The Implications of Regional Variations in Medicare Spending. Part 1: The Content, Quality, and Accessibility of Care FREE

The Editors

Health care spending in the United States is expected to increase dramatically in this decade. By 2011, per capita spending is forecast to increase by 49% in real terms, reaching $9216 per capita or 17% of the gross domestic product (1). The likely consequences of such dramatic growth in health care costs include further increases in the numbers of uninsured persons and reduced public and private spending in other sectors of the economy. Spending growth, however, is seen as an inexorable consequence of the aging of the population and advancing technology (2 - 3). Moreover, the effectiveness of specific interventions in cardiovascular disease, neonatal care, and cancer treatment has been used to argue that the overall gains from increased spending are worth the costs (3) and that any constraints on the expansion of the specialist workforce or on further spending growth could be harmful (2 - 4).

These forecasts and the policy prescriptions that depend on them do not take into account the dramatic regional variations in spending and medical practice observed across the United States (5 - 8). For example, age-, sex-, and race-adjusted spending for traditional (fee-for-service) Medicare in 1996 was $8414 per enrollee in the Miami, Florida, region compared with $3341 in the Minneapolis, Minnesota, region (9). The greater-than-twofold differences observed across U.S. regions are not due to differences in the prices of medical services (7,10) or to apparent differences in average levels of illness or socioeconomic status (10 - 12). Rather, they are due to the overall quantity of medical services provided and the relative predominance of internists and medical subspecialists in high-cost regions (2,13).

The implications for health and health care of these regional differences in resources and spending, although directly relevant to current policy debates, remain relatively unexplored (14). The financial implications are clear: Savings of up to 30% of Medicare spending might be possible, and the Medicare Trust Fund would remain solvent into the indefinite future (10). However, remarkably little is known about whether the increased spending in high-cost regions results in better care or improved health. Although recent studies have found no improvement in mortality (12,15 - 16), they have been criticized because of weak designs (most were cross-sectional and ecologic), inadequate individual-level measures to control for potential differences in case mix, insufficient clinical detail on the process of care to allow inferences on potential causal pathways to be drawn, and limited outcome measures. We designed a research project to address these concerns.

We present our findings in two articles. This article, Part 1, provides an overview of the study design and addresses the question, What are the differences in the content, quality, and accessibility of care across U.S. regions that differ in per capita Medicare spending? The second article, Part 2, asks, Do regions with higher Medicare spending achieve better health outcomes and improved patient satisfaction?

One approach to determining whether the increased spending in some U.S. regions leads to better care and better outcomes would be to conduct a randomized trial. This would ensure that assignment to the treatment and control groups (those receiving more and less spending) was independent of patient characteristics. Logistic barriers to such a trial, however, would be substantial. We therefore conducted a cohort study in four parallel cohorts using a natural randomization approach (17), in which one or more exposure variables allowed assignment of patients into treatment groups (different levels of average spending), as would a randomized trial. An overview of the design is provided in Figure 1.

Because some of the regional differences in Medicare spending are due to differences in illness levels (enrollees in Louisiana are sicker than those in Colorado) and price (Medicare pays more for the same service in New York than in Iowa), we could not use Medicare spending itself as the exposure. We therefore assigned U.S. hospital referral regions (HRRs), and thus the cohort members residing within them, to different exposure levels. We did this by using the End-of-Life Expenditure Index (EOL-EI), a measure reflecting the component of regional variation in Medicare spending that is due to physician practice rather than regional differences in illness or price. Because regional differences in end-of-life spending are unrelated to underlying illness levels, it is reasonable to consider residence in HRRs with different end-of-life spending as a random event. The index was calculated as standardized spending on hospital and physician services provided to a reference cohort distinct from the study cohorts: Medicare enrollees in their last 6 months of life.

We confirmed that the exposure used to assign the HRRs achieved the goals of natural randomization: 1) Study samples assigned to different levels of the exposure [the EOL-EI] were similar in baseline health status, and 2) the actual quantity of services delivered to the individuals within the study samples nevertheless differed substantially across exposure levels and was highly correlated with average per capita Medicare spending in the HRRs. We then followed the cohort members for up to 5 years after study enrollment and compared the processes of care (Part 1) and health outcomes (Part 2) across HRRs assigned to different exposure levels.

