Tài liệu Differences in 4-Year Health Outcomes for Elderly and Poor, Chronically III Patients Treated in HMO and Fee-for-Service Systems ppt

Reprinted from JAMA @ The Journal of the American Medical Association  October 2, 1996 Volume 276  Copyright 1996, American Medical Association Original Contributions Differences in 4-Year Health Outcomes for Elderly and Poor, Chronically III Patients Treated in HMO and Fee-for-Service Systems Results From the Medical Outcomes Study John E. Ware, Jr, PhD; Martha S. Bayliss, MSc; William H. Rogers, PhD; Mark Kosinski, MA; Alvin R. Tarlov, MD Objective To compare physical and mental health outcomes of chronically ill adults, including elderly and poor subgroups, treated in health maintenance orga- nization (HMO) and fee-for-service (FFS) systems. Study Design A 4-year observational study of 2235 patients (18 to 97 years of age) with hypertension, non-insulin-dependent diabetes mellitus (NIDDM), re- cent acute myocardial infarction, congestive heart failure, and depressive disorder sampled from HMO and FFS systems in 1986 and followed up through 1990. Those aged 65 years and older covered under Medicare and low-income patients (200% of poverty) were analyzed separately. Setting and Participants Offices of physicians practicing family medicine, in- ternal medicine, endocrinology, cardiology, and psychiatry, in HMO and FFS sys- tems of care. Types of practices included both prepaid group (72% of patients) and i ndependent practice association (28%) types of HMOs, large multispecialty groups, and solo or small, single-specialty practices in Boston, Mass, Chicago, III, and Los Angeles, Calif. Outcome Measures Differences between initial and 4-year follow-up scores of summary physical and mental health scales from the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) for all patients and practice settings. Results On average, physical health declined and mental health remained stable during the 4-year follow-up period, with physical declines larger for the elderly than for the nonelderly (P<.001). In comparisons between HMO and FFS systems, physical and mental health outcomes did not differ for the average patient; however, they did differ for subgroups of the population differing in age and poverty status. For elderly patients (those aged 65 years and older) treated under Medicare, de- clines in physical health were more common in HMOs than in FFS plans (54% vs 28%; P<.001). In 1 site, mental health outcomes were better (P<.05) for elderly patients in HMOs relative to FFS but not in 2 other sites. For patients differing in poverty status, opposite patterns of physical health (P<.05) and for mental health ( P<.001) outcomes were observed across systems; outcomes favored FFS over HMOs for the poverty group and favored HMOs over FFS for the nonpoverty group. Conclusions During the study period, elderly and poor chronically ill patients had worse physical health outcomes in HMOs than in FFS systems; mental health outcomes varied by study site and patient characteristics. Current health care plans should carefully monitor the health outcomes of these vulnerable subgroups. JAMA, October 2, 1996-Vol 276, No. 13 JAMA. 1996;276:1039-1047 ENROLLMENTS in health mainte- nance organizations (HMOs) have in- creased nearly 10-fold since 1976, and in some regions of the country, half of pri- vately insured Americans are enrolled in HMOs! Policies at the state and fed- eral levels seek to affect a similar shift for those who are publicly insured, in- cluding both Medicare and Medicaid. Congress has signed legislation that will give Medicare patients strong financial incentives to enroll in managed care plans. Yet, as documented in a recent literature analysis,' little is known about health outcomes in HMOs for the elder- ly and the poor, who have historically tended to favor fee-for-service (FFS) over HMO systems. The Medical Outcomes Study (MOS) was fielded to compare 4-year health outcomes for chronically ill patients treated in well-established HMOs and FFS plans serving the same "medical marketplaces" in 3 cities.' To increase the generalizability of results, adults with 4 physical conditions (hypertension, non-insulin-dependent diabetes mellitus [ NIDDM], recent acute myocardial in- farction, and congestive heart failure) and 1 mental condition (depressive dis- From The Health Institute, New England Medical Center (Drs Ware, Rogers, and Tarlov, Ms Bayliss, and Mr Kosinski), Tufts University School of Medicine (Drs Ware and Tarlov), and Harvard School of Public Health ( Drs Ware and Tarlov), Boston, Mass. Reprints: John E. Ware, Jr, PhD, The Health Institute, New England Medical Center, Box 345, 750 Washington St, Boston, MA 02111 ( e-mail: j ohn.ware@es.nemc.org) Chronically III Elderly and Poor Patients-Ware et al  1039 order) were followed. Sampling patients with the same diagnoses across systems of care and measuring them with the same methods allowed more valid com- parisons of outcomes across plans. To better address policy issues, the MOS oversampled the elderly and the poor. Focusing on chronically ill patients and oversampling of the elderly and poor increased the likelihood of detecting dif- ferences in health outcomes because these subgroups account for a dispro- portionate share of health care expen- ditures and are, therefore, prime tar- gets of cost containment. We report here the results of com- paring changes in physical and mental health status between FFS and HMO systems, measured over a 4-year pe- riod. In contrast to previous MOS re- ports of outcomes for the average pa- tient, we focus on outcomes for policy- relevant subgroups-including patients aged 65 years and older covered by Medicare and those near and below the poverty line. Further, results are re- ported for patients across all of the conditions sampled in the MOS and not just for patients with hypertension and NIDDM 4 and mental disorders s,6 METHODS The MOS was an observational study of variations in practice styles and of outcomes for chronically ill adults treated in staff-model and independent practice HMOs vs FFS care in large multispe- cialty groups, small, single-specialty groups, and solo practices serving the same areas. Details of the MOS design, including site selection, sampling, clini- cian and patient recruitment, and data collection methods are documented elsewhere' - " To briefly recap the study design, MOS sites included Boston, Mass, Chicago, Ill, and Los Angeles, Calif, which represent 3 of the 4 US census regions. When sampling began in 1986 and 1987, these cities included well-developed HMO and FFS plans, including 2 of the country's largest HMOs employing salaried physicians and 2 of the largest independent prac- tice association (IPA) networks. In each city, 5 or 6 practice sites were sampled from each group practice HMO. The physician sample included 206 general internists, 87 family practitioners, 42 cardiologists, 27 endocrinologists, and 65 psychiatrists. In HMOs, patients treated by 8 nurse practitioners were also sampled. In addition, patients with a depressive disorder were sampled from the practices of 59 clinical psy- chologists and 9 social workers. Clini- cians averaged 39.6 years of age; 22% were female, and 29% were interna- tional medical graduates. 1040  JAMA, October 2, 1996-Vol 276, No. 13 Patient Sampling and Characteristics Patients followed up longitudinally were selected from 28 257 adults who visited an MOS site in 1986; 71.6% agreed to participate. In 18 794 (92.9%) of the visits, a standardized screening form was completed both by the MOS clinician and the patient. Using criteria docu- mented elsewhere,' clinicians identified patients with hypertension, NIDDM, myocardial infarction within the past 6 months, and congestive heart failure. Patients with depressive disorder were identified independently in a 2-stage screen, which included a patient-com- pleted form and a computer-assisted di- agnostic interview by telephone; 80% of those contacted completed this screen- ing process. Patients were selected for follow-up on the basis of diagnosis and participa- tion in baseline data collection, as docu- mented in detail elsewhere. 