This retrospective cohort hospital-level analysis used Medicare Administrative and Claims data and American Hospital Association (AHA) data. We used Medicare Parts A and B and enrollment information (from the master beneficiary summary data file [MBSF]) from the Chronic Conditions Data Warehouse (CCW) Virtual Research Data Center (VRDC), 2018–2020, to obtain beneficiary-level information. From the enrollment files, we identified Medicare fee-for-service (FFS) eligibility and demographic information. From the Medicare Part A file, we identified the CR-qualifying events and beneficiaries’ history of events. From the Part B institutional and non-institutional claims files, we identified the CR visits to calculate CR metrics. Beneficiaries were followed longitudinally using unique beneficiary identifiers from CCW to calculate the CR metrics.

We obtained hospital-level characteristics from fiscal year 2018 AHA Annual Survey Database for 4,052 hospitals; we used AHA data from fiscal years 2016 (4 hospitals) and 2017 (13 hospitals) when hospitals could not be linked to Medicare using the 2018 data, and 4 hospitals could not be linked at all.²² The AHA dataset is derived from annual voluntary surveys sent to all AHA-registered U.S. hospitals. Survey data from respondent hospitals are merged into yearly datasets. We linked the Medicare claims data to the AHA survey data using the CMS provider number. This research was considered exempt from Institutional Review Board review under 45 Code of Federal Regulations 46.101[b] [5], which covers Department of Health and Human Services research and demonstration projects, which are designed to study, evaluate, or examine public benefit or service programs.

At the hospital-level (Appendix Figure 1A), we included all hospitals who treated patients with a CR-qualifying event in 2018, aged ≥65 years, continuously enrolled in a Medicare FFS plan or until death. We excluded hospitals (1) in U.S. territories, (2) with <20 CR-qualifying events (because CR metrics would be statistically unstable), and (3) that could not be linked with the AHA data.

At the patient-level (Appendix Figure 1B), we included Medicare FFS beneficiaries with an AMI, PCI, or CABG in 2018, aged ≥65 years old as of January 1, 2018, continuously enrolled in a FFS plan from 90 days before the CR-qualifying event (pre-CR period) though the end of the CR treatment period (365 days and 36 weeks after the CR-qualifying event) or until date of death, and without a history of an AMI, PCI, or CABG in the pre-CR period (see Appendix Figure 2 for study timeline of CR-qualifying events). AMI events were identified using International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) diagnosis codes (first or second-listed code) (Appendix Table 1A) according to the Million Hearts^® Outpatient CR Use Surveillance Methodology.^10,21 CABG and PCI procedures were identified using International Classification of Diseases, 10th Revision, Procedure Coding System (ICD-10-PCS) procedural codes (any location) on inpatient claims, or current procedural terminology (CPT) codes (any location) on outpatient or clinician claims (Appendix Table 1A).^10,21

We excluded beneficiaries who (1) died <21 days after the CR-qualifying event; (2) had ≥90 consecutive days in an inpatient acute care hospital, other inpatient hospital, or postacute care setting (extended hospital stay) within the pre-CR period or initiated an extended hospital stay within 21 days after the CR-qualifying event; (3) received hospice care in the pre-CR period or initiated hospice care within 21 days after the CR-qualifying event;^10,16,21 or (4) had end-stage renal disease in the pre-CR period or after the CR-qualifying event.²¹

Standard and intensive CR claims were identified by Healthcare Common Procedure Coding System (HCPCS) codes and revenue center code 0943 within 365 days of the qualifying event index date in alignment with previous research and Medicare benefit guidance (Appendix Table 1B).^10,11,23 For single events, the index date was the inpatient hospital discharge date associated with the CR-qualifying event. For combination events (≥2 CR-qualifying events within 21 days), the index date was that associated with the last event. The number of individual CR sessions on different dates was used to identify number of CR sessions. If ≥1 CR session was identified on the same day, each individual session was counted separately.²¹ The hospital where the CR-qualifying event occurred is referred to as the index hospital.

Hospitals can be conceptualized according to level of cardiac care provided.^18–20 In our analysis, hospitals were stratified into mutually exclusive categories by tier of cardiac care provided: (1) hospitals that performed ≥10 CABG and ≥10 PCI procedures (ie, AMI, CABG, and PCI care available, referred to as “comprehensive” care); (2) hospitals that performed ≥10 PCI procedures but <10 CABG procedures (ie, AMI and PCI care available, referred to as “AMI/PCI” care); and (3) hospitals that performed <10 CABG and <10 PCI procedures (ie, “AMI-only” care). To eliminate potential billing errors, a cutoff of 10 was used to determine whether a hospital performed a procedure.