We sought study samples that would be similarly ill across regions based on the occurrence of an incident illness (acute myocardial infarction [MI], hip fracture, colorectal cancer) or in which we had excellent data for case-mix adjustment (acute MI, Medicare Current Beneficiary Survey [MCBS] sample). We restricted the eligible population to Medicare enrollees between the ages of 65 and 99 years who, at the time of study enrollment, were eligible for both Medicare Parts A and B and were not enrolled in a Medicare health maintenance organization (HMO).

The acute MI cohort was drawn from patients included in the Cooperative Cardiovascular Project, which identified from billing records a national sample of Medicare beneficiaries who were discharged after acute MI between February 1994 and November 1995 (18). We excluded patients with an unconfirmed acute MI (with the same criteria as in previous studies (19)) and included only the first episode of acute MI for a given patient. The hip fracture and colorectal cancer cohorts were selected based on a first admission between 1993 and 1995 for a primary diagnosis of hip fracture or colorectal cancer with resection, using the same International Classification of Diseases, Ninth Revision, Clinical Modification codes as in earlier work (20). Hospitalization rates for acute MI, hip fracture, and colorectal cancer vary little across regions (21), and patients with incident cases of these conditions are likely to be similarly ill in different communities (20). We excluded patients with a previous hospitalization for the same diagnosis in the year before their index stay. The general population cohort was drawn from the access-to- care component of the MCBS, a continuous panel survey that is representative of the Medicare population (22). Our inclusion criteria are detailed in the Appendix.

Trained abstractors working in the Cooperative Cardiovascular Project obtained characteristics of patients in the acute MI cohort from the medical record (18). Quality of the chart review process was monitored by random reabstractions, and percentage agreement was generally very high (93.3% to 94.8%) (23). Missing data for clinical variables were handled by including a specific categorical variable for patients with each missing variable (for example, admission blood pressure missing). Income was defined based on ZIP code of residence by using 1990 U.S. Census data.

For the hip fracture and colorectal cancer cohorts, we coded the presence of specific comorbid conditions based on diagnoses recorded on the discharge abstract, as was done in previous work (24 - 25). Cancer stage was classified as distant versus local or regional because this classification has been found to correspond most closely to reported stage, according to analyses of linked MedicareSurveillance, Epidemiology, and End Results data (26). Data from the 1990 U.S. Census, measured at the level of the ZIP code, were used to provide measures of income, education, disability status, urban or rural residence, employment, marital status, and Hispanic origin. For all three chronic disease cohorts, we used American Hospital Association data to characterize the hospital teaching status (27) and the Medicare claims files of patients' index hospitals to determine the volume of cases of hip fracture, colorectal cancer, and acute MI treated per year.

Data collection and preparation procedures for the MCBS are described elsewhere (22). Because not all respondents completed all survey items, analyses of utilization, access to care, satisfaction, and survival are based on slightly different numbers of respondents (Appendix, Section C). Among the patient attributes, responses were most likely to be missing for income (5%); patients for whom income data were missing were recoded to the lowest-income group after analyses showed them to be similar in other attributes and survival to other patients in that group. The MCBS includes responses from proxies, which represented a maximum of 8.0% of our initial cohort. There were no differences in the proportion of proxy responses across the quintiles of spending, and excluding proxy responses did not alter our findings.

We used information from Medicare enrollment files to determine the percentage of all Medicare enrollees in each HRR who were enrolled in HMOs, to assign patients to one of nine major regions of the country, to determine whether patients had moved within an HRR or across HRRs within 1 or 2 years before their index admission, and to determine when patients should be censored based on loss of Medicare fee-for-service coverage (for utilization analyses) or based on relocation from their original HRR.

We used two approaches to determine cohort members' exposure to different levels of Medicare spending in their regions of residence. Previous research has shown that the dramatic differences in end-of-life treatment across U.S. regions are highly predictive of differences in total spending (13,28) but are not due to differences in case mix or patient preferences (29). Our primary measure of exposure was the EOL-EI, which was calculated as age-sex-raceadjusted spending (measured with standardized national prices) on hospital and physician services provided to Medicare enrollees in their last 6 months of life in each of the 306 U.S. HRRs from mid-1994 to 1997, excluding any members of the study cohorts (Appendix, Sections D and E).