5,1 Inclusion of patients with more than 1 of the 5 conditions, with or without other comor- bidities, allowed for a more generaliz- able study. Of the 3589 eligible patients, 2708 (75.5%) completed a baseline as- sessment. We randomly selected 2235 of these for follow-up, by chronic con- dition and severity of their disease. A patient sample of this size was sufficient to detect clinically and socially relevant differences in health outcomes, defined as an average difference of 2 points or larger on a scale of 0 to 100,"I in a com- parison between HMO and FFS sys- tems. Specifically, the statistical power was greater than 80%, with a at the .05 level for a 2-tailed test. Patients ranged from 18 to 97 years of age, with a mean just under 58 years. At baseline, 36.8% were 65 years of age or older; all but 1 reported being covered by Medicare. (An additional 144 patients aged into this group during the 4-year follow-up.) A slight majority (54%) were female. About 22% were at or below 200% of the poverty line; 16% of those reported being covered by Medicaid. Three of 10 eligible for Medicare were also in the poverty group. Three of 4 had completed at least a 12th grade educa- tion; about 1 in 5 was nonwhite. Patients sampled had the following di- agnoses: hypertension (n=1318), NIDDM (n=441), congestive heart failure (n=215), recent acute myocardial infarction (n=104), and depressive disorder (n=444). (These numbers add to more than 2235 because some patients had more than one condition.) 1,9 As in previous MOS analyses,' FFS patients followed up in this study were significantly older (41.9 vs 32.9 years on average) than HMO pa- tients, were more likely to be female (62.8% vs 57.8%), and were more likely to be in the poverty group (25.4% vs 18.1%). The FFS patients followed were also more likely to have congestive heart failure (11.8% vs 7.3%) and to have had a recent myocardial infarction (8.9% vs 3.4%). As documented in detail else- where (MOS unpublished data; see ac- knowledgment footnote at the end of this article for availability of all MOS un- published data), 99% of patients fol- lowed in both FFS and HMO systems had 1 or more comorbid conditions; the most prevalent conditions were back pain/ sciatica (39% and 37% in FFS and HMO systems, respectively), musculoskel- etal complaints (24% and 22%), derma- titis (17% in each), and varicosities (15% and 14%). Longitudinal Data Collection After screening in the physician's of- fice and enrollment by telephone inter- view, each patient was sent a baseline health survey by mail." The baseline survey was completed, on average, 4 months after the patient's screening visit with an MOS clinician. Four-year follow- up data were obtained for 1574 of the 2235 patients (70.4% of the longitudinal cohort). Patients were lost to follow-up for a variety of reasons including refus- als and failure to contact (n=661; 29.6%); 137 (6.1%) who died during follow-up were included in the analysis. Analysis of initial health status for those lost to follow-up for reasons other than death revealed no differences and loss to follow- up was equally likely in HMO and FFS systems. However, younger and pov- erty-stricken patients were more likely to be lost from both HMO and FFS systems. All analyses of outcomes ad- justed for age, poverty status, and other variables to take into account this po- tential source of bias (see "Statistical Analysis"). Health Status Measures Summary physical and mental health scales constructed from the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) were analyzed (Table 1). These summary measures capture 82% of the reliable variance in the 8 SF-36 health scores estimated us- ing the internal-consistency reliability method The construction of sum- mary measures, score reliability and va- lidity, and normative and other inter- pretation guidelines are documented elsewhere." , " Changes in health were estimated in 2 ways. First, baseline scores were sub- tracted from 4-year follow-up scores, with deaths assigned a follow-up physi- cal health score of 0 (Table 1). Although these average change scores have the advantage of reflecting the magnitude Chronically III Elderly and Poor Patients-Ware et al of change in the metric of the scales, they mask the proportion of patients with follow-up scores that differed from those at baseline. Therefore, individual patients also were classified into 3 change categories: (1) those whose follow-up score did not change more than would be expected by chance ("same" group); (2) those who improved more than would be expected ("better" group); and (3) those whose score declined more than would be expected and those who died ("worse" group) (Table 1). This latter method has the advantage of combining health status and mortality without mak- ing any assumption about the "scale value" of death. Unlikely to be due to measurement error, changes large enough to be labeled better or worse also have been shown to be relevant in terms of a wide range of clinical and social criteria." Estimates of health outcomes for sur- vivors only were substantially biased be- cause deaths were more common among those with congestive heart failure, aged 65 years and older, and under FFS care; deaths were less likely for the clinically depressed group. Differences in survival rates between FFS and HMO systems were insignificant after adjustment for baseline patient characteristics. Thus, al- ternative methods of coding deaths" in estimating outcomes did not affect com- parisons between FFS and HMO sys- tems (MOS unpublished data). Statistical Analysis The goal of the analysis was to com- pare HMO and FFS systems of care in terms of average changes in health sta- tus and in terms of the percentages of patients who were better, the same, or worse at follow-up. These outcomes were estimated for all patients, and separately for subgroups differing in age, poverty status, and initial health. Multivariate statistical methods were used to adjust baseline scores so that the HMO and FFS groups would begin as equal as possible in terms of demographic and socioeconomic characteristics, study site, chronic conditions, disease severity, co- morbid conditions, initial health status, and other design variables (Table 2). Independent regression models were estimated for physical and mental health summary measures, and F tests of sig- nificance determined whether adjusted change scores differed, on average, across HMO and FFS systems. To make sure that the summary measures did not miss a difference concentrated in 1 of the 8 scales, all comparisons between FFS and HMO systems also were replicated for each of the 8 SF-36 scales. Because the summary measures captured all signifi- cant differences, results of their analyses JAMA, October 2, 1996 Vol 276, No. 13 Table 1 Definitions of Baseline and Outcome Health Measures Baseline Physical health 36-Item Short-Form Health Survey (SF-36) Physical Health Summary Scale, standardized to have a mean=50, SD=10 in the general US population. 13 I nternal-consistency reliability=0.91; test-retest reliability=0.89, which exceed the minimum standard suggested for group-level comparisons." Mental health SF-36 Mental Health Summary Scale, standardized to have mean=50, SD=10 in the general US population. 