We calculated hospital-level CR use metrics consistent with the Million Hearts^® Outpatient CR Use Surveillance Methodology²¹: (1) CR enrollment (ie, number of beneficiaries who attended ≥1 CR session within 365 days of the CR-qualifying event out of the number of CR-qualifying events within the measurement period) and (2) time to CR enrollment (ie, time in days from the CR-qualifying event index date or hospital discharge date after index event to the first CR session date). Although they may not be as directly influenced by the index hospital, we also examined variation in the number of CR sessions (ie, number of CR sessions attended within 36 weeks of the first CR session date) and CR completion (ie, number of individuals who attended ≥36 CR sessions within 36 weeks of the first CR session date out of the number of individuals who enrolled in CR). The CR metrics were calculated based on CR encounters at any outpatient facility and attributed to the index hospital where the CR-qualifying event occurred. As such, hospitals without CR programs were included in our calculation of CR metrics if the patient who had a CR-qualifying event at that hospital attended CR at any outpatient facility.

Hospital-level categorical variables included information on geographic location, including urbanicity (urban, rural) and U.S. Census region and division (Northeast [New England, Middle Atlantic], Midwest [East North Central, West North Central], South [South Atlantic, East South Central, West South Central], West [Mountain, Pacific]). We included information on hospital economic model (nonprofit, for-profit, public); teaching status (nonteaching, teaching); service type (general medical and surgical, heart, other specialty treatment); critical care access designation (yes, no), community hospital designation (yes, no); and whether the hospital provided cardiac intensive care (yes, no) and CR (yes, no). Continuous variables included information on total hospital beds and intensive care unit (ICU) beds for cardiovascular disease care. About 15.1% of AHA hospitals did not respond to the annual survey.

Patient-level variables included age at the CR-qualifying event index date (65–70, 71–75, 76–80, 81–85, >85 years), sex (male, female), race and ethnicity (Non-Hispanic White, Non-Hispanic Black, Hispanic, Non-Hispanic Asian, Other/Unknown), and CR-qualifying indication (AMI with CABG and PCI, AMI with CABG, AMI with PCI, AMI only, CABG with PCI, CABG only, and PCI only).

Data are described using standard statistical summary measures, including mean ± standard deviation (SD), median (interquartile range [IQR]; quartile 1 = 25%, quartile 3 = 75%), and range (minimum, maximum). For count metrics (CR enrollment, CR completion), we report percentages for the summary measures. For continuous metrics (time to CR enrollment, number of CR sessions), we report the mean number of days to CR enrollment/CR sessions within each hospital (each individual has a value for number of days to CR enrollment/CR sessions), and the median of the mean number of days to CR enrollment/CR sessions. Chi-square tests were used to compare differences in proportions for categorical variables by tier of cardiac care; Kruskal-Wallis tests were used to compare differences in medians for continuous variables by tier of cardiac care using SAS PROC NPAR1WAY.²⁴

In adjusted analyses, using SAS PROC GLIMMIX, we used generalized linear mixed models with binary distribution and logit function to examine patient- and hospital-level factors associated with CR enrollment and completion (among hospitals with enrollment) and present estimates as odds ratios (ORs). Among hospitals with enrollment, we used similar models but with gamma distribution and log-link function to examine factors associated with time to enrollment and number of sessions, and we present differences in means estimates as means ratios. In supplemental analyses, because enrollment is associated with CR-qualifying indication and hospital performance on CR enrollment may thereby be associated with patient case-mix, we stratified by CR-qualifying indication to examine the association between tier of cardiac care with CR metrics for the relevant comparisons: for patients with AMI only among all 3 tiers of cardiac care (reference = AMI only care), and for patients with PCI only or AMI with PCI among hospitals with comprehensive cardiac care and AMI/PCI care available (reference = AMI/PCI care). In addition, given that comprehensive care hospitals made up the majority of the patient population, we performed a supplemental analysis to stratify the comprehensive care hospitals (N = 983) by hospital size (small [<200 beds], medium [200–499 beds], and large [≥500 beds]) using the hospital bed size variable from the AHA data. All models included patient- and hospital-level factors and accounted for clustering of patients within hospitals. The patient-level factors included age (reference = ≤70), race and ethnicity (reference = non-Hispanic White), sex (reference = male), and CR-qualifying indication (reference = AMI only). The hospital-level factors included hospital economic model (reference = public), region (reference = Midwest), urbanicity (reference = urban), teaching status (reference = nonteaching), cardiac intensive care hospital (reference = yes), and cardiac rehabilitation hospital (reference = yes). A P value < .05 was used to indicate statistical significance. All analyses were performed with SAS (v.9.4; SAS Institute, Inc., Cary, NC).

