Intensive versus Traditional Cardiac Rehabilitation: Mortality & Cardiovascular Outcomes in a 2016–2020 Retrospective Medicare Cohort

Abstract

Background:

Traditional cardiac rehabilitation (CR) improves cardiovascular outcomes and reduces mortality, but less is known about the relative benefit of intensive cardiac rehabilitation (ICR) which incorporates greater lifestyle education through 72 sessions (vs 36 in CR). Our objective was to determine whether ICR is associated with a mortality and cardiovascular benefit compared to CR.

Methods:

Retrospective cohort study of Medicare Fee-For-Service beneficiaries in a 100% sample, claims dataset. Qualifying events were captured from 5/1/2016 to 12/31/2019 and ICR/CR utilization captured from 5/1/2016 to 12/31/2020. Among patients attending at least one day of either CR or ICR, Cox proportional hazards models utilizing a 1-to-5 propensity score match were used to compare utilization and the association of ICR vs. CR participation with 1) all-cause mortality and 2) cardiovascular-related hospitalizations (CVH) or non-fatal cardiac events. Dose-response was assessed by the number of days attended.

Results:

1,277,358 unique patients met at least one qualifying indication for ICR/CR from 2016 to 2019. Of these, 262,579 (20.6%) and 4,452 (0.4%) attended at least one session of CR or ICR, respectively (mean [SD] age, 73.2 [7.8] years; 32.3% female). In the matched sample, including 26,659 total patients (median 2.4-year follow-up), ICR was associated with 12% lower all-cause mortality (multivariable-adjusted hazard ratio, 0.88; 95%CI, 0.78–0.99, p=0.036) compared to CR but no significant difference for CVH or non-fatal cardiac events. The mortality benefit was seen for both ICR and CR per day strata, with each modality demonstrating a clear dose-response benefit.

Conclusions:

ICR is associated with lower mortality than traditional CR among Medicare beneficiaries but no difference in CVH or non-fatal cardiac events. Moreover, ICR and CR demonstrate a dose-response relationship for mortality. Additional studies are needed to confirm these observations and to better understand the mechanisms by which ICR may lead to a reduction in mortality.

INTRODUCTION

Cardiac rehabilitation (CR) is a program of structured, supervised exercise and wellness sessions aimed at improving health following a cardiac event. Traditional CR programs have been shown to improve cardiovascular outcomes, including mortality, for up to five years^1–5 in patients who have had a prior myocardial infarction or percutaneous coronary intervention. After coronary artery bypass surgery (CABG), the survival benefit is seen for up to 10 years.⁶ Even with CR, however, morbidity and mortality rates remain high among patients with established cardiovascular disease.

Intensive cardiac rehabilitation (ICR) programs were developed to expand the benefits of traditional CR by incorporating additional sessions comprised of exercise, education, and/or cohort sessions focusing on nutrition, lifestyle behaviors and stress management.^7–9 In 2010, the Centers for Medicare and Medicaid Services (CMS) approved the outpatient Pritikin Comprehensive Lifestyle Change Program and the Ornish Reversal Program for reimbursement based on achieving reductions in cardiovascular risk factors, reductions in need for cardiovascular medications, and improvements in coronary artery disease progression/outcomes.⁷ In 2014, the Benson-Henry Institute Cardiac Wellness Program became the third CMS-approved ICR program.¹⁰

Despite the growth of these programs, there has not been a nationwide analysis on a large sample, such as the 100% Medicare database, regarding utilization and outcomes of ICR. Understanding the utilization rates and potential outcome benefits of outpatient ICR relative to CR in a real-world setting is critical to determining the value-added effects of ICR as compared to traditional CR. Therefore, the intent of this work was to 1) describe utilization of ICR and CR in a real-world national population of Medicare beneficiaries; 2) compare clinical outcomes between beneficiaries participating in ICR versus CR; and 3) explore if any observed benefit associated with ICR demonstrates a dose-response relationship.