1 3 I nternal-consistency reliability=0.87; test-retest reliability=0.80, which exceed the minimum standard suggested for group-level comparisons." Mean changes Physical health Calculated for all patients as [(score at 4-year follow-up) -(baseline score)], prorated to adjust for unequal ti me intervals. Patients who died during the study were assigned a score of 0 at 4-year follow-up. 16 A score of 0 falls about 1 SD below the worst possible score, a score that was observed among MOS survivors. A score of 0 is also about 1 SD below the worst health state quantified in preliminary studies of an SF-36-based utility index, which combines health status and mortality. Sensitivity analyses with deaths scored 1 SD above and 1 SD below a score of 0 did not change conclusions about differences in health outcomes between fee-for-service and prepaid health maintenance (HMO) plans (MOS unpublished data). Mental health Calculated for surviving patients as [(score at 4-year follow-up) -(baseline score)], prorated to adjust for unequal time intervals. Categories of change Physical health Each patient was classified into 1 of 3 categories, according to the direction and magnitude of change between baseline and 4-year follow-up. Patients whose scores declined by more than 6.5 points were categorized as worse. Those who scores improved by more than 6.5 points were categorized as better. Those whose scores were within 6.5 points at baseline and follow-up were classified as same. Patients who died during the follow-up period were included in the worse group. As documented elsewhere, 3 a change greater than 6.5 is outside of the 95% confidence interval for an individual patient score, as estimated from the SD and score reliability. 13 Differences this large have been shown to be clinically and socially relevant. For example, average improvements in SF-36 Physical Health Summary scores this large or larger were observed following heart valve replacement surgery and total hip arthroplasty; such i mprovements are predictive of a one third decrease in probability of job loss, within the next year, among working patients." Patients who declined enough to be classified as worse in physical health at the end of 4 years were nearly 10 times more likely (0.9I vs 8.1%, P<.001) to die during the subsequent 3 years. Mental health Each surviving patient was classified into 1 of 3 categories according to the direction and magnitude of change between baseline and 4-year follow-up. Patients whose scores declined by more than 7.9 points were categorized as worse, those whose scores improved by more than 7.9 points were categorized as better, and those whose scores were within 7.9 points were classified as same. A change of this amount is outside the 95% confidence interval for an individual patient score. 13 An improvement in mental health nearly this large was observed for the average elderly depressed patient who responded to drug treatment i n comparison with nonresponders 25 are reported here. Results for the 8 SF- 36 scales are documented elsewhere (MOS unpublished data). Multinominal (polytomous) logistic re- gression" methods were used to com- pare categorical changes (better, same, worse) in physical and mental health across HMO and FFS systems for the total sample and for the subgroups. Ad- justed percentages for change catego- ries were generated with statistical ad- justments for the same baseline characteristics used in linear models (Table 2). The X 2 tests of significance were computed to determine whether the percentages across change catego- ries differed between HMO and FFS systems of care. Comparisons of outcomes across sys- tems reported here combine results for IPA "network" and staff-model HMOs. As in previous MOS analyses , 4 there were no significant differences in outcomes for those in WAS and staff-model HMOs in any of the analyses performed and there were no consistent trends suggesting a difference between IPAs and staff-model HMOs. However, because only 28% of prepaid patients were sampled from WAS, the MOS did not have enough statistical power to meaningfully compare outcomes across types of HMOs. To facilitate interpretation, regression models were used to estimate adjusted outcomes for the total sample and for each subgroup in comparing outcomes be- tween FFS and HMO systems. Formal statistical tests for interactions were per- formed to determine whether conclusions about differences between systems were the same across subgroups differing in age (Medicare), poverty status, Medicaid coverage, and initial health. To test for differences in outcomes for groups in bet- ter or worse initial health status, patients were stratified using baseline physical and mental health measures, both for lin- ear and logistic regression models. Thirds of the sample were identified based on whether they were functioning (physi- cally or mentally) higher, lower, or as would be expected at baseline, given their age and medical condition (Table 2). In keeping with the logic of an intention- to-treat analysis, patients were analyzed according to the system from which they were sampled. In support of this decision, the great majority of patients had been in their system 4 years or more at the time of sampling and most who switched did not do so for another 2 years. Thus, more than two thirds of those who switched systems during the 4-year follow-up had been in the type of system they were sampled from for 6 or more years before switching. However, because MOS pa- Chronically III Elderly and Poor Patients-Ware et al  104 1 Table 2 Covariates Used in the Estimation of Regression Adjusted Health Change Scores Main effects System of care Sampled from prepaid health maintenance organization (HMO) or fee-for-service care* Age Age -65 y or age <65 y, classified at baseline Sex Male or female Race White, black, or other minority Poverty status Above or below 200% of poverty, defined as per capita household income in 1986 dollars Medical Outcomes Study (MOS) tracer conditions Hypertension, myocardial infarction (MI), congestive heart failure, non-insulin-dependent diabetes mellitus, depressive disorder Comorbid medical conditionst Asthma, chronic obstructive pulmonary disease, angina (ever), angina (recent, no MI), MI past, other lung disease, back pain/sciatica, hip impairments, rheumatoid arthritis, osteoarthritis, musculoskeletal complaints, other rheumatic disease, colitis, diverticulitis, fistulas, gallbladder disease, irritable bowel disease, liver disease, type I diabetes mellitus, ulcer, kidney disease, benign prostatic hypertrophy, urinary tract infection, varicosities, cancer, dermatitis, anemia I nitial physical or mental health Tertiles of baseline health status estimated from multiple linear regression models that adjusted for age, MOS tracer conditions, and comorbid medical conditions. Initial tertiles labeled as "good," "average," and "ill" health were defined by thirds of the distribution of residuals from each regression model; these patients were, respectively, functioning better than expected, as expected, or worse than expected, given their age and medical condition MOS design variables Study site, cluster sampling of patients within physician offices, seasonality, weights for unequal probability caused by design choices and nonresponse Two-way interaction terms HMO and age ?65 y HMO and poverty status HMO and physical or mental health tertiles Age ?65 y and poverty Age -65 y and physical or mental health tertiles Poverty and physical or mental health tertiles Three-way interaction terms HMO and age -65 and physical or mental health tertiles HMO and poverty and physical or mental health tertiles *Thirty patients (1.9% of those followed) who reported no insurance coverage were included in the fee-for-service group. All were younger than 65 years. Analyses excluding the uninsured group did not change the conclusions from comparisons between systems reported here. tInformation regarding the comorbid medical conditions was obtained from the patient during a structured medical history interview conducted by a trained clinician. If information regarding a condition (or conditions) was missing, an independently derived probability of each diagnosis was substituted. Because of very low prevalence, the following conditions are incorporated into an index of 11 comorbid conditions: angina (ever), other rheumatic disease, colitis, diverticulitis, intestinal fistulas, gallbladder disease, liver disease, benign