Abstract
Objective:
Despite guideline recommendation, cardiac rehabilitation (CR) following cardiac surgery remains underutilized, and the extent of interhospital variability is not well understood. This study evaluated determinants of interhospital variability in CR use and outcomes.
Methods:
This retrospective cohort study included 166,809 Medicare beneficiaries undergoing cardiac surgery who were discharged alive between 07/01/2016 and 12/31/2018. CR participation was identified in outpatient facility claims within a year of discharge. Hospital-level CR rates were tabulated, and multilevel models evaluated the extent to which patient, organizational, and regional factors accounted for interhospital variability. Adjusted 1-year mortality and readmission rates were also calculated for each hospital quartile of CR use.
Results:
Overall, 90,171 (54.1%) participated in at least one CR session within a year of discharge. Interhospital CR rates ranged from 0.0% to 96.8%. Hospital factors that predicted CR use included non-teaching status and lower hospital volume. Before adjusting for patient, organizational, and regional factors, 19.3% of interhospital variability was attributable to the admitting hospital. After accounting for covariates, 12.3% of variation was attributable to the admitting hospital. Patient (0.5%), structural (2.8%), and regional (3.7%) factors accounted for the remaining explained variation. Hospitals in the lowest quartile of CR use had higher adjusted 1-year mortality rates (Q1 = 6.7%, Q4 = 5.2%, p < 0.001) and readmission rates (Q1 = 37.6%, Q4 = 33.9%, p<0.001).
Conclusion:
Identifying best practices among high CR use facilities and barriers to access in low CR use hospitals may reduce interhospital variability in CR use and advance national improvement efforts.
Keywords: Cardiac rehabilitation, cardiac surgery, CABG, SAVR
Central Picture:
Hospital-level cardiac rehabilitation use ranged from zero to near universal use:
Central Message:
In Medicare patients undergoing cardiac surgery, hospital-specific factors affected explainable interhospital variability in CR to a greater extent than patient-level or geographic factors.
Perspective:
In this national sample of Medicare beneficiaries undergoing cardiac surgery, the overall cardiac rehabilitation use varied widely across hospitals. Patient factors contributed to a small proportion of explainable variation as compared to the admitting hospital. Lower rates of adjusted 1-year outcomes were observed in higher quartiles of CR utilization.
INTRODUCTION
Cardiac rehabilitation (CR) is a multifaceted intervention that includes monitored exercise, lifestyle change education, and social support for patients experiencing acute cardiovascular events. Previous studies have demonstrated that the use of exercise-based secondary prevention programs reduce mortality and improve functional status among patients receiving coronary revascularization.1,2 CR has therefore received a Class 1A recommendation for eligible patients undergoing coronary artery bypass (CABG) and surgical aortic valve replacement (SAVR).3,4 Despite its well-studied benefits, CR remains broadly underutilized among cardiac surgical patients.5,6 Furthermore, multiple organizations have been developed to increase CR use, and while no quality metric has been published to date, the Million Hearts Cardiac Rehabilitation Collaborative has set a goal of 70% participation rate among qualifying patients.7
While studies have documented patient-level factors associated with CR use after cardiac surgery, less is known about interhospital variability in CR use and its association with outcomes. Multicenter studies have demonstrated intrahospital variation in CR use after coronary revascularization and identified the discharging hospital as an important contributor to CR use, but this observation has not been replicated on a national scale.8,9 Significant geographic variation in use has also been reported, with higher rates observed within states in the Northern/Central regions and number of facilities per 10,000 people strongly associated with CR use.10 Understanding the hospital-specific factors that contribute to interhospital variation in CR use after cardiac surgery can help stimulate quality improvement efforts to enhance uptake of this consensus guideline therapy.
This cohort study of Medicare beneficiaries undergoing CABG and/or SAVR sought to describe determinants of CR use and quantify the amount of interhospital variation in CR use that was explained by patient, organizational, and regional factors. This study also evaluated the hypothesis that Medicare beneficiaries undergoing CABG and/or SAVR at high CR-use hospitals would be associated with better clinical outcomes at one-year post-discharge.
MATERIAL & METHODS
Data Source and Sample
This study was approved by the University of Michigan’s Institutional Review Board (HUM00175541). Two authors, M.P. Thompson and H. Hou, are fully responsible for the integrity of the data and accuracy of the analysis. The dataset can not be shared with other researchers due to data use agreement restrictions, but statistical code and study protocols can be shared upon request.
