has a TIA or stroke, THEN the medical record should 6 100 document that smoking status was assessed and that smokers were counseled to stop smoking. IF a male vulnerable elder has carotid artery symptoms and receives a 3 100 diagnosis of TIA or nondisabling stroke, and the medical record does not document that the patient is not a candidate for carotid surgery, THEN a carotid artery imaging study should be performed within 4 weeks. IF a vulnerable elder has a presumed stroke, THEN a CT or an MRI of the 2 100 head should be obtained before initiation or continuation of thrombolytic treatment, anticoagulant therapy, or antiplatelet therapy. IF a vulnerable elder receives a diagnosis of acute atherothrombotic ischemic 2 100 stroke or TIA, THEN antiplatelet treatment should be offered within 48 hours after the stroke or TIA, unless the patient is already receiving anticoagulant treatment. IF a vulnerable elder is admitted to the hospital with a diagnosis of acute 250 ischemic or hemorrhagic stroke, THEN he or she should be admitted to a specialized acute or combined acute and rehabilitative stroke unit or transferred to a specialized stroke unit if such a unit is available in the hospital.

Urinary incontinence ALL vulnerable elders should annually have documentation of the presence 363 31 or absence of urinary incontinence. IF a vulnerable elder has new urinary incontinence that persists for more 32 22 than 1 month or urinary incontinence at the time of a new evaluation, THEN a targeted physical examination should be performed that documents 1) a rectal examination and 2) a genital system examination (including a pelvic examination for women). IF a vulnerable elder has new urinary incontinence that persists for more 32 13 than 1 month or urinary incontinence at the time of a new evaluation, THEN a dipstick urinalysis and postvoid residual should be obtained.

W-78 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Appendix Table—Continued

Text of Quality Indicators Patients Pass Triggered, Rates, n† %† IF a vulnerable elder has new urinary incontinence or urinary incontinence at 32 59 the time of a new evaluation, THEN treatment options should be discussed. IF a vulnerable elder has new urinary incontinence that persists for more 32 19 than 1 month or urinary incontinence at the time of a new evaluation, THEN a targeted history should be obtained that documents each of the following: 1) characteristics of voiding, 2) ability to get to the toilet, 3) previous treatment for urinary incontinence, 4) importance of the problem to the patient, and 5) mental status. IF a cognitively intact vulnerable elder who is capable of independent 31 13 toileting has documented stress, urge, or mixed incontinence without evidence of hematuria or high postvoid residual, THEN behavioral treatment should be offered. ALL vulnerable elders should have documentation of the presence or 450 absence of urinary incontinence during the initial evaluation. IF a female vulnerable elder has documented stress urinary incontinence 1 100 caused by isolated ISD or ISD with coexistent hypermobility and she undergoes surgical correction, THEN a sling or artificial sphincter procedure should be used. IF a vulnerable elder undergoes surgery or periurethral injections for urinary 10 incontinence, THEN subtracted cystometry should be performed before the procedure.

Vision care IF a vulnerable elder receives a diagnosis of a cataract, THEN assessment of 102 31 visual function with respect to his or her ability to carry out needed or desired activities should be performed every 12 months. IF a vulnerable elder with diabetes has a retinal examination, THEN the 43 88 presence or degree of diabetic retinopathy should be documented. IF a vulnerable elder receives a diagnosis of a cataract that limits the 22 86 patient’s ability to carry out needed or desired activities, THEN cataract extraction should be offered. IF a vulnerable elder undergoes cataract surgery, THEN a follow-up ocular 18 100 examination should occur within 48 hours and reexamination should occur within 3 months. IF a vulnerable elder has sudden-onset visual changes, eye pain, corneal 10 80 opacity, or severe purulent discharge, THEN the patient should be examined within 72 hours by an ophthalmologist. IF a vulnerable elder who has been prescribed an ocular therapeutic regimen 683 becomes hospitalized, THEN the regimen should be administered in the hospital unless discontinued by an ophthalmologic consultant. IF a vulnerable elder develops progression of a chronic visual deficit that 5 100 now interferes with his or her ability to perform needed or desired activities, THEN he or she should have an ophthalmic examination performed by a person skilled at ophthalmic examination within 2 months. IF a vulnerable elder has a new diagnosis of primary open-angle glaucoma, 30 THEN the initial evaluation of each eye should include the essential components of a comprehensive eye examination and documentation of the optic nerve appearance, visual field testing, and determination of an initial target pressure. IF a vulnerable elder with diabetes receives a diagnosis of macular , 2 100 THEN a dilated eye examination should be performed at least every 6 months.

* ACE ϭ angiotensin-converting enzyme; AF ϭ atrial fibrillation; CAD ϭ coronary artery disease; CBC ϭ complete blood count; CHD ϭ coronary heart disease; COX-2 ϭ cyclooxygenase-2; CT ϭ computed tomography; DNR ϭ do not resuscitate; DSM-IV ϭ Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition; HDL ϭ high-density lipoprotein; INR ϭ international normalized ratio; ISD ϭ intrinsic sphincter deficiency; LDL ϭ low-density lipoprotein; LV ϭ left ventricular; MI ϭ myocardial infarction; MRI ϭ magnetic resonance imaging; NSAID ϭ nonsteroidal anti-inflammatory drug; NYHA ϭ New York Heart Association; TIA ϭ transient ischemic attack; TSH ϭ thyroid-stimulating hormone. † Number of eligible patients ϫ pass rate may not be an integer because partial score was awarded if a patient triggered a quality indicator more than once and received recommended care only some of the time.

www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 W-79 Review Evaluation and Management of the Patient with Pulmonary Arterial Hypertension Lewis J. Rubin, MD, and David B. Badesch, MD

Increased pressure in the pulmonary circulation, or pulmonary development of disease-specific therapies for pulmonary arterial hypertension, is a common disorder that may complicate various hypertension over the past decade underscores the importance of cardiopulmonary conditions, including severe obstructive airways diagnosing pulmonary hypertension early in the course of the disease and left ventricular dysfunction. An increase in pulmonary condition and implementing a treatment strategy that is based on arterial pressure that is not due to coexistent cardiopulmonary the condition’s cause and severity. In this review, the authors disease, known as pulmonary arterial hypertension, may occur in present approaches to the diagnosis and management of pulmo- the absence of a demonstrable cause (idiopathic or familial); as a nary arterial hypertension, using a hypothetical case to highlight complication of systemic conditions, such as connective tissue the key management points. disease, HIV infection, or chronic liver disease; or as a result of the Ann Intern Med. 2005;143:282-292. www.annals.org use of fenfluramine anorexigens, amphetamines, or cocaine. The For author affiliations, see end of text.

ntil recently, management of pulmonary arterial hy- WHAT IS PULMONARY HYPERTENSION, AND WHEN IS Upertension (PAH) was generally ineffective in alleviat- IT CONSIDERED TO BE PRESENT? ing symptoms or improving survival. However, the past Pulmonary hypertension is an elevation in pulmonary decade has witnessed remarkable advances in our under- vascular pressure that can be caused by an isolated increase standing of the pathogenesis of PAH, advances that have in pulmonary arterial pressure or by increases in both pul- led to the development of disease-specific treatments. De- monary arterial and pulmonary venous pressures. The term spite these achievements, PAH remains a challenging con- pulmonary arterial hypertension refers to conditions that dition to diagnose and manage. This article reviews recent share common isolated elevations in pulmonary arterial developments in the diagnosis and management of PAH in the context of a typical case, illustrating the importance of pressure (Table 1), hemodynamically defined as a resting collaboration between the internist and the specialist in mean pulmonary arterial pressure greater than 25 mm Hg patient care. with a normal pulmonary capillary or left atrial pressure (Ͻ15 mm Hg) (1, 2). Pulmonary arterial hypertension that A 33-year-old woman presented to her primary care phy- occurs without a demonstrable cause, formerly known as sician with a 6-month history of gradually progressive exer- primary pulmonary hypertension, may occur sporadically (id- tional dyspnea. She had lightheadedness and near-syncope while iopathic PAH) or as an inherited condition (familial PAH). climbing steps. She was mildly obese but had otherwise previously Mutations in the bone morphogenetic protein receptor II been in good health. Her vital signs were normal, her lungs were gene occur in approximately 50% of families with a history clear, and cardiac examination showed a slightly prominent sec- of familial PAH and in nearly 25% of patients believed to ond heart sound and a systolic murmur over the left-heart border. have sporadic idiopathic PAH (3, 4). Genetic testing and An initial diagnosis of exercise-induced asthma was made, and counseling have been recommended for relatives of pa- use of an inhaled bronchodilator was prescribed. tients with familial PAH (4). Pulmonary arterial hyperten- The patient’s symptoms did not diminish. A chest radio- sion occurs in association with connective tissue diseases, graph showed mild cardiomegaly and a slightly prominent particularly scleroderma (5–9); HIV infection (10–12); main pulmonary artery. Pulmonary function tests showed only sickle-cell disease (13); and chronic liver disease (14, 15). a mild reduction in the diffusing capacity for carbon monox- Pulmonary hypertension is suggested when an echocardio- ide, and an electrocardiogram showed right-axis deviation and gram-derived estimate of pulmonary arterial systolic pres- possible right ventricular hypertrophy. On the basis of these sure exceeds 40 mm Hg at rest. results, echocardiography was done, and the echocardiogram showed right ventricular hypertrophy, right atrial enlargement, flattening of the interventricular septum, and moderate tricus- pid regurgitation with an estimated pulmonary artery systolic pressure of 60 mm Hg. Serologic test results did not suggest See also: connective tissue disease. A ventilation–perfusion lung scan was normal. Overnight oximetry showed mild nocturnal oxy- Web-Only gen desaturation, and a formal sleep study excluded sleep ap- CME quiz nea syndrome. The patient was referred to a regional center Conversion of figure and tables into slides with a provisional diagnosis of pulmonary hypertension.

282 © 2005 American College of Physicians Evaluation and Management of the Patient with Pulmonary Arterial Hypertension Review

WHEN SHOULD PAH BE SUSPECTED? Table 1. Nomenclature and Classification of Pulmonary Although PAH may be asymptomatic, exertional dysp- Hypertension* nea is the most frequently encountered symptom (16). Ac- Pulmonary arterial hypertension cordingly, PAH should be suspected in patients with un- Sporadic explained dyspnea. Angina or syncope is less common and Familial portends a poor prognosis. Peripheral edema or ascites in- Related to: Collagen vascular disease dicates right ventricular failure. The symptoms of PAH are Congenital systemic-to-pulmonary shunts (large, small, repaired, or nonspecific and are similar to those in more common dis- nonrepaired) Portal hypertension eases, such as obstructive lung disease and left-sided heart HIV infection disease. Drugs and toxins A family history of pulmonary hypertension may lead Other (glycogen storage disease, Gaucher disease, hereditary hemorrhagic telangiectasia, hemoglobinopathies, myeloproliferative to early recognition of clinical disease in other persons disorders, splenectomy) (17). A history of use of fenfluramine appetite suppressants Associated with significant venous or capillary involvement (18, 19) and current or previous use of amphetamines or Pulmonary veno-occlusive disease cocaine should be explored, because these factors have been Pulmonary capillary hemangiomatosis implicated in the development of PAH in some users. A Pulmonary venous hypertension history of acute requires a careful Left-sided atrial or ventricular heart disease search for chronic thromboembolic pulmonary hyperten- Left-sided valvular heart disease sion, although this condition may occur in the absence of Pulmonary hypertension associated with hypoxemia symptomatic venous thromboembolic disease (20). Chronic obstructive pulmonary disease Interstitial lung disease Sleep-disordered breathing Alveolar hypoventilation disorders WHEN PAH IS SUSPECTED,HOW DO YOU CONFIRM Long-term exposure to high altitude THE DIAGNOSIS AND ESTABLISH A CAUSE? Pulmonary hypertension due to chronic thrombotic or embolic disease The diagnostic strategy systematically uses testing to Thromboembolic obstruction of proximal pulmonary determine whether the patient’s symptoms are due to PAH Thromboembolic obstruction of distal pulmonary arteries and, if so, the underlying cause (Figure). Screening with Pulmonary embolism (tumor, parasites, foreign material) less invasive, less complex, and lower-risk tests is followed Miscellaneous by specific and direct confirmatory tests. Sarcoidosis, histiocytosis X, lymphangiomatosis, compression of pulmonary The electrocardiogram may provide evidence of hemo- vessels (adenopathy, tumor, fibrosing mediastinitis) dynamically significant pulmonary hypertension, such as * Adapted with permission from reference 1. This classification schema was based right ventricular hypertrophy, right-axis deviation, or right on one proposed at the 3rd World Conference on Pulmonary Hypertension, atrial enlargement. Radiographic signs of pulmonary hy- Venice, Italy, 2003. pertension include enlarged main and hilar pulmonary ar- terial shadows (Ͼ17 mm) with attenuation of peripheral possibility of elevated left-sided filling pressures that may pulmonary vascular markings (pruning). Right ventricular contribute to pulmonary hypertension. enlargement is evidenced by anterior displacement of the The evaluation of PAH includes assessment for an un- right ventricle into the retrosternal space on the lateral derlying autoimmune–collagen vascular disorder, includ- view. The chest radiograph is also useful in showing co- ing physical examination and serologic testing for anti- morbid or causal conditions, such as pulmonary venous nuclear antibodies. However, as many as 40% of patients congestion, chronic obstructive pulmonary disease, or in- with idiopathic PAH have serologic abnormalities (16), terstitial lung disease. usually an antinuclear antibody in a low titer and nonspe- Doppler echocardiography is often the first test with cific pattern. Additional serologic studies may be appropri- results that suggest a diagnosis of pulmonary hypertension ate if initial testing suggests an underlying autoimmune (21). Echocardiography also provides information about disorder. the cause and consequences of pulmonary hypertension. Pulmonary function testing is necessary in the initial Studies in patients with PAH (22–25) have reported good evaluation of patients with suspected pulmonary hyperten- correlations between Doppler-derived estimates of pulmo- sion, primarily to exclude or characterize the contribution nary arterial systolic pressure and direct measurements ob- of underlying airways or parenchymal lung disease. In gen- tained by right-heart catheterization. Echocardiography eral, the degree of pulmonary hypertension in patients with also provides direct evidence about left ventricular systolic chronic obstructive lung disease is less severe than that in and diastolic function, as well as valvular function and patients with PAH, and the presence and severity of pul- morphologic characteristics that can give clues to causes of monary hypertension correlate with the degree of airflow pulmonary hypertension due to elevated pulmonary ve- obstruction and hypoxemia. Approximately 20% of pa- nous pressures. Left atrial enlargement, even in the absence tients with idiopathic PAH have a mild restrictive defect of definite left ventricular dysfunction, should raise the (16). In chronic thromboembolic pulmonary hypertension, www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 283 Review Evaluation and Management of the Patient with Pulmonary Arterial Hypertension

Figure. Guideline for approaching the differential diagnosis of pulmonary hypertension (PH).

Is There a Reason To Suspect PAH? Clinical history (symptoms, risk factors, family history), Do examination, radiography, ECG

No Yes

No further Is PAH Rationale evaluation for Likely? TRV to measure RVSP; RVE; RAE; RV dysfunction PAH Do echocardiography

Yes

Is PH due to Yes Diagnosis of LV systolic, diastolic dysfunction and valvular disease: Left-Heart Disease? Appropriate treatment and further evaluation if necessary, including Do echocardiography right- and left-heart catheterization

No

Is PAH due to Yes Diagnosis of abnormal morphologic characteristics, shunt: CHD? Surgery. Medical treatment of PAH or evaluation for further definition or other contribution, Do echocardiography including right- and left-heart catheterization with contrast

No

Is PAH due to Yes Diagnosis of scleroderma, SLE, other CTD, HIV infection: CTD, HIV? Medical treatment of PAH and further evaluation for other contributing causes, Do serologic tests including right- and left-heart catheterization

No

Is Chronic PE YesIs Chronic PE Confirmed Yes Suspected? and Operable? Do VQ scan Do pulmonary angiography

Anatomic definition (CT and MRI may provide No additional useful but not definitive information): Do thromboendarterectomy if appropriate or medical treatment; clotting evaluation No VQ normal

Is PAH due to Lung Disease Yes Diagnosis of parenchymal lung disease, hypoxemia, or sleep disorder: or Hypoxemia? Medical treatment, oxygen, positive-pressure breathing as appropriate, and further Do PFTs, arterial saturation evaluation for other contributing causes, including right-heart catheterization if necessary

No

What Limitations Are Document exercise capacity regardless of cause of PH: Caused by PAH? Establish baseline and prognosis and document progression/response to treatment with Functional class; serial reassessments 6-minute walk test

What Are the Precise Document PA and RA pressures, PCWP (LV or LA pressure if PCWP unobtainable or

Pulmonary uncertain), transpulmonary gradient, CO, PVR, SvO2, response to vasodilators: Hemodynamic Characteristics? Confirm PAH, or IPAH if no other cause identified; discuss genetic testing and counseling Right-heart catheterization of IPAH family members; refer to section on Therapy

CHD ϭ congenital heart disease; CO ϭ cardiac output; CT ϭ contrast-enhanced computed tomography of the chest; CTD ϭ connective tissue disease; ECG ϭ electrocardiogram; Echo ϭ Doppler transthoracic echocardiogram; IPAH ϭ idiopathic pulmonary arterial hypertension; LA ϭ left atrial; LV ϭ left ventricular; MRI ϭ magnetic resonance imaging; PA ϭ pulmonary arterial; PAH ϭ pulmonary arterial hypertension; PCWP ϭ pulmonary capillary wedge pressure; PE ϭ pulmonary embolism; PFTs ϭ pulmonary function tests; PVR ϭ pulmonary vascular resistance; RA ϭ right atrial; RAE ϭ right atrial enlargement; RV ϭ right ventricular; RVE ϭ right ventricular enlargement; RVSP ϭ right ventricular systolic pressure; SLE ϭ systemic lupus erythematosus; ϭ ϭ ϭ SvO2 mixed venous oxygen saturation; TRV tricuspid regurgitant velocity; VQ ventilation–perfusion. (Adapted with permission from reference 1.)

284 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Evaluation and Management of the Patient with Pulmonary Arterial Hypertension Review a mild to moderate restrictive defect is thought to be due to inantly in the lower lobes is also suggestive of pulmonary parenchymal scarring from previous infarctions (26). In veno-occlusive disease (35). both conditions, the diffusing capacity for carbon monox- Cardiac catheterization is ultimately required to con- ide is often mildly to moderately reduced (16, 26). Mild to firm the presence of pulmonary hypertension, definitively moderate arterial hypoxemia is due to ventilation–perfu- establish its cause, assess severity, and guide therapy. Open sion mismatch and reduced mixed venous oxygen satura- or thoracoscopic lung biopsy entails substantial risk for tion resulting from low cardiac output. Severe hypoxemia hemorrhage, hypoxemia, and death in patients with PAH. is due to right-to-left intracardiac or intrapulmonary Because it has a low likelihood of altering the clinical di- shunting. Twenty percent of patients with systemic sclero- agnosis, routine biopsy is discouraged. Under certain cir- sis have an isolated reduction in diffusing capacity (27); a cumstances, histopathologic diagnosis may be needed diffusing capacity less than 45% to 55% of predicted or when vasculitis, granulomatous or interstitial lung disease, one that is decreasing may signal the development of pul- pulmonary veno-occlusive disease, or bronchiolitis are sug- monary hypertension (28). gested clinically (36). Overnight oximetry may show frequent desaturations and may be the first clue to sleep apnea sufficient to con- Initial evaluation at the referral center included a 6-min tribute to pulmonary hypertension. Nocturnal hypoxemia walk test of 305 m. Right-heart catheterization showed a pul- occurs in more than 75% of patients with idiopathic PAH monary arterial pressure of 65/30 mm Hg (mean, 42 mm without sleep apnea (29). Because hypoxemia is a potent Hg), right atrial pressure of 12 mm Hg, pulmonary capillary pulmonary vasoconstrictor, all patients with unexplained wedge pressure of 6 mm Hg, and cardiac output of 3.2 L/min. pulmonary hypertension require assessment of oxygen sat- Little change was seen in hemodynamic measurements with uration during both sleep and exercise (30). the inhalation of nitric oxide. The patient was thought to have Chronic thromboembolic pulmonary hypertension is idiopathic PAH, falling into New York Heart Association potentially curable and should be sought in all patients (NYHA) functional class III on the basis of her symptoms. with pulmonary hypertension. Ventilation–perfusion lung Treatment options were discussed with the patient and her scanning is the preferred test to evaluate for this condition family. Medical therapy, diet, exercise, travel, altitude expo- (4). Chronic thromboembolic pulmonary hypertension sure, and were all discussed. manifests as at least 1 segmental-sized or larger perfusion There are few data on which to base recommendations defect, which is typically mismatched and larger than ven- about physical activity or cardiopulmonary rehabilitation tilation abnormalities (20, 31, 32). Patchy, nonsegmental in patients with PAH. Potentially hazardous exposure ac- defects are less specific but may be associated with chronic tivities are listed in Table 2. We encourage cautious, grad- thromboembolic pulmonary hypertension. Perfusion scans uated physical activity. Heavy physical activity can precip- tend to underestimate the extent of large-vessel obstruction itate exertional syncope. Hot baths or showers are in this condition (32). Although a normal perfusion scan discouraged because the resultant peripheral vasodilatation essentially excludes surgically accessible chronic thrombo- can produce systemic hypotension and syncope. Excessive embolic disease, scans suggestive of thromboembolic dis- sodium intake can contribute to fluid retention. ease may also be seen in patients with pulmonary arterial We discourage exposure to high altitudes (more than sarcoma, large-vessel pulmonary arteritis, extrinsic vascular approximately 1800 m above sea level), as this may pro- compression, or pulmonary veno-occlusive disease (33). duce hypoxic pulmonary and further Pulmonary angiography is the definitive test for diagnosing compromise oxygen transport. Supplemental oxygen chronic thromboembolic pulmonary hypertension and de- should be used to maintain oxygen saturations greater than termining operability, and it should be done in experi- 91%. Air travel can be problematic for patients with PAH, enced centers when chronic thromboembolic pulmonary as commercial aircraft are typically pressurized to the hypertension remains possible. equivalent of approximately 2400 m above sea level. While Computed tomographic scanning may suggest a cause on commercial aircraft, patients who have borderline oxy- of PAH, such as severe airway or parenchymal lung dis- gen saturations at sea level may require 3 to 4 L/min of eases. A spectrum of abnormalities on computed tomo- supplemental oxygen, and those already using supplemen- graphic scans has been described in chronic thromboem- tal oxygen at sea level should increase their oxygen flow bolic pulmonary hypertension, including right ventricular rate. Because of the potentially devastating effects of respi- enlargement, dilated central pulmonary arteries, chronic ratory , immunization against influenza and thromboembolic material within the central pulmonary ar- pneumococcal pneumonia is recommended. teries, increased bronchial artery collateral flow, variability The hemodynamic changes occurring during preg- in the size and distribution of pulmonary arteries, paren- nancy impose substantial stress in women with PAH, lead- chymal abnormalities consistent with previous infarctions, ing to a 30% to 50% mortality rate (37, 38). Although and mosaic attenuation of the pulmonary parenchyma successful use of long-term intravenous epoprostenol and (34). A ground-glass, mosaic-attenuation pattern predom- inhaled nitric oxide to treat pregnant patients with idio- www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 285 Review Evaluation and Management of the Patient with Pulmonary Arterial Hypertension

Table 2. Potentially Hazardous Activities for Patients with Pulmonary Arterial Hypertension