Although Medicare enrollees identified 6 months before death are identical in terms of their risk for death (which is 100%), they may differ in other ways across HRRs (for example, in illness levels unrelated to risk for death). We therefore repeated the major analyses with an alternative exposure measurean Acute Care Expenditure Index (AC-EI)which was based on differences across HRRs in risk-adjusted spending during an acute care episode, calculated as follows. For each of the four study cohorts, we determined age-, sex-, race, and illness-adjusted spending on physician and hospital services (measured with standardized national prices) provided during the first 6 months after index hospitalization across the 306 HRRs. The AC-EI subsequently used to assign each cohort to different exposure levels was the average of the age-sex-race-illnessadjusted spending during the acute episode of care in the other cohorts (Appendix, Section F).

The EOL-EI measures regional differences in practice at the end of life, while the AC-EI measures regional differences in practice during acute illness (or exacerbation of a chronic illness). Both measures were highly predictive of average age-sex-raceadjusted Medicare spending at the HRR level (r = 0.81 for the EOL-EI and 0.79 for the AC-EI in the acute MI cohort) and of regional differences in utilization. Both exposure measures produced similar results, so we present our findings on utilization based on the EOL-EI. (A sensitivity analysis of our mortality analyses using the AC-EI is presented in Part 2.) For many analyses, we grouped HRRs into quintiles of increasing exposure to Medicare spending based on the EOL-EI.

We used 100% Part A and B Medicare claims for all four cohorts to determine rates of specific physician and hospital services within the first year of follow-up and a 5% sample to examine aggregate physician utilization over the full 5 years of follow-up. The major categories of physician services provided to the Medicare population (MCBS) were defined by using Medicare's BerensonEggers Type of Service classification (cms.hhs.gov/data/betos/default.asp). The acute MI chart review allowed us to describe the proportion of patients in defined clinical subgroups who received specific medications or interventions during the initial hospitalization. Analyses of the quality of care were restricted to patients who were ideal candidates for each therapy (that is, patients with any absolute or potential contraindication to the treatment were excluded), as in the original description of the Cardiovascular Cooperative Project (18). The in-person interviews for the MCBS included specific questions about the types of visits received during the past year, waiting times at these visits, receipt of specific preventive services, and access to care.

All analyses used the patient as the unit of analysis and measured other attributes at progressively higher levels of aggregation where appropriate (ZIP code of residence, hospital of index admission, and HRR of residence). To test whether patients' baseline characteristics differed across HRRs with differing EOL-EI levels, we used logistic regression at the individual level with the attribute (for example, age 65 to 74 years) as a dichotomous dependent variable and the HRR-level EOL-EI as the independent variable. To assess the aggregate impact of any differences in individual attributes on average baseline risk for death across regions of increasing EOL-EI, we used logistic regression to determine each individual's predicted 1-year risk for death as a function of his or her baseline characteristics. The models had modest to excellent predictive ability (c-statistics were 0.61 for the colorectal cancer cohort, 0.68 for the hip fracture cohort, 0.77 for the acute MI cohort, and 0.82 for the MCBS cohort). We used these models to determine the average predicted risk for death across quintiles of EOL-EI. We then determined average predicted 1-year risk for death in each HRR and tested for an association across HRRs using logistic regression.

For the analyses of aggregate utilization, cohort members were censored (no longer followed) if they lost Medicare insurance (either Part A or B eligibility), enrolled in an HMO, or moved out of the original HRR of residence. The average amount of hospital and physician services provided to each cohort member across quintiles of the EOL-EI was calculated by using standardized national prices (Appendix, Section D) and was calculated separately for the acute episode (index admission to 6 months for the three chronic disease cohorts) and for 6 months to 5 years (chronic disease cohorts) or for 5 years (MCBS cohort). To control for baseline differences, we used linear regression, weighting by follow-up time, with the log-transformed utilization of hospital and physician services per person-year as the dependent variable for the individual and the quintile of end-of-life spending as the primary independent variable. Coefficients and confidence intervals were back-transformed to provide estimates of the adjusted relative rates of utilization in the highest compared with the lowest quintile for each cohort.

We analyzed only the first year of follow-up (for which we had 100% physician claims) to compare rates of specific physician services across quintiles. We used Poisson regression to calculate adjusted relative rates of specific services (dependent variable) in quintile 5 compared with quintile 1, controlling for baseline differences (independent variables) (30). We then computed pooled relative rates across the cohorts, using a weighted average of the individual regression coefficients for each cohort and weighting by the inverse of the variance (31).