METHODS

Data Source & Population

Using the Centers for Medicare & Medicaid Services’ (CMS) Virtual Research Data Center (VRDC), we analyzed Medicare fee-for-service claims data from 2016 to 2020 using the 100% national sample. Qualifying events were from 5/1/2016 to 12/31/2019 and ICR/CR utilization from 5/1/2016 to 12/31/2020. Qualifying events were identified by specific International Classification of Disease (ICD-10-CM) codes in the primary diagnostic position, procedural codes, or current procedural terminology (CPT) codes from inpatient or outpatient files (Table S1). Comorbidities and demographic information were obtained using the 100% Master Beneficiary Base File and the 100% Master Beneficiary Summary File: Chronic Conditions Segment, which collates 27 chronic conditions. Patients’ socioeconomic status was determined by dual enrollment within Medicaid, a marker for poverty, and the Social Vulnerability Index (SVI) created by the Centers for Disease Control (CDC)/Agency for Toxic Substances and Disease Registry (ATSDR). Hospital characteristics, including hospital size, teaching status, hospital ownership, location, and hospital region were obtained from the American Hospital Association Annual Survey (2018). A previously validated claims-based frailty index was also utilized to categorize beneficiaries as robust, pre-frail, or frail.^11,12

Beneficiaries were included if alive > 30 days after their index event and had continuous enrollment in Medicare Parts A and B for >12 months after their index event. Exclusion criteria included: 1) enrollment in Medicare Advantage, 2) participation in ICR/CR within 18 weeks prior to the qualifying event, or 3) discharge to a facility for those with inpatient qualifying events. The analysis included only the first qualifying event.

The methods that support the findings of this study are available from the corresponding author upon reasonable request. The VRDC data and other materials are unable to be shared given CMS restrictions.

Qualifying Indications

Beneficiaries were considered eligible for outpatient ICR/CR based on Medicare guidelines: 1) hospitalization for acute myocardial infarction (AMI), 2) structural cardiac procedures, such as coronary artery bypass grafting (CABG), heart valve repair or replacement (surgical or percutaneous), heart or heart-lung transplant, 3) percutaneous coronary intervention (PCI), 4) stable angina, and 5) heart failure with reduced ejection fraction (i.e., left ventricular ejection fraction [LVEF] ≤35%) with no documented heart failure hospitalizations/procedures within the preceding 6 weeks. As Medicare administrative data do not contain certain clinical data (e.g., LVEF), approximate diagnostic criteria were utilized where appropriate, as described previously.⁵

ICR and CR Participation and Utilization

Participation in ICR or CR was defined as ≥1 Medicare fee-for-service claim for either ICR (HCPCS G0422 [with exercise] or G0423 [without exercise]) or CR (CPT 93798 [with continuous EKG monitoring] or CPT 93797 [without EKG monitoring]) within 12 months after the qualifying event. Time to participation was defined as the number of days from the event until rehabilitation initiation. Distance to center was defined as the distance in miles from the patient’s home ZIP code to the ZIP code of the ICR or CR location.

Mortality and Cardiovascular Outcomes

The primary outcome was all-cause mortality to the end of follow-up. The secondary outcome was a non-fatal major adverse cardiovascular event (MACE), defined as any of the ICR/CR eligible criteria or any cardiovascular-related hospitalization (Table S2). For each analysis, the time 0 was the date of the first CR/ICR claim. A sensitivity analysis for the primary analysis was also completed utilizing a 1-year from qualifying event landmark.

A dose-response relationship of ICR/CR with mortality was explored by defining dose as the number of days with an ICR/CR claim. The number of days attended is less biased than the number of claims as claims vary per day between the major ICR providers, namely Ornish and Pritikin (Ornish tends to have a higher number of claims for the same day compared to Pritikin). Furthermore, the duration of treatment over time, indicated by the number of days, better reflects the true dose of CR given multiple visits over a span of days reinforce the exercise and non-exercise components. A sensitivity analysis using the number of claims as a representation of dose was completed. For these dose-response analyses, only beneficiaries alive 1 year after their qualifying event were included to isolate the effect of ICR/CR utilization.

Statistical Analysis

Categorical variables are reported as frequency (percentage); continuous variables are reported as mean (standard deviation) for age and median (IQR) for number of comorbidities and SVI. Due to the large sample size, patients participating in ICR or CR were compared by standardized differences, with an absolute difference greater than 0.1 considered clinically significant. A propensity score was calculated using a logistic model predicting ICR vs CR. The initial non-parsimonious model included age, sex, race, qualifying event type, patient’s geographic region, Medicaid enrollment, SVI, and all Elixhauser comorbidities which includes cancer, psychiatric condition, anemia, asthma, atrial fibrillation, chronic kidney disease, chronic obstructive lung disease, diabetes, glaucoma, hyperlipidemia, hypertension, hypothyroidism, ischemic heart disease, osteoporosis, rheumatoid/osteoarthritis, and stroke/transient ischemic attack. The final model was selected by backward elimination. Sex was forced into the model; the remaining variables were included if the p-value was less than 0.10. The final backward selection model included age, sex, qualifying event type, patient’s geographic region, Medicaid enrollment, SVI, anemia, atrial fibrillation, chronic obstructive pulmonary disease, glaucoma, hypertension, hypothyroidism, and ischemic heart disease. The probability of attending ICR based on this final model was used as the propensity score.