Sources of data for this study included Medicare Beneficiary Summary Files (MBSF), Medicare Provider Analysis and Review (MedPAR), and professional (Carrier) files. The American Hospital Association Annual Surveys (2016) were used to provide hospital characteristics, which were linked using hospital Medicare or national provider identifier numbers. The Distressed Community Index (DCI), a composite measure of local community distress, was linked to Medicare data using beneficiary zip code data. Beneficiaries were eligible for inclusion if they had International Classification of Disease 10th edition (ICD-10) procedure codes for CABG (0210*, 0211*, 0212*, or 0213*), SAVR (02RF*), or CABG+SAVR in MedPAR claims between July 1st, 2016 and December 31st, 2018 and had continual Medicare Part A and B coverage from at least six months prior to surgery through one-year post-discharge (n =179,453 from 1,148 hospitals). Beneficiaries were excluded if they had missing information on covariates (n=5,510), died or were discharged to hospice (n=5,977), or were admitted to a hospital with low volume (fewer than 20 cases, n=1,157 beneficiaries from 126 hospitals) for a total analytic sample of 166,809 beneficiaries from 1,022 hospitals.
Cardiac Rehabilitation Use
Current Procedural Technology (CPT) codes 93797 and 93798 and Healthcare Common Procedural Coding System (HCPCS) codes G0424 and G0423 with revenue center code 943 were used to identify claims for CR utilization in Medicare MBSF and Carrier files.5 The primary CR variable in this analysis was CR use within one year of discharge, which was defined as a beneficiary having at least one CR claim within a year of discharge (yes or no). Hospital-level CR rates (and 95% confidence intervals) were estimated as the proportion of beneficiaries who attended at least one CR session within a year of discharge out of all patients treated with CABG and/or SAVR at that hospital. Hospitals were then ordered by rank from lowest to highest CR rate and grouped into four quartiles with quartile 1 representing the lowest CR use hospitals and quartile 4 representing the highest CR use hospitals. Secondary measures of CR use included the number of days between discharge and the first CR session, the total number of sessions completed, and the completion of all 36 CR sessions (yes or no).
Clinical Outcomes
The primary clinical outcomes were 1-year mortality and 1-year readmission rates. Beneficiaries were categorized as alive or dead at 1-year post-discharge (yes versus no) if they had a date of death present in the MBSF file that occurred within 365 days of discharge. Readmission within one year of discharge was defined as having an acute care hospitalization within one-year of discharge.
Covariates
This study examined multiple beneficiary, hospital, and regional-level covariates. Demographic data such as age, sex (male vs. female), race (White, Black, Asian, Hispanic, North American Native, Unknown, Other), and Medicare and Medicaid dual-eligibility status were drawn from MBSF. Clinical information from the index hospitalization was drawn from the MedPAR data and included procedure type (CABG, SAVR, CABG+SAVR), length of stay (in days), admission type (elective vs. non-elective), discharge location (home vs. extended care facility [ECF]). A Charlson Comorbidity Index (CCI) score was created for each beneficiary based on comorbidities present during the six months prior to admission.11 Claims-based frailty index (CFI), a previously validated measurement of fitness in the elderly, was tabulated and split into quartiles, with Q1 representing the least frail and Q4 representing the most frail.12,13 Claims-based methods that have been described previously were also used to identify postoperative complications, including acute kidney injury, bleeding, pneumonia, cardiac arrest, stroke, myocardial infarction, and sepsis.14 Structural characteristics of hospitals were drawn from the American Hospital Association Annual Surveys and included hospital teaching status (major, minor, or non-teaching), the presence of a system-affiliated CR facility (yes vs. no), hospital ownership type (nonprofit, for-profit, government), and hospital bed size (<100 beds, 100–299, 300–499, and 500+ beds). Regional factors from the DCI data including the Distressed Community Index (DCI) quintile (1st - prosperous, 2nd - comfortable, 3rd - mid-tier, 4th - at risk, 5th - distressed), geographic category of the beneficiary (rural, small town, suburban, urban), and region (Midwest, Northeast, South, West). The distance between the beneficiary and the nearest CR facility zip codes was estimated and categorized as within the same zip code, 0–10 miles, 11–20 miles, and 21+ miles.
Statistical Analysis
Beneficiary, structural, and regional characteristics were described at the beneficiary-level for the overall sample and by quartile of CR use. Hospital characteristics were also described at the hospital level across quartiles of CR participation. Chi-square tests and ANOVA were used to identify significant differences in categorical and continuous variables, respectively. Rates of CR use rates and 95% confidence intervals were estimated for each hospital and displayed in rank-order from lowest to highest rate of CR use.