The specific independent and dependent variables included in each regression are described in Appendix Table 5. We used the REG routine of Stata 6.0 (Stata Corp., College Station, Texas) to perform all regression analyses. For the analyses of the MCBS cohort, we used SUDAAN (Research Triangle Institute, Research Triangle Park, North Carolina) to account for sampling weights and the two-stage design.

The 306 U.S. HRRs were assigned to quintiles of Medicare spending based on the primary exposure measure, the EOL-EI, which averaged $14 644 in quintile 5 (the highest-spending quintile) and $9074 in quintile 1 (the lowest-spending quintile) (Figure 2). Average age-sex-raceadjusted per capita Medicare spending was highly correlated with EOL-EI (r = 0.81); per capita Medicare spending was $6304 in quintile 5 and $3922 in quintile 1. Residents of the highest quintile received 61% more Medicare resources than those in the lowest quintile whether measured by the EOL-EI or by average Medicare spending. Quintiles with a higher expenditure index had more hospital beds and physicians, a relative predominance of large hospitals and teaching hospitals, and a higher proportion of urban residents. The percentage of Medicare beneficiaries enrolled in HMOs was higher in both the lowest and highest quintiles than in the middle quintiles.

Tables 1, 2, 3, and 4 present selected characteristics of each study cohort in each quintile and a test for trend across HRRs that had a higher EOL-EI. Because of the large sample sizes, many differences in the chronic disease cohorts were statistically significant. Notable differences were found in racial composition (more black patients in higher quintiles) and income (higher quintiles had more enrollees in the highest and lowest income categories). Smaller differences across quintiles were apparent in age, sex, comorbid conditions, and cancer stage. For the acute MI cohort, patients in the highest quintile had a higher prevalence of nonQ-wave infarctions and congestive heart failure but were less likely to have creatine kinase levels over 1000 IU/L. For the MCBS cohort, residents of HRRs in the quintiles with a higher EOL-EI were more likely to report being in fair or poor health but were less likely to live in a facility. Few differences were found, however, in activities of daily living or instrumental activities of daily living, smoking, or reported chronic conditions.

Tables 1, 2, 3, and 4 also present the average predicted risk for death in each quintile and a formal test of trend assessing whether HRRs with a higher EOL-EI had higher predicted mortality at baseline. A higher expenditure index was associated with an increased predicted risk for death for the AMI cohort, a lower predicted risk for death for the hip fracture cohort, and no significant difference in predicted mortality across spending levels for the colorectal cancer or MCBS cohorts. These data indicate that the burden of illness in these cohorts is similar across HRRs that differ by more than 60% in both EOL-EI and, as grouped according to the EOL-EI, average per capita Medicare spending.

Figure 3 displays the aggregate utilization of hospital and physician services by the study cohorts across quintiles. The data presented exclude the initial acute episode of care for the acute MI, hip fracture, and colorectal cancer cohorts because utilization rates were, as expected, similar during this period. Differences across the cohorts are apparent: Within each quintile, the MCBS cohort received the least care while the acute MI cohort received the most care. Within each cohort, however, patients who resided in regions with a higher EOL-EI received more medical care. After adjustment for baseline differences in health status, overall use of hospital and physician services was between 52% higher (MCBS cohort) and 77% higher (acute MI cohort) in the highest compared with the lowest quintile. For each of the cohorts, approximately half of the spending within a quintile was for acute inpatient hospital care and half was for physician services (data not shown).

Figure 4 characterizes the content of physician services provided to the Medicare population (MCBS) across quintiles. Only a small proportion of total physician services in any quintile was for major surgical procedures, and the overall rate of major surgery was relatively constant across quintiles. Evaluation and management services (visits), tests, radiology services, and minor procedures made up the vast majority of physician activity and explained the differences in physician practice found across the quintiles.

Figure 5 provides detailed information on differences in the specific services provided to the chronic disease cohorts across HRRs with differing spending levels. We present the pooled relative rates and 95% CIs for all three chronic disease cohorts combined in quintile 5 (highest expenditure index) compared with quintile 1 (lowest expenditure index) because the relative use rates across quintiles were similar for each cohort. For example, although pulmonary function tests were performed twice as frequently overall in patients with acute MI than in those with hip fracture, they were performed nearly three times more often in the highest quintile than in the lowest quintile in all three cohorts (Appendix Table 11).

Rates of major procedures differed little across quintiles and were sometimes lower and sometimes higher in quintile 5. Rates of several minor, relatively nondiscretionary procedures (skin laceration repair, breast biopsy) were also similar across quintiles.