The primary analysis used 1-to-5 matching on the propensity score using a greedy matching algorithm with a caliper of 0.25 standard deviation of the logit of the propensity score.¹³ In addition, exact matching on event type and discharge type (outpatient claim vs home vs home health care) was performed. Matching quality was assessed by standardized differences (Figure S1). A sensitivity analysis used inverse probability weights (IPW) to compare participants in CR to ICR.

In a univariate Chi-square test, the rates of three types of events (death, non-fatal MACE, no event) were compared between ICR and CR in the 1:5 matched subset. The difference in total number of rehabilitation days (collapsed into 4 categories; 36 days, 19–35 days, 10–18 days, and < 10 days) was compared between ICR and CR using Chi-square analysis. Kaplan-Meier plots obtained from the adjusted Cox proportional hazards model were used to compare ICR vs. CR with respect to the primary outcome and main secondary outcome. A multilevel Cox proportional hazards model for ICR and CR was fit, adjusting for within-state correlation. In the sensitivity analysis utilizing IPW, the weights were standardized and ties were handled using the Breslow method. Age, sex, race, SVI, and ICR vs CR were forced into the model. The other variables offered into the models included event type, concomitant Medicaid enrollment, geographical region, frailty score (sensitivity analysis only), and all Elixhauser comorbidities. The final models were selected by backward elimination and variables with p-value less than 0.01 (due to the size of the dataset) were retained.

Proportionality assumptions were tested. We performed several sensitivity analyses. A landmark sensitivity analysis was performed including only patients who survived for 1 year after their initial qualifying indication occurred. Since timing of stable angina and chronic heart failure as eligibility criteria for cardiac rehabilitation are often not clear within administrative datasets compared to the other qualifying indications, a sensitivity analysis was performed excluding these two qualifying indications. To evaluate any potential effect of the SARS-CoV-2 pandemic, a sensitivity analysis excluding 2020 was completed with qualifying events from 5/1/2016 to 12/31/2018 and ICR/CR utilization from 5/1/2016 to 12/31/2019. A sensitivity analysis including frailty was performed, excluding patients from 2016 because 6 months of pre-event claims are needed to calculate a frailty index. In addition, several pre-specified stratified analyses of subgroups were performed (e.g., <75 versus ≥75 years of age, sex (M/F), diabetes (present/absent), qualifying indication, and frailty category).

All analyses were performed within the Medicare VRDC using SAS Enterprise Version 7.15 and summaries were downloaded to R statistical software version 4.5 for graphs. A p-value of <.05 was considered significant. Descriptions of the statistical methods utilized in this analysis can be found in the review by Rossello and González-Del-Hoyo.^14,15 The study was approved by the Washington University in St. Louis Institutional Review Board; the requirement for informed consent was waived due to the de-identified nature of the data.

RESULTS

Study Population

There were 1,277,358 unique patients with at least one qualifying indication for cardiac rehabilitation from 2016 to 2019 who met inclusion criteria. 20.6% (N = 267,579 patients) of these eligible patients completed at least one session of ICR/CR and are thus the cohort for the outcomes analysis. Among the 19.5% that were propensity matched, mean age was 73.2 ± 7.8 years, 32.5% were female, and 91% were Caucasian (Table 1). Among beneficiaries who participated in either ICR or CR, structural cardiac procedures were the most common qualifying event, followed by AMI, chronic heart failure, PCI, and stable angina. Participating beneficiaries most commonly resided in the South, followed by the Midwest, Northeast, and the West (Table 1). Among beneficiaries with information available to incorporate frailty as a covariate, 20.6% were categorized as robust, 74.6% as pre-frail, and 4.8% as frail (Table S3). For beneficiaries with a hospital-based qualifying event, a majority of the cohort (89%) were admitted to an urban hospital (Table S4).

Table 1.