Activity Potential Adverse Effects Recommendations Exposure to high altitude Hypoxemia, pulmonary vasoconstriction, Avoid altitudes Ͼ 1800 m above sea level; use worsening pulmonary hypertension, supplemental oxygen as needed to keep right-heart failure oxygen saturation Ն 91% at all times Air travel Hypoxemia, pulmonary vasoconstriction, Use supplemental oxygen as needed to keep worsening pulmonary hypertension oxygen saturation Ն 91% at all times Heavy exertion Near-syncope, syncope Engage in low-level activity or cautious, graduated exercise, such as walking Bending over and rising quickly Near-syncope, syncope Rise slowly from bending, sitting, or lying positions Use of decongestant medications Vasoconstriction, worsening pulmonary Avoid using decongestants; consider nonsedating hypertension antihistamines or local treatments, such as nasal steroids Use of appetite suppressants or diet pills Worsening pulmonary hypertension Have dietary and nutritional consultation; engage in cautious low-level exercise High sodium intake Fluid retention, right-heart failure Follow 2-g sodium diet Cigarette smoking Worsening of intrinsic lung disease; nicotine Stop smoking (preferably without use of nicotine is a vasoconstrictor and may contribute replacement therapy) to worsening pulmonary hypertension

pathic PAH has been reported (39–42), most experts rec- carbia, and shifts in intrathoracic pressure and associated ommend early termination of the pregnancy (43). changes in cardiac filling pressures. Estrogen-containing contraceptives may increase risk for venous thromboembolism and are not recommended WHAT ARE THE TREATMENTS FOR PAH, AND HOW for women of childbearing potential with PAH. Also, the DOES ONE CHOOSE AND MONITOR THERAPY? endothelin- bosentan may decrease the efficacy of hormonal contraception, and dual mechanical Medical therapy for PAH has recently been addressed barrier contraceptive techniques are recommended in in detail in 2 major consensus documents (43, 44) that use women of childbearing age using this medication. similar evidence-based therapeutic algorithms. Physicians should discuss with their patients the use of General Measures any concomitant medications or herbal preparations. The Two small retrospective studies (45, 46) reported im- use of vasoconstricting sinus or cold medications (such as proved survival with oral anticoagulation in patients with pseudoephedrine) or the use of serotonergic medications idiopathic PAH. On the basis of these reports and the for migraine headaches may be problematic. Concomitant knowledge that microscopic in situ thrombosis can occur, use of glyburide or cyclosporine with bosentan is contrain- anticoagulation with warfarin is recommended. Although dicated, and use of azole-type antifungal agents is discour- little evidence is available to guide such therapy, current aged because of potential drug interactions that may in- consensus suggests a target international normalized ratio crease risk for hepatotoxicity. Patients using warfarin of approximately 1.5 to 2.5 (43, 44). Anticoagulation is should be cautioned about the many drug interactions pos- controversial for patients who have PAH due to other sible with this medication. Bosentan may slightly decrease causes, such as scleroderma or congenital heart disease, be- international normalized ratios in patients using warfarin. cause of a lack of evidence supporting efficacy, an increased Invasive procedures and surgery can be associated with risk for gastrointestinal bleeding in patients with sclero- increased operative and perioperative risks. Patients with derma, and an increased risk for hemoptysis in patients severe PAH are especially prone to vasovagal events leading with congenital heart disease. The relative risks and bene- to syncope, cardiopulmonary arrest, and death. Cardiac fits of anticoagulant therapy should be considered on a output is particularly dependent on heart rate in this set- case-by-case basis. Patients with documented right-to-left ting, and the bradycardia and systemic vasodilatation ac- intracardiac shunting due to an atrial septal defect or companying a vasovagal event can result in hypotension. patent foramen ovale and a history of transient ischemic Heart rate should be monitored during invasive proce- attack or embolic stroke should receive anticoagulation. dures, and an anticholinergic agent should be readily avail- Patients treated with long-term intravenous epoprostenol able. Oversedation can lead to ventilatory insufficiency and generally receive anticoagulation in the absence of contra- precipitate clinical deterioration. Caution should be used indications, partly because of the additional risk for cathe- with laparoscopic procedures in which carbon dioxide is ter-associated thrombosis. used for abdominal insufflation, as absorption can produce Diuretics are indicated for right ventricular volume hypercarbia, which is a pulmonary vasoconstrictor. Anes- overload. However, rapid and excessive diuresis may pre- thesia and intubation can be particularly problematic be- cipitate systemic hypotension and renal insufficiency. Spi- cause they can induce vasovagal events, hypoxemia, hyper- ronolactone, an aldosterone antagonist that is of benefit in

286 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Evaluation and Management of the Patient with Pulmonary Arterial Hypertension Review patients with left-heart failure, is also used by some experts continuous intravenous infusion. Because epoprostenol has to treat right-heart failure. Serum electrolytes and renal a short half-life and has irritant effects on peripheral veins, function should be monitored closely. and because of the risk for rebound worsening with inter- Hypoxemia is a pulmonary vasoconstrictor, and sup- ruption of the infusion, the drug should be administered plemental oxygen should be used to maintain an oxygen through an indwelling central venous . Common saturation greater than 90%. Supplemental oxygen use is side effects include headache, flushing, jaw pain, , more controversial in patients with Eisenmenger complex nausea, a blotchy erythematous rash, and musculoskeletal but may decrease the need for phlebotomy and reduce the pain. Acute overdose can lead to systemic hypotension, and occurrence of neurologic complications (47). long-term overdosage can produce a hyperdynamic circu- Although not extensively studied in PAH, digitalis is latory state with high-output cardiac failure (60). Serious sometimes used for refractory right ventricular failure (48). complications include catheter-related and thrombo- In addition, atrial flutter or other atrial dysrhythmias often sis. Although epoprostenol is approved by the U.S. Food complicate late-stage right-heart dysfunction, and digoxin and Drug Administration for patients in functional NYHA may be useful for rate control. class III and IV with idiopathic PAH or PAH due to scleroderma, it is generally reserved for those with ad- Vasodilator Testing and Calcium-Channel Blockers vanced disease refractory to oral therapies. Because of its Patients with idiopathic PAH who respond to vasodi- complexity, epoprostenol therapy should be given in cen- lators in the short term have improved survival with long- ters experienced with its administration. term use of calcium-channel blockers (46, 49). Various is an orally active analogue short-acting agents, including intravenous epoprostenol or (61). In a 12-week trial done in patients with PAH in adenosine and inhaled nitric oxide, have been used to test NYHA functional class II and III, beraprost increased the short-term response to vasodilators (50, 51). The recently 6-min walking distance without substantially affecting car- formulated consensus definition of a positive short-term diopulmonary hemodynamic variables or survival (62). response to vasodilators in PAH is a decrease of at least 10 However, a longer-term trial (63) found that improvement mm Hg in mean pulmonary arterial pressure to 40 mm Hg was not present after 9 to 12 months. Beraprost is ap- or less, with increased or unchanged cardiac output (43, proved for treatment of PAH in Japan. 44). Most experts feel that true vasoreactivity is uncom- is a stable prostacyclin analogue with a mon, occurring in approximately 10% of patients with half-life of 3 hours, and it can be administered subcutane- idiopathic PAH and rarely in other forms of PAH. Vaso- ously or intravenously. In a multicenter study of subcuta- reactivity testing should be done in experienced centers. neously infused treprostinil in patients with PAH (64), Only patients who have a substantial response to a 6-min walking distance and hemodynamic variables im- short-acting vasodilator should be considered candidates proved modestly with treprostinil compared with placebo. for treatment with oral calcium-channel blockers; treat- Common side effects include pain at the infusion site, ment should be monitored closely because maintenance of headache, diarrhea, nausea, rash, and jaw pain. Treprostinil response is not universal. Agents with negative inotropic is approved by the U.S. Food and Drug Administration for effects, such as verapamil, should be avoided. subcutaneous or intravenous treatment of patients with PAH in NYHA functional class II, III, and IV and is gen- Prostacyclin is a metabolite of pro- erally used when oral therapy has failed to produce benefit. duced in vascular endothelium. It is a potent vasodilator, To receive treprostinil therapy, patients should be referred affecting both the pulmonary and systemic circulations, to experienced centers. and has antiplatelet aggregator effects. A relative deficiency is a stable prostacyclin analogue with a serum of endogenous prostacyclin may contribute to the patho- half-life of 20 to 25 minutes. Although it may be possible genesis of PAH (52, 53). to administer iloprost intravenously, the drug has gained In idiopathic PAH, continuous intravenous infusion attention recently for delivery by inhalation. In idiopathic of epoprostenol improved exercise capacity, as assessed by PAH, short-term inhalation of iloprost resulted in greater the 6-min walk test; cardiopulmonary hemodynamic vari- pulmonary vasodilatation than did nitric oxide (65). For ables; and survival compared with conventional therapy long-term use, the relatively short duration of action of (such as oral vasodilators and anticoagulation) (54). A sim- inhaled iloprost necessitates 6 to 9 inhalations per day (66, ilar study (55) showed that epoprostenol improved exercise 67). In a 3-month multicenter trial, iloprost administered capacity and hemodynamic variables in patients with PAH in a dosage of 2.5 or 5 ␮g 6 or 9 times daily improved the due to the scleroderma spectrum of disease. The beneficial 6-min walking distance and NYHA functional class (68). effects of epoprostenol therapy are sustained for years in Hemodynamic variables measured after iloprost inhalation many patients with idiopathic PAH (56–59), and epopro- were also improved at 3 months. Cough, flushing, and stenol therapy remains the treatment of choice for the most headache occurred more frequently in the iloprost group severely ill patients. than in the group receiving placebo. Inhaled iloprost may Epoprostenol therapy is complicated by the need for be useful as an adjunct to oral therapy. It is approved in www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 287 Review Evaluation and Management of the Patient with Pulmonary Arterial Hypertension

Europe for patients with idiopathic PAH in NYHA func- (type 5 inhibitors) augment the pulmonary vascular re- tional class III and was recently approved by the U.S. Food sponse to endogenous or inhaled nitric oxide in models of and Drug Administration for patients with PAH in NYHA pulmonary hypertension (79, 80). functional class III and IV. Sildenafil is a highly specific phosphodiesterase-5 in- hibitor currently approved for erectile dysfunction. Silde- Endothelin-Receptor Antagonists nafil reduces pulmonary arterial pressure in patients with Endothelin-1 is a potent vasoconstrictor and smooth- PAH in the short term (81) and both augments and pro- muscle mitogen that may contribute to increased vascular longs the effects of inhaled nitric oxide (81, 82), preventing tone and proliferation in PAH (69). Endothelin-1 expres- rebound pulmonary vasoconstriction after withdrawal of sion, production, and concentration in plasma (70) and nitric oxide (83). The combination of sildenafil and aero- lung tissue (71) are elevated in PAH, and levels correlate solized iloprost caused a greater, more prolonged decrease with disease severity (72). Two endothelin-receptor iso- in pulmonary vascular resistance than did either agent forms, endothelin-A (ETA) and endothelin-B (ETB), have alone (84). Several single-center, nonrandomized studies of been identified. Activation of ETA receptors facilitates va- patients with PAH treated with long-term sildenafil ther- soconstriction and proliferation of vascular smooth-muscle apy are promising (85–88). A multicenter clinical trial was cell. The ETB receptors are thought to be principally in- recently completed and presented in abstract form (89). volved in the clearance of endothelin, particularly in the Sildenafil has recently been approved by the U.S. Food and vascular beds of the lung and kidney. Activation of ETB receptors may also cause vasodilatation and nitric oxide Drug Administration for use in PAH. release. Whether it is preferable to block both ETA and ETB receptors or to selectively target ETA receptors is con- The patient began receiving oral bosentan, 62.5 mg twice troversial. It has been argued that selective ETA-receptor daily, and the dosage was titrated to 125 mg twice daily after antagonism may be beneficial for the treatment of PAH 1 month. An echocardiogram 3 months later showed an esti- because of maintenance of the vasodilator and clearance mated pulmonary arterial systolic pressure of 55 mm Hg, and functions of ETB receptors. results of the 6-min walk test had increased by 35 m. The A multicenter study of bosentan, a dual ETA/ETB- patient reported less dyspnea and greater activity tolerance. She receptor antagonist, showed a mean change of 76 m in saw her specialist every 3 months and had follow-up visits with 6-min walking distance (73). Bosentan also improved car- her primary care physician. Two years after starting bosentan diopulmonary hemodynamic variables and NYHA func- therapy, the patient reported gradually worsening fatigue, dysp- tional class compared with placebo. Asymptomatic in- nea on exertion, ankle edema, and an episode of near-syncope. creases in hepatic aminotransferase levels were seen in 2 The 6-min walk test results were slightly worse compared with bosentan-treated patients, but the levels normalized with- baseline, and right-heart catheterization showed a pulmonary out discontinuation of therapy or change of dose. arterial pressure of 75/36 mm Hg (mean, 48 mm Hg). Right Similar improvement in exercise capacity, alleviation atrial pressure was 13 mm Hg, pulmonary capillary wedge of symptoms, and alleviation of clinical worsening were pressure was 7 mm Hg, and cardiac output was 2.7 L/min. seen in a larger multicenter study (74). Because of the risk An indwelling was placed, and long- for hepatotoxicity, the U.S. Food and Drug Administra- term intravenous epoprostenol therapy was initiated. Although tion requires that liver function tests be done at least relatively little evidence supports combined use of bosentan and monthly. Bosentan may also cause anemia, edema, and epoprostenol, bosentan treatment was continued during the teratogenicity. The endothelin antagonists as a class may initiation of epoprostenol therapy. The patient also began re- produce testicular atrophy and male infertility. Bosentan is ceiving diuretic therapy and was referred for evaluation for approved in the United States for patients with PAH in lung transplantation. Eighteen months later, fatigue had less- NYHA class III and IV and in Europe for those in NYHA ened, ankle edema had resolved, and the 6-min walking dis- class III. Two selective ETA-receptor antagonists, sitaxsen- tance was 330 m. The patient decided not to pursue lung tan and ambrisentan, are currently in phase III clinical transplantation at this time. trials and remain investigational agents (75). Interventional and Surgical Therapies Phosphodiesterase-5 Inhibitors Atrial septostomy involves the creation of a right-to- The effects of nitric oxide, a pulmonary arterial vaso- left interatrial shunt to decompress the failing pressure- dilator, depend on its augmentation of cyclic guanosine and volume-overloaded right heart (90). Where advanced monophosphate (cGMP) content in vascular smooth mus- medical therapies are available, atrial septostomy is seen cle. The effects of intracellular cGMP are short-lived be- largely as a palliative procedure or as a stabilizing bridge to cause of the rapid degradation of cGMP by phosphodies- lung transplantation (91). In regions of the world lacking terases (76, 77). Phosphodiesterase-5 is strongly expressed access to advanced medical therapies, atrial septostomy in the lung, and its gene expression and activity are in- may be the best treatment option available (92). Patient creased in chronic pulmonary hypertension (78). Drugs selection, timing, and appropriate sizing of the septostomy that selectively inhibit cGMP-specific phosphodiesterase are crucial to optimizing outcomes.

288 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Evaluation and Management of the Patient with Pulmonary Arterial Hypertension Review

Table 3. Indications for Referral to a Specialized Center for We generally recommend maintaining a euvolemic state in Treatment of Pulmonary Arterial Hypertension* these patients. Peripheral edema, ascites, or marked jugular venous distention suggests right ventricular dysfunction Unexplained dyspnea on exertion with evidence of PAH on echocardiography and volume overload. Patients should regularly monitor Evidence of moderate to severe PAH body weight and contact their internist or pulmonary hy- Estimated pulmonary arterial systolic pressure Ͼ45 mm Hg on pertension specialist if their weight increases by more than echocardiography Symptoms consistent with New York Heart Association functional class II several pounds. Timely intervention limiting fluid and so- Near-syncope or syncope dium intake, or adjusting diuretic dosing, may avert the Absence of substantial left-sided cardiac disease or parenchymal lung need to hospitalize the patient to administer intravenous disease Clinical or echocardiographic evidence of right ventricular dysfunction diuretics. Emergency care in the patient’s home area may Lower-extremity edema also be facilitated by a close relationship with an internist. Ascites Right ventricular enlargement or systolic dysfunction on echocardiography FUTURE DIRECTIONS ϭ * PAH pulmonary arterial hypertension. Considerable interest has been seen in combining ther- apeutic agents with different mechanisms of action, as has Lung transplantation is particularly challenging in been done in the treatment of systemic hypertension and PAH (93) and is usually reserved for patients whose con- cancer. For example, phosphodiesterase inhibitors may en- ditions are deteriorating despite the best available medical hance and prolong the effects of prostanoids. Such combi- therapy. Survival in patients with PAH having lung trans- nations offer the possibility of enhanced efficacy and may plantation is approximately 66% to 75% at 1 year (94). permit individual agents to be used in lower doses, mini- Most centers prefer bilateral lung transplantation for pa- mizing toxicity. To date, only a few small studies of com- tients with PAH (95). bination therapy have been done (97, 98). The development of treatments for PAH has created Future studies targeting newly identified alterations in the challenge of how best to monitor the effects of long- endothelial and smooth-muscle cell function, including va- term therapy. Noninvasive markers of disease severity, such soactive intestinal peptide synthesis and activity (99), an- as results on the 6-min walk test and assessment of NYHA giopoeitin (100), and the serotonin pathway (101), may functional class, are useful in identifying stability or dete- provide novel treatments. rioration (96). Echocardiographic indices of right-heart Drugs that are currently marketed to treat other con- size and function may also be useful in noninvasive mon- itoring of therapy. Repeated cardiac catheterization may be Table 4. Roles of the Internist and the Specialist in the a good option when noninvasive measures suggest deterio- Diagnosis and Coordinated Care of Patients with Pulmonary ration and alternative therapies are being considered, be- Arterial Hypertension cause the choice of therapy may be governed by the sever- ity of hemodynamic abnormality. Biomarkers, such as Internist brain natriuretic peptide and troponin levels, or the results Recognize possible pulmonary arterial hypertension in the patient presenting with unexplained dyspnea on exertion of physiologic studies, such as cardiopulmonary stress test- Initiate appropriate screening evaluation ing (96), are also being used to monitor clinical course. Chest radiography, pulmonary function testing, echocardiography, ventilation–perfusion lung scanning, oximetry Facilitate appropriate referral to specialty center WHAT IS THE ROLE OF THE PRIMARY CARE PHYSICIAN Contact a specialist in pulmonary hypertension Obtain appropriate referrals and approvals from the patient’s insurer IN THE MANAGEMENT OF THE PATIENT WITH PAH? Provide the patient’s records to the specialty center The importance of coordinated care by the general Provide regular follow-up in the patient’s local community Assess volume status, vital signs, and oxygenation internist and the specialist cannot be overemphasized. Monitor laboratory test results, including serum electrolyte levels, renal Early recognition, diagnosis, and referral (Table 3)toa function, and results of liver function tests center with expertise in management of PAH can make a Manage low-dose anticoagulation with warfarin, if indicated Provide emergency care in the patient’s local community substantial difference in outcome. Specialized centers car- ing for patients with PAH are generally led by a cardiolo- Specialist gist or pulmonologist, and a list of centers can be found on Complete the diagnostic evaluation Right-heart catheterization the Web site of the Pulmonary Hypertension Association Pulmonary angiography if indicated (www.phassociation.org). Because of the special expertise Discuss treatment options with the patient required, referral centers often prefer to do the invasive Initiate therapy Provide ongoing monitoring of pulmonary hypertension procedures and initiate therapy. Long-term follow-up can See patient at 3-month intervals be provided in both the internist’s and the specialist’s prac- Physical examination, 6-min walking test, echocardiography tice (Table 4). The internist can play an important role in Monitor right-heart catheterization as appropriate Usually done for clinical deterioration before adding new treatment monitoring the patient’s fluid balance, electrolytes, renal specific to pulmonary hypertension function, liver test results, and levels of anticoagulation. www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 289 Review Evaluation and Management of the Patient with Pulmonary Arterial Hypertension ditions may have beneficial effects in PAH. For example, Prevalence of pulmonary hypertension in limited and diffuse scleroderma. Chest. the 3-hydroxy-3-methylglutaryl coenzyme A reductase in- 1996;110:1515-9. [PMID: 8989070] 6. Ungerer RG, Tashkin DP, Furst D, Clements PJ, Gong H Jr, Bein M, et al. hibitors manifest pleiotropic effects that may be responsi- Prevalence and clinical correlates of pulmonary arterial hypertension in progres- ble for a component of their benefit in arteriosclerotic dis- sive systemic sclerosis. Am J Med. 1983;75:65-74. [PMID: 6859087] ease, and these agents attenuate pulmonary arteriopathy in 7. Tanaka E, Harigai M, Tanaka M, Kawaguchi Y, Hara M, Kamatani N. animal models (102). Similarly, currently available Pulmonary hypertension in systemic lupus erythematosus: evaluation of clinical characteristics and response to immunosuppressive treatment. J Rheumatol. inhibitors (such as aspirin) and newer antithrombotic 2002;29:282-7. [PMID: 11838845] agents may have a role in the treatment of patients with 8. Fagan KA, Badesch DB. Pulmonary hypertension associated with connective PAH, in light of the beneficial effects and inherent risks of tissue disease. In: Peacock AJ, Rubin LJ, eds. Pulmonary Circulation. London: anticoagulation with warfarin in these patients. Finally, Arnold Publishers; 2004:181-90. 9. Love PE, Santoro SA. Antiphospholipid antibodies: anticardiolipin and the given the important structural changes seen in severe PAH, lupus anticoagulant in systemic lupus erythematosus (SLE) and in non-SLE dis- inhibitors of vascular cellular proliferation and angiogenesis orders. Relevance and clinical significance. Ann Intern Med. 1990;112:682-98. warrant clinical investigation. [PMID: 2110431] 10. Opravil M, Pechere M, Speich R, Joller-Jemelka HI, Jenni R, Russi EW, et From the University of California, San Diego, Medical Center, La Jolla, al. HIV-associated primary pulmonary hypertension. A case control study. Swiss HIV Cohort Study. Am J Respir Crit Care Med. 1997;155:990-5. [PMID: California; and University of Colorado Health Sciences Center, Denver, 9117037] Colorado. 11. Petitpretz P, Brenot F, Azarian R, Parent F, Rain B, Herve P, et al. Pulmonary hypertension in patients with human immunodeficiency virus infec- Acknowledgments: The authors thank their colleagues who participated tion. Comparison with primary pulmonary hypertension. Circulation. 1994;89: in the 3rd World Symposium on Pulmonary Artery Hypertension and 2722-7. [PMID: 8205687] who served on the ACCP Evidence-Based Guidelines panel, from whose 12. Mehta NJ, Khan IA, Mehta RN, Sepkowitz DA. HIV-related pulmonary work they have drawn in the preparation of this manuscript. hypertension: analytic review of 131 cases. Chest. 2000;118:1133-41. [PMID: 11035689] 13. Gladwin MT, Sachdev V, Jison ML, Shizukuda Y, Plehn JF, Minter K, et Potential Financial Conflicts of Interest: Consultancies: L.J. Rubin al. Pulmonary hypertension as a risk factor for death in patients with sickle cell (Actelion, Pfizer Inc., Schering, CoTherix, United Therapeutics, Myo- disease. N Engl J Med. 2004;350:886-95. [PMID: 14985486] gen), D.B. Badesch (GlaxoWellcome/GlaxoSmithKline, Actelion, Berlex, 14. Kuo PC, Plotkin JS, Johnson LB, Howell CD, Laurin JM, Bartlett ST, et Astra Merck, AstraZeneca, Myogen, Intermune, Forest Laboratories, En- al. Distinctive clinical features of portopulmonary hypertension. Chest. 1997; cysive, Exhale Therapeutics/CoTherix, Pfizer, Scios, MondoBiotech, PR 112:980-6. [PMID: 9377962] Pharmaceuticals); Honoraria: L.J. Rubin (Actelion, Myogen), D.B. Ba- 15. Castro M, Krowka MJ, Schroeder DR, Beck KC, Plevak DJ, Rettke SR, et desch (Glaxo Wellcome/GlaxoSmithKline, Actelion, Pfizer Inc., En- al. Frequency and clinical implications of increased pulmonary artery pressures in cysive, United Therapeutics); Stock ownership or options (other than mu- liver transplant patients. Mayo Clin Proc. 1996;71:543-51. [PMID: 8642882] 16. Rich S, Dantzker DR, Ayres SM, Bergofsky EH, Brundage BH, Detre KM, tual funds): D.B. Badesch (Johnson & Johnson); Expert testimony: L.J. et al. Primary pulmonary hypertension. A national prospective study. Ann Intern Rubin (Appetite suppressant litigation), D.B. Badesch (Appetite sup- Med. 1987;107:216-23. [PMID: 3605900] pressant litigation); Grants received: L.J. Rubin (Actelion, Pfizer Inc., 17. Grunig E, Janssen B, Mereles D, Barth U, Borst MM, Vogt IR, et al. CoTherix, Myogen, National Institutes of Health), D.B. Badesch (Glaxo Abnormal pulmonary artery pressure response in asymptomatic carriers of pri- Wellcome/GlaxoSmithKline, United Therapeutics, Boehringer In- mary pulmonary hypertension gene. Circulation. 2000;102:1145-50. [PMID: gelheim, Actelion, ICOS Corp./Texas Biotechnologies, Encysive, Pfizer 10973844] Inc., Myogen, CoTherix, Lilly/ICOS, American Lung Association, 18. Abenhaim L, Moride Y, Brenot F, Rich S, Benichou J, Kurz X, et al. American Heart Association, National Institutes of Health, Scleroderma Appetite-suppressant drugs and the risk of primary pulmonary hypertension. In- Foundation). ternational Primary Pulmonary Hypertension Study Group. N Engl J Med. 1996;335:609-16. [PMID: 8692238] 19. Rich S, Rubin L, Walker AM, Schneeweiss S, Abenhaim L. Anorexigens Requests for Single Reprints: Lewis J. Rubin, MD, University of Cal- and pulmonary hypertension in the United States: results from the surveillance of ifornia, San Diego, Medical Center, 9300 Campus Point Drive, M/C North American pulmonary hypertension. Chest. 2000;117:870-4. [PMID: 7381, La Jolla, CA 92037-7381. 10713017] 20. Fedullo PF, Auger WR, Kerr KM, Rubin LJ. Chronic thromboembol- Current author addresses are available at www.annals.org. ic pulmonary hypertension. N Engl J Med. 2001;345:1465-72. [PMID: 11794196] 21. Bossone E, Rubenfire M, Bach DS, Ricciardi M, Armstrong WF. Range of tricuspid regurgitation velocity at rest and during exercise in normal adult men: References implications for the diagnosis of pulmonary hypertension. J Am Coll Cardiol. 1. Diagnosis and management of pulmonary arterial hypertension: ACCP evi- 1999;33:1662-6. [PMID: 10334439] dence-based clinical practice guideline. Chest. 2004;125(Suppl):1S-92S. 22. Brecker SJ, Gibbs JS, Fox KM, Yacoub MH, Gibson DG. Comparison of 2. Rubin LJ. Primary pulmonary hypertension. N Engl J Med. 1997;336:111-7. Doppler derived haemodynamic variables and simultaneous high fidelity pressure [PMID: 8988890] measurements in severe pulmonary hypertension. Br Heart J. 1994;72:384-9. 3. Thomson JR, Machado RD, Pauciulo MW, Morgan NV, Humbert M, [PMID: 7833199] Elliott GC, et al. Sporadic primary pulmonary hypertension is associated with 23. Denton CP, Cailes JB, Phillips GD, Wells AU, Black CM, Bois RM. germline mutations of the gene encoding BMPR-II, a receptor member of the Comparison of Doppler echocardiography and right heart catheterization to as- TGF-beta family. J Med Genet. 2000;37:741-5. [PMID: 11015450] sess pulmonary hypertension in systemic sclerosis. Br J Rheumatol. 1997;36:239- 4. McGoon M, Gutterman D, Steen V, Barst R, McCrory DC, Fortin TA, et 43. [PMID: 9133938] al. Screening, early detection and diagnosis of pulmonary arterial hypertension: 24. Shapiro SM, Oudiz RJ, Cao T, Romano MA, Beckmann XJ, Georgiou D, ACCP evidence-based clinical practice guidelines. Chest. 2004;126(Suppl):14S- et al. Primary pulmonary hypertension: improved long-term effects and survival 34S. [PMID: 15249493] with continuous intravenous epoprostenol infusion. J Am Coll Cardiol. 1997;30: 5. Battle RW, Davitt MA, Cooper SM, Buckley LM, Leib ES, Beglin PA, et al. 343-9. [PMID: 9247503]