prostatic hypertrophy, varicosities, cancer, and type I diabetes mellitus. tients were more likely to switch from an HMO than from an FFS plan (20% vs 15%; P<.01), estimates of outcomes could have been biased. This potential source of bias was evaluated by comparing rates of switching within elderly and poverty subgroups along with average outcomes for those who did and did not switch. As documented elsewhere (MOS unpublished data), the relative probability of switch- ing from an HMO observed within the elderly and poverty subgroups was com- parable to that for the total sample. Fur- ther, baseline scores and average changes in physical and mental health did not dif- fer significantly for those who did and did not switch plans within either subgroup ( MOS unpublished data). Thus, conclu- sions about system differences in health outcomes are not likely to have been bi- ased by the intention-to-treat method of analysis used in this study. To evaluate whether differences in rates of loss to follow-up were a source of bias in comparisons of outcomes between systems, these rates were compared for the total sample and separately for the elderly and poverty subgroups. As docu- mented in detail elsewhere (MOS unpub- lished data), follow-up rates did not dif- 1042  JAMA, October 2, 1996-Vol 276, No. 13 fer between the 2 system cohorts for the total sample (71% vs 70% for FFS and HMO, respectively), among the elderly (both 74%), or for those in poverty (62% vs 60%). Baseline physical health scores for those followed up and lost to follow- up did not differ between FFS and HMO cohorts in analyses of the total sample or for elderly or poverty subgroups. To de- termine whether those lost and followed for health status outcomes had equal sur- vival probabilities, survival was moni- tored for all study participants for 7 years after baseline. Survival probabilities did not differ for those followed up and those lost to follow-up. As documented in de- tail elsewhere (MOS unpublished data), mental health scores for those lost to follow-up were significantly (P<.001) lower at baseline for both FFS and HMO cohorts. The same pattern was observed for elderly and poverty subgroups, with a significant difference favoring FFS over HMO for the poverty group (P<.05) ( MOS unpublished data). However, as documented in the tables cited in the "Results," adjusted physical and mental health scores for the follow-up samples analyzed here did not differ at baseline in comparisons between FFS and HMO co- horts within the total follow-up sample, the elderly subgroup, or the poverty sub- group. To test whether differences in patient outcomes between FFS and HMO sys- tems could be explained by the specialty of their regular physicians, these dif- ferences were also estimated with sta- tistical adjustment for physician special- ties. Estimates of outcomes for each system were equivalent with and with- out adjustment for specialty and are re- ported here without adjustment. To facilitate interpretation, all tables of results include 95% confidence inter- vals around average change scores and all differences associated with a chance probability of .05 or less were consid- ered statistically significant. Significance tests were not adjusted for multiple com- parisons. We hypothesized that the MOS sample would score below 50, the norm for the general population, on both measures at baseline, and they did. Because there are good arguments for hypothesizing better or worse outcomes across HMO and FFS systems over the 4-year follow- up period, we used 2-tailed tests of sig- nificance throughout. RESULTS Adjusted physical and mental health scores were virtually identical at base- line for patients sampled from HMO and FFS systems (Table 3). In relation to pub- lished norms for the US general popula- tion," MOS patients scored at the 24th and 35th percentiles for physical and men- tal health, respectively, indicating sub- stantially more physical impairment and emotional distress than experienced by the great majority of adults. During the 4-year follow-up, average changes in physical and mental health were indis- tinguishable between HMO and FFS sys- tems. Physical health scores declined about 3 points in both systems, lowering the average patient to the 19th percentile at follow-up. Mental health improved slightly in both systems, raising the av- erage to about the 38th percentile. The MOS had sufficient statistical power to detect differences in health outcomes as small as 1 to 2 points be- tween HMO and FFS systems of care. According to published interpretation guidelines for the SF-36 Health Sur- vey," differences of this amount or smaller are rarely clinically or socially relevant. Thus, there is a basis for con- fidence that an important average dif- ference in health outcomes between HMO and FFS systems was not missed. Analyses of change scores categorized as better, same, or worse confirmed these results for physical and mental health for the average patient. How- Chronically III Elderly and Poor Patients-Ware et al Table 3 Physical and Mental Health Outcomes for Patients Treated in Prepaid and Fee-for-Service Systems, Groups Differing in Age and Poverty Status *HMO indicates health maintenance organization. Scores are adjusted for demographics, chronic disease, and design factors. The 4-year change scores for physical health (but not mental health) include deaths scored at 0 at 4-year follow-up. tThe X2 statistics for categorical change refer to the results shown below and i ndicate whether the patterns of change are equal across the following pair of rows. $Significance tests for average scores indicate whether the mean score in 1 row differed from the mean score for the other row. §If the 95% confidence interval (CI) does not include 0, then average change scores are larger than expected by chance (P<.05). II P<.001. ~P=.01. Table 4 Physical and Mental Health Outcomes in Prepaid and Fee-for-Service Systems for Elderly and Nonelderly Patients ever, the categorical analyses called at- tention to substantial variation in out- comes. Physical health scores at follow- up differed (from those at baseline) for 45% of patients; about 30% declined and 15% improved, more than would be ex- pected due to measurement error. The reverse pattern-improvement more of- ten than decline-was observed for men- tal health scores (Table 3). Variations in Outcomes for Elderly and Poverty Groups The average adjusted physical decline was greater for elderly than nonelderly patients (0=-5.8 vs -1.9; P<.001); 36% and 26% of elderly and nonelderly pa- tients, respectively, scored worse at fol- low-up than at baseline (P<.001) (Table 3). Elderly patients scored higher in men- JAMA, October 2, 1996 Vol 276, No. 13 *Scores are adjusted for demographics, chronic disease, and design factors. The 4-year change scores for physical health (but not mental health) include deaths scored at 0 at 4-year follow-up. tThe X 2 statistics for categorical change refer to the results shown below and indicate whether the patterns of change are equal across the following pair of rows. $Significance tests for average scores indicate whether the mean score for the health maintenance organization (HMO) group differs from the mean score for the fee-for-service (FFS) group. §If the 95% confidence interval (CI) does not include 0, then average change scores are larger than expected by chance (P<.05). II P=.001. TP=.03. #P-05. **P<.001. tt P-01. tal health than nonelderly at baseline (P<.001); nonelderly patients improved significantly over time while the elderly did not. Both poverty and nonpoverty groups declined in physical health (0=-3.6 and -2.9, respectively), which are not sig- nificantly different amounts. Mental health improved significantly for non- poverty patients but did not improve for those in the poverty group. Differences in Outcomes by System: Elderly and Nonelderly Although adjusted baseline scores were equivalent for elderly and nonel- derly patients