Sequential hierarchical logistic regression models were created to quantify explainable interhospital variability. The benefit of multilevel logistic regression has been described previously.15–17 Briefly, comparing the intraclass correlation coefficients (ICC) between sequentially nested models can determine the percent of interhospital variation that was attributable to the added covariates in each level of the model. These models included a random intercept for the discharging hospital.18 The first model included the hospital-specific random intercept (Model 0) only, with sequential additions of beneficiary demographic and clinical covariates (Model 1), structural covariates (Model 2), and regional covariates (Model 3). The hospital-specific effect is meant to represent processes of care used by a given hospital that were not captured in our dataset. Each of the models was estimated in a two-stage approach. The first stage was to estimate beneficiary-level predicted probability estimates of CR use from a logistic regression model including the set of covariates specified for a specific model. The second stage was to create a hierarchical logistic regression model of CR use that included the beneficiary-level predicted probability of CR use and hospital random intercept. For each of the models (Model 0 - Model 3), the ICC was estimated from the hierarchical model using the variance estimate from the hospital random effect and can be interpreted as the percentage of interhospital variation that was attributed to the hospital level.
RESULTS
Of the 166,809 beneficiaries in this study undergoing CABG and/or SAVR, 90,171 (54.1%) participated in at least one CR session within a year of discharge. There were significant differences in patient, organizational, and regional factors across hospital CR quartiles (Table 1). Beneficiaries in low-use hospitals were more likely to be racial/ethnic minorities, have Medicaid dual eligibility, be treated in a for-profit hospital, be further away from the nearest CR facility, live in more distressed communities, and reside in regions other than the Midwest.
Table 1:
Patient-level characteristics by hospital quartile of CR use
| Quartile of CR use | ||||||
|---|---|---|---|---|---|---|
| Characteristics | Overall | Q1 - low | Q2 | Q3 | Q4 - high | p-value |
| Patients, n | 166,809 (100.0%) | 34,764 (20.8%) | 44,570 (26.7%) | 48,423 (29.0%) | 39,052 (23.4%) | |
| CR use, n (%) | 90,171 (54.1%) | 9,880 (28.4%) | 21,179 (47.5%) | 29,315 (60.5%) | 29,797 (76.3%) | <0.001 |
| Age, mean (SD) | 73.8 (5.7) | 73.7 (5.8) | 73.9 (5.7) | 73.8 (5.6) | 73.8 (5.6) | <0.001 |
| Female, n (%) | 48,433 (29.0%) | 10,566 (30.4%) | 13,084 (29.4%) | 13,828 (28.6%) | 10,955 (28.1%) | <0.001 |
| Race/ethnicity, n (%) | <0.001 | |||||
| White | 150,272 (90.1%) | 29,806 (85.7%) | 39,950 (89.6%) | 44,089 (91.0%) | 36,427 (93.3%) | |
| Black | 6,970 (4.2%) | 2,022 (5.8%) | 2,232 (5.0%) | 1,900 (3.9%) | 816 (2.1%) | |
| Asian | 1,920 (1.2%) | 711 (2.0%) | 514 (1.2%) | 408 (0.8%) | 287 (0.7%) | |
| Hispanic | 1,519 (0.9%) | 702 (2.0%) | 329 (0.7%) | 336 (0.7%) | 152 (0.4%) | |
| North American Native | 802 (0.5%) | 244 (0.7%) | 179 (0.4%) | 192 (0.4%) | 187 (0.5%) | |
| Unknown | 2,870 (1.7%) | 524 (1.5%) | 736 (1.7%) | 859 (1.8%) | 751 (1.9%) | |
| Other | 2,456 (1.5%) | 755 (2.2%) | 630 (1.4%) | 639 (1.3%) | 432 (1.1%) | |
| Medicaid dual eligibility, n (%) | 16,307 (9.8%) | 5,347 (15.4%) | 4,422 (9.9%) | 3,878 (8.0%) | 2,660 (6.8%) | <0.001 |
| Procedure type, n (%) | <0.001 | |||||
| CABG | 118,002 (70.7%) | 25,720 (74.0%) | 31,249 (70.1%) | 33,755 (69.7%) | 27,278 (69.9%) | |
| SAVR | 28,741 (17.2%) | 5,271 (15.2%) | 7,917 (17.8%) | 8,629 (17.8%) | 6,924 (17.7%) | |
| CABG + SAVR | 20,066 (12.0%) | 3,773 (10.9%) | 5,404 (12.1%) | 6,039 (12.5%) | 4,850 (12.4%) | |
| Length of stay (days), mean (SD) | 9.5 (6.6) | 10.0 (6.8) | 9.6 (6.8) | 9.5 (6.6) | 8.9 (6.0) | <0.001 |
| Transferred in, n (%) | 25,107 (15.1%) | 5,012 (14.4%) | 7,416 (16.6%) | 7,386 (15.3%) | 5,293 (13.6%) | <0.001 |
| Elective procedure, n (%) | 100,299 (60.1%) | 19,434 (55.9%) | 26,401 (59.2%) | 29,283 (60.5%) | 25,181 (64.5%) | |