290 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Evaluation and Management of the Patient with Pulmonary Arterial Hypertension Review

25. Hinderliter AL, Willis PW 4th, Barst RJ, Rich S, Rubin LJ, Badesch DB, 327:76-81. [PMID: 1603139] et al. Effects of long-term infusion of prostacyclin (epoprostenol) on echocardio- 47. Sandoval J, Aguirre JS, Pulido T, Martinez-Guerra ML, Santos E, Alvarado graphic measures of right ventricular structure and function in primary pulmo- P, et al. Nocturnal oxygen therapy in patients with the Eisenmenger syndrome. nary hypertension. Primary Pulmonary Hypertension Study Group. Circulation. Am J Respir Crit Care Med. 2001;164:1682-7. [PMID: 11719310] 1997;95:1479-86. [PMID: 9118516] 48. Rich S, Seidlitz M, Dodin E, Osimani D, Judd D, Genthner D, et al. The 26. Morris TA, Auger WR, Ysrael MZ, Olson LK, Channick RN, Fedullo PF, short-term effects of digoxin in patients with right ventricular dysfunction from et al. Parenchymal scarring is associated with restrictive spirometric defects in pulmonary hypertension. Chest. 1998;114:787-92. [PMID: 9743167] patients with chronic thromboembolic pulmonary hypertension. Chest. 1996; 49. Ogata M, Ohe M, Shirato K, Takishima T. Effects of a combination ther- 110:399-403. [PMID: 8697841] apy of anticoagulant and vasodilator on the long-term prognosis of primary pul- 27. Owens GR, Fino GJ, Herbert DL, Steen VD, Medsger TA Jr, Pennock BE, monary hypertension. Jpn Circ J. 1993;57:63-9. [PMID: 8437343] et al. Pulmonary function in progressive systemic sclerosis. Comparison of 50. Sitbon O, Brenot F, Denjean A, Bergeron A, Parent F, Azarian R, et al. CREST syndrome variant with diffuse scleroderma. Chest. 1983;84:546-50. Inhaled nitric oxide as a screening vasodilator agent in primary pulmonary hy- [PMID: 6628005] pertension. A dose-response study and comparison with prostacyclin. Am J Respir 28. Steen V, Medsger TA Jr. Predictors of isolated pulmonary hypertension in Crit Care Med. 1995;151:384-9. [PMID: 7842196] patients with systemic sclerosis and limited cutaneous involvement. Arthritis 51. Schrader BJ, Inbar S, Kaufmann L, Vestal RE, Rich S. Comparison of the Rheum. 2003;48:516-22. [PMID: 12571862] effects of adenosine and nifedipine in pulmonary hypertension. J Am Coll Car- 29. Rafanan AL, Golish JA, Dinner DS, Hague LK, Arroliga AC. Nocturnal diol. 1992;19:1060-4. [PMID: 1552096] hypoxemia is common in primary pulmonary hypertension. Chest. 2001;120: 52. Christman BW, McPherson CD, Newman JH, King GA, Bernard GR, 894-9. [PMID: 11555526] Groves BM, et al. An imbalance between the excretion of and 30. Atwood CW Jr, McCrory D, Garcia JG, Abman SH, Ahearn GS. Pulmo- prostacyclin metabolites in pulmonary hypertension. N Engl J Med. 1992;327: nary artery hypertension and sleep-disordered breathing: ACCP evidence-based 70-5. [PMID: 1603138] clinical practice guidelines. Chest. 2004;126:72S-77S. [PMID: 15249496] 53. Tuder RM, Cool CD, Geraci MW, Wang J, Abman SH, Wright L, et al. 31. Azarian R, Wartski M, Collignon MA, Parent F, Herve P, Sors H, et al. Prostacyclin synthase expression is decreased in lungs from patients with severe Lung perfusion scans and hemodynamics in acute and chronic pulmonary em- pulmonary hypertension. Am J Respir Crit Care Med. 1999;159:1925-32. bolism. J Nucl Med. 1997;38:980-3. [PMID: 9189155] [PMID: 10351941] 32. Ryan KL, Fedullo PF, Davis GB, Vasquez TE, Moser KM. Perfusion scan 54. Barst RJ, Rubin LJ, Long WA, McGoon MD, Rich S, Badesch DB, et al. findings understate the severity of angiographic and hemodynamic compromise A comparison of continuous intravenous epoprostenol (prostacyclin) with con- in chronic thromboembolic pulmonary hypertension. Chest. 1988;93:1180-5. ventional therapy for primary pulmonary hypertension. The Primary Pulmonary [PMID: 3371097] Hypertension Study Group. N Engl J Med. 1996;334:296-302. [PMID: 33. Bailey CL, Channick RN, Auger WR, Fedullo PF, Kerr KM, Yung GL, et 8532025] al. “High probability” perfusion lung scans in pulmonary venoocclusive disease. 55. Badesch DB, Tapson VF, McGoon MD, Brundage BH, Rubin LJ, Wigley Am J Respir Crit Care Med. 2000;162:1974-8. [PMID: 11069842] FM, et al. Continuous intravenous epoprostenol for pulmonary hypertension due 34. King MA, Ysrael M, Bergin CJ. Chronic thromboembolic pulmonary hy- to the scleroderma spectrum of disease. A randomized, controlled trial. Ann In- pertension: CT findings. AJR Am J Roentgenol. 1998;170:955-60. [PMID: tern Med. 2000;132:425-34. [PMID: 10733441] 9530043] 56. Barst RJ, Rubin LJ, McGoon MD, Caldwell EJ, Long WA, Levy PS. 35. Swensen SJ, Tashjian JH, Myers JL, Engeler CE, Patz EF, Edwards WD, et Survival in primary pulmonary hypertension with long-term continuous intrave- al. Pulmonary venoocclusive disease: CT findings in eight patients. AJR Am J nous prostacyclin. Ann Intern Med. 1994;121:409-15. [PMID: 8053614] Roentgenol. 1996;167:937-40. [PMID: 8819387] 57. Shapiro SM, Oudiz RJ, Cao T, Romano MA, Beckmann XJ, Georgiou D, 36. Pietra GG, Capron F, Stewart S, Leone O, Humbert M, Robbins IM, et al. et al. Primary pulmonary hypertension: improved long-term effects and survival Pathologic assessment of vasculopathies in pulmonary hypertension. J Am Coll with continuous intravenous epoprostenol infusion. J Am Coll Cardiol. 1997;30: Cardiol. 2004;43(Suppl):25S-32S. [PMID: 15194175] 343-9. [PMID: 9247503] 37. McCaffrey RN, Dunn LH. Primary pulmonary hypertension in pregnancy. 58. McLaughlin VV, Shillington A, Rich S. Survival in primary pulmonary Obstet Gynecol Surv. 1964;19:567-91. [PMID: 14179852] hypertension: the impact of epoprostenol therapy. Circulation. 2002;106:1477- 38. Weiss BM, Zemp L, Seifert B, Hess OM. Outcome of pulmonary vascular 82. [PMID: 12234951] disease in pregnancy: a systematic overview from 1978 through 1996. J Am Coll 59. Sitbon O, Humbert M, Nunes H, Parent F, Garcia G, Herve P, et al. Cardiol. 1998;31:1650-7. [PMID: 9626847] Long-term intravenous epoprostenol infusion in primary pulmonary hyperten- 39. Elkayam U, Dave R, Bokhari SW. Primary pulmonary hypertension and sion: prognostic factors and survival. J Am Coll Cardiol. 2002;40:780-8. [PMID: pregnancy. In: Elkayam U, Gleivher N, eds. Cardiac Problems in Pregnancy. 12204511] New York: Wiley-Liss; 1998:183-90. 60. Rich S, McLaughlin VV. The effects of chronic prostacyclin therapy on 40. Badalian SS, Silverman RK, Aubry RH, Longo J. Twin pregnancy in a cardiac output and symptoms in primary pulmonary hypertension. J Am Coll woman on long-term epoprostenol therapy for primary pulmonary hypertension. Cardiol. 1999;34:1184-7. [PMID: 10520810] A case report. J Reprod Med. 2000;45:149-52. [PMID: 10710749]. 61. Okano Y, Yoshioka T, Shimouchi A, Satoh T, Kunieda T. Orally active 41. Stewart R, Tuazon D, Olson G, Duarte AG. Pregnancy and primary pul- prostacyclin analogue in primary pulmonary hypertension [Letter]. Lancet. 1997; monary hypertension: successful outcome with epoprostenol therapy. Chest. 349:1365. [PMID: 9149701] 2001;119:973-5. [PMID: 11243988] 62. Galie N, Humbert M, Vachiery JL, Vizza CD, Kneussl M, Manes A, et al. 42. Decoene C, Bourzoufi K, Moreau D, Narducci F, Crepin F, Krivosic- Effects of beraprost sodium, an oral prostacyclin analogue, in patients with pul- Horber R. Use of inhaled nitric oxide for emergency Cesarean section in a monary arterial hypertension: a randomized, double-blind, placebo-controlled woman with unexpected primary pulmonary hypertension. Can J Anaesth. 2001; trial. J Am Coll Cardiol. 2002;39:1496-502. [PMID: 11985913] 48:584-7. [PMID: 11444454] 63. Barst RJ, McGoon M, McLaughlin V, Tapson V, Rich S, Rubin L, et al. 43. Badesch DB, Abman SH, Ahearn GS, Barst RJ, McCrory DC, Simonneau Beraprost therapy for pulmonary arterial hypertension. J Am Coll Cardiol. 2003; G, et al. Medical therapy for pulmonary arterial hypertension: ACCP evidence- 41:2119-25. [PMID: 12821234] based clinical practice guidelines. Chest. 2004;126:35S-62S. [PMID: 15249494] 64. Simonneau G, Barst RJ, Galie N, Naeije R, Rich S, Bourge RC, et al. 44. Galie N, Seeger W, Naeije R, Simonneau G, Rubin LJ. Comparative analy- Continuous subcutaneous infusion of treprostinil, a prostacyclin analogue, in sis of clinical trials and evidence-based treatment algorithm in pulmonary arterial patients with pulmonary arterial hypertension: a double-blind, randomized, pla- hypertension. J Am Coll Cardiol. 2004;43:81S-88S. [PMID: 15194183] cebo-controlled trial. Am J Respir Crit Care Med. 2002;165:800-4. [PMID: 45. Fuster V, Steele PM, Edwards WD, Gersh BJ, McGoon MD, Frye RL. 11897647] Primary pulmonary hypertension: natural history and the importance of throm- 65. Hoeper MM, Olschewski H, Ghofrani HA, Wilkens H, Winkler J, Borst bosis. Circulation. 1984;70:580-7. [PMID: 6148159] MM,etal.A comparison of the acute hemodynamic effects of inhaled nitric 46. Rich S, Kaufmann E, Levy PS. The effect of high doses of calcium-channel oxide and aerosolized iloprost in primary pulmonary hypertension. German PPH blockers on survival in primary pulmonary hypertension. N Engl J Med. 1992; Study Group. J Am Coll Cardiol. 2000;35:176-82. [PMID: 10636277] www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 291 Review Evaluation and Management of the Patient with Pulmonary Arterial Hypertension

66. Olschewski H, Ghofrani HA, Schmehl T, Winkler J, Wilkens H, Hoper hypertension. Circulation. 2001;104:1218-22. [PMID: 11551870] MM, et al. Inhaled iloprost to treat severe pulmonary hypertension. An uncon- 85. Ghofrani HA, Wiedemann R, Rose F, Olschewski H, Schermuly RT, trolled trial. German PPH Study Group. Ann Intern Med. 2000;132:435-43. Weissmann N, et al. Combination therapy with oral sildenafil and inhaled ilo- [PMID: 10733442] prost for severe pulmonary hypertension. Ann Intern Med. 2002;136:515-22. 67. Hoeper MM, Schwarze M, Ehlerding S, Adler-Schuermeyer A, Spie- [PMID: 11926786] kerkoetter E, Niedermeyer J, et al. Long-term treatment of primary pulmonary 86. Sastry BK, Narasimhan C, Reddy NK, Raju BS. Clinical efficacy of silde- hypertension with aerosolized iloprost, a prostacyclin analogue. N Engl J Med. nafil in primary pulmonary hypertension: a randomized, placebo-controlled, dou- 2000;342:1866-70. [PMID: 10861321] ble-blind, crossover study. J Am Coll Cardiol. 2004;43:1149-53. [PMID: 68. Olschewski H, Simonneau G, Galie N, Higenbottam T, Naeije R, Rubin 15063421] LJ, et al. Inhaled iloprost for severe pulmonary hypertension. N Engl J Med. 87. Bharani A, Mathew V, Sahu A, Lunia B. The efficacy and tolerability of 2002;347:322-9. [PMID: 12151469] sildenafil in patients with moderate-to-severe pulmonary hypertension. Indian 69. Kim H, Yung GL, Marsh JJ, Konopka RG, Pedersen CA, Chiles PG, et al. Heart J. 2003;55:55-9. [PMID: 12760589] Endothelin mediates pulmonary vascular remodelling in a canine model of 88. Ghofrani HA, Schermuly RT, Rose F, Wiedemann R, Kohstall MG, chronic embolic pulmonary hypertension. Eur Respir J. 2000;15:640-8. [PMID: Kreckel A, et al. Sildenafil for long-term treatment of nonoperable chronic 10780753] thromboembolic pulmonary hypertension. Am J Respir Crit Care Med. 2003; 70. Rubens C, Ewert R, Halank M, Wensel R, Orzechowski HD, Schultheiss 167:1139-41. [PMID: 12684251] HP, et al. Big endothelin-1 and endothelin-1 plasma levels are correlated with the 89. Ghofrani HA, Barst J, Badesch DB, Burgess G, Fleming T, Galie N, et al. severity of primary pulmonary hypertension. Chest. 2001;120:1562-9. [PMID: Sildenafil use in pulmonary arterial hypertension: a multinational, randomized, 11713135] double-blind, placebo-controlled trial. Chest. 2004 [Abstract]. 71. Giaid A, Yanagisawa M, Langleben D, Michel RP, Levy R, Shennib H, et 90. Sandoval J, Rothman A, Pulido T. Atrial septostomy for pulmonary hyper- al. Expression of endothelin-1 in the lungs of patients with pulmonary hyperten- tension. Clin Chest Med. 2001;22:547-60. [PMID: 11590848] sion. N Engl J Med. 1993;328:1732-9. [PMID: 8497283] 91. Rothman A, Sklansky MS, Lucas VW, Kashani IA, Shaughnessy RD, 72. Galie N, Grigioni F, Bacchi-Reggiani L, et al. Relation of endothelin-1 to Channick RN, et al. Atrial septostomy as a bridge to lung transplantation in survival in patients with primary pulmonary hypertension. Eur J Clin Invest. patients with severe pulmonary hypertension. Am J Cardiol. 1999;84:682-6. 1996;26:273. [PMID: 10498139] 73. Channick RN, Simonneau G, Sitbon O, Robbins IM, Frost A, Tapson VF, 92. Sandoval J, Gaspar J, Pulido T, Bautista E, Martínez-Guerra ML, Zeballos et al. Effects of the dual endothelin-receptor antagonist bosentan in patients with M, et al. Graded balloon dilation atrial septostomy in severe primary pulmonary pulmonary hypertension: a randomised placebo-controlled study. Lancet. 2001; hypertension. A therapeutic alternative for patients nonresponsive to vasodilator 358:1119-23. [PMID: 11597664] treatment. J Am Coll Cardiol. 1998;32:297-304. [PMID: 9708453] 74. Rubin LJ, Badesch DB, Barst RJ, Galie N, Black CM, Keogh A, et al. 93. Christie JD, Kotloff RM, Pochettino A, Arcasoy SM, Rosengard BR, Lan- Bosentan therapy for pulmonary arterial hypertension. N Engl J Med. 2002;346: dis JR, et al. Clinical risk factors for primary graft failure following lung trans- 896-903. [PMID: 11907289] plantation. Chest. 2003;124:1232-41. [PMID: 14555551] 75. Barst RJ, Langleben D, Frost A, Horn EM, Oudiz R, Shapiro S, et al. 94. Mendeloff EN, Meyers BF, Sundt TM, Guthrie TJ, Sweet SC, de la Sitaxsentan therapy for pulmonary arterial hypertension. Am J Respir Crit Care Morena M, et al. Lung transplantation for pulmonary vascular disease. Ann Med. 2004;169:441-7. [PMID: 14630619] Thorac Surg. 2002;73:209-1799. [PMID: 11834012] 76. Beavo JA, Reifsnyder DH. Primary sequence of cyclic phospho- diesterase isozymes and the design of selective inhibitors. Trends Pharmacol Sci. 95. Pielsticker EJ, Martinez FJ, Rubenfire M. Lung and heart-lung transplant 1990;11:150-5. [PMID: 2159198] practice patterns in pulmonary hypertension centers. J Heart Lung Transplant. 77. Ahn HS, Foster M, Cable M, Pitts BJ, Sybertz EJ. Ca/CaM-stimulated and 2001;20:1297-304. [PMID: 11744413] cGMP-specific phosphodiesterases in vascular and non-vascular tissues. Adv Exp 96. Hoeper MM, Oudiz RJ, Peacock A, Tapson VF, Haworth SG, Frost AE, et Med Biol. 1991;308:191-7. [PMID: 1666264] al. End points and clinical trial designs in pulmonary arterial hypertension: clin- 78. Braner DA, Fineman JR, Chang R, Soifer SJ. M&B 22948, a cGMP ical and regulatory perspectives. J Am Coll Cardiol. 2004;43:48S-55S. [PMID: phosphodiesterase inhibitor, is a pulmonary vasodilator in lambs. Am J Physiol. 15194178] 1993;264:H252-8. [PMID: 8381611] 97. Hoeper MM, Taha N, Bekjarova A, Gatzke R, Spiekerkoetter E. Bosentan 79. Cohen AH, Hanson K, Morris K, Fouty B, McMurty IF, Clarke W, et al. treatment in patients with primary pulmonary hypertension receiving nonparen- Inhibition of cyclic 3’-5’-guanosine monophosphate-specific phosphodiesterase teral prostanoids. Eur Respir J. 2003;22:330-4. [PMID: 12952269] selectively vasodilates the pulmonary circulation in chronically hypoxic rats. J 98. Humbert M, Barst RJ, Robbins IM, Channick RN, Galie N, Boonstra A, Clin Invest. 1996;97:172-9. [PMID: 8550830] et al. Combination of bosentan with epoprostenol in pulmonary arterial hyper- 80. Hanson KA, Ziegler JW, Rybalkin SD, Miller JW, Abman SH, Clarke tension: BREATHE-2. Eur Respir J. 2004;24:353-9. [PMID: 15358690] WR. Chronic pulmonary hypertension increases fetal lung cGMP phosphodies- 99. Petkov V, Mosgoeller W, Ziesche R, Raderer M, Stiebellehner L, Vonbank terase activity. Am J Physiol. 1998;275:L931-41. [PMID: 9815111] K, et al. Vasoactive intestinal peptide as a new drug for treatment of primary 81. Michelakis E, Tymchak W, Lien D, Webster L, Hashimoto K, Archer S. pulmonary hypertension. J Clin Invest. 2003;111:1339-46. [PMID: 12727925] Oral sildenafil is an effective and specific pulmonary vasodilator in patients with 100. Du L, Sullivan CC, Chu D, Cho AJ, Kido M, Wolf PL, et al. Signaling pulmonary arterial hypertension: comparison with inhaled nitric oxide. Circula- molecules in nonfamilial pulmonary hypertension. N Engl J Med. 2003;348: tion. 2002;105:2398-403. [PMID: 12021227] 500-9. [PMID: 12571257] 82. Lepore JJ, Maroo A, Pereira NL, Ginns LC, Dec GW, Zapol WM, et al. 101. Eddahibi S, Humbert M, Fadel E, Raffestin B, Darmon M, Capron F, et Effect of sildenafil on the acute pulmonary vasodilator response to inhaled nitric al. Serotonin transporter overexpression is responsible for pulmonary artery oxide in adults with primary pulmonary hypertension. Am J Cardiol. 2002;90: smooth muscle hyperplasia in primary pulmonary hypertension. J Clin Invest. 677-80. [PMID: 12231108] 2001;108:1141-50. [PMID: 11602621] 83. Atz AM, Wessel DL. Sildenafil ameliorates effects of inhaled nitric oxide 102. Nishimura T, Vaszar LT, Faul JL, Zhao G, Berry GJ, Shi L, et al. Sim- withdrawal. Anesthesiology. 1999;91:307-10. [PMID: 10422958] vastatin rescues rats from fatal pulmonary hypertension by inducing apoptosis of 84. Wilkens H, Guth A, Konig J, Forestier N, Cremers B, Hennen B, et al. neointimal smooth muscle cells. Circulation. 2003;108:1640-5. [PMID: Effect of inhaled iloprost plus oral sildenafil in patients with primary pulmonary 12963647]

292 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Current Author Addresses: Dr. Rubin: University of California, San Dr. Badesch: University of Colorado Health Sciences Center, Box Diego, Medical Center, 9300 Campus Point Drive, La Jolla, CA 92037- C-272, 4200 East Ninth Avenue, Denver, CO 80262. 7381.

W-80 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Perspective Promoting Informed Choice: Transforming Health Care To Dispense Knowledge for Decision Making Steven H. Woolf, MD, MPH; Evelyn C.Y. Chan, MD, MS; Russell Harris, MD, MPH; Stacey L. Sheridan, MD, MPH; Clarence H. Braddock III, MD, MPH; Robert M. Kaplan, PhD; Alex Krist, MD; Annette M. O’Connor, RN, MScN, PhD; and Sean Tunis, MD, MSc

Ours is an era in which patients seek greater engagement in training, or trained third parties who function as impartial decision health care choices, increasing the demand for high-quality infor- counselors. Controlled studies are needed to determine which mation about clinical options. Providing support for informed model is best, but none appears to be ideal. choice is not straightforward, however, because of challenges The health care system cannot truly support informed decision faced by clinicians, health systems, and consumers. Greater use of making without correcting the underlying obstacles that impede written or electronic tools can help to clarify choices for patients, patient access to needed information. New information technology but decision aids cannot replace the human element in facilitating solutions, training programs, and reimbursement schemes are nec- informed choice. The ideal solution is to couple information with essary. Patient demand for guidance will only increase as clinical high-quality decision counseling to help patients understand the options multiply and the world of information continues its rapid potential risks, benefits, and uncertainties of clinical options and growth. Today’s health care system is unprepared for the conver- to assist them in selecting the option that best accommodates gence of these 2 burgeoning domains, and the need to address their personal preferences. Decision counseling can be offered by systemic deficiencies will grow more urgent over time. 3 types of providers: clinicians who lack formal informed-choice Ann Intern Med. 2005;143:293-300. www.annals.org training (“usual care”), clinicians with formal informed-choice For author affiliations, see end of text.

atients need good information to make good choices, In this article, we outline patients’ expanding needs for Pbut supplying that needed information is not some- decision support and the challenges clinicians and health thing that physicians and hospitals do well. Among the systems face in meeting those needs. We then explore 2 great ironies of the modern health care system is how categories of solutions to facilitate informed choice: ex- poorly it delivers knowledge at a time when society enjoys panded information resources for decision support and the unprecedented access to information. Americans can ob- coupling of information with decision counseling. We con- tain so many facts with the click of a button yet must clude that no current model can succeed without a major struggle to gather well-tailored information about their transformation in system design to make knowledge a key clinical options or ways to care for their own health. Con- commodity accessible to all participants. sumers encounter a system designed to deliver the material commodities of care (such as tests and drugs) but not knowledge. According to the results of a Commonwealth Fund survey published in 2004 (1), the frequency (33%) THE PATIENT’S EXPANDING ROLE IN MAKING with which sick patients in the United States leave the DECISIONS physician’s office without getting important questions an- For millions of Americans who lack health insurance swered is the highest among the 5 countries studied (the or a regular clinician—problems that disproportionately others being Australia, Canada, New Zealand, and the plague the poor and minorities—informed choice is a mar- United Kingdom). ginal concern eclipsed by the larger priority of gaining ac- The mounting need for consumer information is, in cess to care (4). However, among those with established many ways, a contemporary phenomenon. Generally access (and even for those without), deeper engagement in speaking, patients of previous generations merely needed to decision making is increasing for several reasons: increased decide whether to seek medical attention and whether to patient autonomy, broader access to information, expand- follow their physician’s advice; it was not their place to ing clinical options, rising costs, ascendancy of chronic ill- decide which options were best. Physicians tended to fill a ness, complex tradeoffs, and greater accommodation of paternalistic role, maintaining exclusive purview over med- personal values. ical knowledge. The expectation of both providers and consumers of health care was that the physician knew what was best (2). Today, roles and expectations for information have shifted, giving way to the newer model of informed See also: choice and active patient participation in care (3). This dynamic is placing new demands on both the provider and Web-Only the health care system, perhaps too quickly for either entity Conversion of tables into slides to respond.