in comparisons between HMO and FFS systems (Table 4), changes in physical and mental health scores over time for the elderly in HMO and FFS plans were significantly dif- ferent from those for the nonelderly (F=2.1, P<.05, and X2=35.6, P<.001 for physical health; F=1.3, P>.05, and Xz=25.9, P<.01 for mental health) (Table 4). Physical health outcomes were, on average, more favorable for nonelderly patients in HMOs, while physical health outcomes were more favorable for el- derly patients in FFS. Although we could say with statistical confidence that the patterns of average change scores were different across HMO and FFS systems for elderly and nonel- derly patients, only pairwise comparisons between categories of changes were sig- nificant for the elderly (Table 4). The analysis of change categories also revealed that physical health was much less stable over time for elderly patients in HMOs Chronically III Elderly and Poor Patients-Ware et al  104 3 Physical Health* Mental Health* No. Average Scores Baseline* 4-y At 95% CI§ Categorical Change, */.t Worse Same Better Average Scores Baseline* 4-y At 95% CI§ Categorical Change, %t Worse Same Better Total sample 2235 45.0 -3.0 -3.8 to -2.2 29 56 15 48.5 1.1 0.3 to 1.9 15 63 22 Service system Prepaid (HMO) 1073 44.9 -3.1 -4.3 to -1.9 30 X 2 =1 .5 55 15 47.9 1.2 0.0 to 2.4 14 X 2 =1 .3_ 64 22 Fee-for-service 1162 45.2 -3.0 -4.2 to -1.8 27 57 15 49.0 1.0 -0.4 to 2.4 16 63 21 Age Elderly 822 43.5§ -5.8T -7.0 to -4.6 36 X 2 =14.1 11 53 11 50.3T 0.7 -0.5 to 1.9 15 X 2 4.3 65 20 Nonelderly 1413 45.7 -1.9 -2.9 to -0.9 26 58 17 47.7 1.3 0.3 to 2.3 15 63,. 22 Poverty status Poverty 489 44.4 -3.6 -5.2 to -2.0 33 X 2 =4.6 51 17 47.6 0.7 -1.1 to 2.5 17 ' X 2 =1:6 60 23 Nonpoverty 1746 45.2 -2.9 -3.7 to -2.1 27 58 15 48.8 1.2 0.4 to 2.0 15 64 21 Physical Health* Mental Health* Average Scores Categorical Change, %t Average Scores Categorical Change, %t No. Baseline (SE) 4-y A$ 95% CI§ Worse Same Better Baseline (SE) 4-y At 95% CI§ Worse Same Better Elderly 822 X 2 =19.211 X 2 =7.1$ Prepaid (HMO) 346 43.4 (0.7) -7.0 -8.8 to -5.2 54 37 9 50.1 (0.8) 1.3 -0.5 to 3.1 14 60 26 Fee-for-service 476 43.5 (0.7) -5.0 -6.6 to -3.4 28 63 9 50.6 (0.8) 0.2 -1.6 to 2.0 14 73 13 Nonelderly 1413 X 2 =2.3 X z =2.6 Prepaid (HMO) 727 45.8 (0.5) -1.2 -2.6 to 0.2 23 62 16 46.9 (0.6) 1.5 0.1 to 2.9 12 68 20 Fee-for-service. 686 45.6 (0.5) -2.4 -3.8 to -1.0 29 57 15 48.5 (0.5) 1.1 -0.7 to 2.9 16 64 19 Test for equivalence of differences in outcomes between prepaid and fee-for- service systems among elderly vs nonelderly subgroups 1`6,5,8=2.1# X 2 =35.6 ** F6,,3s3=1.3 X 2 =2~.;9tt, Table 5~Physical and Mental Health Outcomes in Prepaid and Fee-for-Service Systems for Poverty and Nonpoverty Groups *Scores are adjusted for demographics, chronic disease, and design factors. The 4-year change scores for physical health (but not mental health) include deaths scored at o at 4-year follow-up. HMO indicates health maintenance organization. tThe )( 3 statistics for categorical change refer to the results shown below and indicate whether the patterns of change are equal. *Significance tests for average scores indicate whether the mean score for the HMO group differs from the mean score for the fee-for-service group. §lt the 95% confidence i nterval (CI) does not i nclude 0, then average change scores are larger than expected by chance (P<05). II P=.01. 1P=.02. #P<.001. **P=.03. Table 6 Physical and Mental Health Outcomes in Prepaid and Fee-for-Service Systems for Initially III Patients in the Poverty Group Average Scorest  Categorical Change, %#  Average Scorest  Categorical Change, %3 No.  Baseline (SE)  4y A  95% CI§  Worse  Same  Better  Baseline (SE)  4y A  95% CI§  Worse  Same  Better X 3 =10.911  X 2 -4.1 Prepaid (HMO)  90  35.21(0.8)  -2.0#  -5.1 to 1.1  33  45  22**  37.1 (0.9)  4.5  -1.4 to 10.4  16  55  29 Fee-for-service  126  32.1 (1.0)  5.4  2.1 to 8.7  5  38  57  37.5 (0.8)  5.9  2.2 to 9.6  16  34  49 *Scores are adjusted for demographics, chronic disease, and design factors. The 4-year change scores for physical health (but not mental health) include deaths scored at 0 at 4-year follow-up. tSignificance tests for average scores indicate whether the mean score for the health maintenance organization (HMO) group differs from the mean score for the fee-for-service group. $The Xz statistics for categorical change refer to the results shown below and indicate whether the patterns of change are equal across the following pair of rows. §If the 95 1 /6 confidence i nterval (CI) does not include 0, then average change scores are larger than expected by chance (P<.05). II P=.006. 1P 014. #P<.001. **P=.04. compared to those in FFS (37% vs 63%, respectively, stayed the same; X 2 =19.2, P<001). The elderly treated in HMOs were nearly twice as likely to decline in physical health over time (54% vs 28%, P<001) (Table 4). The difference in physi- cal health outcomes favoring FFS over HMOs was statistically significant for el- derly patients regardless of their initial health (MOS unpublished data). Physical health outcomes favoring FFS over HMOs for the elderly were also apparent in all 3 study sites (MOS unpublished data). Average changes in mental health for elderly and nonelderly patients did not favor 1 system over the other (P>.05). However, analyses of mental health change categories for elderly patients favored HMOs over FFS; the elderly were twice as likely to improve in an HMO (26% vs 13% for FFS; X 2 =7.1, P<03). This result was due entirely to the better performance of HMOs in 1 study site. A formal test for a statistical interaction between plan and site re- vealed that mental health outcomes in 1044  JAMA, October 2, 1996-Vol 276, No. 13 Physical Health' AAeMaI 1t Nlth* HMOs differed significantly across the three sites (F=2.44, P<01). Differences in Outcomes of Poverty and Nonpoverty Groups by System As shown in Table 5, comparisons of physical and mental health outcomes across HMO and FFS systems produced different patterns of results for poverty and nonpoverty groups (F=2.7, P<.01, and X 2 = 24.2, P<.02 for physical health; F=4.2, P<.001, and X 2 =23.0, P<.03 for mental health). Only the pairwise com- parisons between HMO and FFS sys- tems for poor patients who were in ill health at baseline were significant (Table 6). Those in HMOs experienced an av- erage decline of -2.0 in physical health; those in FFS improved 5.4 points, on average (P<.001). Comparison of cat- egorical changes for poor patients in ini- tial ill health also favored FFS plans, with 57% scoring better at follow-up in FFS versus 22% in HMOs (X 2 =10.2, P<006). To determine whether Medicaid sta- tus accounted for differences observed in outcomes for the poor, HMO and FFS systems were compared among Medic- aid patients (n=216). Medicaid patients in HMOs did not differ from Medicaid patients in FFS plans in health status at baseline or in health outcomes, as docu- mented elsewhere (MOS unpublished data), and there were no noteworthy trends. However, because of the rela- tively small sample of Medicaid patients, the MOS did not have sufficient preci- sion to rule out an important difference among Medicaid patients favoring ei- ther system. COMMENT Limitations Limitations of the MOS have been discussed extensively,"" but some limi- tations and potential sources of bias war- rant special emphasis here. Analyses of 4-year health outcomes have been a long time coming because of the many meth- odological challenges faced by the MOS. Do results apply to current health care? If cost-containment pressures have in- Chronically.Ul_EldeJly.and Poor Patents-Ware et-al Physical Health* Mental Health* s Average Scores Categorical Change, %t Average Scores Categorical Change, %t No. Baseline (SE) 4y A4 95% CI§ Worse Same Better Baseline (SE) 4y A* 95% CI§ Worse Same Better Poverty 489 X 2 =4.1 X 2 =4.3 Prepaid (HMO) 295 43.3 (0.9) -4.0 -6.2 to -1.8 32 58 9 47.2(l.0) -0.4 -3.9 to 3.1 14 71 14 Fee-for-service 194 