| Discharge location, n (%) | 26,077 (64.7) | <0.001 | ||||
| Home | 108,856 (65.3%) | 22,558 (64.9%) | 28,763 (64.5%) | 31,613 (65.3%) | 25,922 (66.4%) | |
| ECF/other | 57,953 (34.7%) | 12,206 (35.1%) | 15,807 (35.5%) | 16,810 (34.7%) | 13,130 (33.6%) | <0.001 |
| Postoperative complications, n (%) | 39,563 (23.7%) | 8,394 (24.1%) | 10,387 (23.3%) | 11,322 (23.4%) | 9,460 (24.2%) | <0.001 |
| Charlson Comorbidity Index, n (%) | ||||||
| 0 | 23,745 (14.2%) | 4,594 (13.2%) | 6,287 (14.1%) | 6,894 (14.2%) | 5,970 (15.3%) | <0.001 |
| 1–2 | 70,594 (42.3%) | 14,503 (41.7%) | 18,873 (42.3%) | 20,330 (42.0%) | 16,888 (43.2%) | |
| 3–4 | 39,666 (23.8%) | 8,332 (24.0%) | 10,782 (24.2%) | 11,651 (24.1%) | 8,901 (22.8%) | |
| 5+ | 32,804 (19.7%) | 7,335 (21.1%) | 8,628 (19.4%) | 9,548 (19.7%) | 7,293 (18.7%) | |
| Claims-based Frailty Index (CFI) by quartile, n (%) | <0.001 | |||||
| Q1 (least frail) | 41,697 (25.0%) | 7,611 (21.9%) | 10,953 (24.6%) | 12,221 (25.2%) | 10,912 (27.9%) | |
| Q2 | 41,703 (25.0%) | 8,455 (24.3%) | 10,926 (24.5%) | 12,135 (25.1%) | 10,187 (26.1%) | |
| Q3 | 41,705 (25.0%) | 8,987 (25.9%) | 11,109 (24.9%) | 12,143 (25.1%) | 9,466 (24.2%) | |
| Q4 (most frail) | 41,704 (25.0%) | 9,711 (27.9%) | 11,582 (26.0%) | 11,924 (24.6%) | 8,487 (21.7%) | |
The median hospital-level CR participation rate for the 1,022 hospitals in this sample was 52.5%, with quartiles defined as follows: Q1: 0.0% - 38.8%, Q2: 38.8% - 54.2%, Q3: 54.2% - 67.0%, Q4: 67.0% - 96.8% (Figure 1). Hospital characteristics for the 1,022 hospitals in the sample can be seen in Supplemental Table 1. Hospitals were more likely to be nonprofit in high-use quartiles, but differences in hospital characteristics were otherwise not significant. The number of sessions completed within 1 year and the odds of completing all 36 sessions were not significantly different across quartiles; however, the duration to the first session decreased with greater-use quartiles (Table 2).
Figure 1:
Interhospital variation in CR use with patient-level CR participation rates
Table 2:
CR measures by hospital quartile of CR use
| CR Measure | CR Use Quartile | Mean (SD) or N (%) | Adjusted Difference or Odds Ratio (95% CI) | p-value |
|---|---|---|---|---|
| Days to First CR Session | Q4 - High | 41.2 (34.8) | −26.6 (−29.4, −23.9) | <0.001 |
| Q3 | 54.2 (40.2) | −14.2 (−16.8, −11.6) | <0.001 | |
| Q2 | 62.0 (44.8) | −7.8 (−10.4, −5.2) | <0.001 | |
| Q1 - Low | 68.5 (52.9) | Ref | - | |
| Number of CR Sessions Completed | Q4 - High | 26.3 (11.5) | −0.5 (−1.0, 0.1) | 0.13 |
| Q3 | 26.7 (11.5) | 0.2 (−0.4, 0.7) | 0.49 | |
| Q2 | 26.5 (11.8) | −0.3 (−0.8, 0.3) | 0.304 | |
| Q1 - Low | 26.6 (12.4) | Ref | - | |
| Completion of 36 CR Sessions | Q4 - High | 8,595 (28.9) | 0.83 (0.71, 0.96) | 0.01 |
| Q3 | 9,369 (32.0) | 1.02 (0.89, 1.18) | 0.767 | |
| Q2 | 6,705 (31.7) | 0.91 (0.79, 1.05) | 0.189 | |
| Q1 - Low | 3,297 (33.4) | Ref | - |
Models adjusted for age, sex, race, dual eligibility, procedure type, LOS elective procedure, home discharge, postoperative complications, CCI, hospital teaching status, system affiliation, ownership, bed size, distance to CR facility, DCI, region, and geographic category
Results of the hierarchical logistic regression model including all beneficiary, hospital, and regional covariates predicting patient-level CR use can be found in Supplemental Table 2. Covariates that were independently associated with increased relative odds of CR use included male sex (versus female), elective procedure status, discharge to home, non-teaching hospital status, smaller bed size categories, and living in non-urban locations. Covariates that were independently associated with lower relative odds of CR use included older age, racial/ethnic minority group status, undergoing SAVR or CABG+SAVR, presence of a complication after surgery, increasing comorbidity, for-profit hospital status, increasing distance from nearest CR facility, increasing DCI quintile, and living in a region other than the Midwest. Patient-level distribution of hospital and regional factors are found in Supplemental Table 3.