Across the United States, 2,212 hospitals treated individuals aged ≥65 years with a CR-qualifying event in 2018 (Appendix Figure 1A). Within the 2,212 hospitals, 240,947 beneficiaries had a CR-qualifying event in 2018 and were alive for ≥21 days after the event without an extended hospital stay, hospice care, or end-stage renal disease (Appendix Figure 2B).

Characteristics of U.S. hospitals overall and by tier of cardiac care are presented in Table 1. Overall, nearly all hospitals included in the analysis were of general medical and surgical service type (99.1%) and non-critical access hospitals (98.6%). Most hospitals were under non-profit ownership (70.1%) and nonteaching institutions (89.5%). Geographically, most hospitals were in a major metropolitan area (83.7%).

By tier of cardiac care, 44.4% (n = 983) of hospitals provided comprehensive care, while 31.2% (n = 690) provided AMI/PCI care, and 24.4% (n = 539) provided AMI-only care. More comprehensive care hospitals were in metropolitan areas (95.4%), whereas more AMI-only care hospitals were in rural areas (8.7%). Compared with AMI-only and AMI/PCI care hospitals, a higher percentage of comprehensive care hospitals were cardiac intensive care (70.0%) and cardiac rehabilitation hospitals (81.2%). An individual-level analytic cohort summary table is presented in Appendix Table 2.

Hospital-level CR metrics overall and by tier of cardiac care are presented in Figure 1. Overall, across all hospitals, there was variation in CR enrollment (median 19.6%, IQR = 7.0%, 32.8%). Among hospitals with CR enrollment (n = 1,866), the median (of the mean) time to enrollment was 55.0 days (IQR = 41.0, 71.0) and number of CR sessions was 26.0 (IQR = 23.0, 29.0). The median CR percent completion was 26.0% (IQR = 10.5%, 41.2%). By tier of cardiac care, comprehensive care hospitals served 189,674 patients (78.7%), AMI/PCI care hospitals served 41,679 patients (17.3%), and AMI-only care hospitals served 9,594 patients (4.0%). There was variation in CR performance metrics within each tier of cardiac care. For example, the median percent CR enrollment was 30.7% (IQR = 20.7%, 41.3%) for comprehensive care, 18.6% (IQR = 9.5%, 27.7%) for AMI/PCI care, and 0.0% (IQR = 0.0%, 7.7%) for AMI-only care hospitals. The median (of the mean) time to CR enrollment was 56.0 days (IQR = 43.0, 68.0) for comprehensive care, 54.0 days (IQR = 37.0, 73.0) for AMI/PCI care, and 64.0 days (IQR = 38.0, 122.0) for AMI-only care hospitals. The CR metrics by tier of cardiac care and U.S. Census division are presented in Appendix Figures 3–6. Appendix Tables 3–6 present all summary measures, including number of hospitals, number of patients served in each hospital, mean (SD), median (IQR), minimum, and maximum, for all hospital-level CR metrics by tier of cardiac care and division.

Patient- and hospital-level factors associated with CR metrics are presented in Table 2. Compared with AMI only, other CR-qualifying indications were highly associated with higher odds of CR enrollment: CABG only (OR = 15.51, 95% CI, 14.99, 16.06); CABG with PCI (OR = 15.40, 95% CI, 12.40, 19.14); AMI with CABG and PCI (OR = 12.62, 95% CI, 10.92, 14.58); AMI with CABG (OR = 11.87, 95% CI, 11.34, 12.41); AMI with PCI (OR = 6.39, 95% CI, 6.20, 6.59); PCI only (OR = 3.94, 95% CI, 3.81, 4.07). CR-qualifying indications other than AMI only were also highly associated with fewer mean days to CR enrollment, higher mean number of CR sessions, and higher odds of CR completion. Characteristics associated with inferior CR metrics (ie, lower odds of CR enrollment, more mean days to CR enrollment, fewer mean number of CR sessions, and/or lower odds of CR completion) included patient-level factors such as older age (vs ≤70), female sex (vs male), non-White race or ethnicity (vs Non-Hispanic White), and hospital-level factors such as for-profit ownership (vs public), U.S. Census regions other than the Midwest (vs Midwest), rural location (vs urban), and medium or large hospital size (vs small).