45.1 (0.8) -3.3 -5.7 to -0.9 36 46 18 47.9 (0.8) 1.3 -1.2 to 3.8 17 57 26 Nonpoverty 1746 X2 =2.34 Xz=2.59 Prepaid (HMO) 879 45.3 (0.5) -2.2 -3.6 to -0.8 24 62 13 47.9 (0.5) 1.4 0.2 to 2.6 11 70 18 Fee-for-service 867 45.1 (0.4) -3.4 -4.6 to -2.2 30 57 12 49.5 (0.5) 1.0 -0.8 to 2.8 16 66 18 Test for equivalence of differences in out- comes between pre- paid and fee-for- service systems among poverty vs non poverty subgroups Fa,1s1e=2.711 X 2 =24.21 Fa,3s3 - 4.2# X 2 =23.0** creased since MOS data collection ended in the early 1990s, high-risk patient groups may be at an even greater risk today. If information systems for moni- toring and improving the quality of care are better now and if health promotion and disease prevention initiatives are more successful in HMOs, MOS results may not apply to current health care. The MOS was not a randomized trial; such trials are rare in health care policy research.'a' 9 Although quasi-experimen- tal methods 2 ° achieved equivalent aver- age baseline health status scores for nearly all pairwise - comparisons between FFS and HMO systems of care, unmea- sured risk factors could have biased es- timates of differences in outcomes. Fur- ther, differences in outcomes that occurred "on the watch" of the FFS and HMO systems are not necessarily their responsibility. Structural and process differences in care beyond their control, such as arrangements for home health and long-term care, may account in part - for MOS findings. The MOS monitored outcomes in only 3 large urban cities; results should not be generalized to HMO or FFS plans in other cities or rural areas. Although the MOS represented 5 chronic conditions and many patients had comorbid condi- tions such as angina, back pain/sciatica, lung disease, and osteoarthritis, these patients do not necessarily represent other conditions or results of care pro- vided by other medical specialties. All patients had a regular source of care. All patients were being actively treated when the MOS began, and only three fourths who agreed to participate were followed up longitudinally. Two potential sources of bias in esti- mates of health outcomes-plan switch- ing and loss to follow-up-were system- atically studied. Patient loss to follow- up is an unlikely source of bias in comparisons of outcomes between sys- tems because adjusted physical health scores at baseline did not differ between FFS and HMO cohorts followed within the total sample or for elderly or pov- erty subgroups (Tables 3 through 5). Further, all study participants were fol- lowed up through 1993 to determine their survival." Seven years after baseline, those included and not included in this 4-year analysis were equally likely to have survived (MOS unpublished data). Two of 10 HMO patients switched to an FFS plan by the end of the 4-year follow-up. Comparisons between sys- tems could have been biased had these rates differed within elderly or poverty subgroups or had switchrs experienced different outcomes than nonswitchers. However, rates of switching did not dif- fer for elderly or poverty subgroups, JAMA, October 2, 1996-Vol 276, No. 13 and system differences in physical and mental health outcomes were indistin- guishable for those who stayed in the same system, in comparison with those who switched (MOS unpublished data). Thus, it is unlikely that conclusions about system differences in outcomes were bi- ased by switching. Because more than two thirds of patients who switched sys- tems during the follow-up period had been in their system at least 6 years before switching, we adhered to the logic of intent to treat and analyzed patients according to the systems from which they were sampled. The finding that MOS patients were significantly more likely to switch from an HMO than to an HMO (20% vs 15%; X 2 =7.3, P<.01) is surprising given that most MOS patients were aged 60 years or older, all were chronically ill, and financial incentives were beginning to favor HMOs over FFS during the MOS. The dynamics of switch- ing and their implications for monitor- ing current health outcomes warrant fur- ther study. Although the MOS achieved the de- sired statistical precision for overall HMO vs FFS comparisons, confidence intervals were too large for meaningful interpretation of some comparisons that yielded insignificant differences in out- comes. Examples include comparisons between IPAs, the fastest growing form of HMO, and staff-model HMOs; Med- icaid and non-Medicaid groups could not be compared with precision, and com- parisons between plans within sites were relatively imprecise, although the dif- ference in 1 site was large enough to reach significance. (This difference would not have been significant with an ad- justment for multiple comparisons.) For many comparisons, the MOS cannot rule out large differences in outcomes in ei- ther direction. Interpretation of Results The success of HMOs in reducing health care utilization has been docu- mented in numerous studies? , ' 9 With few exceptions, the best-designed and most recent studies show that HMOs achieve lower hospital admission rates, shorter hospital stays, rely on fewer subspecial- ists, and make less use of expensive tech- nologies. Results from FFS-HMO com- parisons of utilization rates in the MOS ,, " are consistent with previous studies, and extend that evidence to the population of adults with chronic conditions, for whom health outcomes are reported here. Rarely have the same studies ad- dressed health outcomes .2,18,21-23 Results from the MOS lead us to sev- eral conclusions about health outcomes for the chronically ill adults who were treated in HMO and FFS systems of care during the years of the MOS. First, similarities in health outcomes between systems previously reported' for the av- erage MOS patient with hypertension or NIDDM do not appear to hold for elderly patients covered by Medicare or for those in poverty. Elderly patients sampled from an HMO were more likely (than those sampled from an FFS plan) to have a poor physical health outcome in all 3 sites studied. Second, patients in the poverty group and particularly those most physically limited appear to be at a greater risk of a decline in health in an HMO than similar patients in an FFS plan. Finally, MOS results suggest the need for caution in generalizing conclu- sions about outcomes across study sites. Mental health outcomes for Medicare patients differed significantly across HMOs, suggesting that their perfor- mance relative to FFS plans may de- pend on site. Previous studies ; ' -21 that found no dif- ferences in health outcomes between FFS and HMO plans followed patients for only 1 year. Were these studies too brief to draw conclusions about health outcomes? Supporting this explanation, significant differences in health outcomes observed between the FFS and HMO systems after 4 years of follow-up in the MOS were not statistically significant after 1 year. The importance of a longer follow-up is underscored by the obser- vation that the 4-year statistical models reported here explained twice as much of the variance in patient outcomes as did the same models in analyses of 1- and 2-year outcomes (MOS unpub- lished data). Thus, follow-up periods longer than i year may be required to detect differences in outcomes for groups differing in chronic condition, age, in- come, and across different health care systems. Future Outcomes Studies Our results raise many questions that the MOS was not designed to address. What are the "clinical" correlates of changes in patient-assessed functional health and well-being? What can health care plans do to improve outcomes, and what specific treatments have been linked to physical and mental health out- comes as measured by the SF-36 Health Survey? Adverse medical events were too rare for meaningful comparison be- tween plans in the MOS and were moni- tored only during the first 2 years of follow-up' However, these events were significantly related to health outcomes, as