Figure 2 shows the proportion of hospital variation in CR utilization explained by each sequential model. In Model 0 (null model), 19.3% of the hospital variation in CR use was attributed to the admitting hospital. The addition of patient-level covariates in Model 1 accounted for 0.5% of the variation in CR use across hospitals. Addition of hospital-level covariates in Model 2 explained another 2.8% of the hospital-level variation in CR use. Model 3 incorporated regional factors, which accounted for 3.7% of explained variation. In total, 12.3% of the variation in CR use across hospitals was attributed to the admitting hospital, 0.5% was attributed to beneficiary-level covariates, 2.8% was attributed to hospital-level covariates, 3.7% was attributed to regional-level covariates, and 80.7% remained unexplained by the factors in the model.
Figure 2:
Explanation of hospital variance in CR rates
Rates of all-cause one-year mortality and readmissions across quartiles of CR utilization are shown in Figure 3. One-year mortality and readmission rates were 5.8% and 34.3%, respectively. Adjusted mortality rates trended downwards across quartiles (Q1 = 6.7%, Q4 = 5.2%, p<0.001), as did readmissions (Q1 = 37.6%, Q4 = 33.9%, p<0.001). Relative to beneficiaries in the lowest CR quartile hospitals and after multivariable adjustment, one-year mortality and readmissions were significantly lower for beneficiaries treated in quartiles 2 through 4 with the strongest effects occurring in quartile 4 (Supplemental Table 4). A subgroup analysis excluding patients who died within 30 days of discharge (n = 2409) shows no substantive changes to the associations between hospital CR quartile and one-year mortality and readmissions (Supplemental Table 4). Figure 4 (graphical abstract) summarizes this study’s results. Correlation coefficients between variables in the final model can be found in Supplemental Figure 1 and indicate minimal collinearity between variables.
Figure 3:
Crude and adjusted rates of clinical outcomes by hospital quartile of CR use
Figure 4:
Graphical Abstract describing study design and results
DISCUSSION
This study evaluated the factors at patient, hospital, and geographic levels that contributed to variability in CR utilization in a large, diverse sample of cardiac surgical patients eligible for CR. There were several notable findings from this study. Substantial variation in hospital-level CR use was observed across the nation, with some hospitals attaining no utilization and others achieving near universal enrollment. This finding builds on previous statewide research showing 10-fold variation in hospital-level CR participation rates for patients undergoing aortic valve replacement.19 Approximately half of all patients in the sample participated in at least one CR session, and rates of CR participation varied widely across all hospitals. These findings reflect the large discrepancies and overall low uptake in CR that still prevails throughout the US.4,20 This study found that out of the 1,022 hospitals in the sample, only 178 (17.4%) met the contemporary suggested target of 70% CR utilization.7
Patient and hospital characteristics varied between hospital-level quartiles of CR use. Hospitals that were nonprofit, relatively smaller in volume, and not major academic centers were associated with higher rates of CR utilization. While patients were more often white and male in the entire sample, these trends were more pronounced across higher-use quartiles. Conversely, the lowest hospital-level quartile had a higher proportion of patients from more urbanized, distressed communities and more often contained patients with higher comorbidity profiles. The odds of CR utilization also decreased incrementally with increasing distance between patient and CR facility zip codes (as compared with patients with CR facility in the same zip). Risk-adjusted one-year mortality and readmissions decreased incrementally from low to high use quartiles, indicating a dose-response association between hospital-level CR use and clinical outcomes. Additionally, after excluding patients who died within 30 days of discharge, adjusted differences in one-year outcomes across quartiles of CR use were similar to original estimates.