© 2005 American College of Physicians 293 Perspective Promoting Informed Choice

Increased Patient Autonomy Complex Tradeoffs The culture of consumerism in the United States en- Patients face a more difficult task in weighing benefits courages the public to exercise control over life choices (5). and harms (17). Yesterday’s patients confronted tradeoffs This trend and the ethical imperative to respect patient with less difficulty by relying on the intuitive judgment of autonomy have shifted the locus of control in the clini- clinicians. Today’s patients expect clinicians to help them cian–patient relationship toward a patient-centered model to understand sophisticated probability data used to weigh that eschews paternalism and invites patients to engage the tradeoffs of a therapeutic option, such as the number of actively in the decision-making process (6, 7). patients per 1000 who benefit from an intervention versus Broader Access to Information the number who are harmed (18, 19). This increased desire for control arises in an era of Greater Accommodation of Personal Values 24-hour news, direct-to-consumer advertising, search en- For the growing number of decisions for which the gines, and high-speed Internet connectivity. Patients have “best choice” depends on personal preferences (20), pa- grown accustomed to accessing information and are ac- tients must consider how procedures will affect their lives quiring the tools to research clinical options and review and must cope with the scientific uncertainties surround- personal medical data. The physician is no longer the sole ing outcomes. For this kind of analysis, patients require purveyor of medical knowledge (8). informed decision making (21), which involves a level of Expanding Clinical Options counseling that goes beyond the offhand advice that clini- Advances in medications and biotechnology have cians conventionally offer in busy practice (22). On topics yielded a complex menu of choices for conditions that ranging from screening tests (23) to surgery and end-of-life once had a single standard of care. The sophistication of care (24–27), patients cannot properly weigh the benefits these procedures gives patients the added task of processing and harms without examining the evidence in light of per- scientific nomenclature, data, and technological concepts, a sonal values (28). To do so, they require the more substan- special challenge for those patients with literacy, numeracy, tive support offered by informed decision making and or language barriers (9–12). shared decision making (19, 29–32), as shown in Table 1. The intensity of informed decision making must be cali- Rising Costs brated to the type of clinical decision, as others have elu- Consumers must consider the economic implications cidated (29). of clinical choices as they encounter higher insurance co- Promoting informed decision making is motivated not payments and deductibles (13). Medical savings accounts only by moral arguments (that beneficence requires knowl- and defined contribution plans are promoted on the edge of what the patient wants) but also by economic and premise that patients will pay closer attention the cost of legal considerations (33). The high costs taxing the U.S. health care services. health care system stem largely from costly procedures that Ascendancy of Chronic Illness often have weak supporting evidence; some of these proce- Active engagement is vital for the growing number of dures might be deferred if patients knew more about pos- patients with chronic diseases (14). Self-management, a sible complications and other tradeoffs (34). A notable ex- centerpiece of effective long-term care, requires greater ac- ample is intensive care and other interventions at the end cess to information (15, 16). of life, which are often inconsistent with the preferences of patients (35, 36). Teno and associates (36) reported that 60% of seriously ill Medicare beneficiaries preferred com- Table 1. Components of Informed Decision Making fort care over aggressive interventions, but only 41% of these patients believed their care reflected this preference. Themes Addressed in Informed Components of Shared Decision Unwanted, costly interventions might become less com- Decision Making* Making† mon if they were preceded by well-informed discussions The patient’s role in decision making Understanding the risks associated with the condition with patients and loved ones. The clinical issue or nature of the Understanding the options, including discussion the risks, benefits, alternatives, and uncertainties The alternatives for management of Weighing personal values regarding CHALLENGES TO INFORMED DECISION MAKING the patient’s condition potential benefits and harms Even when information is readily available, the public The potential benefits and risks of Participating in decision making at proposed management options the level desired faces difficulties with informed decision making (37–39). The uncertainties Although people generally want to be educated about The patient’s understanding The patient’s preferences health care options and appreciate having the freedom to participate in decision making, not everyone wants this * Reference 21. Copyrighted 1999, American Medical Association, all rights re- role (40–43). Patients face cognitive and emotional chal- served. † Proposed by U.S. Preventive Services Task Force (29). Reprinted with permis- lenges in vetting complex decisions (19). Minorities and sion from the American Journal of Preventive Medicine. disenfranchised patients often arrive at the clinical encoun-

294 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Promoting Informed Choice Perspective ter with less knowledge about certain topics (44) and are per and Mettler (58), envisions clinicians using handheld less likely to be actively engaged in decision making (45–47). devices or electronic medical records to “prescribe” tailored Clinicians also face barriers to implementing informed educational materials and e-mail hyperlinks to relevant decision making (48). The busy pace of patient care leaves Web sites (59). These tools, however, are rarely organized little time for long discussions and detailed presentations of in a format to support decision making. options and statistics. Few clinicians can quote accurate data or divorce themselves from personal biases to ensure a balanced presentation of options. Many lack the time or Decision Aids aptitude to consider patients’ risk profiles, to predict pref- Greater use of decision aids and interactive software erences, or to help patients apply these values to select the technology could help. Decision aids are available in a va- best personal choice (49–52). The best approach to in- riety of formats—print publications, decision boards, vid- formed decision making varies by patient, requiring clini- eos, audio-guided workbooks, and Web applications—and cians to adjust to the individual. Low literacy and cultural help to clarify choices by providing information about the barriers intensify the challenges in communicating facts condition and possible treatment options, probabilities of and eliciting preferences (53, 54). Clinicians, caught in a relevant outcomes, to clarify values, and coaching struggle for economic survival, receive little reimbursement in the steps of decision making (60–63). Controlled trials for this effort. have shown that decision aids increase patient participa- The health care system as a whole faces its own diffi- tion; reduce decisional conflict and indecision; and im- culties in implementing informed decision making because prove indicators of decision quality, such as knowledge, it is not well designed for this task. Health care has been perceptions of probabilities, and concordance between val- slow to respond to society’s appetite for communication ues and choices (64). and has not reorganized itself to provide high-quality in- Decision aids offer clinicians a validated format for formation about options and outcomes. The system is not presenting facts that surpasses conventional advice in terms equipped to inform patients in a manner that is timely, of balance, accuracy, and consistency. They also offer a easily understood, and jargon-free, nor does it encourage medium for expanding counseling beyond the time con- people to consider consequences, to ask questions, to clar- straints of busy office visits. Patients can study decision aids ify values, and to express preferences. at their leisure, contemplate their preferences, and return for another appointment for further discussion. The Internet has spawned a new generation of deci- POTENTIAL SOLUTIONS AND THEIR LIMITATIONS sion aids (65). The Cochrane Inventory of Patient Deci- Thus, the problem at hand is an expanding need to sion Aids lists 50 Web-based tools (66). For example, the help patients navigate complex decisions set against the Ottawa Personal Decision Guide uses interactive technol- limitations of the health care system in delivering the sa- ogy to help people assess decision-making needs, make lient facts and guidance. We explore 2 potential solutions: plans, and track progress (67). CollaborativeCare.net (68) information resources for decision support, and the cou- uses textual information and online videos to present op- pling of these tools with decision counseling. tions and tradeoffs for 13 “crossroad” decisions faced by Information Resources for Decision Support women with breast cancer. The Information Superhighway However, electronic or print material is not the only Much of the information that patients require is avail- answer, in part because of limitations in existing products. able amid the vast resources of the Internet and other me- The current generation of decision aids and software tools dia. The Internet is especially helpful because of its acces- is of variable quality (61, 69) and cannot fully accommo- sibility, convenience, and capability for interactive tailoring date patients’ questions and information needs. Experts are of information. However, the quality of Web sites and not certain how to frame decisions accurately and how to chat-room information is suspect (55), and patients do not present numerical information to achieve clarity, objectiv- always recognize embedded advocacy and sponsor biases. ity, and balance (70–75). Decision aids are even less useful The sheer volume of available information is itself a prob- for patients with literacy or language barriers (76), and lem, making it difficult for patients to locate crucial facts. they may not perform well cross-culturally (77). Disadvantaged persons are especially at risk for falling into No electronic platform is likely to replace the human the digital divide when their physical or financial limita- being’s capacity for guidance: the innate sensitivity to the tions prevent them from easily accessing the Internet and needs and desires of the patient and the ability to employ other new technologies (56). interactive dialogue and nonverbal cues to communicate Nonetheless, the Internet has done much to connect facts, values, emotions, and advice. In a randomized trial patients with relevant knowledge. For example, MED- involving women with menorrhagia, Kennedy and associ- LINEplus, a service of the National Library of Medicine, ates (78) demonstrated that decision aids had no effect on offers a rich collection of information on more than 600 hysterectomy rates or patient satisfaction unless they were conditions (57). Information therapy, as advanced by Kem- coupled with nurse interviews aimed at clarifying values. www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 295 Perspective Promoting Informed Choice

Web sites and decision aids cannot function as partners in may not pursue such training for themselves or their staff, such decisions. Some experts believe that the benefit pa- especially if they see little to gain from their time and tients obtain from human involvement in decision support financial investment. The time required to support in- is less about cognitive learning than about the interaction formed decision making (for example, maintaining a li- and relationships that such counseling engenders. brary of decision aids [83]) might not be feasible for busy Finally, decision aids of any form are unlikely to offer practices. The overhead expenditures associated with this optimal guidance unless they are integrated into medical model, some of which may not be reimbursed by health care. Clinicians possess vital information for determining plans, can also strain tight budgets. Finally, despite the best the best choice for an individual, such as risk factors and informed-choice training, clinicians may still have diffi- medical history, treatments attempted in the past and their culty shedding conflicts of interest, biases, and preferences outcomes, and the availability of local resources; these de- when presenting options. tails are unknown to off-the-shelf aids or counselors with weak collaborative ties to the clinician. What seems like the best choice under nominal conditions might be a poor Counseling by a Trained Third Party: Decision Counselors choice when contextual and clinical circumstances are con- A third solution is for patients to turn to decision sidered. A strategy of decision support that ignores this counselors outside their physician’s practice for help with context could propagate poor advice and frustrate care. informed decision making. Decision counselors, a new Information Coupled with Decision Counseling type of health care professional in many communities, can Counseling by Clinicians without Informed-Choice Training help patients to understand options, to consider the prob- One solution, often suggested for physicians who lack ability of benefits and harms and the supporting evidence, the time or skills to facilitate personally informed decision to explore beliefs and fears, to determine the desired level making, is to start a discussion and then refer patients to of control in making decisions, and to find motivation to nurses or other office staff who have more time to distrib- engage with the primary clinician. We envision offices of ute educational materials and answer patients’ questions. decision counselors that provide a quiet environment to This model allows informed decision making to be inte- use high-speed Internet workstations, a complete library of grated into patient care within the immediate environs of decision aids, and other patient education materials. the physician. Decision counselors offer certain qualities that clini- The strategy has its drawbacks, however, because most cians may lack: a talent for assembling the best educational practices cannot afford to dedicate staff time to patient resources for patients without the interference of compet- counseling or to fund training in informed choice. Typi- ing agendas and specialty bias, and the expertise to guide cally they must call on personnel with competing clinical patients in recognizing and applying personal preferences. duties and inconsistent skills. Consequently, what patients Although these advisors may not possess the clinical exper- experience as informed decision making varies in content tise that is the forte of specialists (on whose advice patients across personnel and practices. Aside from compromising would still rely), they function as highly skilled knowledge patient education, this inconsistency makes reimbursement brokers. Their libraries of decision aids can help patients by health plans unlikely. review the key evidence to consider when weighing op- tions. Skilled counselors can coach patients to understand preferences, become engaged in care, and express their Counseling by Clinicians with Informed-Choice Training preferences to clinicians. Their focus on facilitating in- Clinicians or their staff may opt to undergo formal formed decision making makes decision counselors con- informed-choice training, emphasizing the communication ceptually distinct from thematically related professions, and negotiation skills required for the 7 elements in Table such as informationists, decision analysts, ethicists, patient 1 (21, 79). A Cochrane review found that such training navigators, and patient advocates (84, 85). Two models for can significantly increase the patient-centeredness of con- this form of counseling can be envisioned: autonomous sultations (80). Raising the informed decision-making and coordinated. skills of the average clinician would spare patients from Autonomous third-party counseling would occur inde- needing to look outside the patient–clinician relationship pendently from primary clinicians. An example is the for help from an unfamiliar advisor. Patients trust the pri- counseling provided by Health Dialog (Boston, Massachu- mary clinician, who has the benefit of knowing both the setts), under contract with employers and health plans. patient and the specialty in detail. Formal informed-choice Health Dialog furnishes “health coaches,” who motivate training, which could begin as early as medical (81) or patients to participate in treatment selection, prepare for nursing school and be credentialed through certification, discussions with physicians, weigh the implications of op- might qualify for sufficient reimbursement to subsidize the tions, and translate decisions into action (68, 86). These extended sessions and staff time that informed decision consultations are not always formally coordinated with the making imposes on practices (82). primary clinician, however (82). The disadvantage of this model is that many clinicians In the second model, coordinated third-party counsel-

296 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Promoting Informed Choice Perspective ing, decision counselors collaborate with primary clini- clinical care, Health Dialog uses claims and pharmacy data cians. Patients who face difficult “close-call” decisions (20) to identify patients facing decisions that might benefit and who would benefit from decision counseling might be from counseling. It distributes to primary care providers referred by their providers (or self-referred) for such coun- registries of patients with multiple chronic conditions and seling to ensure that their choices are well-informed and those with gaps in care that suggest inattention to impor- tailored to personal preferences. These counselors may tant decisions. have offices located conveniently within hospitals and out- Decision counseling offers a mechanism for assisting patient office buildings, enabling them to serve both indi- patients in ways that busy clinicians cannot. Centralizing vidual practices and integrated delivery systems. Primary this service within a decision counselor’s office provides clinicians would detail pertinent clinical background and greater consistency, efficiency, and quality control than is recommended options on requisitions. Decision counselors possible when informed decision making is conducted at would reciprocate with written reports and direct commu- primary practices. Decision counseling excels over imper- nication. A referral to a decision counselor might be re- sonal decision aids and Web sites by offering a human face quired for certain tests or procedures, just as genetic coun- and personal assistance to patients who might be over- seling is mandatory for certain genetic tests. For high-risk whelmed in their efforts to extract trustworthy information or costly procedures, Medicare and private health plans from the sea of available knowledge. Counselors can em- might make reimbursement contingent on previous con- ploy special approaches and materials to assist patients with sultation with a decision counselor. low literacy or with language or cultural barriers (89). This model has been successfully pioneered by the The most negative aspect of third-party decision coun- Center for Shared Decision Making at Dartmouth-Hitch- seling is that it appropriates the role of the primary clini- cock Medical Center (87). Located in a suite on the main cian. The model introduces triangulation that can under- floor of the hospital, the center offers consultations with a mine trust, continuity, and other valued elements of the decision-support nurse and lends videotapes and other de- patient–clinician relationship, potentially weakening the cision aids to patients to review at home. The center pre- healing power of the caretaker conversation (90). Decision pares patients for clinical encounters while also helping counselors who collaborate with the clinician as part of an clinicians by including relevant clinical data in reports that integrated plan of care can mitigate this problem, as gen- summarize patients’ preferences and decision needs. Using erally occurs with genetic counseling. If coordination is this service has become routine for hospital physicians; al- lacking, however, inconsistencies in the guidance offered most all spine surgeons at Dartmouth-Hitchcock ask pa- by the clinician and counselor may confuse the patient and tients to examine decision aids before consenting to sur- trigger resistance from the clinician. The intrusion of a gery. Other hospitals engaged in an ongoing multicenter third party can also introduce inefficiency, as when extra trial have adopted this approach (88). office visits become necessary to resolve confusion over In an effort to coordinate decision counseling with treatment plans.

Table 2. Advantages and Disadvantages Associated with 3 Models of Decision Counseling*

Variable Untrained Clinician Trained Clinician Third-Party Decision Provider Provider Counselor* Disadvantages Provider bias ŒŒƒ Variable skill and quality of counseling Œƒƒ Large time and financial expenditure needed to obtain training NA Œ NA Overhead expenditures for allocating staff time for counseling ŒŒƒ Triangulation of patient–clinician relationship ƒƒŒ Lack of clinical expertise ƒƒŒ Unfamiliarity with pertinent clinical data ƒƒŒ Confusion and inefficiency if not coordinated with primary clinician NA NA Œ Uncertain professional and legal liability ƒƒŒ

Advantages Occurs in context of trusted patient–clinician relationship ŒŒƒ Integrated with primary care ŒŒƒ Less demanding on primary clinician 7 ƒŒ Consistent, efficient delivery of patient-centered counseling ƒ 7 Œ Improved prospects for reimbursement ƒŒŒ High-quality counseling techniques ƒ 7 Œ Talent and resources of “knowledge broker” ƒƒŒ Clinical autonomy and independent analysis ƒƒŒ Ideal, centralized physical environment for counseling ƒƒŒ

*Œ ϭ increased likelihood; … ϭ decreased likelihood; 7 ϭ highly variable; NA ϭ not applicable. www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 297 Perspective Promoting Informed Choice

Decision counselors may also overlook pertinent clin- effects on health care costs and other resources can be ical issues. They may lack the content expertise and train- sweeping, with ripple effects that influence benefits pack- ing of specialists and the firsthand knowledge of patients’ ages, health insurance premiums, and access to care. personal histories to properly individualize potential bene- One may safely predict that the passage of time will fits, barriers, and alternatives. Although their training may only increase the demand for information and guidance as make them the best at presenting options with consistency, clinical options increase (both in number and in technical they cannot fully rid themselves of biases or achieve the complexity) and as the stunning expansion of information standardization and convenience that decision aids and continues. Today’s health care system is unprepared for the Web sites offer. Finally, it is unclear who is professionally intersection of these 2 realities. The system clings to an and legally responsible for adverse outcomes resulting from outdated model—relying on busy clinicians to keep their actions (or inaction) propagated by decision counselors. patients informed—a holdover from an earlier time when a physician’s impromptu advice was sufficient and when WHICH APPROACH IS BEST? there was little concern about its inherent incompleteness or bias (98). The traditional office or bedside conversation Each of the aforementioned models has advantages remains an important anchor for patients but is inadequate and disadvantages (Table 2); which model is best is un- for facilitating thoughtful analysis of tradeoffs and connect- clear. Studies suggest that decision aids can reduce the use ing patients to the world’s best information. of some procedures (91–93), but one cannot predict which Systems of care cannot support informed choice without models for informed decision making will decrease or in- redesigning their infrastructure. New information systems are crease utilization or will optimize the quality of decisions. needed to link patients with the best resources and decision The ideal goal is to improve health outcomes linked to aids available. Clinicians require training programs to raise patient values and satisfaction while minimizing costs and their skill levels. Imaginative models for redesigned office care litigation. Case reports describe adverse legal consequences from offering informed decision making (94). Controlled and restructured reimbursement schemes are needed to pro- studies are ultimately needed to determine the effect of vide sufficient time and incentive to counsel patients. Current each model on the quality of decisions; health outcomes; reimbursement incentives reward costly procedures and liability; costs; and acceptability to patients, clinicians, and rushed visits; they discourage the counseling that ensures that systems of care. procedures are warranted in the first place and that gives pa- tients the self-management tools on which the effectiveness of those treatments often depends (15). CONCLUSION Today’s leaders may dismiss informed choice as too Regardless of which of the decision-counseling models great a luxury for major monetary outlays or infrastructure is most effective, none seems capable of delivering the in- changes, but the underlying problem will not go away. formation and guidance needed for informed choice with- Perhaps nothing is changing more dramatically in health out introducing discontinuity or disruption in patient care. care than the increased volume and influence of informa- Less disruptive solutions are unlikely to satisfy patients’ tion. Patients face a growing need for assistance in knowl- information needs. The ultimate long-term solution, then, edge management and for access to professionals who are is not merely to choose among the aforementioned models qualified for this task. As the volume of information and but to correct the larger defects in the current health care the complexity of choices increase, this need will only grow infrastructure that impede the delivery of the information more urgent. patients need when they need it (95). The design of the current system reflects an outdated service model that From Virginia Commonwealth University, Richmond, Virginia; Univer- viewed medicine’s primary product as therapeutics and the sity of Texas–Houston School of Medicine, Houston, Texas; University delivery of information as a secondary service to enhance of North Carolina School of Medicine, Chapel Hill, North Carolina; Stanford University School of Medicine, Stanford, California; UCLA patient satisfaction. School of Public Health, Los Angeles, California; University of Ottawa, Modern thinkers understand that information is Ottawa, Ontario, Canada; and Centers for Medicare & Medicaid Ser- power, as Francis Bacon presciently observed long ago vices, Baltimore, Maryland. (96). Berwick considers knowledge to be the primary com- modity of health care (97). Experts know that, for many Acknowledgments: The authors thank Angela Coulter, PhD; Joanne conditions, the success or failure of clinicians’ work de- Lynn, MD; Albert G. Mulley Jr., MD; James Weinstein, DO, MSc; and pends on what patients understand. Informed choice is the anonymous reviewers for their helpful comments on earlier drafts of important not only as a moral duty to patients, who de- this manuscript. serve to know how options affect their health, but to soci- ety as a whole. The costly consumption of health services at Potential Financial Conflicts of Interest: Grants received: R.M. Kaplan the macro level originates at the micro level in decisions (Foundation for Informed Medical Decision Making), A. Krist (Ameri- made by individual clinicians and patients. If informed can Academy of Family Physicians), A.M. O’Connor (Foundation for decision making shifts the dynamics of these choices, the Informed Medical Decision Making).