In stratified analyses by CR-qualifying indication for the relevant comparisons (Appendix Tables 7–8), 37.1% of patients served by the hospitals included in our study had an AMI only (n = 89,284), 24.5% had an AMI with PCI (n = 59,039), and 18.5% had a PCI only (n = 44,645) (Appendix Table 2). Among patients with AMI only, after controlling for patient- and hospital-level factors, hospitals with more comprehensive cardiac care were associated with higher odds of CR enrollment (comprehensive care: OR = 2.01, 95% CI: 1.76, 2.29; AMI/PCI care: OR = 1.56, 95% CI, 1.38, 1.77 vs AMI-only), fewer mean days to CR enrollment (comprehensive care: −17.48, 95% CI, −22.97, −6.34 and AMI/PCI care: −13.98, 95% CI, −20.08, −2.20 vs AMI-only), but lower odds of CR completion (comprehensive care: OR = 0.68, 95% CI, 0.50, 0.93) (Appendix Table 8). Among patients with PCI only, after controlling for patient- and hospital-level factors, hospitals with more comprehensive cardiac care were associated with higher odds of CR enrollment (comprehensive care: OR = 1.12, 95% CI: 1.04, 1.20 vs AMI/PCI care) (Appendix Table 8).

In a subset analyses of comprehensive care hospitals stratified by hospital size (Appendix Tables 9–10, Appendix Figure 7), small hospitals had slightly higher percent CR enrollment (31.9% vs 30.6% medium and 30.1% large) and shorter time to enrollment (47 days vs 56 medium and 59 large) (Appendix Figure 7). In adjusted analyses, relative to small hospitals, medium hospitals had lower enrollment (OR = 0.84, 95% CI, 0.81, 0.87) and longer time to enrollment (11.40; 95% CI, 8.8, 14.4) as did large hospitals (CR enrollment: 0.84 95% CI, 0.80, 0.87; time to enrollment: 11.70; 95% CI, 8.9, 15.0) (Appendix Table 8).

This study was the first national, hospital-level analysis of CR metrics among Medicare beneficiaries. In a large cohort of hospitals (n = 2,212) serving more than 240,000 Medicare FFS beneficiaries eligible for CR, using CMS claims and AHA data, we showed that just under half of hospitals provided comprehensive cardiac care (40%), but served over three quarters of eligible patients (78%), and a quarter of hospitals provided no procedural care and served only 4% of eligible patients. We presented hospital-level CR use benchmarks nationally, with median CR enrollment at 19.6%. Among hospitals with enrollment, median time to enrollment was 55 days, number of sessions attended was 26.0, and completion was 26.0%. Our innovative study design highlighted wide variation in CR performance across all hospitals nationally, and also within each tier of cardiac care. This information is important because it suggests where efforts might be focused, considering limited healthcare resources. Comprehensive cardiac care hospitals have higher percent enrollment and participation, but they also serve the most patients; efficient use of resources may suggest learning first from high-performers among these centers.

This study was unique in its analysis of hospital-level metrics by tier of cardiac care, which can inform hospitals of their performance relative to peer hospitals. Previous analyses of the Michigan Value Collaborative statewide multipayer registry demonstrated that hospital-level factors may be associated with CR enrollment¹² and participation.^13,14 Our study expanded on prior analyses by the inclusion of AMI events and examining hospital performance at a national level. Our levels of cardiac care were delineated similarly to the American Heart Association and The Joint Commission’s new certification for hospitals, where, as of 2022, hospitals can be certified according to level of cardiac care provided, ranging from AMI-only care (ie, Level III: Acute Heart Attack Ready) to comprehensive cardiac care (ie, Comprehensive Cardiac Center).^18–20 While our data were collected before this certification, future analyses could examine whether the certifications (and potential related improvements in cardiac care) might be associated with better CR metrics.

