Cohort

The cohort consisted of Medicare beneficiaries who had an index hospitalization in 1997 in a US nonfederal, acute care hospital with “a qualifying coronary diagnosis or procedure.” Inclusion was based on a principal discharge diagnosis code (with the use of the International Classification of Diseases, Ninth Revision code for AMI [410.xx] or a procedure code for CABG [36.1x]). For patients with >1 qualifying admission during 1997, the earliest admission was considered the “index” admission.

Additional inclusion criteria were age ≥65 years at the time of admission, an index hospitalization stay of ≤31 days and alive at 30 days after discharge (used to identify reasonable CR candidates), and uninterrupted enrollment in fee-for-service payment (not in capitated health plan) and entitlement under Medicare Part A and Part B during the 12 months after the index hospitalization discharge date. Patients with index hospitalizations in Puerto Rico or US territories were excluded because of small numbers of patients.

Data Sources

The primary data source was Medicare’s National Claims History File. For qualifying patients, we linked inpatient claims with Medicare’s master enrollment database to obtain information on date of birth, sex, race, date of death (where applicable), residence zip code, enrollment status over time, entitlements (Part A and Part B), and group health plan membership. Census 2000 data were linked to the patient’s residence zip code statistics as proxies for socioeconomic, educational, and disability statuses. We used American Hospital Association and Medicare data to determine hospital characteristics for index admissions.

CR Services Use

The use of CR services was defined as any Medicare payment in hospital outpatient claims for at least 1 CR session (Current Procedure Terminology codes 93797 and 93798) within 1 year after discharge from the index hospitalization. We also evaluated how soon after discharge from the index hospitalization CR was initiated. We characterized CR intensity by the number of sessions received within 1 year and the number of days they spanned.

Predictors of CR Use

Predictors of CR use were identified with the use of Andersen and Aday’s classic behavioral model of health services utilization that focuses on predisposing, enabling, and illness characteristics of patients.^17,18 CR candidates were classified into 2 main groups: AMI or CABG surgery without AMI. Patients with AMI were further classified into 3 subgroups: medical treatment only, percutaneous coronary intervention (PCI) without CABG, or CABG. Patients who received both CABG and PCI were classified as CABG.

Patient demographic and comorbidity characteristics were examined with the use of information from the claims data. We identified 25 comorbidity groups considered related to CR use that resembled Charlson’s groupings using diagnostic and procedure codes from the index hospitalization (or any hospitalization within 1 year before) and DxCG software (DxCG, Inc, Boston, Mass, 2001¹⁹). For patients aged 65 years, the qualifying age for Medicare benefits for the elderly, on average the claims file contained only half a year of prehospitalization data to identify comorbid conditions. These participants represented only ≈6% of the entire study cohort.

We also examined distance from the patient’s residence to the nearest CR facility. Distance to CR was defined as the shortest distance (in miles) from the patient’s zip code centroid to the nearest available CR facility within the state (located by its exact latitude and longitude). We assumed that patients did not cross state borders to receive CR.

Patient socioeconomic characteristics were inferred by Medicaid dual eligibility and census data. Enrollment in Medicaid was the only patient-level indicator of low income available in the claims file. We also assigned variables to each patient reflecting the proportions of people within the patient’s zip code: residing in urban areas; living under the poverty line according to race and age groups (65 to 74 and ≥75 years); having some college education according to sex and race; having any disability (according to race, sex, and age group); and household median income by age group of the head of household. We created overall quintiles for each of these indicators. We used 5 hospital characteristics: availability of cardiac catheterization, angioplasty, and open heart surgery; number of beds; and medical school affiliation.

Statistical Analyses

We first performed univariate and bivariate analyses. We used univariate analyses to determine the proportion of patients receiving CR. We employed bivariate analyses to describe differences in CR use by patient demographics, comorbidities, characteristics of the index hospitalization and hospital in which it occurred, patient zip code, census region, and state (50 states and the District of Columbia). We computed t tests for continuous variables and χ² tests for categorical variables.