298 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Promoting Informed Choice Perspective

Requests for Single Reprints: Steven H. Woolf, MD, MPH, Depart- 25. National Institutes of Health Consensus Development Conference statement: ments of Family Medicine, Epidemiology, and Community Health, Vir- adjuvant therapy for breast cancer, November 1-3, 2000. J Natl Cancer Inst ginia Commonwealth University, West Hospital, 1200 East Broad Monogr. 2001:5-15. [PMID: 11773285] 26. Kaplan RM. The significance of quality of life in health care. Qual Life Res. Street, P.O. Box 980251, MCV Station, Richmond, VA 23298-0251; 2003;12 Suppl 1:3-16. [PMID: 12803306] e-mail, [email protected]. 27. Cohen LM, Germain MJ, Poppel DM. Practical considerations in dialysis withdrawal: “to have that option is a blessing”. JAMA. 2003;289:2113-9. Current author addresses are available at www.annals.org. [PMID: 12709469] 28. Woolf SH. Shared decision-making: the case for letting patients decide which choice is best [Editorial]. J Fam Pract. 1997;45:205-8. [PMID: 9299998] References 29. Sheridan SL, Harris RP, Woolf SH. Shared decision making about screening 1. Commonwealth Fund. First Report and Recommendations of the Common- and chemoprevention. a suggested approach from the U.S. Preventive Services wealth Fund’s International Working Group on Quality Indicators: A Report to Task Force. Am J Prev Med. 2004;26:56-66. [PMID: 14700714] Health Ministers of Australia, Canada, New Zealand, the United Kingdom, and 30. Frosch DL, Kaplan RM. Shared decision making in clinical medicine: past the United States. New York: Commonwealth Fund; 2004. research and future directions. Am J Prev Med. 1999;17:285-94. [PMID: 2. Laine C, Davidoff F. Patient-centered medicine. A professional evolution. 10606197] JAMA. 1996;275:152-6. [PMID: 8531314] 31. Charles C, Gafni A, Whelan T. Decision-making in the physician-patient 3. Hibbard JH. Moving toward a more patient-centered health care delivery encounter: revisiting the shared treatment decision-making model. Soc Sci Med. system. Health Aff (Millwood) 2004;Suppl Web Exclusive:VAR133-5. 1999;49:651-61. [PMID: 10452420] 4. Committee on the Consequences of Uninsurance, Board on Health Care 32. Briss P, Rimer B, Reilley B, Coates RC, Lee NC, Mullen P, et al. Promot- Services, Institute of Medicine. Care Without Coverage: Too Little, Too Late. ing informed decisions about cancer screening in communities and healthcare Washington, DC: National Academy Press; 2002. systems. Am J Prev Med. 2004;26:67-80. [PMID: 14700715] 5. Hibbard JH. Engaging health care consumers to improve the quality of care. 33. Wennberg JE, Fisher ES, Skinner JS. Geography and the debate over Medi- Med Care. 2003;41:I61-70. [PMID: 12544817] care reform. Health Aff (Millwood). 2002;Suppl Web Exclusives:W96-114. 6. Reiser SJ. The era of the patient. Using the experience of illness in shaping the [PMID: 12703563] missions of health care. JAMA. 1993;269:1012-7. [PMID: 8429582] 34. Lomas SJ, Rice CL. Cutting healthcare costs without rationing at the bed- 7. Coulter A. The Autonomous Patient: Ending Paternalism in Medical Care. side: preserving the doctor-patient fiduciary relationship. Healthc Pap. 2001;2: London, Nuffield Trust; 2002. 38-44; discussion 69-75. [PMID: 12811144] 8. Kassirer JP. The next transformation in the delivery of health care [Editorial]. 35. Covinsky KE, Fuller JD, Yaffe K, Johnston CB, Hamel MB, Lynn J, et al. N Engl J Med. 1995;332:52-4. [PMID: 7990866] Communication and decision-making in seriously ill patients: findings of the 9. Wills CE, Holmes-Rovner M. Patient comprehension of information for SUPPORT project. The Study to Understand Prognoses and Preferences for shared treatment decision making: state of the art and future directions. Patient Outcomes and Risks of Treatments. J Am Geriatr Soc. 2000;48:S187-93. Educ Couns. 2003;50:285-90. [PMID: 12900101] [PMID: 10809474] 10. Gwyn R, Elwyn G, Edwards A, Mooney A. The problematic of decision- 36. Teno JM, Fisher ES, Hamel MB, Coppola K, Dawson NV. Medical care sharing: deconstructing ‘cholesterol’ in a clinical encounter. Health Expect. 2003; inconsistent with patients’ treatment goals: association with 1-year Medicare re- 6:242-54. [PMID: 12940797] source use and survival. J Am Geriatr Soc. 2002;50:496-500. [PMID: 11943046] 11. Schwartz LM, Woloshin S, Black WC, Welch HG. The role of numeracy in 37. Coulter A. Partnerships with patients: the pros and cons of shared clinical understanding the benefit of screening mammography. Ann Intern Med. 1997; decision-making. J Health Serv Res Policy. 1997;2:112-21. [PMID: 10180362] 127:966-72. [PMID: 9412301] 38. Woolf SH. The logic and limits of shared decision making [Editorial]. J Urol. 12. Woloshin S, Schwartz LM, Moncur M, Gabriel S, Tosteson AN. Assessing 2001;166:244-5. [PMID: 11435879] values for health: numeracy matters. Med Decis Making. 2001;21:382-90. 39. Kravitz RL, Melnikow J. Engaging patients in medical decision making [PMID: 11575488] [Editorial]. BMJ. 2001;323:584-5. [PMID: 11557690] 13. Gabel JR, Whitmore H, Rice T, Lo Sasso AT. Employers’ contradictory 40. McKinstry B. Do patients wish to be involved in decision making in the views about consumer-driven health care: results from a national survey. Health consultation? A cross sectional survey with video vignettes. BMJ. 2000;321:867- Aff (Millwood). 2004;Suppl Web Exclusives:W4-210-8. [PMID: 15452008] 71. [PMID: 11021866] 14. Trends in aging—United States and worldwide. MMWR Morb Mortal 41. Robinson A, Thomson R. Variability in patient preferences for participating Wkly Rep. 2003;52:101-4, 106. [PMID: 12645839] in medical decision making: implication for the use of decision support tools. 15. Bodenheimer T, Lorig K, Holman H, Grumbach K. Patient self-manage- Qual Health Care. 2001;10 Suppl 1:i34-8. [PMID: 11533436] ment of chronic disease in primary care. JAMA. 2002;288:2469-75. [PMID: 42. Mazur DJ, Hickam DH. Patients’ preferences for risk disclosure and role in 12435261] decision making for invasive medical procedures. J Gen Intern Med. 1997;12: 16. Norris SL, Lau J, Smith SJ, Schmid CH, Engelgau MM. Self-management 114-7. [PMID: 9051561] education for adults with type 2 diabetes: a meta-analysis of the effect on glycemic 43. Deber RB, Kraetschmer N, Irvine J. What role do patients wish to play in control. Diabetes Care. 2002;25:1159-71. [PMID: 12087014] treatment decision making? Arch Intern Med. 1996;156:1414-20. [PMID: 17. Cuervo LG, Clarke M. Balancing benefits and harms in health care [Edito- 8678709] rial]. BMJ. 2003;327:65-6. [PMID: 12855496] 44. Chan EC, Vernon SW, O’Donnell FT, Ahn C, Greisinger A, Aga DW. 18. Nuovo J, Melnikow J, Chang D. Reporting number needed to treat and Informed consent for cancer screening with prostate-specific antigen: how well are absolute risk reduction in randomized controlled trials. JAMA. 2002;287:2813-4. men getting the message? Am J Public Health. 2003;93:779-85. [PMID: [PMID: 12038920] 12721144] 19. McNutt RA. Shared medical decision making: problems, process, progress. 45. Blackhall LJ, Murphy ST, Frank G, Michel V, Azen S. Ethnicity and JAMA. 2004;292:2516-8. [PMID: 15562133] attitudes toward patient autonomy. JAMA. 1995;274:820-5. [PMID: 7650806] 20. Kassirer JP, Pauker SG. The toss-up [Editorial]. N Engl J Med. 1981;305: 46. Cooper-Patrick L, Gallo JJ, Gonzales JJ, Vu HT, Powe NR, Nelson C, et 1467-9. [PMID: 7300866] al. Race, , and partnership in the patient-physician relationship. JAMA. 21. Braddock CH 3rd, Edwards KA, Hasenberg NM, Laidley TL, Levinson W. 1999;282:583-9. [PMID: 10450723] Informed decision making in outpatient practice: time to get back to basics. 47. Perkins HS, Geppert CM, Gonzales A, Cortez JD, Hazuda HP. Cross- JAMA. 1999;282:2313-20. [PMID: 10612318] cultural similarities and differences in attitudes about advance care planning. J 22. Kaplan RM, Ganiats TG, Frosch DL. Diagnostic and treatment decisions in Gen Intern Med. 2002;17:48-57. [PMID: 11903775] US healthcare. J Health Psychol. 2004;9:29-40. [PMID: 14683567] 48. Holmes-Rovner M, Valade D, Orlowski C, Draus C, Nabozny-Valerio B, 23. Sieber WJ, Kaplan RM. Informed adherence: the need for shared medical Keiser S. Implementing shared decision-making in routine practice: barriers and decision making. Control Clin Trials. 2000;21:233S-40S. [PMID: 11018581] opportunities. Health Expect. 2000;3:182-191. [PMID: 11281928] 24. Hayden M, Pignone M, Phillips C, Mulrow C. Aspirin for the primary 49. Cotler SJ, Patil R, McNutt RA, Speroff T, Banaad-Omiotek G, Ganger prevention of cardiovascular events: a summary of the evidence for the U.S. DR, et al. Patients’ values for health states associated with hepatitis C and phy- Preventive Services Task Force. Ann Intern Med. 2002;136:161-72. [PMID: sicians’ estimates of those values. Am J Gastroenterol. 2001;96:2730-6. [PMID: 11790072] 11569703] www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 299 Perspective Promoting Informed Choice

50. Bruera E, Willey JS, Palmer JL, Rosales M. Treatment decisions for breast understanding of number needed to treat and other common risk reduction carcinoma: patient preferences and physician perceptions. Cancer. 2002;94:2076- formats. J Gen Intern Med. 2003;18:884-92. [PMID: 14687273] 80. [PMID: 11932912] 75. Edwards A, Evans R, Elwyn G. Manufactured but not imported: new direc- 51. Teutsch C. Patient-doctor communication. Med Clin North Am. 2003;87: tions for research in shared decision making support and skills. Patient Educ 1115-45. [PMID: 14621334] Couns. 2003;50:33-8. [PMID: 12767582] 52. Devereaux PJ, Anderson DR, Gardner MJ, Putnam W, Flowerdew GJ, 76. Kim SP, Knight SJ, Tomori C, Colella KM, Schoor RA, Shih L, et al. Brownell BF, et al. Differences between perspectives of physicians and patients Health literacy and shared decision making for prostate cancer patients with low on anticoagulation in patients with atrial fibrillation: observational study. BMJ. socioeconomic status. Cancer Invest. 2001;19:684-91. [PMID: 11577809] 2001;323:1218-22. [PMID: 11719412] 77. Chan EC, Haynes MC, O’Donnell FT, Bachino C, Vernon SW. Cultural 53. Institute of Medicine. Speaking of Health: Assessing Health Communication sensitivity and informed decision making about prostate cancer screening. J Com- Strategies for Diverse Populations. Washington, DC: National Academy Pr; munity Health. 2003;28:393-405. [PMID: 14620963] 2002. 78. Kennedy AD, Sculpher MJ, Coulter A, Dwyer N, Rees M, Abrams KR, et 54. Institute of Medicine. Health Literacy: A Prescription to End Confusion. al. Effects of decision aids for menorrhagia on treatment choices, health out- Washington, DC: National Academy Press; 2004. comes, and costs: a randomized controlled trial. JAMA. 2002;288:2701-8. 55. Impicciatore P, Pandolfini C, Casella N, Bonati M. Reliability of health [PMID: 12460093] information for the public on the World Wide Web: systematic survey of advice 79. Towle A, Godolphin W. Framework for teaching and learning informed on managing fever in children at home. BMJ. 1997;314:1875-9. [PMID: shared decision making. BMJ. 1999;319:766-71. [PMID: 10488010] 9224132] 80. Lewin SA, Skea ZC, Entwistle V, Zwarenstein M, Dick J. Interventions for 56. Brodie M, Flournoy RE, Altman DE, Blendon RJ, Benson JM, Rosenbaum providers to promote a patient-centred approach in clinical consultations (Coch- MD. Health information, the Internet, and the digital divide. Health Aff (Mill- rane Review). In: The Cochrane Library, Issue 4, 2003. Chichester, UK: Wiley; wood). 2000;19:255-65. [PMID: 11192412] 2003. 57. Miller N, Lacroix EM, Backus JE. MEDLINEplus: building and maintain- 81. Yedidia MJ, Gillespie CC, Kachur E, Schwartz MD, Ockene J, Chepaitis ing the National Library of Medicine’s consumer health Web service. Bull Med AE, et al. Effect of communications training on medical student performance. Libr Assoc. 2000;88:11-7. [PMID: 10658959] JAMA. 2003;290:1157-65. [PMID: 12952997] 58. Kemper DW, Mettler M. Information Therapy: Prescribed Information as a 82. O’Connor AM, Llewellyn-Thomas HA, Flood AB. Modifying unwarranted Reimbursable Medical Expense. Washington, DC: Center for Information Ther- variations in health care: shared decision making using patient decision aids. apy; 2002. Health Aff (Millwood). 2004;Suppl Web Exclusive:VAR63-72. [PMID: 59. About Information Therapy. Washington, DC: Center for Information 15471770] 83. Godolphin W, Towle A, McKendry R. Evaluation of the quality of patient Therapy. Accessed at www.informationtherapy.org/about_ix.html on 12 April information to support informed shared decision-making. Health Expect. 2001; 2005. 4:235-42. [PMID: 11703497] 60. Barry MJ. Health decision aids to facilitate shared decision making in office 84. Davidoff F, Florance V. The informationist: a new health profession? [Edi- practice. Ann Intern Med. 2002;136:127-35. [PMID: 11790064] torial]. Ann Intern Med. 2000;132:996-8. [PMID: 10858185] 61. O’Connor AM, Lagare F, Stacey D. Risk communication in practice: the 85. Ferguson T. Online patient-helpers and physicians working together: a new contribution of decision aids. BMJ. 2003;327:736-40. [PMID: 14512487] partnership for high quality health care. BMJ. 2000;321:1129-32. [PMID: 62. Foundation for Informed Medical Decision Making. Boston, MA: Founda- 11061737] tion for Informed Medical Decision Making, Inc. Accessed at www.fimdm.org/ 86. Health Dialog: Support for Decisions That Matter Most. Boston: Health index.html on 12 April 2005. Dialog Services Corporation. Accessed at www.healthdialog.com/Health_ 63. O’Connor AM, Rostom A, Fiset V, Tetroe J, Entwistle V, Llewellyn- Dialog_QA.pdf on 12 April 2005. Thomas H, et al. Decision aids for patients facing health treatment or screening 87. Dartmouth-Hitchcock Center for Shared Decision Making. Lebanon, NH: decisions: systematic review. BMJ. 1999;319:731-4. [PMID: 10487995] Dartmouth-Hitchcock Medical Center; 2005. Accessed at www.dhmc 64. O’Connor AM, Stacey D, Entwistle V, Llewellyn-Thomas, H, Rovner D, .org/webpage.cfm?site_idϭ2&org_idϭ108&gsec_idϭ0&sec_idϭ0&item_idϭ2486 Holmes-Rovner M, et al. Decision aids for people facing health treatment or on 12 April 2005. screening decisions. Cochrane Database Syst Rev. 2003;(2):CD001431 [PMID: 88. Birkmeyer NJ, Weinstein JN, Tosteson AN, Tosteson TD, Skinner JS, 12804407] Lurie JD, et al. Design of the Spine Patient outcomes Research Trial (SPORT). 65. Frosch DL, Kaplan RM, Felitti VJ. A randomized controlled trial comparing Spine. 2002;27:1361-72. [PMID: 12065987] internet and video to facilitate patient education for men considering the prostate 89. Pignone M, DeWalt DA, Sheridan S, Berkman N, Lohr KN. Interventions specific antigen test. J Gen Intern Med. 2003;18:781-7. [PMID: 14521639] to improve health outcomes for patients with low literacy. A systematic review. J 66. Cochrane Inventory of Patient Decision Aids. Ottawa: Ottawa Health Re- Gen Intern Med. 2005;20:185-92. [PMID: 15836553] search Institute. Accessed at http://204.187.39.28/cochinvent.php on 12 April 90. Cassell EJ. The Nature of Suffering and the Goals of Medicine. Oxford: 2005. Oxford Univ Pr; 1994. 67. Ottawa Personal Decision Guide. Ottawa: Ottawa Health Research Institute. 91. Wagner EH, Barrett P, Barry MJ, Barlow W, Fowler FJ Jr. The effect of a Accessed at http://204.187.39.28/decguide.html on 12 April 2005. shared decisionmaking program on rates of surgery for benign prostatic hyperpla- 68. CollaborativeCare.net. Boston: Health Dialog. Accessed at www sia. Pilot results. Med Care. 1995;33:765-70. [PMID: 7543638] .collaborativecare.net/ on 12 April 2005. 92. Morgan MW, Deber RB, Llewellyn-Thomas HA, Gladstone P, Cusimano 69. Billings J. Promoting the dissemination of decision aids: an odyssey in a RJ, O’Rourke K, et al. Randomized, controlled trial of an interactive videodisc dysfunctional health care financing system. Health Aff (Millwood). 2004;Suppl decision aid for patients with ischemic heart disease. J Gen Intern Med. 2000; Web Exclusive:VAR128-32. [PMID: 15471781] 15:685-93. [PMID: 11089711] 70. Misselbrook D, Armstrong D. Patients’ responses to risk information about 93. Volk RJ, Spann SJ. Decision-aids for prostate cancer screening. J Fam Pract. the benefits of treating hypertension. Br J Gen Pract. 2001;51:276-9. [PMID: 2000;49:425-7. [PMID: 10836773] 11458479] 94. Merenstein D. A piece of my mind. Winners and losers. JAMA. 2004;291: 71. Bogardus ST Jr, Holmboe E, Jekel JF. Perils, pitfalls, and possibilities in 15-6. [PMID: 14709561] talking about medical risk. JAMA. 1999;281:1037-41. [PMID: 10086441] 95. Institute of Medicine. Crossing the Quality Chasm: A New Health System 72. Feldman-Stewart D, Brundage MD, McConnell BA, MacKillop WJ. Prac- for the 21st Century. Washington, DC: National Academy Pr; 2001. tical issues in assisting shared decision-making. Health Expect. 2000;3:46-54. 96. Bacon F. The Proficience and Advancement of Learning, 1605. [PMID: 11281911] 97. Berwick D. Knowledge always on call: for docs, practicing medicine will 73. Schapira MM, Nattinger AB, McHorney CA. Frequency or probability? A mean providing information more than providing care. Clin Leadersh Manag qualitative study of risk communication formats used in health care. Med Decis Rev. 2000;14:250-2. [PMID: 11210208] Making. 2001;21:459-67. [PMID: 11760103] 98. Eddy DM. Clinical policies and the quality of clinical practice. N Engl J 74. Sheridan SL, Pignone MP, Lewis CL. A randomized comparison of patients’ Med. 1982;307:343-7. [PMID: 7088099]

300 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Current Author Addresses: Dr. Woolf: Departments of Family Medi- Dr. Braddock: Stanford Center for Biomedical Ethics, Stanford Univer- cine, Epidemiology, and Community Health, Virginia Commonwealth sity School of Medicine, 251 Campus Drive, MS5475, Stanford, CA University, West Hospital, 1200 East Broad Street, P.O. Box 980251, 94305. MCV Station, Richmond, VA 23298-0251. Dr. Kaplan: Department of Health Services, UCLA School of Public Dr. Chan: The University of Texas-Houston School of Medicine, 6431 Health, Box 951772, Los Angeles, CA 90095-1772. Fannin, 1.122 MSB, Houston, TX 77030. Dr. Krist: Department of Family Medicine, Virginia Commonwealth Dr. Harris: Cecil G. Sheps Center for Health Services Research, Univer- University, 3825 Charles Stewart Drive, Fairfax, VA 22033. sity of North Carolina School of Medicine, CB 7590, Chapel Hill, NC Dr. O’Connor: Ottawa Health Research Institute, University of Ottawa, 27599-7590. 1053 Carling Avenue, Ottawa, ON K1Y4E9 Canada. Dr. Sheridan: 5039 Old Clinic Building, CB 7110, University of North Dr. Tunis: Centers for Medicare and Medicaid Services, 7500 Security Carolina, Chapel Hill, NC 27599. Boulevard, Baltimore, MD 21244-1850.

www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 W-81 Editorial Trying To Predict the Future for People with Diabetes: A Tough but Important Task

ith growing health care costs, policymakers and than the $50 000 threshold for rapid adoption (7). Eddy Whealth care providers need information on the cost- and colleagues critique the model in an online Appendix effectiveness of interventions. This issue contains an eco- (1). Speaking as a coauthor of the DPPRG cost-effective- nomic evaluation (1) of a prevention policy for diabetes—a ness analysis, I believe that many of Eddy and colleagues’ condition that affects more than 18 million persons in the criticisms are based on inaccurate descriptions of the United States at a cost of approximately $132 billion an- DPPRG model. However, my focus is to understand why nually. the 2 models have such different policy implications. To assist policymakers, researchers have developed Although these 2 analyses gave different values for models that simulate the progression of diabetes, expendi- cost-effectiveness, several clinically relevant results were tures on diabetes care, and effects of interventions. The similar. First, both analyses predicted that the lifestyle in- outputs of these models include costs and health outcomes, tervention would substantially reduce the proportion of such as length of life (often expressed as quality-adjusted patients at high risk who developed diabetes. Second, both life-years [QALYs]), a measure that considers the quality of analyses predicted that the lifestyle intervention would de- life in each health state. Two principal types of diabetes lay the onset of diabetes. The Archimedes model predicted models exist. Most researchers use a Markov model, which that the time required for 50% of patients with IGT to comprises disease states (for example, normal, impaired develop diabetes would increase from 7 years to 14 years. glucose tolerance [IGT], and diabetes) represented in a The corresponding figures from the DPPRG study were 8 computer program. The computer simulates transitions years and 18 years, respectively. Third, both analyses pre- from one disease state to another as chance events. A sec- dicted that the lifestyle intervention will lead to fewer com- ond novel type of model, named Archimedes (2, 3), uses plications, longer life, and improved quality of life. Fourth, object-oriented computer programming and complex dif- both analyses suggested that the cost of the lifestyle inter- ferential equations to simulate pathophysiologic processes vention exceeds the savings from lower rates of diabetes (for example, hepatic glucose output after a meal) that complications. change over time and can lead to disease. For a technical Why do the 2 studies differ in the cost per QALY of explanation of the Archimedes model, I refer to the accom- the lifestyle intervention and metformin therapy? This panying editorial in this issue (4). My editorial focuses on question is crucial because the different costs per QALY the clinical and policy aspects of the Archimedes model. lead to different policy recommendations. To facilitate a The Archimedes model calculates the changes in costs comparison, I examined the results for the base-case co- and quality-adjusted length of life (in QALYs) from imple- horts (Archimedes model, no intervention; DPPRG analy- menting the lifestyle intervention of the Diabetes Preven- sis, standard lifestyle) for both models. The Archimedes’ tion Program (DPP) study (5) or prescribing metformin to time horizon (30 years) is shorter than the DPPRG time a cohort with IGT like the enrollees in the DPP. Eddy and horizon (the time from IGT diagnosis until death). This colleagues (1) took a societal perspective and used time difference is important. A shorter time horizon will always horizons of 10, 20, and 30 years. Compared with no in- be less favorable if some patients benefit only after decades tervention, the cost per QALY of beginning the intensive of the intervention, as Eddy and colleagues show by com- lifestyle intervention or metformin for IGT was $62 600 paring the 10-, 20-, and 30-year time horizon analyses. and $35 400, respectively, over 30 years (1). The cost per Because the DPPRG analysis followed the cohort longer QALY for an intensive lifestyle intervention started after but projected roughly the same life expectancy as the the onset of diabetes was $24 500. Archimedes model (approximately 24 years), one can infer What do these results imply for health care policy? that the Archimedes model projected less morbidity and Acceptable cost-effectiveness depends on a society’s will- mortality during its shorter 30-year time horizon. As the ingness to pay. Interventions that cost less than $50 000 complication rate increases, the absolute benefit from an per QALY are reasonable to consider for rapid adoption. effective intervention increases. Thus, the Archimedes model results imply that we should Why are the complication rates higher in the DPPRG not rush to adopt the DPP lifestyle intervention (6). model? In the Archimedes model, the microvascular disease Another model, developed by the DPP Research cumulative incidence rates were very low. The lifestyle in- Group (DPPRG), leads to a different conclusion. It used a tervention reduced the base-case cohort’s 30-year cumula- Markov model, lifetime horizon, and societal perspective tive incidence of kidney failure from 0.07% to 0.03% and to determine the cost-effectiveness of the same interven- the need for amputation from 0.03% to 0.02%. Compa- tions for IGT. The cost per QALY was $8800 for the rable rates from the DPPRG were from 1.0% to 0.6% and lifestyle intervention and $29 900 for metformin, both less from 1.9% to 1.3%, respectively. These differences are www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 301 Editorial Trying To Predict the Future for People with Diabetes probably too large to be due solely to the shorter time Michael M. Engelgau, MD, MS horizon used in the Archimedes model. National Center for Chronic Disease Prevention and Health Pro- What accounts for this difference in microvascular dis- motion, Centers for Disease Control and Prevention ease rates in the 2 studies? One reason is different assump- Atlanta, GA 30341 tions for the rates of progression of glycemia. For the base Disclaimer: case, the Archimedes analysis modeled the FPG level to The findings and conclusions in this editorial are those of the author and do not necessarily represent the views of the funding increase at a rate of 0.1 to 0.2 mmol/L (2.0 to 3.0 mg/dL) agency or the Diabetes Prevention Program Research Group. per year from onset of diabetes until it reached 10.0 to 11.1 mmol/L (180.0 to 200.0 mg/dL), taking about 20 to Grant Support: Dr. Engelgau is a federal employee of the Centers for 30 years. This is much longer than the empirical observa- Disease Control and Prevention. tion of roughly 10 years from onset until clinical diagnosis is made (8–10), which is the assumption used by the Potential Financial Conflicts of Interest: None disclosed. DPPRG. Slower rates of glycemic progression imply slower Requests for Single Reprints: Michael M. Engelgau, MD, MS, Divi- microvascular complication rates. This is consistent with sion of Diabetes Translation, Mailstop K-10, 4770 Buford Highway NE, results from a validation study reported by the Archimedes Atlanta, GA 30341; e-mail, [email protected]. analysis that predicted a substantially lower cumulative in- cidence of retinopathy than that observed (15% in the Ann Intern Med. 2005;143:301-302. Archimedes model vs. 30% in the comparison trial) (see Eddy and colleagues’ Appendix Table 5 [1]). Assumptions and modeling of cardiovascular out- References comes will have a greater effect on life expectancy than 1. Eddy DM, Schlessinger L, Kahn R. Clinical outcomes and cost-effectiveness of strategies for managing people at high risk for diabetes. Ann Intern Med. microvascular outcome assumptions. Here, the difference 2005;143:251-64. between the Archimedes model and the DPPRG analysis 2. Eddy DM, Schlessinger L. Archimedes: a trial-validated model of diabetes. was mostly due to the different time horizons. In blinded Diabetes Care. 2003;26:3093-101. [PMID: 14578245] 3. Schlessinger L, Eddy DM. Archimedes: a new model for simulating health predictions of the results of the Collaborative Atorvastatin care systems—the mathematical formulation. J Biomed Inform. 2002;35:37-50. Diabetes Study (CARDS) (11), both models predicted the [PMID: 12415725] observed rates reasonably well. 4. Brandeau ML. Modeling complex medical decision problems with the In my editorial, I have tried to dissect the 2 models to Archimedes model. Ann Intern Med. 2005;143:303-4. 5. Knowler WC, Barrett-Connor E, Fowler SE, Hamman RF, Lachin JM, show why the cost per QALY was so much higher in the Walker EA, et al. Reduction in the incidence of type 2 diabetes with lifestyle Archimedes model despite the 2 models projecting several intervention or metformin. N Engl J Med. 2002;346:393-403. [PMID: similar qualitative conclusions. Different assumptions 11832527] 6. Laupacis A, Feeny D, Detsky AS, Tugwell PX. How attractive does a new about the rate of glycemic progression and a different time technology have to be to warrant adoption and utilization? Tentative guidelines horizon (which allows a longer time in which events could for using clinical and economic evaluations. CMAJ. 1992;146:473-81. [PMID: occur) are probably the principal causes of the differences. 1306034] The lower rate of complications in the Archimedes analysis 7. Herman WH, Hoerger TJ, Brandle M, Hicks K, Sorensen S, Zhang P, et al. The cost-effectiveness of lifestyle modification or metformin in preventing type 2 means that the interventions will have a smaller effect on diabetes in adults with impaired glucose tolerance. Ann Intern Med. 2005;142: the outcomes of complications than in the DPPRG analy- 323-32. [PMID: 15738451] sis. A smaller effect of the interventions, with roughly the 8. Harris MI, Klein R, Welborn TA, Knuiman MW. Onset of NIDDM occurs at least 4-7 yr before clinical diagnosis. Diabetes Care. 1992;15:815-9. [PMID: same costs, would translate into lower cost-effectiveness in 1516497] the Archimedes analysis. 9. Thompson TJ, Engelgau MM, Hegazy M, Ali MA, Sous ES, Badran A, et al. The Archimedes model is a new, novel, and welcome The onset of NIDDM and its relationship to clinical diagnosis in Egyptian addition to the diabetes care–modeling efforts. However, adults. Diabet Med. 1996;13:337-40. [PMID: 9162609] 10. Ferrannini E, Nannipieri M, Williams K, Gonzales C, Haffner SM, Stern its inputs and assumptions need more refinement and MP. Mode of onset of type 2 diabetes from normal or impaired glucose tolerance. transparency before we can understand it well enough to Diabetes. 2004;53:160-5. [PMID: 14693710] use it in setting national diabetes prevention policy. As my 11. Colhoun HM, Betteridge DJ, Durrington PN, Hitman GA, Neil HA, Livingstone SJ, et al. Primary prevention of cardiovascular disease with atorva- editorial shows, understanding why the Archimedes model statin in type 2 diabetes in the Collaborative Atorvastatin Diabetes Study differs from other models is an exacting and difficult task. (CARDS): multicentre randomised placebo-controlled trial. Lancet. 2004;364: However, whoever said that interpreting economic studies 685-96. [PMID: 15325833] would be easy and straightforward? (12, 13). Trying to 12. Neumann PJ. Why don’t Americans use cost-effectiveness analysis? Am J Manag Care. 2004;10:308-12. [PMID: 15152700] predicting the future is a tough job—but we should still 13. Tunis SR. Economic analysis in healthcare decisions [Editorial]. Am J Manag try to do it and try to do it well. Care. 2004;10:301-4. [PMID: 15152698]