Patient Characteristics in Matched Analysis

	Total Cohort	ICR	CR	Standardized difference
	N = 26,659	N = 4,452	N = 22,207
Age	73.2 (7.8)	73.2 (7.7)	73.1 (7.8)	0.01
Female	32.5%	32.5%	32.4%	0.00
Race				−0.03
White	90.6%	91.2%	90.5%
Black	4.9%	4.5%	5.0%
Other	4.5%	4.2%	4.5%
Medicaid	5.1%	5.0%	5.2%	−0.01
# Comorbidities	7(6,9)	7(6,9)	7(6,9)	−0.02
Social Vulnerability Index (percentiles)	0.4 (0.2,0.6)	0.4 (0.2,0.6)	0.4 (0.2,0.6)	−0.04
Qualifying Event				0.00
Acute Myocardial Infarction	42.7%	42.7%	42.8%
Cardiac Surgery	46.5%	46.5%	46.6%
Heart Failure	4.1%	4.1%	4.1%
Percutaneous Coronary Intervention	2.8%	2.8%	2.8%
Stable Angina	3.8%	4.0%	3.7%
Outpatient Qualifying Event	6.4%	6.5%	6.3%	0.00
Region				0.32
Midwest	30.0%	30.5%	29.9%
Northeast	6.1%	6.2%	6.1%
South	57.4%	57.0%	57.5%
West	6.4%	6.3%	6.5%
Comorbidities
Anemia	48.8%	48.6%	48.9%	0.00
Asthma	8.4%	8.6%	8.3%	−0.01
Atrial Fibrillation	25.0%	25.3%	25.0%	−0.01
Cancer	13.4%	13.1%	13.4%	0.01
Chronic Kidney Disease	47.0%	46.3%	47.1%	0.02
Chronic Obstructive Pulmonary Disease	19.8%	19.9%	19.8%	0.00
Diabetes	42.7%	42.2%	42.9%	0.01
Glaucoma	8.6%	8.7%	8.6%	0.00
Hip fracture	0.3%	0.4%	0.3%	0.01
Hyperlipidemia	92.8%	92.3%	92.9%	0.02
Benign prostatic hyperplasia	20.1%	19.0%	20.3%	0.03
Hypertension	94.1%	94.4%	94.1%	−0.01
Hypothyroidism	22.0%	21.9%	22.0%	0.00
Ischemic Heart Disease	97.6%	97.7%	97.6%	−0.01
Osteoporosis	6.2%	6.0%	6.2%	0.01
Psychiatric Conditions	24.7%	25.6%	24.6%	−0.02
Rheumatoid or Osteoarthritis	42.2%	42.4%	42.2%	0.00
Stroke or Transient Ischemic Attack	6.7%	6.5%	6.8%	0.01
Distance to CR/ICR center (miles)	6.5 (2.6,13.4)	7.9 (3.9,15.2)	6.3 (2.3,13)	0.19
Time from index event to initiation of rehabilitation (days)	37 (22,59)	34 (20,59)	37 (23,59)	−0.05

Continuous variables (except age) are shown as median (interquartile range). Categorical variables are shown as number (%). Standardized differences > 0.1 or < −0.1 considered clinically significant. Social Vulnerability Index interpreted as “x” proportion of counties are more vulnerable in the nation (lower the score, the more vulnerable). Abbreviations: CR = cardiac rehabilitation, ICR = intensive cardiac rehabilitation

ICR and CR Utilization

Among all patients with a qualifying event, 0.4% of patients completed at least 1 session of ICR and 20.6% of patients completed at least 1 session of CR within one year. Within the matched analysis, ICR participants had a similar rate of dual Medicaid enrollment (ICR 5.0%, CR 5.2%) and a similar number of comorbidities compared with CR participants (median [IQR] ICR 7 [6,9], CR 7 [6,9]). While the median patient travel distance to ICR locations was longer compared to CR locations (ICR 7.9 miles [3.9, 15.2], CR 6.3 miles [2.3, 13]), there was no significant difference in time to first session (ICR 34 days [20, 59], CR 37 days [23, 59]).