hypothesized. Declines in SF-36 physi- cal health scores were significantly more likely among patients who experienced a new myocardial infarction, weight loss sufficient to warrant a physician visit, Chronically III Elderly and Poor Patients-Ware et al-  1045 and chest pain sufficient to require hos- pitalization (MOS unpublished data). These preliminary MOS results are con- sistent with published studies that have linked SF-36 health scores to disease severity and to treatment response, in- cluding severity of soft-tissue injuries" and changes in hematocrit among chronic dialysis patients 2 5 The SF-36 studies of outcomes have also linked treatment to outcomes including drug treatment for depression among the elderly ,26 total knee replacement 2',21 heart valve re- placement surgery ,21 use of aerosol in- halers in treating asthma, 3 ° intermit- tent vs maintenance drug therapy for duodenal ulcer," elective hip arthro- plasty, 32 elective coronary revascular- ization," and various other elective sur- gical procedures 34 Three dozen such studies using the SF-36 are cited else- where .15 Identification of the clinical correlates of changes in physical and mental health status warrants high pri- ority in outcomes and effectiveness re- search." Future studies should address whether variations in the quality of care explain differences in outcomes across systems. The MOS patients in HMOs reported fewer financial barriers and better coordination of services in com- parisons with equivalent FFS pa- tients. 12,3 s Analyses of primary care qual- ity criteria indicated that those in FFS systems experienced shorter treatment queues and better comprehensiveness and continuity of care and rated the qual- ity of their care more favorably. 12,3' D o such variations in process account for differences in outcomes? Practice-level analyses in progress have linked scores for primary care process indicators 12 to 4-year health outcomes, as defined here, supporting this hypothesis. These and other associations warrant further study to determine which practice styles and specific treatments are most likely to i mprove health outcomes. Because many of the structural and process indicators being relied on to evaluate the quality of current health care have not been shown to predict outcomes, targeted monitor- ing efforts are required to discern health outcomes. The MOS has demonstrated the fea- sibility and usefulness of readily avail- able patient-based assessment tools, such as the SF-36 Health Survey, in monitoring outcomes across diverse pa- tient populations and practice settings. The SF-36 summary measures of physi- cal and mental health reduce the num- ber of comparisons necessary to moni- tor outcomes while retaining the option of analyzing the 8-scale SF-36 health profile on which they are based. The reporting of results in change catego- 1046  JAMA, October 2, 1996-Vol 276, No. 13 ries in terms of better, same, and worse may simplify the reporting of outcomes to diverse audiences and may make re- sults easier for them to understand. More practical data collection and processing systems-under development-and ad- vances in understanding of the specific treatments that improve health scores the most and the clinical and social rel- evance of those improvements will in- crease their usefulness in improving pa- tient outcomes." Policy Implications The MOS results reported here and previously' for the average chronically ill patient constitute good news for those who consider HMOs as a solution to ris- ing health care costs. Outcomes were equivalent for the average patient be- cause those who were younger, rela- tively healthy, and relatively well-off financially did at least as well in HMOs as in the FFS plans. However, our re- sults sound a cautionary note to policy- makers who expect overall experience to date with HMOs to generalize to spe- cific subgroups, such as Medicare ben- eficiaries or the poor. Patients who were elderly and poor were more than twice as likely to decline in health in an HMO than in an FFS plan (68% declined in physical health in an HMO vs 27% for FFS; P<.001) (MOS unpublished data). An implication for future evaluations of changes in health care policies is that high-risk groups, including the elderly and poor who are chronically ill, should be oversampled when outcomes are monitored to achieve the statistical pre- cision necessary to rule out harmful health effects. Medicaid coverage did not explain the differences in physical or mental health outcomes observed for the poor in MOS comparisons between FFS and HMO systems. Only 1 in 5 poor were covered under Medicaid. Further, when out- comes for MOS patients covered and not covered under Medicaid were com- pared, there were no significant differ- ences between FFS and HMO plans and there were no noteworthy trends (MOS unpublished data). Poverty status, as opposed to Medicaid beneficiary status, was the better marker of risk of a poor health outcome in an HMO. This is not a new finding. The Health Insurance Experiment also observed that some health outcomes were less favorable over a 5-year follow-up for low-income pa- tients in poor health in 1 HMO com- pared with equivalent patients under FFS care." Final Comment In this article, the MOS has docu- mented variations in health outcomes for chronically ill patients that cannot be explained in terms of measurement error. For elderly Medicare patients and for poor patients, variations in outcomes during a 4-year period extending through 1990 were linked to FFS and HMO systems of care (the latter were predominantly staff-model HMOs). Other explanatory factors included prac- tice site, suggesting that health out- comes should be monitored on an ongo- ing basis, by particular HMO and by marketplace. Outcomes did not differ across systems for those covered under Medicaid and could not be explained in terms of the specialty training of phy- sicians. The contrast between results reported here for high-risk patients vs results reported previously for the average patient' underscore the hazard in generalizing about outcomes on the basis of averages. This is why quality i mprovement initiatives focus on var- iations rather than only on usual per- formance." Patient-based assessments of outcomes are likely to add signifi- cantly to the evidence used in informing the public and policymakers regarding which health care plans perform best- not just in terms of price, but in overall quality and effectiveness. Indications in the text of "MOS unpublished data" refer to 16 pages of additional documents that are available at http://www.sf-36.com on the Internet. These data are also available from the National Auxiliary Publications Service, document 05340. Order from NAPS, c/o Microfiche Publications, PO Box 3513, Grand Central Station, New York, NY 10163-3513. Remit in advance, in US funds only, $7.75 for photocopies or $5 for microfiche. Outside the United States and Canada, add postage of $4.50. The postage charge for any microfiche order is $1.50. Collection of 4-year health outcome data and preparation of this article were supported by grant 91-013 from the Functional Outcomes Program of the Henry J. Kaiser Family Foundation, at The Health Institute, New England Medical Center, Boston, Mass (John E. Ware, Jr, PhD, principal in- vestigator). Design and implementation of the MOS were sponsored by the Robert Wood Johnson Foundation, Princeton, NJ; the Henry J. Kaiser Family Foundation, Menlo Park, Calif; and the Pew Charitable Trusts, Philadelphia, Pa. Previously re- ported analyses were sponsored by the National Institute on Aging, Bethesda, Md; the Agency for Health Care Policy and Research; and the National Institute of Mental Health, Rockville, Md. Partici- pating plans, professional organizations who as- sisted in recruitment, and our many colleagues who contributed to the success of the MOS are acknowl- edged elsewhere.