302 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Editorial Modeling Complex Medical Decision Problems with the Archimedes Model

ow often should women 40 to 50 years of age un- resource utilization, such as hospital bed occupancy. Un- Hdergo mammography? Could an HIV vaccine with like Markov models, which typically allow for a finite only limited effectiveness be a cost-effective means of slow- number of states and allow transitions between states to ing the spread of HIV infection? What is the best work-up occur at discrete time intervals (for example, every month), for a patient who has a suspected pulmonary embolism? the Archimedes model allows for an infinite number of In the ideal world, one would like free, fast, and eth- patient states (by using continuous variables) that can ical clinical trials that would provide the evidence needed change at any point in time. The model can be used to to answer such questions. In the real world, these clinical evaluate a variety of interventions, including disease pre- trials may be too time-consuming, too expensive, unethi- vention, screening, diagnosis, treatment, and care. cal, or even impossible to perform. How, then, can we How does the Archimedes model simulate physiologic obtain useful answers about appropriate methods for dis- processes and their interactions? The model uses differen- ease prevention and treatment? We need a structured tial equations to describe each of the elements mentioned framework that uses the best evidence, imperfect as it may earlier. A differential equation models the change in a vari- be, and captures relevant complexities and interactions. able, such as blood pressure, as a continuous function of Decision models fulfill these requirements. variables and parameters. As an example, a differential A Markov model is one way to represent a decision equation in the Archimedes model might express the problem. A Markov model represents a dynamic process change in a physiologic state (such as serum insulin level) (for example, a patient’s health over time) by using a set of as a function of various variables and parameters in the distinct states (for example, benign tumor, malignant tu- model (such as current serum insulin level, functioning of mor, or death) and transition probabilities that describe the pancreas, and an individual’s inherited propensity to how the disease process transitions from one state to an- develop insulin resistance). Such an equation—and every other over time (for example, the probability that a benign other equation in the model—must be fitted with empiri- tumor will become malignant). The analyst enters the cal data. The Archimedes model comprises scores of such states, the transition probabilities, and the outcomes into a differential equations. When the model simulates a large computer program that simulates the time course and out- cohort of patients, the computational requirements are comes for each of many patients and calculates average very large. values. Markov models are commonly used to analyze Others have developed large-scale models that capture medical problems. A recent MEDLINE search on “Markov the interaction of many individual people in their environ- model” generated almost 3000 hits. ment. Some of these models have simulated urban traffic The article by Eddy and colleagues in this issue (1) systems (4), the spread of epidemics (5), and bioterror evaluates the potential cost-effectiveness over 30 years of a threats to a community (6). Like those models, the lifestyle modification program used in the Diabetes Pre- Archimedes model captures information at a high level of vention Program (DPP) for a group of high-risk people. At detail, such as interaction of patients with the health care the core of the analysis is a unique and groundbreaking system. Unlike those models, Archimedes also captures a mathematical model. Like a Markov model, it represents very fine level of detail, such as the functioning of specific the progression of disease in individual patients, but it does organs in each individual. so by simulating the underlying physiologic processes that How can one validate the output of such an ambi- cause disease, as well as patient behavior, resource utiliza- tious, complex model? First, the model designer must ver- tion, and other factors. The authors named the model after ify every single equation that is part of the model. Does the Greek scientist Archimedes. each equation properly reflect the underlying biological or Other researchers have developed models of physio- other process? Is each equation consistent with existing logic processes to analyze medical decision problems (2, 3); data? Do the interactions among the equations in the the Archimedes model is unique because it explicitly mod- model properly capture the effects of interactions among els a variety of physiologic processes (for example, func- biological or other processes? One-way and multivariate tioning of the pancreas, heart, and other organs) and their sensitivity analyses are useful at this level of validation. In a interactions. The model describes a patient’s health status sensitivity analysis, the analyst varies the value of one or (such as body mass index or serum insulin level) at any more variables and observes the effects on the outputs of moment, including diseases the patient might have or de- the model (for example, the rate of progression of glucose velop (for example, diabetes or coronary artery disease). intolerance). The next step in validating a complex model The Archimedes model also captures patient behavior, like the Archimedes model is to use it to simulate a clinical such as seeking medical help for symptoms, and health care trial (such as for diabetes prevention or heart disease pre- www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 303 Editorial Modeling Complex Medical Decision Problems with the Archimedes Model vention) and compare its results with the results from sim- the results from the carefully validated Archimedes model ilar real-life clinical trials (7, 8). Do the health outcomes in help to identify the intervention that is most likely to be the simulated intervention patients or controls match the the best one. That is the purpose of a clinical trial, and that health outcomes of the actual intervention patients or con- is the purpose of a model-based medical policy analysis. trols? Once a model has successfully predicted the results of Margaret L. Brandeau, PhD previously published clinical trials, the next step is to pro- Stanford University School of Engineering vide the model with baseline data from a trial that is in Stanford, CA 94305 progress and use the model to predict the results of the trial (as has been done for the Archimedes model [7, 8]). Yet Potential Financial Conflicts of Interest: None disclosed. another step is to use the model to evaluate interventions Requests for Single Reprints: Margaret L. Brandeau, PhD, Department that have never been tested in a clinical trial. A key element of Management Science and Engineering, Terman Building, Stanford of validation at this stage is to ask whether the model’s University, Stanford, CA 94305-4026; e-mail, [email protected]. predictions make intuitive sense. If they do not, it is im- portant to investigate whether the model is incorrect or Ann Intern Med. 2005;143:303-304. whether it is showing us something new that is correct but that had not been obvious a priori. In addition, informa- References tion about model details—and, if possible, the model it- 1. Eddy DM, Schlessinger L, Kahn R. Clinical outcomes and cost-effectiveness self—should be in the public domain (1, 9, 10) so that of strategies for managing people at high risk for diabetes. Ann Intern Med. others in the scientific community can validate the model. 2005;143:251-64. It is, of course, virtually impossible to fully validate 2. Potter LK, Greller LD, Cho CR, Nuttall ME, Stroup GB, Suva LJ, et al. every aspect of a complex model such as the Archimedes Response to continuous and pulsatile PTH dosing: a mathematical model for parathyroid hormone receptor kinetics. Bone. 2005;May 24 [Epub ahead of model. The foregoing steps can go far toward validating print]. [PMID: 15921971] the output of such a model, but they cannot prove with 3. Wein LM, D’Amato RM, Perelson AS. Mathematical analysis of antiretroviral certainty that every simulation using the model has its therapy aimed at HIV-1 eradication or maintenance of low viral loads. J Theor Biol. 1998;192:81-98. [PMID: 9628841] equivalent in reality. Some gaps between the model’s pre- 4. Los Alamos National Laboratories Computer and Computational Sciences dictions and reality are inevitable: Many medical processes, Division. Transportation Analysis and Simulation System (TRANSIMS). Ac- including those captured in the Archimedes model, are too cessed at www.lanl.gov/delphi/projects/transims.shtml on 29 June 2005. complex, too uncertain, and too variable across patients to 5. Los Alamos National Laboratories Computer and Computational Sciences Division. Epidemiology System (EpiSims). Accessed at www.ccs.lanl.gov/ccs5/ model exactly. apps/epid.shtml on 29June 2005. However, the purpose of a medical decision model is 6. Trahan MW, Sobel AL. Transnational and Homeland Defense: Software not to predict events with certainty; rather, it is to identify Tools for Biological Threat Reduction. Proceedings of the Workshop on United the best decision in light of the uncertainty that exists at Science and Technology for Biological Threat Reduction. Santa Fe, NM: Univ of New Mexico Center for Advanced Studies; 2001. Available at www. the time that a decision must be made. The analysis by foxdreams.com/work/papers/Software_Tools_for_Biological_Threat_Reduction. Eddy and colleagues (1) shows that implementing the life- pdf#searchϭ‘large%20scale%20simulation%20model%20individuals%20 style modification program used in the DPP for a cohort of behavior%20health’. people at high risk for diabetes is likely to be an inefficient 7. Eddy DM, Schlessinger L. Validation of the Archimedes diabetes model. Diabetes Care. 2003;26:3102-10. [PMID: 14578246] use of funds. According to the results from the Archimedes 8. Eddy DM, Schlessinger L. Archimedes: a trial-validated model of diabetes. model, taking metformin or delaying lifestyle modification Diabetes Care. 2003;26:3093-101. [PMID: 14578245] until after a high-risk person develops diabetes would be a 9. Schlessinger L, Eddy DM. Archimedes: a new model for simulating health more cost-effective use of funds than modifying lifestyle care systems—the mathematical formulation. J Biomed Inform. 2002;35:37-50. [PMID: 12415725] before development of diabetes. Although we have no em- 10. Schlessinger L, Eddy DM, Kahn R. Equations for the Archimedes Model of pirical evidence about the effects of the DPP lifestyle in- Diabetes and Its Complications. Technical Report, Parts A and B. 2004. Avail- tervention over 30 years (the lack of long-term follow-up is able at www.archimedesmodel.com. a principal reason for carrying out a model-based analysis), © 2005 American College of Physicians

304 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Editorial Improving Patient Care Can Set Your Brain on Fire

hen I was an intern and resident, I was learning so postoperative complication rates but not other complica- Wmuch so fast that I sometimes felt my brain was on tion rates, length of stay, or cost, presumably because the fire. One thing I learned was that much of what was orthopedists were already providing a high level of care (6). known about good medical care was not being used. I was The article by Higashi and colleagues in this issue (7) providing care to people whose problems were avoidable. is another favorite. It describes a cohort study of 372 com- My teachers in medical school taught me how to prevent munity-living people 65 years of age or older who were at these things, and the textbooks were absolutely clear that increased risk for functional decline or death and were con- achieving this goal was possible, but there they were every tinuously enrolled in 1 of 2 managed care organizations, 1 day—unnecessary catastrophes. I concluded that having in the northeastern United States and the other in the medical knowledge about how to prevent, diagnose, and southwestern United States, from July 1998 to July 1999. treat disease is not enough. Trained nurses abstracted medical charts to identify the Patients are part of the problem. Some don’t know presence or absence of 160 quality indicators that were what to do. Others know what to do but can’t afford it, measures of the process of care. The authors calculated a and still others are unwilling or unable to do it. But pa- quality score by dividing the number of quality indicators tients are only part of the problem. In the United States, that people received by the number that they should have too many people don’t have access to care and the costs are received. The mean number of possible quality indicators too high for many who have access. These problems are too was 21 per person, and people received 53% of the quality big for clinicians to solve alone. indicators. In this vulnerable group, 86 people (23%) died Another part of the problem is the way clinicians use, during the 3-year study. When the authors divided the or don’t use, existing medical knowledge to care for pa- people into those who received more and those who re- tients. None of us knows everything. Even when we know ceived fewer than the median number of quality indicators, enough, we may be rushed or tired or distracted or we may 18% of those who received more indicators died and 28% not have the right resources to do the best job possible. We of those who received fewer indicators died (P ϭ 0.02). can always do better, but we need more help than that. When the authors grouped people into deciles according to Many thoughtful people have concluded that we need their quality scores, survival increased as the quality score more and better systems to support us. increased. In analyses that adjusted for potential confound- To encourage the development of those systems, An- ers, the relative risk for dying was 0.80 (95% CI, 0.65 to nals introduced a new section in June 2003 with the help 0.98) for each 10% increase in the quality score. Higashi of the federal Agency for Healthcare Research and Quality and colleagues concluded that “better performance on pro- (AHRQ): Improving Patient Care (1). This section fea- cess quality measures is strongly associated with better sur- tures papers about quality improvement and patient safety: vival among community-dwelling vulnerable older adults.” “. . .[It is] about the organization of practice rather than the If these findings are valid, they are important because clinical content of care.” We have published almost 40 pa- we don’t know much about the relationship between the pers in this new section. We’ve learned, as expected, that process of care and patient outcomes in the real world it’s difficult to conduct randomized, controlled trials of outside interventional research studies, and most real- quality improvement and patient safety, and the papers world quality improvement efforts measure the process of have various formats and study designs. We’re not discour- care because it is easier than measuring outcomes. There- aged. Consider what’s been learned in astronomy and pa- fore, most quality improvement efforts assume that a better leontology without randomized trials. process of care will lead to better patient outcomes. Hi- I have favorites. A systematic review found an inverse gashi and colleagues’ study examines that fundamental as- relationship between years in practice and the quality of sumption. care (2). A cross-sectional comparison showed higher-qual- Are these findings valid? The central issue is whether ity care in the Veterans Health Administration than in physicians decided not to do some quality indicators when community practices, presumably because the Veterans patients were sicker because concentrating on more serious Health Administration established performance measures problems made more sense or, since many quality indica- and monitored performance (3). Semistructured group in- tors provide long-term benefits, sicker patients were un- terviews showed that physicians do not understand the likely to survive to realize benefits. If physicians did not do problems and concerns of deaf and hard-of-hearing pa- some quality indicators because patients were sicker, then tients (4). Case examples illustrated the range of current patients probably died because they were already sicker and innovations in the delivery of care, and Berwick described not because physicians omitted quality indicators. what care would be like if they were combined in 1 place Higashi and colleagues were well aware of this issue (5). A randomized, controlled trial found that adding a and addressed it in several ways. First, they eliminated the hospitalist to a team of orthopedic surgeons reduced minor obvious cases. They identified quality indicators that were www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 305 Editorial Improving Patient Care irrelevant to patients with advanced dementia, poor prog- important. They are valid because the alternative hypoth- noses, and preferences for less aggressive care. When they esis doesn’t stand up to scrutiny. They are important be- analyzed the data, they did not count these quality indica- cause they provide evidence that quality improvement ef- tors for the few patients in these categories. forts that focus on the process of care improve patient Second, the authors identified patient attributes that outcomes, and they remind us that if vulnerable patients are known to identify sicker patients and used these at- want the best we have to offer, we should do everything we tributes to adjust their analyses for the fact that some pa- can for them, not just what seems practical. tients were sicker. They adjusted for patient age and sex, Higashi and colleagues’ study is a good example of self-reported health and function, the number of office vis- how to improve patient care. We want more good exam- its and hospitalizations, and the number of prespecified ples. Readers, find the problems, imagine the solutions, do comorbid conditions found during chart review. the studies, and send us your manuscripts. I think you’ll Third, they looked for relationships between patient find that improving patient care can set your brain on fire. attributes that identify sicker patients and the level of qual- Trust me. You’ll like it. ity, because those relationships should be present if sicker patients received fewer quality indicators. They found that Sankey V. Williams, MD patient age and self-reported health and function were not University of Pennsylvania related to quality. In separate material sent to the editors, Philadelphia, PA 19104 the authors also showed few differences in quality scores when they compared patients with a specific comorbid Potential Financial Conflicts of Interest: None disclosed. condition and patients without that condition. When dif- Current Author Address: Sankey V. Williams, MD, University of Penn- ferent, the quality scores were higher for patients with the sylvania, 1220 Blockley Hall, 423 Guardian Drive, Philadelphia, PA comorbid condition, which is the opposite of what would 19104. have been found if sicker patients were the cause of the relationship between quality indicators and survival. Ann Intern Med. 2005;143:305-306. Together, these observations mean that if physicians were identifying sicker patients, they were not using pa- tient age, sex, self-reported function, self-reported health, References the number of office visits and hospitalizations, the num- 1. Sox HC. Improving patient care [Editorial]. Ann Intern Med. 2003;138:996. ber of comorbid conditions, or individual comorbid con- [PMID: 12809457] ditions to do so. If physicians were identifying sicker pa- 2. Choudhry NK, Fletcher RH, Soumerai SB. Systematic review: the relation- tients, they were using something other than the patient ship between clinical experience and quality of health care. Ann Intern Med. 2005;142:260-73. [PMID: 15710959] attributes that have worked best in previous research. 3. Asch SM, McGlynn EA, Hogan MM, Hayward RA, Shekelle P, Rubenstein In correspondence with the editors, the authors pro- L, et al. Comparison of quality of care for patients in the Veterans Health Ad- vided additional information. If physicians were identify- ministration and patients in a national sample. Ann Intern Med. 2004;141:938- ing sicker patients by using patient attributes other than 45. [PMID: 15611491] 4. Iezzoni LI, O’Day BL, Killeen M, Harker H. Communicating about health those studied, the physicians would have to be so good at care: observations from persons who are deaf or hard of hearing. Ann Intern Med. identifying sicker patients that the association between the 2004;140:356-62. [PMID: 14996677] physicians’ predictions about which patients would die and 5. Berwick DM. My right knee. Ann Intern Med. 2005;142:121-5. [PMID: which patients actually died was stronger than the associa- 15657160] 6. Huddleston JM, Long KH, Naessens JM, Vanness D, Larson D, Trousdale tion between quality scores and which patients actually R, et al. Medical and surgical comanagement after elective hip and knee arthro- died. We don’t know much about how good physicians are plasty: a randomized, controlled trial. Ann Intern Med. 2004;141:28-38. [PMID: at predicting death in the next few months or years, but 15238368] 7. Higashi T, Shekelle PG, Adams JL, Kamberg CJ, Roth CP, Solomon DH, what we do know says that they’re not very good at it (8, 9) et al. Quality of care is associated with survival in vulnerable older patients. Ann and that they don’t trust their own predictions (10) even Intern Med. 2005;143:274-81. when they use the best predictors. 8. Poses RM, Smith WR, McClish DK, Huber EC, Clemo FL, Schmitt BP, et Patients may have declined some quality indicators al. Physicians’ survival predictions for patients with acute congestive heart failure. Arch Intern Med. 1997;157:1001-7. [PMID: 9140271] when they were sicker. Patients also may have known 9. Walter LC, Brand RJ, Counsell SR, Palmer RM, Landefeld CS, Fortinsky things about how sick they were that were not measured by RH, et al. Development and validation of a prognostic index for 1-year mortality the patient attributes that the authors studied, but I doubt in older adults after hospitalization. JAMA. 2001;285:2987-94. [PMID: this. There’s always another potential confounder lying 11410097] 10. Christakis NA, Iwashyna TJ. Attitude and self-reported practice regarding around, and I won’t bet on that one. prognostication in a national sample of internists. Arch Intern Med. 1998;158: Where does all this leave us? The findings need to be 2389-95. [PMID: 9827791] tested in additional studies. For the present, I think the findings in Higashi and colleagues’ study are valid and © 2005 American College of Physicians

306 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org On Being a Doctor Lessons from a Patient

isa was a young woman in her 20s. I thought she might don’t think you need any more Percocets. In fact, I think Lhave had bipolar disorder. Early on, after offering my you are addicted to them at this point.” diagnosis, I seemed to have time for her. She received my We’d been here before. She deftly replied that she only counsel with appreciation. I listened with at least some needed enough to get her to Friday, as she was seeing a attention as she described the horrors of her brief life, specialist who would assume care of the pain portion of her which, needless to say, had not been improved by her psy- care. I already felt bad about my abrupt phone diagnosis chiatric diagnosis. Whenever a patient like Lisa shares with and blunt pronouncement of addiction, but I saw my me her tale of misery, I experience the following arc, per- chance here. I took the bait and gave her a couple more haps common to most doctors: first, horror at how cruel days’ worth with the understanding that I would no longer and harsh the world can be, then appreciation of the rela- prescribe pain medication for her. tively insignificant hurts I have experienced in comparison, When she called me the next day, my first thought was followed finally by compassion and a compunction to one of annoyance. But I did notice something different. “make it all better.” Finally, once reality sets in, I realize She sounded happy. Was she was manic or high on the that what I have to offer is very unlikely to help in the end. painkillers? As I listened, I learned that my diagnosis of I often feel that I am simply not giving enough. addiction had struck her to the core—not because she I am often driven to believe that I will be the doctor to hadn’t heard it before, but because coming from “the only look these patients in the eye and tell them how to fix their doctor she’s ever trusted,” it caused her to think. She re- lives, limit their hurt, and restore order to the chaos swirl- lated how she cried, not out of anger, but out of realization ing around them. In this way, Lisa and I connected. I that I was right. She had tracked down an addiction spe- looked forward to her visits and she genuinely seemed to cialist whom I had introduced her to a few admissions ago, enjoy my counsel. I did continually prescribe painkillers to had set up a visit that very day, and had her first group her because she had an obviously painful and dramatic session scheduled for later in the week. I asked why she injury from an accident, but I felt sure that I couldn’t be sounded so happy. She bubbled, “I realize I can get clean “played” by this young girl, that I would be able to rein it and may get it together at last.” In my cynicism I was all in down the road. I had done this before and I was no already trying to visualize how this was a ploy to get more pushover. medication from me. She continued, “And I don’t want But she was always at least 1 step ahead of me, offering anymore Percocet. I wanted to thank you for all the time up a seemingly endless stream of shifting organic sources of you’ve spent with me. I know I’ve been a pain in the neck. pain requiring me to postpone the day of reckoning. Either No other doctor has given me as much time and attention she was truly cursed, or her endless series of lost prescrip- as you have. Thanks for being my doctor. I’m glad I’m tions, abusive family members, stolen possessions, money your patient.” woes, and reported mistreatment by emergency depart- I think I spluttered something, but my head was spin- ment doctors was inspired manipulation. ning as I hung up the phone. Here I was feeling bad that I For a while, I would mechanically try to rein in the had been shortchanging her of late (I had been), but yet excesses, but, in the end, I would write and rewrite pre- she felt thankful and saw me as a trustworthy source of scriptions and, between office visits, phone in whatever counsel. In the days since, I have tried to make sense of this medications she wanted. When I did say “no,” I would seeming contradiction. Just as I was again falling victim to invariably get burned—later learning about renal stones on my inherent cynicism—“She’ll never quit, lots of addicts ultrasound or the like. This was the patient you could make lots of promises they cannot keep”—I received a note never be sure about. So, I wrote the prescriptions, made from her. In her message, she told me that she had joined the referrals, listened to her stories in person and on the Narcotics Anonymous and was off pills. She again thanked phone. Did she call me so often because of my sage counsel me for “everything” and signed the note as “your patient.” and steadying influence in her chaotic life or merely to While this still could have been an elaborate plot to control me? play me for more pills, I don’t think so. She may or may Slowly, steadily, I began to resent her. The constant not remain off the pills. But for me, this isn’t the critical interruptions, white lies, black lies, and endless refusal to point. Her life must be so terrible, her resources so thin, take my advice fueled my plan. I would wait for the next her connections to others so tenuous that even my half- “hard data” that would enable me to proclaim the end of hearted efforts seemed Herculean to her. Even as I have the current organic cause of pain and then steadfastly de- attempted to fathom some of the pain described by pa- cline any more pleas for narcotics. This opportunity pre- tients, I don’t think I am really able to understand and see sented itself the next week. I was almost eager for her call. the world from their perspective. I simply have been too It came, and I cut her off and told her, “I don’t think you fortunate in ways they have not. have renal stones now and your fractures are better, so I What have I learned? That I should not underestimate www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 307 On Being a Doctor Lessons from a Patient my role as “doctor” in someone’s life. Even when not dra- the case). But that is not central here. It is the act, the matic or seemingly curative, simple attention and caring go caring, and the connection that matter. Most patients seem a long way. I think I will be more up-front and direct with to forgive me my days of distraction and inattention. They patients who may be abusing substances. It may be awk- may see, at heart, a person who cares and is trying to do ward, it may cause some heat, but by dancing around it, what he can. That seems to be enough for most. Maybe however well-intentioned, I am not doing my duty to next week, Lisa will ask for pills, come in high, or tell me those who call themselves my patients. I owe them my about her difficult life despite my counsel. Whatever the caring and concern, but also my honesty. Ignoring the outcome, though, I know she will see me for what I have issue because it will prolong a visit, be awkward, or lead to always wanted to be: her doctor. more phone calls is not fair to them. The path of least resistance for me may be a path of destruction for them. William Rifkin, MD Sometimes in trying to be liked or thought well of as a Waterbury Hospital doctor, I have fed their delusions. There is a fine line be- Waterbury, CT 06721 tween this and too much paternalism, but I think I owe them my sincere, honest, and judgment-free engage- Requests for Single Reprints: William Rifkin, MD, Department of ment—even if that’s hard to provide. Medicine, Waterbury Hospital, 64 Robbins Street, Waterbury, CT In the end, I have found some of that meaning I’ve 06721. been searching for, quite inadvertently. I cannot “fix” my patients or even help them very much (as often seems to be Ann Intern Med. 2005;143:307-308.