ICR and CR Outcomes

Over a median of 2.4 (1.5, 3.4) years of follow-up among all patients, 24,888 Medicare beneficiaries died (ICR 322 vs CR 24,566) and 26,820 had non-fatal MACE (ICR 428 vs CR 26,392). In unadjusted analyses, participation in ICR significantly was associated with a lower rate of death (3.1 [2.8,3.5] vs 3.6 [3.4,3.7] per 100 patient years, p = 0.036) but not a lower rate of non-fatal MACE (4.7 [4.3,5.2] vs 4.5 [4.3,4.7] per 100 patient years, p = 0.45). In multivariable Cox proportional hazards analyses, ICR was associated with lower all-cause mortality (HR 0.88, 95% CI 0.78–0.99, p=0.036, Figure 1A, Table 2), with a number needed-to-treat of 59, but no difference in MACE (HR 1.05, 0.94–1.17, p=0.41, Figure 1B). Furthermore, in those patients who survived at least 1 year (to avoid immortal time bias), ICR remained an independent predictor of survival (HR = 0.83, 0.72–0.95) even after adjusting for numbers of days of CR or ICR (i.e., 10–18 days vs. 19–35 days vs. 36+ days).

Figure 1. Adjusted Survival and Non-Fatal major adverse cardiac event (MACE) Analysis, Medicare Beneficiaries Participating in ICR versus CR, 2016–2020.

5-to-1 matched survival analysis (A) and non-fatal MACE analysis (B). There was a significant survival advantage of ICR vs. CR in terms of survival (p < 0.036), but not in non-fatal MACE (p = 0.41).

Abbreviations: CR = cardiac rehabilitation, ICR = intensive cardiac rehabilitation

Table 2:

Risk of Morality by Phenotypic Variables in a Multivariable Analysis Among Medicare Beneficiaries Participating in CR or ICR, 2016–2020

	Hazard Ratio	Lower 95% Confidence Interval	Upper 95% Confidence Interval	P value
ICR vs CR	0.88	0.78	0.99	0.04
Age, per year	1.06	1.05	1.06	<0.001
Female vs Male	0.80	0.73	0.88	<0.001
Race vs White				<0.001
Black	1.57	1.33	1.86	<0.001
Other	0.91	0.72	1.15	0.42
Medicaid Enrollment	1.65	1.39	1.97	<0.001
Qualifying Event vs Acute Myocardial Infarction
Cardiac Surgery	0.68	0.62	0.75	<0.001
Chronic Heart Failure	1.75	1.50	2.04	<0.001
Percutaneous Coronary Intervention	1.19	0.96	1.48	0.11
Stable Angina	0.72	0.55	0.94	0.02
Cancer	1.29	1.16	1.43	<0.001
Psychiatric Conditions	1.53	1.40	1.67	<0.001
Anemia	1.49	1.36	1.62	<0.001
Atrial Fibrillation	1.41	1.29	1.54	<0.001
Chronic Kidney Disease	1.68	1.53	1.85	<0.001
Chronic Obstructive Pulmonary Disease	1.81	1.66	1.98	<0.001
Diabetes	1.40	1.28	1.53	<0.001
Stroke or Transient Ischemic Attack	1.31	1.15	1.50	<0.001

Abbreviations: CR = cardiac rehabilitation, ICR = intensive cardiac rehabilitation

Findings were similar in all sensitivity analyses, as well as in the inverse probability weighted dataset and in the landmark analysis at 1 year after the qualifying event. Exclusion of stable angina and heart failure as qualifying events did not materially alter the results. Exclusion of patients in 2020 also resulted in similar findings. Findings were consistent across prespecified subgroups (Figure 2), with no significant interactions within the groups (e.g., ICR did not have significantly different effect on patients with diabetes vs. no diabetes).

Figure 2. Association of ICR with Mortality in Selected Subgroups.

Adjusted subgroup analysis from 2016 to 2019 (A) and frailty subgroup analysis from 2017 to 2020 (B). None of the interaction p-values were significant, sex (p = 0.85), age (p = 0.37), diabetes (p = 0.38), qualifying event (p = 0.27), and frailty (p = 0.08).