As was found in the general population sample, differences in utilization across quintiles were largely due to increased use of evaluation and management services and associated tests, imaging and minor procedures, and use of the hospital as a site of care. Rates of outpatient physician office visits averaged 1.27 (95% CI, 1.26 to 1.28) times higher in quintile 5 than in quintile 1. Inpatient visits, however, were 2.13 (CI, 2.12 to 2.14) times higher, and inpatient specialist consultations were 2.36 (CI, 2.33 to 2.39) times higher. The proportion of patients seeing more than 10 different physicians during the first year after their index admission was 2.97 (CI, 2.84 to 3.11) times higher in quintile 5. More frequent physician contact was associated with more frequent use of diagnostic tests and minor procedures. Patients in quintile 5 spent more time in the hospital (rate ratio, 1.52 [CI, 1.50 to 1.54]) and in the intensive care unit (rate ratio, 1.55 [CI, 1.50 to 1.60]).

Some of the most dramatic differences were found in rates of services provided to severely ill patients. For example, among patients in their last 6 months of life, intensive care unit days were 2.28 (CI, 2.18 to 2.38) times higher in quintile 5 than in quintile 1 and the use of vena cava filters, feeding tubes, and emergency intubation were all more than 2.3 times as frequent in quintile 5 as in quintile 1.

Regions with higher expenditure indices did not provide better quality of care on most measures (Table 5). Among patients in whom the specific treatment was recommended, patients with acute MI in the highest quintile were no more likely to receive acute reperfusion, were less likely to receive aspirin at admission or discharge and angiotensin-converting enzyme inhibitors in the setting of a low ejection fraction, and were more likely to receive -blockers. Because some preventive services (for example, influenza vaccination) may be provided in nonreimbursed settings, the results for preventive services are based on patient reports from the MCBS. Although mammography was performed as frequently in high as in low quintiles, influenza and pneumococcal immunization and Papanicolaou smears were provided less frequently in HRRs with higher expenditure indices. Lack of better-quality care in HRRs in the highest quintile was not related to the greater predominance of teaching or large hospitals (Figure 6). Major teaching hospitals had somewhat higher quality on several of the measures, but the differences in quality across quintiles were small and inconsistent.

Although the absolute differences in access to care were small, the findings suggest a general pattern of slightly lower access to care in HRRs with higher expenditure indices (Table 6). Patients with acute MI who lived in regions with higher expenditure indices were significantly less likely to receive exercise testing and angiography, and a slightly smaller percentage saw a physician within 30 days of discharge. Larger differences emerged in the type of specialists seen during the first 30 days: Those in HRRs with lower expenditure indices were more likely to see family practitioners, and those in HRRs with higher expenditure indices were more likely to see medical subspecialists.

In HRRs with higher expenditure indices, a slightly smaller proportion of patients in the general population (MCBS) reported having a usual source of care. Some differences in the site of physician visits were seen, with more frequent outpatient and office visits in regions with high expenditure indices. Waiting times for emergency department, outpatient facility, and office visits were significantly longer in higher-spending regions. Finally, on the basis of one of three traditional measures of access (having a problem but not seeing a physician), HRRs with a higher expenditure index provided significantly worse access to care.

We conducted a cohort study in four distinct samples of Medicare enrollees to compare the content, quality, and accessibility of care across 306 U.S. HRRs with substantially different spending levels. The primary exposure variable in this study, the EOL-EI, was intended to measure the component of regional variation in Medicare spending that is unrelated to regional differences in illness or price. The goal was to ensure assignment of HRRs (and the patients within them) to treatment groups that were similar in baseline health status but that differed in subsequent treatment. The validity of the approach was confirmed by our finding that illness levels in each of the four study cohorts differed little across quintiles but that health care utilization rates and spending (for all four study samples) increased steadily and substantially as the expenditure index for a given HRR increased. Regardless of the measure used to characterize spending, residents of the highest-spending quintile received about 60% more care than residents of the lowest-spending quintile.

We compared the content, quality, and accessibility of care across regions with different levels of spending. The differences in spending across HRRs were largely due to more frequent use of the hospital as a site of care, more frequent physician visits, greater use of medical subspecialists, and more frequent diagnostic tests and minor procedures. The quality of care was no better in higher-spending regions, and access to care was slightly worse on most measures.