We next estimated a multiple logistic regression model to identify patient and hospital predictors of CR use. Covariates in the model were as follows: patient demographics and comorbid conditions, characteristics of the index hospitalization and inpatient facility, socioeconomic and disability characteristics of the patient’s zip code, distance to nearest CR facility, and state indicators. We adjusted for clustering of patients within their index hospital through generalized estimating equations using the GENMOD procedure in SAS software, version 9.1 (SAS Institute Inc, Cary, NC). A single correlation (exchangeable option) affecting any pair of patients within each cluster (hospital) was used, and adjusted odds ratios (ORs) of CR use were obtained for each variable included in the model.

Geographic Variation

To quantify geographic variation in CR use, we estimated a random intercept hierarchical logistic regression model. The random intercepts represented the underlying log-odds of CR use for each state and were assumed to vary across states. We adjusted for patient demographic and socioeconomic characteristics using the same predictors as in the generalized estimating equations model. The model was fitted with the use of the SAS GLIMMIX procedure. We then calculated both state-adjusted estimates of CR use and standardized state-specific rates of CR use using the methodology developed by one of the authors (S.T.N.).²⁰ The adjusted state-specific CR use rate was estimated as the average of the predicted individual probabilities of all the CR candidates living in each state. The expected state rate was calculated as the average of the predicted individual probabilities as if those individuals were living in an average state (through the exclusion of the effect of the state-specific random effect). The standardized state-specific rate of CR use was then estimated as the adjusted, state-specific CR rate divided by the expected CR rate for that state, multiplied by the national unadjusted CR rate.

Each state’s 95% confidence interval (CI) on its adjusted rate was examined to determine whether it excluded the national CR rate. If it did, then we concluded that the state had higher or lower rates than expected.

The authors had full access to and take full responsibility for the integrity of the data. All authors have read and agree to the manuscript as written.

Use of CR

Our study cohort consisted of 267 427 patients, of whom 18.7% (49 877) received at least 1 session of outpatient CR after hospital discharge (Table 1). Recipients of CR received an average of 24 sessions (SD 12). Men were more likely to receive CR than women (22.1% versus 14.3%); use was inversely related to age; and whites were more than twice as likely as nonwhites to receive CR. Only 5.2% of people dually eligible for Medicaid and Medicare received CR compared with 20.3% of those who were not.

Overall, CR was used in 13.9% of patients hospitalized for AMI and in 31.0% of those who underwent CABG surgery. Only 11.1% of patients with an AMI and no subsequent revascularization procedure during the index hospitalization received CR. Patients were more likely to receive CR if they had been admitted from home (19.3%) than if they had been transferred from another acute care hospital (13.2%) or nursing home (5.6%). Index admissions to hospitals with cardiac catheterization, angioplasty, and open heart surgery capabilities increased CR use to 22.4% from 13.8% in hospitals with none of these capabilities. Slightly higher CR rates were observed in hospitals affiliated with medical schools (20.5%) versus those not affiliated (17.1%).

Women, older people, nonwhites, and patients receiving Medicaid received somewhat fewer sessions on average. Patients initiated CR an average of 55 days after discharge from the index hospitalization, with 25% initiating therapy within 21 days and >25% initiating >2 months after discharge. Women, nonwhites, and Medicaid recipients began ≈1 week later than their complementary groups (P<0.0001), but there was no association of age with the timing of CR initiation.

Comorbidities

Commonly reported comorbidities among patients with a coronary diagnosis or procedure (Table 2) were hypertension (57%), congestive heart failure (37%), diabetes mellitus (26%), arrhythmias (33%), chronic pulmonary disease (21%), and musculoskeletal conditions (mainly arthritis) (18%). Overall, CR users had fewer comorbidities than nonusers (2.1 versus 2.7; P<0.0001) among the 25 comorbidity groups considered. In a bivariate analysis, patients with congestive heart failure, diabetes mellitus with complications, cerebrovascular disease, chronic pulmonary disease, or renal disease had moderate reductions in any CR use (0.69 to 0.77), whereas patients with dementia or metastatic malignancies were very unlikely to receive CR.

Other Patient-Related Predictors of CR Use

Older individuals, women, and nonwhites were less likely to receive CR than their comparison groups (Table 3). For example, men and women aged 75 to 84 years were only 87% and 69%, respectively, as likely to receive CR as men aged 65 to 74 years. Sex differences increased with age. Whites were 33% more likely to receive CR than nonwhites after adjustment for age and sex (OR=1.33 versus 1.00).