Abbreviations: CR = cardiac rehabilitation, ICR = intensive cardiac rehabilitation

Dose Response in ICR and in CR groups

There was a significant difference in the distribution of days attended between ICR and CR (p < 0.001, Table S5). ICR attendees were more likely to complete 36 days of attendance but were also more likely to attend fewer than 10 days. Among beneficiaries who were alive 1 year after the qualifying event, we found a dose-response mortality benefit for both ICR and CR. In a multivariate Cox proportional hazards model for ICR alone, there was a stepwise reduction in mortality for 10–18 days of participation (HR 0.65, 0.45–0.95), 19–35 days of participation (HR 0.63, 0.45–0.88), and a full 36 days of participation (HR 0.52, 0.37–0.73), p = 0.0012, Figure 3A) compared to 1–9 days of participation. For CR alone, there was a similar stepwise reduction in mortality for 10–18 days of participation (HR 0.88, 0.75–1.02), 19–35 days of participation (HR 0.67, 0.59–0.76), and a full 36 days of participation (HR 0.60, 0.52–0.69, p<0.0001, Figure 3B) compared to 1–9 days of participation. A sensitivity analysis utilizing the number of claims, as opposed to numbers of days, as a representation of dose yielded similar hazard ratios in multivariable analysis (Table S6).

Figure 3. Adjusted Dose-Response Relationship of ICR and CR with Mortality.

Adjusted analysis with 1–9 days as the baseline comparator. Patients included if alive 1 year after first ICR/CR claim. ICR adjusted for age, sex, race, and comorbidities. CR adjusted for age, sex, race, qualifying indication, Medicaid enrollment, geography, and comorbidities. For both ICR (A) and CR (B) there was a stepwise reduction in mortality for 10–18 days, 19–35 days and 36 days of participation (p<0.0001 for all), compared to 1–9 days of participation.

DISCUSSION

In this study, we report two major findings. First, compared to CR, ICR is associated with a 12% lower all-cause mortality rate. In this analysis, however, ICR and CR were associated with similar rates of MACE. These major findings were consistent across subgroups and sensitivity analyses. Second, we demonstrated a dose-response relationship in which higher ICR or CR attendance was related to lower mortality. Consistent with prior findings, we also demonstrated that a low percentage of patients use CR and an even lower percentage use ICR in a nationwide 100% Medicare sample.

Indeed, the primary finding of this study is that attendance at ICR was associated with better survival than attendance at CR over a median of 2.4 years of follow-up, which is particularly noteworthy given the known impressive benefits of CR on survival.^1,2,6,16 Patients attending CR or ICR generally receive similar type and length of exercise interventions. The extra sessions for ICR (done on the same days as the exercise sessions) are devoted to mental health, diet education, smoking cessation, mindfulness activities, and support groups. Smaller, nonrandomized, retrospective and prospective studies have shown differences between ICR and CR.^9,17 For example, a small observational study conducted prior to CMS’s approval of outpatient ICR demonstrated reduced hospitalization rates and trends towards reduced Medicare costs and reduced all-cause mortality with intensive lifestyle modification programs that were the forebears of current ICR programs.⁹ More recent studies have demonstrated that patients in ICR had better post-rehabilitation scores on fitness, physical function, diet quality, and greater decreases in body mass index compared with those in CR.^17,18, While these benefits could explain, at least in part, the improved survival seen with ICR over CR, future studies are needed to explore these effects.

The survival advantage of ICR can be quantified by the ‘number needed to treat’ (NNT). For the primary outcome of all-cause mortality, 59 patients would need to be treated with ICR to save 1 life in comparison with CR. For context, this NNT is similar to those reported in recent cardiovascular outcomes trials for GLP-1 receptor agonists and SGLT-2 inhibitors, which had NNTs of 45–104 for a compositive primary outcome of mortality and MACE.¹⁹ This survival advantage needs to be confirmed in a randomized trial comparing ICR to CR. In the interim, expanding access to ICR should be a priority.

There was evidence of a dose-response relationship, with a higher number of ICR days being associated with better clinical outcomes. To our knowledge, this is the first study to examine the effect of ‘dose’ of ICR on mortality, though previous studies have aimed to define a specific dose of CR needed to achieve the most benefit. A meta-analysis of CR suggested that at least 12 sessions should be completed to reduce all-cause mortality.²⁰ Additionally, some studies suggest that there may not be an upper bound of benefit related to CR, indicating that more sessions may continuously be associated with decreasing MACE rates.²¹ We similarly found progressively greater benefit up to the maximum number of allowed sessions for both ICR and CR. The findings of the post-hoc sensitivity analysis in Table S6, the number of claims as a representation of dose, suggest that the benefits of ICR over CR are likely, in part, a function of more sessions (i.e., the hazard ratios for CR and ICR when matched for number of sessions are similar). There are limitations, however, to the strength of the conclusions drawn from these findings given 1) claims as a representation of dose was a post-hoc sensitivity analysis, 2) there are qualitative differences regarding what is included in each session of ICR and CR, and 3) there are major differences in how sessions are apportioned within different program (as described in Methods), biasing this claims-based analysis. These findings underscore the need for additional study into how to improve access, insurance coverage, and broader implementation of and adherence to both CR and ICR.