A significant finding in this study is that although traditional patient factors do indeed predict CR use at the patient level, they account for a small percentage of explainable variation between hospitals. Despite significant differences across quartiles in baseline characteristics of patients undergoing CABG and/or SAVR, variation was attributable more to care processes related to the hospitals themselves, as demonstrated by the null model. After adjustment, structural factors, such as bed size, ownership status, and teaching status still explained only a small portion of interhospital variation. Furthermore, these findings disclose regional variability that helps illustrate why some hospitals perform better than others in downstream CR use, such as distance to the nearest facility, regional level socioeconomic distress, and overall geographic region. The majority of explainable variation was attributed to the hospital-specific effect, suggesting the need for further research to elucidate the care processes unique to high-performing hospitals that lead to higher or lower CR enrollment, utilization, and completion.
This study’s findings have implications for interventions aimed at improving CR utilization. If hospitals intend to improve CR use, understanding the practice patterns of hospitals with high vs. low CR use may be more useful than addressing traditional risk factors. Furthermore, significant differences were seen across quartiles in the time it took to enroll patients in CR, yet there were negligible differences in adherence across quartiles. In fact, low CR use hospitals were found to have modestly higher completion rates, indicating that enrollment and completion must be thought of as distinct quality targets. Interventions to increase CR adherence may therefore be beneficial in both high- and low-performing facilities, as attending a greater number of sessions has been correlated directly with the magnitude of improvement in clinical outcomes.21,22 Regional quality improvement collaboratives, such as the Michigan Cardiac Rehab Network, may be well suited to identify and disseminate best-practices that lead to increased CR participation.23
Less than a fifth of the inter-hospital variability was explained by this study’s model, indicating that there are still significant gaps in knowledge that may be due to specific processes of care. Studies have shown higher levels of both referral and enrollment rates in hospitals implementing specific strategies like automatic referrals and CR liaisons.24,25 This study’s findings provide additional support for the establishment of multi-center collaboration networks to facilitate CR uptake, which has the capacity to identify barriers experienced by hospitals with low CR use.23 Additionally, a correlation was identified between hospital-level CR enrollment and one-year outcomes, which is in line with findings of a previous study.26 While the benefits of greater CR participation could be conferred to patients treated in high-performing hospitals, this finding should be interpreted as associative rather than causal and further studies are needed to identify a direct link between CR use and reduced hospital-level mortality. Nevertheless, current evidence supports increasing hospital-level CR use to improve quality and outcomes after cardiac surgery.
This study has limitations to consider. First, the Medicare data in this sample may not be generalizable to patients under 65 or insured by non-Medicare products. However, a prior study by our team has shown that only modest differences in the rates of CR use exist between Medicare and private insurance products.9 Secondly, the administrative claims dataset does not contain detailed clinical factors that may be important for risk-adjusting hospital-level rates of CR use. Claims data also do not capture referral rates; however, previous work has identified referral to CR following CABG to be greater than 90%.26 Similarly, DCI data at the zip code level lack the granularity of other measures of community distress, such as neighborhood-level data, which may lead to within-zip confounders and may further limit generalizability. Third, this study focused solely on CABG/SAVR because they represent the highest volume of cardiac surgery cases in the Medicare cohort, and future studies are needed to expand this study’s findings to other eligible conditions. Fourth, hospital-specific care processes were not available in our data and thus could not be included in the model, which may contribute to unexplained variation. Fifth, our dataset is limited to the 2016–2018 time period, and future studies should include contemporary CR practices, particularly in the wake of the decline in CR use following the Covid pandemic.27 However, we note that trends in CR use have changed very little over the past two decades, and rates have rebounded significantly (albeit incompletely) following the decline during the pandemic.27,28
CONCLUSIONS
Wide variation in hospital-level variation in CR use was observed among Medicare beneficiaries undergoing cardiac surgery, and this variation was minimally attributable to traditional patient factors that predict whether patients will use CR. Instead, these findings suggest a strong hospital component in the explainable variation, indicating the need to examine factors that drive differences in practice patterns between hospitals attaining high rates of CR use vs those with lower rates.
Supplementary Material
Supplemental Table 1: Hospital characteristics by quartile of CR use
Supplemental Table 2: Patient- and hospital-level predictors of patient-level CR use
Supplemental Table 3: Patient-level distribution of hospital and regional factors
Supplemental Table 4: Adjusted 1-year outcomes excluding patients who died within 30 days of discharge.
Supplemental Figure 1: Correlation matrix for variables in logistic regression model predicting cardiac rehab use.
ACKNOWLEDGMENTS
The Distressed Communities Index data are from the Economic Innovation Group’s Distressed Community Index files (2015–2019). The findings expressed in this publication are solely those of Michael P. Thompson and not necessarily those of The Economic Innovation Group. The Economic Innovation Group does not guarantee the accuracy or reliability of, or necessarily agree with, the information provided herein.