CABG surgery during the index hospitalization was a strong predictor of CR use whether or not it was performed after an AMI (OR=3.5). Patients who received PCI after an AMI were nearly 2 times more likely to receive CR than those with no revascularization procedure (OR=1.8).

Distance to the nearest CR facility was an important predictor of CR use in a multivariable analysis, with use declining monotonically as distance increased (Table 3). For example, patients living in the farthest quintile were 71% less likely to participate in CR than those living in the quintile closest to a CR facility (adjusted OR=0.29).

CR use was also associated with the zip code characteristics of the patient’s residence including degree of urbanization, income, proportion of the population at or below the poverty level, and proportion with college education (not shown). Patients living in zip codes with the highest levels of urbanization and poverty were 36% and 17% less likely to use CR than those living in the most rural or least impoverished areas, respectively (P<0.001). Conversely, patients living in zip codes with the highest levels of median household income and education were 23% and 33% more likely to use CR than those living in zip codes with the lowest income and education (P<0.001).

Hospital Predictors

Patients transferred from a skilled nursing facility or long-term care facility for their index hospitalizations were less likely to receive CR (OR=0.72) than those admitted from home. Patients from smaller hospitals were more likely to participate in CR (OR=1.27), as were those hospitalized in facilities not affiliated to medical schools (OR=1.33) compared with patients in hospitals with the opposite characteristics.

Geographic Variations

More than a 9-fold geographic variation in CR use was found among states in all unadjusted, adjusted-smoothed (shrinkage), and standardized rates, with rates ranging from 6.6% in Idaho to 53.5% in Nebraska after multivariable adjustment (Table 4). Large regional variations are evident in the Figure, with the highest-use states clustered in the north central region. The rate of CR use by state was strongly positively correlated with the number of CR facilities per 10 000 people aged ≥65 years (r=0.82, P<0.001).

• Download figure
• Open in new tab
• Download powerpoint

Figure. Standardized rates of CR by state.

Importance

This is the largest and most comprehensive study of which we are aware to examine the use of CR in older patients with CHD. We found that only 13.9% of Medicare beneficiaries with AMIs and 31.0% of those who had undergone CABG surgery received CR services after index hospitalizations in 1997. For AMI, this rate is higher than the 7% found in elderly patients in 1990²¹ but lower than the 29% rate found in a 2001 survey of patients aged 65 to 79 years in 19 states and the District of Columbia.²² Our study found lower use rates in women, nonwhites, older patients, and “dual eligibles” who were also receiving Medicaid. Our study also documents the deterrent effects on CR use of lower mean income, less education, and higher prevalence of disability. Patients with more concomitant illnesses were less likely to receive CR.

Geographic variations in CR use are especially striking. Use rates were >4-fold higher in north central states (Nebraska, Iowa, North and South Dakota, Minnesota, and Wisconsin) than in southern states. In seeking explanations, we found no correlations between these state variations in CR use with indicators of health consciousness (eg, smoking rates) or quality of care (eg, use of pneumococcal vaccine or use of β-blockers after an AMI) among elders in 1997.²³ Instead, interviews with CR staff suggest the role of factors such as the training and attitudes of physicians and CR staff, abundance of training programs for CR staff, the application of standing orders for CR, and procedures and data systems for initiating and tracking referral and entry into CR.²⁴

Higher rates of CR use in patients undergoing CABG surgery than in those with AMIs probably reflect the high salience of the surgical procedure to the patient and systematic referral by cardiac surgeons. Our finding of a strong deterrent effect on CR use of a greater distance from the patient’s residence to a CR facility, even after correction for patient and hospital characteristics in multivariable analyses, is consistent with other studies.^25,26

Because distance to the nearest CR facility was an important predictor of CR use, payers may wish to explore the feasibility of reimbursing community- or home-based CR programs as supplements or alternatives to facility-based programs, particularly in rural and sparsely populated areas. Available evidence indicates that such programs are safe and equally effective, at least for patients who are at low or moderate risk of complications after AMIs or revascularization procedures.^6,27

Study Limitations

Our study’s main limitations relate to its heavy reliance on Medicare claims data and its focus on 1997 hospitalizations. Medicare claims have strengths and limitations. They provide excellent information on the principal diagnoses for hospital admissions, major diagnostic and treatment procedures received, and some information on comorbid conditions. However, claims lack important clinical data such as left ventricular ejection fraction, body weight, smoking habits, and lipid levels, and they do not accurately distinguish treatment complications, such as cardiac arrhythmias or cardiac arrests, from preexisting conditions.