What is less clear is why there was a reduction in overall mortality but not in non-fatal MACE. ICR has been approved by Medicare since 2010 based on observed reductions in cardiovascular disease risk factors (low density lipoprotein cholesterol, triglycerides, body mass index, systolic blood pressure, and diastolic blood pressure), reductions in need for medications (cholesterol, blood pressure, diabetes), and improvements in coronary artery disease progression/outcomes.^9,22,23 These improvements in cardiovascular health would be expected to improve both mortality and cardiovascular events. Indeed, one non-randomized, unadjusted, retrospective single-site study of ICR demonstrated an improvement in MACE vs CR (11% vs. 17%) within 12 ± 4.8 months.²⁴ One possibility is that MACE among Medicare patients in ICR were less likely to be fatal, perhaps suggesting a gain in resiliency associated with ICR participation. Another possibility is that, due to the enhanced education, patients may have been more likely to seek medical attention earlier when symptoms developed. This would have led to more CV hospitalizations while likely reducing mortality. A post-hoc analysis evaluating all-cause hospitalizations support this possibility. When determining the effect of cardiac vs non-cardiac related hospitalization, ICR utilization was associated with a reduction in non-cardiac hospitalizations. In competing risk models using the Fine-Gray method, cardiac hospitalization was not significantly reduced HR 1.03 (0.92–1.15, p = 0.612) but non-cardiac related hospitalizations were, HR 0.911 (0.86–0.97, p = 0.002). These results were consistent when cause-specific hazards models were employed instead of Fine-Gray, censoring the other type of hospitalization. Thus, the lower rate of non-MACE hospitalizations in the ICR group may have contributed to the lower mortality; however, additional research is needed to examine this hypothesis. There are also multiple other benefits that have been documented with outpatient ICR that could theoretically improve mortality without affecting MACE, such as improvements in fitness, physical function, depression (especially in patients with moderate-severe depression), cardiac self-efficacy, diet quality, and body mass index.^25,26 Improvements in overall mortality might also be via improvements in other obesity- and diet-related diseases.²⁷ Future studies are needed to further investigate this potential connection.

The low utilization rates of CR have been extensively documented; recently ICR has also been shown to be underutilized and not widely available.²⁸ Despite well-known improvements in health outcomes and quality of life,^1,2,29–34 only 12–24% of eligible Medicare beneficiaries attend at least one session of CR.^1,4,5 There are also well-documented variations in utilization by sex, race, age, and socioeconomic status, with consistently lower utilization among women and beneficiaries who were non-white, older, and dually enrolled in Medicare and Medicaid.^1,4,35 We previously reported low use of both ICR and CR in a much smaller (5%) Medicare sample from 2012 to 2016,²⁸ with only 1 in 1,000 qualifying beneficiaries attending even one session of ICR and less than one in six attending CR, despite there being a Class IA indication for ICR and CR for the above-listed, Medicare-approved conditions except chronic heart failure, which carries a Class 2A recommendation. Here, we confirmed those findings in a more contemporary and larger, nationally-representative sample, suggesting only minor progress has been made in improving access and increasing uptake of ICR or CR, despite their well-studied benefits and efforts to increase their utilization and days of attendance, such as early referral and automatic order sets.³⁶

Limitations

There are limitations to the present study. This is an observational, non-randomized study; thus, there is a risk of residual confounding. For events that lack extreme event proportions, a set of sensitivity analyses can assess the presence and direction of a relationship, even in the face of an unmeasured confounder.³⁷ Reassuringly, numerous pre-specified sensitivity analyses and incorporation of socioeconomic factors have demonstrated consistent findings in the direction and magnitude of the effect of ICR on mortality.

To further evaluate the effects of unmeasured confounding, we have determined the E-value for the mortality analysis.³⁸ The E-value represents the minimum strength of association that an unmeasured confounder would need to have with both the treatment and the outcome to fully explain away a specific treatment–outcome association, conditional on the measured covariates. In this analysis, the E-value for mortality was 1.53 (CI 1.1 – NA), thus implying that a confounding variable would need to be associated with both the exposure (ICR/CR) and mortality by a factor of approximately 1.53 to nullify the observed association. Notably, all differences between the ICR and CR groups for the measured variables, as shown in Table 1, have risk ratios of less than 1.5, with most being less than 1.1. Therefore, the E-value of 1.53 suggests that confounding variables are unlikely to explain the association with mortality in this analysis.