308 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Letters

COMMENTS AND RESPONSES References 1. Screening for abdominal aortic aneurysm: recommendation statement. Ann Intern Med. 2005;142:198-202. [PMID: 15684208] 2. Fleming C, Whitlock EP, Beil TL, Lederle FA. Screening for abdominal aortic Screening for Abdominal Aortic Aneurysms aneurysm: a best-evidence systematic review for the U.S. Preventive Services Task Force. Ann Intern Med. 2005;142:203-11. [PMID: 15684209] TO THE EDITOR: The U.S. Preventive Services Task Force (USP- 3. Ashton HA, Buxton MJ, Day NE, Kim LG, Marteau TM, Scott RA, et al. The STF) recommended screening for abdominal aortic aneurysms Multicentre Aneurysm Screening Study (MASS) into the effect of abdominal aortic (AAAs) with ultrasonography only in male smokers 65 to 75 years of aneurysm screening on mortality in men: a randomised controlled trial. Lancet. 2002; age. Other subsets were excluded in part because of “good evidence 360:1531-9. [PMID: 12443589] that screening and early treatment result in important harms, includ- 4. Lederle FA, Johnson GR, Wilson SE, Chute EP, Littooy FN, Bandyk D, et al. ing an increased number of with associated morbidity and Prevalence and associations of abdominal aortic aneurysm detected through screening. mortality, and psychological harms” (1). I question the evidence Aneurysm Detection and Management (ADAM) Veterans Affairs Cooperative Study regarding harms. Group. Ann Intern Med. 1997;126:441-9. [PMID: 9072929] First, the companion review on this topic by Fleming and col- 5. Frydman G, Walker PJ, Summers K, West M, Xu D, Lightfoot T, et al. The value leagues (2) concluded that “screening does not appear to be associ- of screening in siblings of patients with abdominal aortic aneurysm. Eur J Vasc Endo- ated with significant physical or psychological harms.” Second, the vasc Surg. 2003;26:396-400. [PMID: 14512002] harm related to morbidity and mortality of “unnecessary” operations 6. Kent KC, Zwolak RM, Jaff MR, Hollenbeck ST, Thompson RW, Schermerhorn requires an unstated assumption that small AAAs identified by ML, et al. Screening for abdominal aortic aneurysm: a consensus statement. J Vasc screening would be inappropriately repaired before they reached a Surg. 2004;39:267-9. [PMID: 14718853] size of substantial rupture risk. The members of the USPSTF do not provide evidence for this assumption, nor do they indicate the mag- IN RESPONSE: The USPSTF released its recommendations on nitude of this effect on their conclusions. In fact, evidence from screening for AAA on 1 February 2005. The USPSTF recommends randomized trials of AAA screening indicates that more than 90% of 1-time screening for men 65 to 75 years of age who have ever subsequent elective AAA repairs were performed at recommended smoked. The recommendation is based on the high prevalence of size criteria (3). AAA in this population, their high risk for AAA rupture, and the The USPSTF made separate recommendations for men on the good-quality evidence from large population-based screening trials basis of smoking, even though randomized trials favor screening for showing that screening leads to decreased AAA-specific mortality in all men (2). To do so, the USPSTF relied on separate analyses of men. The USPSTF makes no recommendation for or against screen- AAA prevalence, based on risk factors such as smoking. Clearly, ing men 65 to 75 years of age who have never smoked because the screening is more cost-effective if the screened population has a balance of harms and benefits is too close to call. Prevalence of AAA higher prevalence of AAA, so the impact of smoking is an important is lower in men who have never smoked than in those who have ever consideration. It is unclear, however, why the USPSTF did not eval- smoked and, therefore, the potential for benefit in this population is uate women on the basis of smoking history. Female smokers have lower when balanced against important harms. Finally, the USPSTF the same AAA prevalence as male nonsmokers (1.9% for 3-cm AAAs recommends against routine primary care screening for women on [4]), yet they were bundled into a grade D recommendation against the basis of 3 important pieces of evidence: 1) The prevalence of screening for all women while male smokers received a neutral grade AAA in women is low compared with its prevalence in men (1.3% C recommendation. I do not believe that risk factor analysis should vs. 7.6%); 2) the peak prevalence of AAA in women is 10 years later be differentially applied on the basis of sex. than it is for men and AAAs in women therefore occur at ages when Finally, it is disappointing that the USPSTF ignored the impor- there are important competing causes of mortality; and 3) the avail- tance of a family history of AAA in its overall recommendations. able trial evidence shows no benefit from screening and repairing Most studies have found that first-degree relatives of patients with AAAs in women (1, 2). AAA have a much higher prevalence of small AAAs (25% to 43% in Contrary to Dr. Cronenwett’s assertion, the USPSTF did not brothers, 6% to 16% in sisters [5]) than the 5.9% prevalence in male base its concerns about AAA screening on psychological harms or the smokers (4), for whom the USPSTF issued a grade B recommenda- harms of unnecessary surgeries. The USPSTF considered both the tion for screening. benefits and harms of screening and early intervention associated I believe that the USPSTF recommendations were too conser- with AAA screening. The USPSTF carefully assessed the evidence of vative in not recommending AAA screening for all men older than harms and concluded that the harms of surgery for AAAs greater age 64 years, for female smokers in this age group, and for men or than 5.5 cm are important ones: 4% to 6% in-hospital mortality women in this age group whose sibling or parent had an AAA. The rates and 32.4% major complication rates (2). Higher rates of com- Society for Vascular Surgery and the Society for Vascular Medicine plications are anticipated in older women, that is, at the ages when and Biology have recommended more comprehensive screening that women develop AAAs. Therefore, it is likely that the net benefit of reflects these concerns (6). screening these women for AAA is, at best, zero. It is for these reasons that the USPSTF recommends against screening women for Jack L. Cronenwett, MD AAA. According to USPSTF methodology (3), the demonstration of Dartmouth-Hitchcock Medical Center no net benefit is sufficient for the USPSTF to recommend against Lebanon, NH 03756 providing a preventive service; no evidence of net harms is needed. Dr. Cronenwett points out that AAA prevalence in female Potential Financial Conflicts of Interest: None disclosed. smokers is about the same as it is in male nonsmokers. He disagrees

© 2005 American College of Physicians 309 Letters with the USPSTF for making no recommendation for or against room. A total of 10.1% of patients in the low blood pressure group screening men who have never smoked for AAA and for recommend- died compared with 6.3% in the usual blood pressure group, a result ing against screening women. The combination of low prevalence, that is statistically significant. However, there is no mention of in- late age of onset, and the negative results from the available screening creased deaths and no attempt to explain why this finding occurred. trial in women explains the reasons for not screening women for Perhaps death is outweighed by the clinical benefit of a delay in AAA (1, 2). The Task Force recognizes, however, that individualiza- dialysis, although it would be interesting to see how patients would tion of care is still required. For example, a clinician may choose to accept that reality in a “shared decision-making” process. In any discuss screening in the unusual circumstance in which a healthy event, I would ask that the authors provide an additional graph for female smoker in her early 70s has a first-degree family history of their Figure 3 detailing the cumulative probability of death over AAA that required surgery. time, along with the appropriate statistical analysis for this outcome. The USPSTF found poor-quality evidence that family history is associated with an increased risk for AAA. Studies by Larcos and Chester B. Good, MD, MPH Webster and colleagues (4, 5), which involved ultrasonographic ex- VA Pittsburgh Healthcare System amination of relatives of patients with AAA, found the prevalence of Pittsburgh, PA 15240 AAA to be much lower—0% and 4%, respectively—than that re- ported by Frydman and associates (6). Small sample sizes in all stud- ies of relatives of patients with AAA translated into large confidence Potential Financial Conflicts of Interest: None disclosed. intervals. In conclusion, the USPSTF used the best available evidence on Reference the risk factors for AAA and the benefits and harms of screening and 1. Sarnak MJ, Greene T, Wang X, Beck G, Kusek JW, Collins AJ, et al. The effect of treatment to recommend in favor of screening men 65 to 75 years of a lower target blood pressure on the progression of kidney disease: long-term follow-up age who have ever smoked. In such cases, there is evidence that AAA of the modification of diet in renal disease study. Ann Intern Med. 2005;142:342-51. screening can save lives. [PMID: 15738453]

Ned Calonge, MD, MPH TO THE EDITOR: Sarnak and colleagues (1) concluded that a low Diana Petitti, MD, MPH target blood pressure slowed the progression of nondiabetic kidney U.S. Preventive Services Task Force disease. However, Ruggenenti and associates (2) recently reported Rockville, MD 20852 that no additional benefit from further blood pressure reduction by felodipine (dihydropyridine calcium-channel blockers) could be Potential Financial Conflicts of Interest: None disclosed. shown in patients with nondiabetic proteinuric nephropathies receiv- ing background angiotensin-converting enzyme (ACE) inhibitor References therapy. As Sarnak and colleagues pointed out, more participants in 1. Scott RA, Bridgewater SG, Ashton HA. Randomized clinical trial of screening for the low target blood pressure group received ACE inhibitors in their abdominal aortic aneurysm in women. Br J Surg. 2002;89:283-5. [PMID: 11872050] study (1). Angiotensin-converting enzyme inhibitors are known to 2. Fleming C, Whitlock EP, Beil TL, Lederle FA. Screening for abdominal aortic slow progression of kidney disease independently of their blood pres- aneurysm: a best-evidence systematic review for the U.S. Preventive Services Task Force. Ann Intern Med. 2005;142:203-11. [PMID: 15684209] sure–lowering effect (3). In the African-American Study of Kidney 3. Translating evidence into recommendations. In: Current Methods of the U.S. Pre- Disease and Hypertension, targeting antihypertensive therapy at a ventive Services Task Force: A Review of the Process. Available at www.ahcpr.gov/ mean blood pressure of 92 mm Hg compared with usual targets of clinic/ajpmsuppl/harris3.htm. 102 to 107 mm Hg did not slow progression of hypertensive ne- 4. Larcos G, Gruenewald SM, Fletcher JP. Ultrasound screening of families with phrosclerosis (4). In that study, an identical proportion of patients in abdominal aortic aneurysm. Australas Radiol. 1995;39:254-6. [PMID: 7487760] the usual and lower blood pressure groups were taking ACE inhibi- 5. Webster MW, Ferrell RE, St Jean PL, Majumder PP, Fogel SR, Steed DL. Ultra- tors (4). In the study by Ruggenenti and associates (2), the propor- sound screening of first-degree relatives of patients with an abdominal aortic aneurysm. tion of participants receiving ramipril in both the conventional (di- J Vasc Surg. 1991;13:9-13. [PMID: 1987400] astolic blood pressure Ͻ90 mm Hg) and intensified (target blood 6. Frydman G, Walker PJ, Summers K, West M, Xu D, Lightfoot T, et al. The value pressure 130/80 mm Hg) blood pressure groups was identical in of screening in siblings of patients with abdominal aortic aneurysm. Eur J Vasc Endo- order to measure the pure effect of lowering blood pressure by felo- vasc Surg. 2003;26:396-400. [PMID: 14512002] dipine on the progression of kidney disease. Therefore, although Sarnak and colleagues’ sensitivity analysis did not affect their results, their conclusion that blood pressure reduction confers an additional Target Blood Pressure and Kidney Disease renoprotective effect remains questionable.

TO THE EDITOR: Sarnak and colleagues (1) presented an observa- Yujiro Kida, MD, PhD tional follow-up of the Modification of Diet in Renal Disease Tsurumi University School of Dental Medicine and Tokyo Medical (MDRD) Study. They reported that after a total of 10 years, the and Dental University hazard ratio for time to progression to kidney failure in the low Yokohama 230-8501, Japan blood pressure group was 0.68 (95% CI, 0.57 to 0.82) compared with the usual blood pressure group. However, in presenting this optimistic report, they seem to ignore the 2-ton gorilla in the living Potential Financial Conflicts of Interest: None disclosed.

310 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Letters

References Mark J. Sarnak, MD, MS 1. Sarnak MJ, Greene T, Wang X, Beck G, Kusek JW, Collins AJ, et al. The effect of Andrew S. Levey, MD a lower target blood pressure on the progression of kidney disease: long-term follow-up Tufts-New England Medical Center of the modification of diet in renal disease study. Ann Intern Med. 2005;142:342-51. Boston, MA 02111 [PMID: 15738453] 2. Ruggenenti P, Perna A, Loriga G, Ganeva M, Ene-Iordache B, Turturro M, et al. Tom Greene, PhD Blood-pressure control for renoprotection in patients with non-diabetic chronic renal Cleveland Clinic Foundation disease (REIN-2): multicentre, randomised controlled trial. Lancet. 2005;365:939-46. Cleveland, OH 44195 [PMID: 15766995] 3. Jafar TH, Schmid CH, Landa M, Giatras I, Toto R, Remuzzi G, et al. Angiotensin- Potential Financial Conflicts of Interest: None disclosed. converting enzyme inhibitors and progression of nondiabetic renal disease. A meta- analysis of patient-level data. Ann Intern Med. 2001;135:73-87. [PMID: 11453706] Reference 4. Wright JT Jr, Bakris G, Greene T, Agodoa LY, Appel LJ, Charleston J, et al. Effect 1. Jafar TH, Schmid CH, Landa M, Giatras I, Toto R, Remuzzi G, et al. Angiotensin- of blood pressure lowering and antihypertensive drug class on progression of hyperten- converting enzyme inhibitors and progression of nondiabetic renal disease. A meta- sive kidney disease: results from the AASK trial. JAMA. 2002;288:2421-31. [PMID: analysis of patient-level data. Ann Intern Med. 2001;135:73-87. [PMID: 11453706] 12435255]

CLINICAL OBSERVATIONS IN RESPONSE: Dr. Good raises the concern that deaths in our study may have been higher in the group with the lower target blood High-Output Heart Failure Associated with Anagrelide pressure. It is important to note, however, that the deaths to which Therapy for Essential Thrombocytosis Dr. Good refers include only those that occurred before development of kidney failure (see the legend to our Figure 2). Consideration of TO THE EDITOR: Background: Congestive heart failure (CHF) is deaths that occurred only before kidney failure may result in an usually associated with disorders causing low cardiac output; disor- informative censoring bias. That is, those patients who reach kidney ders associated with high cardiac output rarely cause CHF (1, 2). failure first may be more likely to die; however, the deaths that occur Medications have not previously been thought to cause high-output after kidney failure are not included in the comparison. Our Figure heart failure. New-onset CHF and cardiovascular death have rarely 2 also does not provide data on follow-up time, which limits the been associated with therapy with anagrelide (3–5), a novel treat- comparison. In fact, median follow-up time was longer in the lower ment for essential thrombocytosis. The mechanism of cardiac dys- blood pressure group because of a delay in reaching kidney failure. A function related to anagrelide has not previously been reported. longer follow-up time will of course allow more deaths to occur. Objective: To report a case of high-output heart failure associ- Without censoring for kidney failure, there were 101 and 107 ated with anagrelide use in a patient with essential thrombocytosis. deaths in the low and usual blood pressure groups, respectively. Case Report: A 34-year-old man was referred for evaluation of When Cox regression analyses were used, the adjusted hazard ratio pulmonary hypertension. He had received a diagnosis of essential for mortality was 0.97 (P Ͼ 0.2) for the low target blood pressure thrombocytosis at age 25 years, and his course was marked by recur- group compared with the usual blood pressure group. Thus, there is rent gastrointestinal bleeding and arterial thromboembolic disease. no evidence to support the contention that the low blood pressure Four years before the evaluation reported here, he began treatment target results in a higher death rate. with anagrelide. In the year before the patient’s cardiac evaluation, Of note, we presented the results of the composite of kidney the dose of anagrelide was titrated upward to 8 mg/d, with a subse- ϫ 9 failure and mortality. This incorporates the competing risk for death quent reduction in platelet count from 1800 10 cells/L to ϫ 9 into the kidney failure models. The results were consistent with the 400 10 cells/L. Over the next several months, he noted palpita- kidney failure models and demonstrated a benefit of the lower blood tion, pedal edema, increasing abdominal girth, exertional dyspnea, and muscle wasting. Furosemide was given with a subsequent 20- pressure target. pound weight loss, but the palpitations, dyspnea, and ascites per- We agree with Dr. Kida that we cannot disprove the possibility sisted. Echocardiography performed after onset of symptoms re- that ACE inhibitor use may have had an effect on the outcome. As vealed mild left ventricular dilatation (5.9 cm) and moderate left noted, however, we believe this is unlikely for the following reasons. atrial dilatation (5.3 cm) and right atrial dilatation. Left ventricular First, adjustment for ACE inhibitor use did not diminish the benefit wall motion and ejection fraction were normal. There was mild tri- of the lower blood pressure target. Second, the benefit achieved in cuspid regurgitation with a peak jet velocity of 3.5 m/s. Cardiac the low blood pressure target group (a hazard ratio of 0.68 with only catheterization revealed moderate pulmonary hypertension and a 19% difference in use of ACE inhibitors) is much greater than prompted referral to a subspecialty pulmonary hypertension clinic. would be expected if one compared these results with those of other Physical examination revealed a blood pressure of 120/60 mm Hg, a nondiabetic ACE inhibitor trials that observed a similar hazard ratio pulse of 80 beats/min, oxygen saturation of 97% on room air, mod- with a 100% difference in use of ACE inhibitors (1). erate elevation of jugular venous pressure, no S3 or murmur, a mod- We believe that the MDRD Study differs from the studies men- erately enlarged pulsatile liver, and 2ϩ pedal edema. Leukocyte tioned by Dr. Kida because it involves longer follow-up and more count was 18.4 ϫ 109 cells/L, hemoglobin level was 11%, hemato- kidney failure outcomes. We propose that it may take time for the crit was 0.34, platelet count was 962 ϫ 109 cells/L, brain natriuretic benefit of a lower blood pressure goal to be appreciated. peptide level was 326 ng/L, and thyroid-stimulating hormone level www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 311 Letters was 3.29 mU/L (within normal range). Results of serologic tests for the patient was asymptomatic and taking hydroxyurea therapy for HIV infection were negative. Electrocardiography showed first-de- essential thrombocytosis. gree atrioventricular block. A pulmonary ventilation–perfusion scan Discussion: High-output heart failure is an uncommon syn- showed low probability of pulmonary embolism, and computed to- drome that has been described in association with anemia, thyrotox- mography of the pulmonary arteries yielded negative results with no icosis, systemic arteriovenous fistulas, Paget disease, and nutritional evidence of acute or chronic pulmonary embolism. Repeated right- deficiencies, such as beriberi heart disease (1, 2). High cardiac output heart catheterization showed evidence of CHF related to high cardiac is also known to occur in association with certain types of renal and output (Figure, Table). hepatic disease, pregnancy, and treatment with certain vasodilators After a discussion with the patient’s hematologist, anagrelide (for example, epoprostenol), but the clinical syndrome of CHF is therapy was discontinued. Within 1 week, symptoms had improved rarely seen under these circumstances. dramatically and signs of CHF regressed. Diuretics were withdrawn. Anagrelide, a thrombocytopenic agent, is a quinazoline deriva- Repeated right-heart catheterization 6 weeks later revealed resolution tive that inhibits cyclic nucleotide phosphodiesterase type IV and is of all hemodynamic abnormalities (Figure, Table). Six months later, approved for use in essential thrombocytosis and for control of

Figure. Intracardiac pressure tracings before and after discontinuation of anagrelide.

312 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Letters

Table. Serial Hemodynamic Findings*

Site 8 March 2004 26 April 2004

Blood Pressure, Oxygen Blood Pressure, Oxygen mm Hg Saturation, % mm Hg Saturation, % Superior vena cava – 86 – 78 Inferior vena cava – 85 – 77 Right atrium 17 82 2 78 Right ventricle 50/24 80 25/4 – Pulmonary artery† 50/20 (37) 81 25/10 (16) 78 Pulmonary artery wedge 17 – 4 – Arterial 120/86 99 140/80 98

* Cardiac output was 15 on 8 March and 8.9 on 26 April; cardiac index was 6.9 on 8 March and 4.2 on 26 April. † Values in parentheses are means.

thrombocytosis associated with vera. In some studies, References commonly reported cardiovascular side effects included palpitations, 1. Braunwald E, Zipes D, Libby P. Heart Disease—A Textbook of Cardiovascular tachycardia, and fluid retention or edema (3). In a study of 577 Medicine. Philadelphia: WB Saunders; 2001. patients treated with anagrelide, 14 developed CHF; of these, 2 died 2. Anand IS, Florea VG. High output cardiac failure. Curr Treat Options Cardiovasc suddenly (4). In another study of 942 patients taking anagrelide for Med. 2001;3:151-159. [PMID: 11242561] 3. Silverstein MN, Petitt RM, Solberg LA Jr, Fleming JS, Knight RC, Schacter LP. thrombocytosis, 15 died of cardiac causes (5). None of the heart Anagrelide: a new drug for treating thrombocytosis. N Engl J Med. 1988;318:1292-4. failure cases in either of these studies was studied in detail. [PMID: 3362187] The present case is remarkable for 2 major features: marked 4. Anagrelide, a therapy for thrombocythemic states: experience in 577 patients. elevation of cardiac output and normal left ventricular ejection frac- Anagrelide Study Group. Am J Med. 1992;92:69-76. [PMID: 1731512] tion in the setting of clinical CHF, and nearly immediate resolution 5. Petitt RM, Silverstein MN, Petrone ME. Anagrelide for control of thrombocythe- of clinical and hemodynamic abnormalities after withdrawal of mia in polycythemia and other myeloproliferative disorders. Semin Hematol. 1997;34: anagrelide. The observation of high cardiac output as a part of the 51-4. [PMID: 9025162] pathophysiologic characteristics of this case is not entirely unex- 6. Packer M, Carver JR, Rodeheffer RJ, Ivanhoe RJ, DiBianco R, Zeldis SM, et al. pected. Anagrelide is known to have positive inotropic activity in Effect of oral milrinone on mortality in severe chronic heart failure. The PROMISE animals (4). As a phosphodiesterase inhibitor, it may have actions in Study Research Group. N Engl J Med. 1991;325:1468-75. [PMID: 1944425] common with such drugs as amrinone, milrinone, and enoximone, phosphodiesterase type 3 inhibitors known for their positive inotro- pic effects and for their adverse effect on mortality when used over the long term in patients with heart failure (6). Subungual Splinter Hemorrhages: A Clinical Window to Cardiac disease is a major cause of death (including sudden Inhibition of Vascular Endothelial Growth Factor Receptors? death) in patients treated with anagrelide. Since the adverse hemo- dynamic effects were reversible in our case, clinicians should consider TO THE EDITOR: Background: Over the past few years, cancer ther- discontinuing anagrelide therapy when even mild cardiac side effects apy has been profoundly changed by the development of drugs tar- are noted. geting specific kinases involved in the proliferation of tumor or en- dothelial cells. Some kinase inhibitors exhibit antiangiogenic activity by targeting vascular endothelial growth factor receptors (VEGFRs). Peter J. Engel, MD Objective: We report the results of tests of 2 such inhibitors: Heide Johnson, RN SU11248 (Pfizer, New York, New York), which inhibits VEGFR-2, The Ohio Heart Health Center and The Lindner Center for platelet-derived growth factor receptor (PDGFR), KIT, and flt3, and Research and Education BAY 439006 (Onyx Pharmaceuticals, Emeryville, California, and Cincinnati, OH 45219 Bayer, Leverkusen, Germany), which targets VEGFR-2 and VEGFR-3, PDGFR, and RAF gene products. Methods and Findings: In a phase I study, we treated patients Robert P. Baughman, MD with various types of cancer with escalating doses of SU11248. Pa- University of Cincinnati Medical Center tients with renal-cell carcinoma received BAY 439006 in a phase III Cincinnati, OH 45267 study. During systematically performed prospective dermatologic ex- aminations, we observed uncommon side effects, such as hair depig- mentation in patients treated with SU11248 (1). Unexpectedly, the Arthur I. Richards, MD most common side effect associated with these drugs was multiple, Oncology Hematology Care, Inc. painless distal subungual splinter hemorrhages (Figure), which devel- Cincinnati, OH 45219 oped in 2 to 4 weeks in 30% of patients taking SU11248 and in more than 60% of patients taking BAY 439006. The hemorrhages Potential Financial Conflicts of Interest: None disclosed. were not associated with thrombotic or embolic clinical events. www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 313 Letters

Figure. Subungual splinter hemorrhages. Subungual splinter hemorrhages are not usually drug-related and have not been reported with other kinase inhibitors. SU11248 and BAY 489006 both target receptors with angiogenic activity: VEGFR and PDGFR. However, subungual splinter hemorrhages are seen only rarely with imatinib, a potent PDGFR inhibitor. Thus, if we correlate the action spectrum of the kinase inhibitors with the observed side effect, it seems that subungual splinter hemorrhages could result from elective VEGFR blocking. Indeed, VEGFR may be constitutively involved in the contin- uous renewal of the delicate spiral capillaries that sustain frequent microinjuries at the extremities of the fingers. Blocking these recep- tors might prevent the physiologic repair of traumatized nail bed capillaries and result in subungual splinter hemorrhages. According to this hypothesis, subungual splinter hemorrhages could directly reflect the anti-VEGFR properties of these new agents, and nail beds might offer an interesting clinical monitoring window to the anti- angiogenic effects of drugs. Inhibitors of epidermal growth factor receptor also frequently target the skin, resulting in a diffuse follicular rash. There is now strong evidence that the cutaneous rash is associated with objective response and prolonged survival, supporting the role of cutaneous response as a surrogate marker of activity (5). In our report, if mul- tiple subungual splinter hemorrhages reflect the antiangiogenic effect of the drug, they could also be associated with an antitumor thera- peutic benefit. Conclusions: We remind other investigators to examine their patients’ nails to identify subungual splinter hemorrhages in this clinical setting. Through collaboration, data could be collected from a large number of patients, possibly allowing researchers to deter- mine whether multiple subungual splinter hemorrhages are indeed a signature of anti-VEGFR agents and whether they are associated with a better tumor response.