Sources of Funding:
This study was funded as part of a career development award for Dr. Thompson receives from the Agency for Healthcare Research and Quality (AHRQ, Grant # 1K01HS027830).
Glossary of Abbreviations:
- CABG
coronary artery bypass graft
- SAVR
surgical aortic valve replacement
- CR
cardiac rehabilitation
- CCI
Charlson Comorbidity Index
- LOS
length-of-stay
- DCI
Distressed Community Index
- ICC
intraclass correlation coefficient
- MBSF
Medicare Beneficiary Summary Files
- MedPAR
Medicare Provider Analysis and Review
Footnotes
Disclosures: Drs. Thompson, Pagani, Sukul, and Likosky receive partial salary support from Blue Cross Blue Shield of Michigan as part of the Value Partnerships portfolio. Dr. Francis D. Pagani is an ad hoc, non-compensated scientific advisor for Medtronic, Abbott, FineHeart, and CH Biomedical. Outside of this work, Dr. Donald S. Likosky receives extramural support from the Agency for Healthcare Research and Quality and the National Institutes of Health and is a consultant to the American Society of ExtraCorporeal Technology. Dr. Steven J. Keteyian receives grant support from the National Heart, Lung, and Blood Institute (R33HL143099). The views of this manuscript do not represent the VA.
Informed Consent: The University of Michigan Institutional Review Board approved this study as ‘Not Regulated’ (HUM00175541). Approved 2/3/2020
REFERENCES
- 1.Taylor RS, Brown A, Ebrahim S, et al. Exercise-based rehabilitation for patients with coronary heart disease: systematic review and meta-analysis of randomized controlled trials. Am J Med. 2004;116(10):682–692. [DOI] [PubMed] [Google Scholar]
- 2.Clark AM, Hartling L, Vandermeer B, McAlister FA. Meta-analysis: secondary prevention programs for patients with coronary artery disease. Ann Intern Med. 2005;143(9):659–672. [DOI] [PubMed] [Google Scholar]
- 3.Neravetla SR, Thourani VH. 2011 ACCF/AHA guideline for coronary artery bypass graft surgery: Executive summary: A report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines. Yearbook of Cardiology. 2012;2012:155–158. doi: 10.1016/j.ycar.2012.02.094 [DOI] [Google Scholar]
- 4.Do Simon M, Do Korn K, Cho L, Do Raymond C. Cardiac rehabilitation: a class 1 recommendation. Cleve Clin J Med. 2018;85:551. [DOI] [PubMed] [Google Scholar]
- 5.Keteyian SJ, Jackson SL, Chang A, et al. Tracking Cardiac Rehabilitation Utilization in Medicare Beneficiaries: 2017 UPDATE. J Cardiopulm Rehabil Prev. 2022;42(4):235–245. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Ritchey MD, Maresh S, McNeely J, et al. Tracking Cardiac Rehabilitation Participation and Completion Among Medicare Beneficiaries to Inform the Efforts of a National Initiative. Circ Cardiovasc Qual Outcomes. 2020;13(1):e005902. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Ades PA, Keteyian SJ, Wright JS, et al. Increasing Cardiac Rehabilitation Participation From 20% to 70%: A Road Map From the Million Hearts Cardiac Rehabilitation Collaborative. Mayo Clin Proc. 2017;92(2):234–242. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Aragam KG, Dai D, Neely ML, et al. Gaps in referral to cardiac rehabilitation of patients undergoing percutaneous coronary intervention in the United States. J Am Coll Cardiol. 2015;65(19):2079–2088. [DOI] [PubMed] [Google Scholar]
- 9.Thompson MP, Yaser JM, Hou H, et al. Determinants of Hospital Variation in Cardiac Rehabilitation Enrollment During Coronary Artery Disease Episodes of Care. Circ Cardiovasc Qual Outcomes. 2021;14(2):e007144. [DOI] [PubMed] [Google Scholar]
- 10.Suaya JA, Shepard DS, Normand SLT, Ades PA, Prottas J, Stason WB. Use of Cardiac Rehabilitation by Medicare Beneficiaries After Myocardial Infarction or Coronary Bypass Surgery. Circulation. 2007;116(15):1653–1662. doi: 10.1161/circulationaha.107.701466 [DOI] [PubMed] [Google Scholar]
- 11.Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373–383. [DOI] [PubMed] [Google Scholar]
- 12.Kim DH, Schneeweiss S, Glynn RJ, Lipsitz LA, Rockwood K, Avorn J. Measuring Frailty in Medicare Data: Development and Validation of a Claims-Based Frailty Index. J Gerontol A Biol Sci Med Sci. 2018;73(7):980–987. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Rockwood K, Song X, MacKnight C, et al. A global clinical measure of fitness and frailty in elderly people. CMAJ. 2005;173(5):489–495. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Hirji SA, Percy ED, McGurk S, et al. Incidence, Characteristics, Predictors, and Outcomes of Surgical Explantation After Transcatheter Aortic Valve Replacement. J Am Coll Cardiol. 2020;76(16):1848–1859. [DOI] [PubMed] [Google Scholar]