We acknowledge that there is not a readily apparent explanation for the observed favorable association between ICR and mortality but no association with non-fatal MACE. Despite studies of CR demonstrating reduced MACE relative to no CR, there are no high-quality prior studies comparing ICR to CR with respect to MACE. There is a risk of residual confounding that we attempted to minimize with numerous statistical methods, sensitivity analyses and the inclusion of > 150 ICD-10 codes to capture MACE events. We also found no clinically significant differences in the any of the comorbidities or markers of socioeconomic status (e.g., dual enrollment in Medicaid and SVI) which therefore fails to explain the apparent discordance between the mortality and MACE outcomes.

ICR requires a greater time commitment compared to CR and may not be a feasible option for many Americans, especially those who are working full-time or have other personal obligations. In clinical practice, however, the decision on ICR versus CR is often dictated by insurance coverage and geographic access to ICR. There is also debate about the relative importance and mechanism of benefit of exercise vs. non-exercise sessions, a topic outside the scope of this manuscript.

Administrative data lack clinical granularity (e.g., LVEF, laboratory data, smoking history, markers of frailty such as gait speed or handgrip strength). The dataset is restricted to Medicare Part A & Part B and does not include patients with Medicare Advantage or patients with private insurance. Thus, the generalizability of our findings to younger, privately insured individuals is unclear.

We were unable to distinguish between cardiovascular and non-cardiovascular mortality or to quantify ICR and CR referrals, as this information is not captured in claims data. Thus, we cannot determine the extent to which non-participation was due to lack of referral or lack of attendance among patients who were appropriately referred. Furthermore, smoking is an important driver of failure to enroll and non-adherence. Given that smoking is not adequately captured in claims, is often under-coded, and lacks time specificity, it was not included within the analysis.

It is possible that the number of beneficiaries with heart failure who qualified for CR was overestimated, since referral to CR in patients with heart failure requires LVEF ≤ 35%, whereas the ICD code for systolic heart failure includes a broader range of LVEFs. Similarly, adjudication for chronic stable angina within claims data is difficult. Reassuringly, sensitivity analysis excluding both heart failure and stable angina did not materially alter our main results. Finally, our study period included the SARS-CoV-2 pandemic, which decreased participation in CR broadly and likely influenced rates of death from cardiovascular disease due to both the competing risk of death from SARS-CoV-2 and effect on health systems (e.g., avoidance or over capacity) at the peak of the pandemic.

CONCLUSIONS

ICR has a very low utilization rate among Medicare beneficiaries with a qualifying indication, but its use is associated with lower mortality than traditional CR. Strategies to improve access to ICR programs may hold potential for improving long-term outcomes in patients with cardiovascular disease. Additional studies are needed to confirm these observations in other populations, including a randomized trial of ICR versus CR, and to better understand the mechanisms whereby ICR may lead to a reduction in mortality.

What is Known:

• Traditional cardiac rehabilitation (CR) has been approved by the Centers for Medicare & Medicaid (CMS) since 1982 and has been shown to reduce mortality.

• Intensive cardiac rehabilitation (ICR) and CR have both been underutilized.

What the Study Adds:

• In a large, longitudinal, and national retrospective analysis, ICR was associated with 12% lower all-cause mortality over a median 2.4 years of follow-up. The direction and magnitude of benefit was similar across numerous pre-specific sensitivity analyses.

• There was no difference in non-fatal cardiac events or cardiovascular hospitalizations between ICR and CR.

• For both ICR and CR, attendance at a higher number of sessions was associated with lower mortality.

ABBREVIATIONS

AMI

acute myocardial infarction

ATSDR

Agency for Toxic Substances and Disease Registry

CABG

coronary artery bypass graft

CDC

Centers for Disease Control

CMS

Centers for Medicare & Medicaid Services

CPT

current procedural terminology

CR

cardiac rehabilitation

ICD-CM

International Classification of Disease – Clinical Modification

ICR

intensive cardiac rehabilitation

LVEF

left ventricular ejection fraction

MI

myocardial infarction

PCI

percutaneous coronary intervention

SVI

Social Vulnerability Index

VRDC

Virtual Research Data Center