Caroline Robert, MD, PhD Sandrine Faivre, MD, PhD Eric Raymond, MD, PhD Top. Multiple-digit subungual splinter hemorrhages in a patient with Jean-Pierre Armand, MD renal-cell carcinoma treated for 6 weeks with BAY 439006. Bottom. Bernard Escudier, MD Epiluminescence microscopic view of a patient’s nail showing discrete subungual splinter hemorrhages. Institute Gustave Roussy 94800 Villejuif Cedex, France Discussion: First described in 1920 in patients with infective endocarditis, subungual splinter hemorrhages are well known to in- Potential Financial Conflicts of Interest: None disclosed. ternists (2). They appear as black or red lines that look like wood splinters under the nails. They are formed and confined in the epi- dermis of the nail bed and consist of a mass of blood in a layer of References squamous cells adhering to the undersurface of the nail. Depending 1. Robert C, Spatz A, Faivre S, Armand JP, Raymond E. Tyrosine kinase inhibition on their origin, they may occur in clusters, and their pattern of and grey hair [Letter]. Lancet. 2003;361:1056. [PMID: 12660081] distribution, color, location (proximal or distal), and painful or pain- 2. Young JB, Will EJ, Mulley GP. Splinter haemorrhages: facts and fiction. J R Coll less character vary. They were initially thought to be a typical sign of Physicians Lond. 1988;22:240-3. [PMID: 3230540] bacterial endocarditis but were subsequently reported to occur in 3. Mujic F, Lloyd M, Cuadrado MJ, Khamashta MA, Hughes GR. Prevalence and various conditions, such as the antiphospholipid syndrome, severe clinical significance of subungual splinter haemorrhages in patients with the antiphos- rheumatoid arthritis, thromboangiitis obliterans, mitral stenosis, or pholipid syndrome. Clin Exp Rheumatol. 1995;13:327-31. [PMID: 7554559] when arterial are used (2, 3). They can also be seen in 4. Monk BE. The prevalence of splinter haemorrhages. Br J Dermatol. 1980;103: healthy patients, in whom trauma seems to be the most common 183-5. [PMID: 7426414] cause (2, 4). Traumatic subungual splinter hemorrhages occur where 5. Perez-Soler R, Chachoua A, Hammond LA, Rowinsky EK, Huberman M, Karp D, the nail plate separates from the nail bed; delicate specialized spiraled et al. Determinants of tumor response and survival with erlotinib in patients with capillaries beneath the nail bed can easily rupture. non–small-cell lung cancer. J Clin Oncol. 2004;22:3238-47. [PMID: 15310767]

314 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Letters

Figure. Microscopic examination of the liver showing invasion Disseminated Aspergillosis Mimicking Hepatic of hepatic sinusoids by fungal hyphae associated with Veno-Occlusive Disease dilatation of sinusoids and atrophy of hepatic cords.

TO THE EDITOR: Background: The incidence of invasive aspergillosis is increasing in parallel with the number of immunosuppressed pa- tients (1). Although Aspergillus is not known as a pathogen with liver tropism, 11% of patients who die of or with invasive aspergillosis have liver involvement (2). The clinical patterns of hepatic aspergil- losis vary and depend on the extent of invasion and the anatomic site of the liver involved. To our knowledge, invasion of sinusoids by fungal hyphae resulting in a syndrome mimicking hepatic veno- occlusive disease has not been previously reported. Case Report: A 53-year-old woman was admitted to our hospital with a 10-day history of productive cough and fever (temperature, Յ38.6 °C). Chest radiograph and computed tomography of the tho- rax revealed infiltrates in both apices without cavitation. The patient underwent bronchoscopy, and the bronchoalveolar lavage cultures grew Aspergillus fumigatus. Amphotericin B was initiated at a daily dose of 1 mg/kg of body weight. Two weeks after initiation of treatment, the patient had remarkable clinical and radiographic improvement. She was doing well until the twenty-third day of treatment, when her hos- (Periodic acid–Schiff stain; original magnification, ϫ 400.) pital course was complicated with staphylococcal sepsis related to a sub- clavian vein catheter. The line was removed, amphotericin B was with- drawn, and dicloxacillin and gentamicin therapy was started. Three days amphotericin B (AmBisome, Fujisawa Healthcare, Inc., Deerfield, later, the patient became afebrile and her clinical condition stabilized. Illinois), 5 mg/kg per day. Two weeks after initiation of treatment, One week after withdrawal of amphotericin B, while the patient she had remarkable clinical and laboratory improvement. The am- was receiving antistaphylococcal treatment, she developed abdominal photericin B regimen was continued for 1 month, and the patient distention, pain in the right upper quadrant of the abdomen, and a was discharged home in good condition while taking itraconazole temperature of 38 °C. On physical examination, jaundice, tender (400 mg/d). On follow-up visits, the patient had no clinical or lab- hepatosplenomegaly, and ascites were present. Laboratory evaluation oratory evidence of active infection and her absolute CD4 cell count revealed the following: hemoglobin level, 74 g/L; leukocyte count, was 0.212 ϫ 109 cells/L. Itraconazole therapy was discontinued after ϫ 9 ϫ 9 12 10 cells/L; platelet count, 60 10 cells/L; bilirubin level, 1 year. ␮ 236 mol/L (13.8 mg/dL); alkaline phosphatase level, 87 U/L; ala- Discussion: Disseminated aspergillosis occurs in different patient nine aminotransferase level, 44 U/L; total serum protein level, 65 groups (1). Our patient had persistently low CD4 cell counts g/L; albumin level, 3.0 g/L; globulin level, 3.5 g/L; prothrombin (Ͻ 0.3 ϫ 109 cells/L) without any other recognized immune defect, time, 12 seconds (control, 11 seconds); creatinine concentration, 115 leading to the diagnosis of idiopathic CD4 lymphopenia (3). Appar- ␮ mol/L (1.3 mg/dL); pH, 7.43; PCO2, 24 mm Hg; and PO 2,77mm ently, the profound immune dysfunction of the host, as indicated by Hg. The patient’s condition deteriorated: She was intubated and trans- the low CD4 cell count, and the inadequate initial antifungal treat- ferred to the intensive care unit. Computed tomography of the chest ment were the main factors that contributed to hematogenous dis- showed extensive infiltrates in both lung fields and bilateral pleural effu- semination of Aspergillus from a pulmonary focus to the liver. sion; computed tomography of the abdomen revealed hepatomegaly, Characteristically, Aspergillus forms abscesses in the liver with splenomegaly, and ascites without focal lesions. Doppler ultrasonogra- vascular invasion, thrombosis, and infarction of tissue in the sur- phy of hepatic and portal veins and inferior vena cavography did not rounding area (2). When the abscesses are small, patients may have reveal any obstruction. The ascitic fluid was turbid and hemorrhagic and no symptoms and modest liver function abnormalities. However, showed a protein level of 40 g/L and 0.8 ϫ 109 leukocytes/L with 80% when the abscesses are large and associated with hepatocellular dam- on examination. All cultures were negative for bacteria and age, patients have pronounced symptoms and marked liver function fungi. Liver biopsy revealed invasion of hepatic sinusoids, central veins, abnormalities (2). It is of interest that invasion of particular anatomic and the portal tract by fungal hyphae associated with sinusoidal dilata- sites by Aspergillus produces several distinct syndromes. As reported tion and atrophy of hepatic cords (Figure). previously, invasion of the hepatic veins by fungi may produce the Results of serologic tests for HIV-1, HIV-2, and human T-cell Budd–Chiari syndrome (4). In our patient, obstruction of hepatic lymphotrophic virus types 1 and 2 were negative twice. Levels of venous outflow by masses of fungal hyphae occurred at the level of immunoglobulins, including IgG subclasses and IgE, were within sinusoids and central hepatic veins, resulting in a syndrome charac- normal limits. Leukocyte chemotaxis, results of the nitroblue tetra- terized by tender hepatomegaly, splenomegaly, ascites, and jaundice, zolium test, and complement components C3, C4, and total hemo- mimicking hepatic veno-occlusive disease. Hepatic veno-occlusive lytic activity CH50 were normal. Bone marrow biopsy was not re- disease was first described in South Africa and was linked to the vealing. The CD4 cell count was 0.196 ϫ 109 cells/L, and the CD4/ ingestion of pyrrolizidine alkaloids contained in Senecio tea. More CD8 ratio was 0.45. The patient began receiving liposomal recently, veno-occlusive disease has been seen in liver and hemato-

www.annals.org 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 315 Letters poietic cell transplantation as well as in association with different spectrum of the disease in 98 patients. Medicine (Baltimore). 1970;49:147-73. chemotherapeutic agents (5). [PMID: 4913991] Conclusion: This case illustrates a not previously recognized 3. Smith DK, Neal JJ, Holmberg SD. Unexplained opportunistic infections and form of hepatic aspergillosis that had all the clinical characteristics of CD4ϩ T-lymphocytopenia without HIV infection. An investigation of cases in the ϩ veno-occlusive disease and occurred in a patient with idiopathic United States. The Centers for Disease Control Idiopathic CD4 T-lymphocytopenia CD4 lymphopenia. It also highlights the importance of liver biopsy Task Force. N Engl J Med. 1993;328:373-9. [PMID: 8093633] to establish the diagnosis and distinguish hepatic aspergillosis from 4. Young RC. The Budd-Chiari syndrome caused by Aspergillus; two patients with other causes of veno-occlusive disease. vascular invasion of the hepatic veins. Arch Intern Med. 1969;124:754-7. [PMID: 5353485] 5. Wadleigh M, Ho V, Momtaz P, Richardson P. Hepatic veno-occlusive disease: George L. Daikos, MD pathogenesis, diagnosis and treatment. Curr Opin Hematol. 2003;10:451-62. [PMID: Vassiliki Syriopoulou, MD 14564177] George Aperis, MD Christos Toubanakis, MD George Petrikkos, MD CORRECTION University of Athens 115-27 Athens, Greece Correction: Recommendations for the Diagnosis and Maria Demonakou, MD Treatment of the Acute Porphyrias Sismonoglion General Hospital A review on recommendations for diagnosis and treatment of acute 151-26 Athens, Greece porphyrias (1) contained an error. In the Key Summary Points on page 440, under “Treatment of the Acute Attack,” the unit of hemin Potential Financial Conflicts of Interest: None disclosed. in the third paragraph should be 3 to 4 mg/kg daily.

References Reference 1. Denning DW. Invasive aspergillosis. Clin Infect Dis. 1998;26:781-803. [PMID: 1. Anderson KE, Bloomer JR, Bonkovsky HL, Kushner JP, Pierach CA, Pimstone NR, 9564455] et al. Recommendations for the diagnosis and treatment of the acute porphyrias. Ann 2. Young RC, Bennett JE, Vogel CL, Carbone PP, DeVita VT. Aspergillosis. The Intern Med. 2005;142:439-50. [PMID: 15767622]

316 16 August 2005 Annals of Internal Medicine Volume 143 • Number 4 www.annals.org Annals of Internal Medicine Summaries for Patients Benefits and Harms of Warfarin plus Aspirin after Acute Coronary Events

Summaries for Patients are a service What is the problem and what is known about it so far? provided by Annals to help patients Clogged heart arteries can cause acute coronary events, such as heart attacks or worsening better understand the complicated chest pain (unstable angina). Many middle-aged and older adults have these and often mystifying language of life-threatening events. They are a major cause of medical care and hospitalization in modern medicine. developed countries. Adults who survive the events often have high risks for blood clots in the heart’s arteries and for repeated events. Doctors usually prescribe aspirin to help prevent further clots and repeated events. They may also prescribe a “blood thinner” The full report is titled “Warfarin (warfarin) or both aspirin and warfarin. Warfarin requires frequent monitoring and plus Aspirin after Myocardial sometimes causes problems with bleeding. Doctors sometimes wonder if giving both Infarction or the Acute Coronary aspirin and warfarin leads to more harms than benefits. Syndrome: Meta-Analysis with Estimates of Risk and Benefit.” It Why did the researchers do this particular study? is in the 16 August 2005 issue of To summarize literature about the benefits and harms of warfarin plus aspirin for people Annals of Internal Medicine who have had an acute coronary event. (volume 143, pages 241-250). Who was studied? The authors are M.B. Rothberg, 5938 adults included in 10 randomized trials. All had either a heart attack or unstable C. Celestin, L.D. Fiore, E. angina. None received stent procedures to open up blocked blood vessels. Lawler, and J.R. Cook. How was the study done? The authors looked for studies published between 1990 and October 2004. They selected randomized trials that compared regular-intensity warfarin (international normalized ratio Ͼ 2.0) plus aspirin with aspirin alone in patients with acute coronary events. They combined data from these studies to estimate chances of benefits (reductions in heart attacks, ischemic strokes, and heart artery procedures) and harms (major bleeding events). They then applied numbers from the combined benefit and harm data (rate ratios) to a large population-based study that included many patients with different risks for heart events and bleeding. The researchers estimated groups of people for whom warfarin plus aspirin might be more helpful than harmful.

What did the researchers find? Compared with aspirin alone, warfarin plus aspirin decreased annual rates of heart attacks, ischemic strokes, and heart artery procedures. Warfarin plus aspirin increased annual rates of major bleeding. Patients with several risk factors, such as older age, diabetes, heart failure, and decreased kidney function, had the highest risks for heart attacks and strokes. Older patients who previously had a stroke, gastrointestinal bleeding episode, kidney disease, or atrial fibrillation were at higher risk for bleeding. Numbers of deaths did not differ between treatments. In patients with low or average risk for bleeding, the numbers of heart and stroke events prevented with combination therapy exceeded the numbers of major bleeding episodes that it caused.

What were the limitations of the study? Doctors treat many patients with acute coronary events with stents. They don’t use warfarin in these patients because of bleeding risks associated with the stent procedure. This review’s findings do not apply to patients treated with stents. Also, the researchers sometimes used assumptions when they applied data to different patient groups.

What are the implications of the study? Benefits of warfarin plus aspirin may outweigh harms for some patients with acute coronary events who are not stented and do not have high bleeding risks.

Summaries for Patients are presented for informational purposes only. These summaries are not a substitute for advice from your own medical provider. If you have questions about this material, or need medical advice about your own health or situation, please contact your physician. The summaries may be reproduced for not-for-profit educational purposes only. Any other uses must be approved by the American College of Physicians.

I-14 © 2005 American College of Physicians Annals of Internal Medicine Summaries for Patients The Outcomes and Costs of Diabetes Prevention with a Diet and Exercise Program or Metformin: A Computer Model

Summaries for Patients are a service What is the problem and what is known about it so far? provided by Annals to help patients Type 2 diabetes mellitus interferes with the body’s ability to store energy from food. The better understand the complicated result is high blood sugar levels that can lead to complications, such as blindness, kidney and often mystifying language of failure, and heart disease. Many people with type 2 diabetes have moderately high levels of modern medicine. blood sugar for years before the levels reach true diabetes levels and symptoms begin. This condition is called abnormal glucose tolerance or “prediabetes.” A large study showed that 2 different methods can prevent or delay the development of diabetes in people with The full report is titled “Clinical prediabetes. The first method is an intensive diet and exercise program supervised by a Outcomes and Cost-Effectiveness clinic. The other is a small daily dosage of a diabetes medication called metformin. Because of Strategies for Managing People many Americans have prediabetes, using 1 of these prevention methods in all people who at High Risk for Diabetes.” It is in might benefit would be expensive. However, caring for diabetes and its complications is the 16 August 2005 issue of also expensive. It is important to know whether the costs of diabetes prevention are worth Annals of Internal Medicine the potential benefits. A previous analysis using a computer model estimated that, (volume 143, pages 251-264). compared with no prevention, the diet and exercise program would cost about $8800 and The authors are D.M. Eddy, L. metformin about $29,000 per quality-adjusted year of life saved. Schlessinger, and R. Kahn. Why did the researchers do this particular study? To estimate whether the costs of using an intensive clinic-based diet and exercise program or metformin to prevent type 2 diabetes is worth the potential benefits. Who was studied? Rather than studying actual patients, the researchers used computers to simulate what would happen to a virtual group of adults with prediabetes. How was the study done? The researchers used published information to estimate what might happen (and how much it would cost) if doctors prescribed an intensive clinic-based diet and exercise program or metformin for adults with prediabetes. They put these estimates into a computer model called the Archimedes model and calculated how much each strategy would cost per year of life saved. The Archimedes model includes more detailed biological and administrative information than models typically used to conduct these types of analyses. What did the researchers find? Both the diet and exercise program and the metformin program would benefit people with prediabetes. However, the program provides more than twice the degree of benefit as metformin. The program is more expensive than metformin, costing approximately $672 a year. Compared with no program at all, and taking its benefits into account, it would cost society about $62,600 per quality-adjusted year of life saved. The diet and exercise program would save money if its yearly cost could be decreased from $672 to $100. What were the limitations of the study? We can never be certain how well a computer model represents what would happen with real patients over long time periods. Much of the information on use of health care services was based on data from Kaiser Permanente and may not accurately represent other health care settings. What are the implications of the study? These authors confirm that a clinic-based diet and exercise program has important benefits for people with prediabetes. However, they estimate that the cost to society of a formal clinic-based program per amount of benefit is higher than previous estimates.

Summaries for Patients are presented for informational purposes only. These summaries are not a substitute for advice from your own medical provider. If you have questions about this material, or need medical advice about your own health or situation, please contact your physician. The summaries may be reproduced for not-for-profit educational purposes only. Any other uses must be approved by the American College of Physicians.

I-22 © 2005 American College of Physicians Annals of Internal Medicine Summaries for Patients Estimating Outcome in Patients with HIV-Related Lymphoma

Summaries for Patients are a service What is the problem and what is known about it so far? provided by Annals to help patients Human immunodeficiency virus (HIV) is the cause of AIDS, an illness that interferes with better understand the complicated the ability to fight off infection and certain types of cancer. Since 1996, combinations of and often mystifying language of drugs, known as highly active antiretroviral therapy, or HAART, have been prescribed to modern medicine. decrease infections and cancer and lengthen life in many people with AIDS. Lymphoma is a type of cancer that develops in lymph nodes, collections of immune cells that are present throughout the body. Lymphoma occurs in people who do not have The full report is titled, “A HIV infection or AIDS, but it is more common in people with HIV or AIDS than in the Prognostic Index for Systemic general population. AIDS-Related Non-Hodgkin The International Prognostic Index (IPI) is an index that helps doctors to predict Lymphoma Treated in the Era of survival of people with lymphoma. The IPI considers patient age, cancer stage, and Highly Active Antiretroviral laboratory tests. It is not known how well the IPI predicts survival in people with Therapy.” It is in the 16 August HIV-related lymphoma who are receiving HAART. Since HIV-related lymphoma is a 2005 issue of Annals of Internal common complication of HIV infection, it would be helpful to have a way to predict how Medicine (volume 143, pages patients with this cancer will do. 265-273). The authors are M. Why did the researchers do this particular study? Bower, B. Gazzard, S. Mandalia, To see whether the IPI or other factors could help to predict survival in people with T. Newsom-Davis, C. Thirlwell, HIV-related lymphoma. T. Dhillon, A.M. Young, T. Powles, A. Gaya, M. Nelson, and Who was studied? J. Stebbing. 111 patients with HIV infection and lymphoma diagnosed since 1996.

How was the study done? The researchers collected information on the factors that make up the IPI, other patient characteristics, and laboratory values and developed mathematical models to see which combination of factors did the best job of predicting how long people lived.

What did the researchers find? The IPI index plus a laboratory value called CD4 cell count could separate patients into 4 groups that had 1-year survival rates of 82%, 47%, 20%, and 15%.

What were the limitations of the study? The study included only 111 patients who were taking a variety of different combinations of HIV drugs. It was not able to determine whether the type of HIV treatment was related to survival.

What are the implications of the study? The IPI and CD4 cell count can help doctors predict the prognosis of patients with HIV-related lymphoma.

Summaries for Patients are presented for informational purposes only. These summaries are not a substitute for advice from your own medical provider. If you have questions about this material, or need medical advice about your own health or situation, please contact your physician. The summaries may be reproduced for not-for-profit educational purposes only. Any other uses must be approved by the American College of Physicians.

I-28 © 2005 American College of Physicians Annals of Internal Medicine Summaries for Patients Association of Quality of Care with Survival of Elderly Managed Care Patients

Summaries for Patients are a service What is the problem and what is known about it so far? provided by Annals to help patients Evaluations of the quality of medical care often focus on processes of care, such as how better understand the complicated often patients receive certain tests or treatments. Some people believe that a better way to and often mystifying language of measure quality of care would be to look at patient outcomes, such as how often they modern medicine. become sick enough to need hospitalization or how long they live. Unfortunately, it is often more difficult to measure patient outcomes than to measure processes of care. First, measuring patient outcomes requires following the patient for a long period of time. The full report is titled “Quality Second, patient outcomes depend on other factors, such as patient age or their baseline of Care Is Associated with level of health, in addition to the quality of the medical care they receive. Theoretically, Survival in Vulnerable Older there should be a link between recommended processes of care and outcomes (such as Patients.” It is in the 16 August patient survival), but few studies have examined this link. 2005 issue of Annals of Internal Medicine (volume 143, pages Why did the researchers do this particular study? 274-281). The authors are T. To see whether patients who received higher-quality care lived longer than similar patients Higashi, P.G. Shekelle, J.L. who received lower-quality care. Adams, C.J. Kamberg, C.P. Roth, Who was studied? D.H. Solomon, D.B. Reuben, L. 372 patients who were at least 65 years of age and were enrolled in 1 of 2 managed care Chiang, C.H. MacLean, J.T. organizations from July 1, 1998, through July 31, 1999. All patients lived in the Chang, R.T. Young, D.M. Saliba, community but were at high risk for worsening health over the next 2 years. and N.S. Wenger. How was the study done? The researchers used a set of 236 measures of quality of care that their group had developed for use in older patients. The measures were process-of-care measures, meaning that they looked at whether the patient had received a certain test, screening, or treatment. Every quality-of-care measure did not apply to every patient. For example, the quality-of-care measures for high blood pressure applied only to patients who had high blood pressure. The researchers reviewed patients’ medical records to determine the number of applicable quality-of-care measures met for each patient and followed patients for 3 years to see whether patients died. They looked for a relationship between the number of quality-of-care measures met and whether a patient was alive after 3 years.

What did the researchers find? The researchers found that the fewer quality-of-care measures that a patient’s care met, the more likely a person was to die during the 3 years of follow-up.

What were the limitations of the study? This study cannot prove that poor quality of care causes death. Health care providers may have refrained from providing some types of care to patients who had a high chance of dying or to patients who declined some types of care.

What are the implications of the study? This study suggests that, in older patients at high risk for worsening health, better performance on process measures of quality of care are associated with better patient survival.

Summaries for Patients are presented for informational purposes only. These summaries are not a substitute for advice from your own medical provider. If you have questions about this material, or need medical advice about your own health or situation, please contact your physician. The summaries may be reproduced for not-for-profit educational purposes only. Any other uses must be approved by the American College of Physicians.

© 2005 American College of Physicians I-33