Our analysis has several limitations. First, we had only a limited number of measures of quality and access and studied only four cohorts. The consistency of our findings across cohorts and measures, however, strengthens the causal inferences that can be drawn. Other large studies have also documented substantial differences in resource use that are unrelated to quality of care (32 - 35).

Second, we must address concerns about potential unmeasured differences in health status. It is highly unlikely that the 60% differences in utilization observed across quintiles of spending could be due to residual confounding by unmeasured illness levels. In each of the four cohorts, we found that patients' predicted risk for death differed little across regions of differing spending levels. Moreover, crude and adjusted utilization analyses yielded nearly identical results. To account for the greater than 60% difference, therefore, an unmeasured confounder would have to be a more powerful predictor of utilization than those we measured (including self-assessed health status or the severity of an acute MI) and at the same time not be correlated with the available measures (because crude and adjusted analyses yielded similar results). The nearly 60% increased utilization observed in higher-spending regions was found in every subgroup of the study samples (Appendix Tables 12, 13, and 14).

Third, our findings cannot prove that the strong association we observed between capacity (the supply of hospital beds and medical specialists) and utilization is entirely causal. Differences in the malpractice environment could contribute to the differences in practice we observed. State-level differences in malpractice, however, are associated with a less than 10% difference in utilization (36). In addition, while it is possible that Medicare enrollees in high-spending regions prefer a more specialist-intensive pattern of practice, neither preferences nor greater fears of malpractice provide a compelling justification for the differences in public expenditures.

Previous research, however, suggests a causal relationship between supply and utilization. Use of physician services is strongly associated with the local workforce composition (7) and supply (9). Chronically ill patients are more likely to receive care in the hospital in communities with more beds (8,12,20,37). In addition, the local bed supply, rather than patient preferences, explained the differences in end-of-life care among patients in the Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments (SUPPORT) (29). Finally, it has been shown that physicians adapt their admission and discharge decisions to the availability of intensive care unit beds, admitting more patients with less severe illness and extending length of stay when more beds are available (38). It appears likely that physicians in all regions are simply managing their patients with available resources and that inpatient management and subspecialist consultation are easier in regions where these resources are readily available.

Regional differences in Medicare spending are due almost entirely to use of discretionary services that are sensitive to the local supply of physicians and hospital resources: more frequent physician visits, greater use of specialists, and greater use of the hospital and intensive care unit as sites of care. Policymakers and purchasers concerned with resurgent growth in health care spending will need to focus on these supply-sensitive services. As we discuss in greater detail in Part 2, however, our study provides little guidance on the potential impact of reducing the use of such services, and caution is warranted as policies are developed to control health care spending.

Nevertheless, for the Medicare population, it appears that neither greater local availability of physicians and hospital beds nor the more inpatient-based and specialist-oriented pattern of practice that result are associated with improved access to care, better-quality care, or (as is reported in Part 2) better health outcomes or satisfaction. These findings call into question the notion that additional growth in health care spending is primarily driven by advances in science and technology and that spending more will inevitably result in improved quality of care.

The Appendix was developed to provide interested readers with additional detail on the methods of the study as well as supplementary findings referred to in the body of the papers that could not be included there because of space constraints. Section B provides an expanded discussion of the rationale for our study design and its relationship to instrumental variables analysis. Section C describes in greater detail our study populations, exclusions applied, and data quality. Section D describes in detail the rationale behind the approach and the methods used to calculate spending and utilization rates using measures free of bias that could be introduced because of differences in wages, prices, or policy payments to physicians or hospitals. Section E describes in greater detail the End-of-Life Expenditure Index (EOL-EI), the primary exposure used in the analysis, including the study population within which it was calculated and how members of each study cohort were excluded from the sample used to calculate the index used as the exposure for that cohort. Section F describes the motivation, methodology, and results of our sensitivity analysis using the Acute Care Expenditure Index.

In addition, the Appendix also includes supplementary tables that present additional detail on individual patient attributes (Appendix Tables 1, 2, 3, and 4), a table that lists specifically which variables are included in each of the major models used in the analyses (Appendix Table 5), the main models examining survival (Appendix Tables 6, 7, 8, and 9) and change in functional status (Appendix Table 10), a table presenting specific procedure rates for each chronic disease cohort and for all three cohorts combined (Appendix Table 11), and tables summarizing overall health care utilization rates across quintiles for each chronic disease cohort (Appendix Tables 12, 13, and 14).