Medication use is not generally available to researchers from Medicare claims. The fact that our study focused on index hospitalizations for AMI or CABG surgery during 1997 means that it does not reflect subsequent changes in the standards for care for AMIs, newer medications for CHD, advances in cardiac surgery, or the increased use and sophistication of PCI. Although changes in medical practice may have affected the use of CR, Medicare’s eligibility criteria for AMI and CABG remained unchanged until 2006. On balance, we believe that our findings closely mirror recent CR use patterns in Medicare beneficiaries. Other, less important, study limitations are its restriction to Medicare beneficiaries with both Part A and Part B coverage who were continuously enrolled in fee-for-service Medicare. Hence, we cannot generalize our findings to individuals who did not have Part B or who were enrolled in health maintenance organizations, but they constitute small shares of Medicare beneficiaries. In 1997, 97% of 32.2 million beneficiaries enrolled in Part A also had Part B,²⁸ and ≈85% of Medicare beneficiaries were under fee-for-service regulations.²⁹

Effects of the Underuse of CR

The low CR utilization rates we have documented are discouraging in light of the considerable evidence that supports the effectiveness of CR. Meta-analyses of controlled studies have found 15% to 28% reductions in all-cause mortality and 26% to 31% reductions in cardiac mortality. In addition, studies have documented substantial reductions in morbidity and decreases in cardiac risk factors.^1–5 If it is assumed that the CR use rate in Nebraska (53.5%) was achieved in all other states, 93 000 additional Medicare beneficiaries would have received CR, and cardiac mortality would have decreased 26% to 31% in these individuals. Cost-effectiveness analyses suggest that achieving these gains would be highly cost-effective.^30,31

Opportunities to Increase the Use of CR

Increased use of CR might be achieved by improving methods of referring patients to CR facilities after their hospitalizations, implementing quality indicators, and increasing reimbursement rates for these services. Opportunities to increase referrals include using automatic referrals after qualifying hospital admissions,³² creating Web-based referral opportunities, and learning lessons from states that currently demonstrate high utilization rates. For example, striking increases in referrals, from 27% to 62%, were achieved by 1 Web-based referral opportunity.³³ A high overall referral rate, however, may not eliminate disparities reflecting lower use, for example, among women,^34,35 nonwhites,³⁶ or the very old.

Referral to, enrollment in, and completion of CR programs have been proposed as quality indicators in cardiovascular care.³⁷ Such measures might be considered by organizations such as the American College of Cardiology, American Heart Association, Agency for Health Care Research and Quality, National Committee for Quality Assurance, and Joint Commission on Accreditation of Health Care Organizations as means to increase appropriate CR use. They might also be adopted by Medicare in its pay-for-reporting program and the pay-for-performance initiative for hospitals that it is currently developing. Rewards could be given both to the hospital from which the patient was discharged and to the responsible physician.^38,39 Quality indicators might reflect both referral rates to CR and the completion of a specified number (eg, 24) of CR sessions within 90 days after a hospital discharge. Lessons from recent pay-for-performance demonstrations suggest the importance of aligning incentives between physicians and hospitals, incorporating case-mix adjustment, and rewarding improvement as well as excellent performance.⁴⁰

Finally, increased reimbursement rates for CR could serve as positive incentives. Medicare expanded eligibility for CR services in 2006 to include PCI and other indications but did not change levels of reimbursement for the service.¹⁶ The midpoint reimbursement rate by Medicare for a phase II CR session was $15.50 in 2001⁴¹ and was $34 in 2006 (CMS, unpublished data, 2006). Some CR providers argue that the current rate does not fully cover costs and is a deterrent to CR use. CMS could reassess reimbursement levels against their resource costs and compare the merits of reimbursement per session versus packaged reimbursement per program completed. Separate payments for key components, such as nutritional counseling and stress management, might also be considered.

In conclusion, this study has found low national utilization rates of CR after AMI and CABG surgery and remarkable cross-state variations in use. Lower use rates were found in women, nonwhites, dual eligibles, and the very old and in persons with more comorbidities, with lower socioeconomic status, or who live farther from a CR facility.