` The authors acknowledge the thorough and constructive suggestions received from Allyson Ross Davies, PhD, Kathleen Lohr, PhD, Edward Perrin, PhD, Dana Safran, SeD, and anonymous JAMA peer reviewers; and gratefully acknowledge the editing and typing assistance of Orna Feldman, Sharon Ployer, Rebecca Voris, and Andrea Molina. References 1. Group Health Association of America. Patterns in HMO Enrollment. Washington, DC: Group Health Association of America; June 1995. 2. Miller RH, Luft HS. Managed care plan perfor- mance since 1980: a literature analysis. JAMA.1994; 271:1512-1519. Chronically ,lf - Elderly and Poor Patients-Ware et al 3. Tarlov AR, Ware JE, Greenfield S, Nelson EC, Perrin E, Zubkoff M. The Medical Outcomes Study: an application of methods for monitoring the re- sults of medical care. JAMA. 1989;262:925-930. 4. Greenfield S, Rogers W, Mangotich M, Carney MF, Tarlov AR. Outcomes of patients with hyper- tension and non-insulin-dependent diabetes melli- tus treated by different systems and specialties: results from the Medical Outcomes Study. JAMA. 1995;274:1436-1474. 5. Wells KB, Hays RD, Burnam MA, Rogers W, Greenfield S, Ware JE. Detection of depressive disorder for patients receiving prepaid or fee-for- service care: results from the Medical Outcomes Study. JAMA. 1989;262:3298-3302. 6. Rogers WH, Wells KB, Meredith LS, Sturm R, Burnam A. Outcomes for adult outpatients with depression under prepaid or fee-for-service financ- ing. Arch Gen Psychiatry. 1993;50:517-525. 7. Stewart AL, Ware JE, eds. Measuring Func- tioning and Well-being: The Medical Outcomes Study Approach. Durham, NC: Duke University Press; 1992. 8. Kravitz RL, Greenfield S, Rogers WH, et al. Dif- ferences in the mix of patients among medical spe- cialties and systems of care: results from the Medical Outcomes Study. JAMA. 1992;267:1617-1623. 9. Stewart AL, Greenfield S, Hays RD, et al. Func- tional status and well-being of patients with chronic conditions: results from the Medical Outcomes Study. JAMA. 1989;262:907-913. 10. Berry S. Methods of collecting health data. In: Stewart AL, Ware JE, eds. Measuring Function- ing and Well-being: The Medical Outcomes Study Approach. Durham, NC: Duke University Press; 1992:48-64. 11. Greenfield S, Nelson E C, Zubkoff M, et al. Varia- tions in resource utilization among medical special- ties and systems of care: results from the Medical Outcomes Study. JAMA. 1992;267:1624-1630. 12. Safran D, Tarlov AR, Rogers W. Primary care performances in fee-for-service and prepaid health care systems: results from the Medical Outcomes Study. JAMA. 1994;271:1579-1586. 13. Ware JE, Kosinski M, Keller SK. SF-35 Physi- cal and Mental Health Summary Scales: A User's Manual. Boston, Mass: The Health Institute, New England Medical Center; 1994. 14. Ware JE, Kosinski M, Bayliss MS, McHorney CA, Rogers WH, Raczek A. Comparison of meth- ods for scoring and statistical analysis of SF-36 Health Profiles and Summary Measures: summary JAMA, October 2, 1996-Vol 276, No. 13 of results from the Medical Outcomes Study. Med Care. 1995;33(suppl 4):AS264-AS279. 15. McHorney CA, Ware JE,RaczekAE.TheMOS 36-Item Short-Form Health Survey (SF-36), 11: psychometric and clinical tests of validity in mea- suring physical and mental health constructs. Med Care. 1993;31:247-263. 16. Diehr P, Patrick D, Hedrick S, et al. Including deaths when measuring health status over time. Med Care. 1994;32(suppl 4):AS164-AS172. 17. STATA Reference Manual: Release 3.1, Vol- ume 3. 6th ed. College Station, Tex: STATA Corp; 1993: 3-16. 18. Ware JE, Brook RH, Rogers WH, et al. Com- parison of health outcomes at a health maintenance organization with those of fee-for-service care. Lan- cet. 1986;1:1017-1022. 19. Manning WG, Leibowitz A, Goldberg GA, Rog- ers WH, Newhouse JP. A controlled trial of the effect of a prepaid group practice on use of services. N Engl J Med. 1984;310:1505-1510. 20. Cook TD, Campbell DT. The design and con- duct of quasi-experiments and true experiments in field settings. In: Dunnette MD, ed. Handbook of Industrial and Organizational Psychology. Chi- cago, Ill: Rand McNally College Publishing Co; 1976: 223-326. 21. Lurie N, Moscovice IS, Finch M, Christianson JB, Popkin MK. Does capitation affect the health of the chronically mentally ill? results from a random- ized trial. JAMA. 1992;267:3300-3304. 22. Retchin SM, Clement DG, Rossiter LF, Brown B, Brown R, Nelson L. How the elderly fare in HMOs: outcomes from the Medicare competition demonstrations. Health Sere Res. 1992;27:651-669. 23. Clement DG, Retchin SM, Brown RS, Stegall MH. Access and outcomes of elderly patients en- rolled in managed care. JAMA.1994;271:1487-1492. 24. Beaton DE, Bombardier C, HoggJohnson S. Choose your tool: a comparison of the psychometric properties of five generic health status instruments in workers with soft tissue injuries. Qiual Life Res. 1994;3:50-56. 25. Beusterien KM, Nissenson AR, Port FK, Kelly M, Steinwald B, Ware JE. The effects of recombi- nant human erythropoietin on functional health and well-being in chronic dialysis patients. J Am See Nephrol. 1996;7:1-11. 26. Beusterien K, Steinwald B, Ware JE. Useful- ness of the SF-36 health survey in measuring health outcomes in the depressed elderly. J Geriatr Psy- chiatry Neural. 1996;9:1-9. Printed and Published in the United States of America 27. Kantz ME, Harris WJ, Levitsky K, Ware JE, Davies AR. Methods for assessing condition-spe- cific and generic functional status outcomes after total knee replacement. Med Care. 1992;30(suppl 5):MS240-MS252. 28. Hawker G, Melfi C, Paul J, Green R, Bombar- dier C. Comparison of a generic (SF-36) and a dis- ease-specific (WOMAC) instrument in the measure- ment of outcomes after knee replacement surgery. J Rheumatol. 1995;22:1193-1196. 29. Phillips RC, Lansky DJ. Outcomes manage- ment in heart valve replacement surgery: early experience. J Heart Valve Dis. 1992;1:42-50. 30. Okamoto LJ, Noonan M, Kirchdoerfer LJ, Bayer JG, Kellerman DJ, Saiers JA. Quality of life in patients with severe asthma: baseline health pro- file and effects of fluticasone propionate aerosol. Ann Allergy Asthma Immunol. 1996;76:1-7. 31. Rampal P, Martin C, Marquis P, Ware JE, Bon- fils S. A quality of life study in five hundred and eighty-one duodenal ulcer patients. Scand J Gas- troenterol. 1994;29(suppl):44-51. 32. Stucki G, Liang MH, Phillips C, Katz JN. The Short Form-36 is preferable to the SIP as a generic health status measure in patients undergoing elec- tive total hip arthroplasty. Arthritis Care Res. 1995; 8:174-181. 33. Krumholz HM, McHorney CA, Clark L, Levesque M, Baim DS, Goldman L. Changes in health after elective percutaneous coronary revas- cularization: a comparison of generic and specific measures. Med Care. 1996;34:754-759. 34. Temple PC, Travis B, Sachs L, Strasser S, Cho- ban P, Flancbaum L. Functioning and well-being of patients before and after elective surgical proce- dures. J Am Coll Surg. 1995;181:17-25. 35. ShielyJ-C, Bayliss MS, Keller SD, Tsai C, Ware JE. SF-36 Health Survey Annotated Bibliography: First Edition, 1988-1995. Boston, Mass: The Health Institute, New England Medical Center. In press. 36. Roper WL, Winkenwender W, Hackbarth GM, Krakauer H. Effectiveness in health care: an ini- tiative to evaluate and improve medical practice. N Engl J Med. 1988;319:1197-1202. 37. Rubin H, Gandek B, Rogers WH, Kosinski Y, McHorney C, Ware JE. Patient's ratings of outpa- tient visits in different practice settings: results from the Medical Outcomes Study. JAMA. 1993; 207:835-840. 38. Davies AR, Halpern R. Health Care Outcomes: An Introduction. Irving, Tex: Voluntary Hospitals of America Inc; 1993. Chronically III Elderly and Poor Patients-Ware et al  1047 . Association Original Contributions Differences in 4-Year Health Outcomes for Elderly and Poor, Chronically III Patients Treated in HMO and Fee -for- Service Systems Results. physical and mental health outcomes of chronically ill adults, including elderly and poor subgroups, treated in health maintenance orga- nization (HMO) and fee -for- service