As seen in previous studies,^{12–14,25,26} patient case-mix did not fully explain variation in hospital performance as we observed variation in CR metrics across hospital tiers even after stratifying by CR-qualifying indication. For example, for patients with AMI only, controlling for patient- and hospital-level factors, it was notable that comprehensive and AMI/PCI hospitals had nearly 2-times higher odds of CR enrollment, and about 14–17 fewer mean days to enrollment than AMI-only care hospitals. AMI-only hospitals served almost 10,000 patients. Given that AMI-only hospitals were the lowest-performing hospitals across all metrics for patients with AMI, and all hospital tiers performed lower than the national 70% CR participation target,^27,28 our findings can encourage hospitals to evaluate their processes for improving CR enrollment and participation. The responsibility to refer and enroll qualifying patients into CR lies heavily with the inpatient referring physician and the care coordination staff responsible for educating the patient about the benefits of CR and facilitating enrollment in a program. For example, top performing hospitals are known to facilitate enrollment in nonemergency transportation programs, vocational rehabilitation services, and/or case management programs to optimize the likelihood of enrollment and completion of CR. Once enrolled, the responsibility to optimize the number of sessions attended and the completion of the program largely lies with the patient, the outpatient hospital CR program staff, and, ideally, the patient’s primary physician or cardiologist who receives the 30-day reports. Because there may be a continued role for the index hospital in providing wraparound services to address patient barriers to participation, in addition to examining enrollment and time to enrollment, we also included participation and completion metrics in this analysis.

Hospitals engaging in measuring performance and quality improvement efforts can refer to strategies in the Million Hearts^®/ AACVPR Cardiac Rehabilitation Change Package, 2nd Edition to improve CR utilization and delivery.¹⁷ This may begin with establishing hospital champions and engaging hospital administrators and the care team to review their performance and initiate steps to improve their post cardiac-event care processes.¹⁷ Another strategy is to incorporate CR referral into routine hospital processes of care, such as including automatic, opt-out referral to CR in electronic health records, and developing standard processes for informing a CR program of a referred patient or empowering patients to self-refer to CR. Tools and resources to inform and equip hospital teams to incorporate CR referral into routine hospital care processes and to standardize the CR referral process with automatic referrals and care coordination were developed through the Agency for Healthcare Research Quality TAKEheart initiative. Additional care process improvements are informed by the 2018 American College of Cardiology (ACC)/American Heart Association (AHA) Task Force on performance measures that describes CR-specific performance and quality measures and recommends all eligible inpatients (prior to discharge) and outpatients be referred to an outpatient CR program.¹ Early appointments to outpatient CR at hospital discharge (ie, an orientation appointment within 10 days) is another approach that has significantly improved attendance at orientation compared with the median time from hospital discharge to enrollment (ie, an orientation appointment at 35 days).²⁹ Other ways to optimize CR care coordination include developing infrastructure for organizing inpatient CR liaisons to schedule the first outpatient CR visit or delivery models to accommodate individual needs (eg, case management or supportive services) and navigating cost-sharing barriers.^17,30 Such coordination of care could particularly help underserved populations, such as women and patients with lower socioeconomic status.^31,32 Ultimately, the availability of hospital-level performance data may help spur a broad array of strategies that have varying degrees of impact and investment to improve the delivery of clinical guideline-directed care for cardiac patients.³³

Our study is not without limitations. First, our findings are not generalizable to all hospitals, as we excluded hospitals who did not treat patients who met our inclusion criteria. Second, our metrics were derived entirely from CMS data and included only adults over age 65; CR metrics may differ among Medicare Advantage, commercially insured, or younger patient populations, so the observed metrics in this report may differ from true metrics among all patients served by each hospital. In addition, hospital-level CR metrics may have been differentially affected due to differing patient population demographics across hospitals (i.e., some hospitals may serve a larger percentage of younger patients, and therefore their true metrics may have been more divergent from the observed results compared to other hospitals that serve an older patient population). Finally, AHA data on the location of CR centers may not align with the availability of programs as assessed by localized hospital networks and quality improvement organizations. Future initiatives include scaling-up work by the Michigan Value Collaborative³⁴ by creating individual hospital-level reports of CR metrics for each hospital in the United States, to be used by hospitals for quality improvement efforts.