- 15.Merlo J, Chaix B, Ohlsson H, et al. A brief conceptual tutorial of multilevel analysis in social epidemiology: using measures of clustering in multilevel logistic regression to investigate contextual phenomena. J Epidemiol Community Health. 2006;60(4):290–297. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Leyland AH, Groenewegen PP. Apportioning Variation in Multilevel Models. Springer; 2020. [Google Scholar]
- 17.Bamdad MC, Brown CS, Kamdar N, Weng W, Englesbe MJ, Lussiez A. Patient, Surgeon, or Hospital: Explaining Variation in Outcomes after Colectomy. J Am Coll Surg. 2022;234(3):300–309. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Xian Y, Chen AY, Thomas L, et al. Sources of Hospital-Level Variation in Major Bleeding Among Patients With Non–ST-Segment Elevation Myocardial Infarction. Circ Cardiovasc Qual Outcomes. Published online 2014. Accessed September 27, 2018. https://www.ahajournals.org/doi/10.1161/CIRCOUTCOMES.113.000715 [DOI] [PubMed] [Google Scholar]
- 19.Guduguntla V, Yaser JM, Keteyian SJ, et al. Variation in Cardiac Rehabilitation Participation During Aortic Valve Replacement Episodes of Care. Circ Cardiovasc Qual Outcomes. 2022;15(7):e009175. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Doll JA, Hellkamp A, Ho PM, et al. Participation in Cardiac Rehabilitation Programs Among Older Patients After Acute Myocardial Infarction. JAMA Intern Med. 2015;175(10):1700–1702. [DOI] [PubMed] [Google Scholar]
- 21.Hammill BG, Curtis LH, Schulman KA, Whellan DJ. Relationship between cardiac rehabilitation and long-term risks of death and myocardial infarction among elderly Medicare beneficiaries. Circulation. 2010;121(1):63–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Martin BJ, Hauer T, Arena R, et al. Cardiac rehabilitation attendance and outcomes in coronary artery disease patients. Circulation. 2012;126(6):677–687. [DOI] [PubMed] [Google Scholar]
- 23.Thompson MP, Yaser JM, Forrest A, Keteyian SJ, Sukul D. Evaluating the Feasibility of a Statewide Collaboration to Improve Cardiac Rehabilitation Participation: THE MICHIGAN CARDIAC REHAB NETWORK. J Cardiopulm Rehabil Prev. 2022;42(6):E75–E81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Gravely-Witte S, Leung YW, Nariani R, et al. Effects of cardiac rehabilitation referral strategies on referral and enrollment rates. Nat Rev Cardiol. 2010;7(2):87–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Sukul D, Seth M, Thompson MP, et al. Hospital and Operator Variation in Cardiac Rehabilitation Referral and Participation After Percutaneous Coronary Intervention: Insights From Blue Cross Blue Shield of Michigan Cardiovascular Consortium. Circ Cardiovasc Qual Outcomes. 2021;14(11):e008242. [DOI] [PubMed] [Google Scholar]
- 26.Bauer TM, Yaser JM, Daramola T, et al. Cardiac Rehabilitation Reduces 2-Year Mortality After Coronary Artery Bypass Grafting. Ann Thorac Surg. 2023;116(5):1099–1105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Varghese MS, Beatty AL, Song Y, et al. Cardiac Rehabilitation and the COVID-19 Pandemic: Persistent Declines in Cardiac Rehabilitation Participation and Access Among US Medicare Beneficiaries. Circ Cardiovasc Qual Outcomes. 2022;15(12):e009618. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Peters AE, Keeley EC. Trends and Predictors of Participation in Cardiac Rehabilitation Following Acute Myocardial Infarction: Data From the Behavioral Risk Factor Surveillance System. J Am Heart Assoc. 2017;7(1). doi: 10.1161/JAHA.117.007664 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplemental Table 1: Hospital characteristics by quartile of CR use
Supplemental Table 2: Patient- and hospital-level predictors of patient-level CR use
Supplemental Table 3: Patient-level distribution of hospital and regional factors
Supplemental Table 4: Adjusted 1-year outcomes excluding patients who died within 30 days of discharge.
Supplemental Figure 1: Correlation matrix for variables in logistic regression model predicting cardiac rehab use.