Author + information
- Received July 20, 2017
- Revision received October 10, 2017
- Accepted November 3, 2017
- Published online January 30, 2018.
- Justin M. Bachmann, MD, MPHa,∗ (, )
- Meredith S. Duncan, MAa,
- Ashish S. Shah, MDb,
- Robert A. Greevy Jr., PhDc,
- JoAnn Lindenfeld, MDa,
- Steven J. Keteyian, PhDd,
- Randal J. Thomas, MD, MSe,
- Mary A. Whooley, MDf,
- Thomas J. Wang, MDa and
- Matthew S. Freiberg, MD, MSca
- aDivision of Cardiovascular Medicine, Department of Medicine, Vanderbilt University Medical Center, Nashville, Tennessee
- bDepartment of Cardiac Surgery, Vanderbilt University Medical Center, Nashville, Tennessee
- cDepartment of Biostatistics, Vanderbilt University Medical Center, Nashville, Tennessee
- dDivision of Cardiovascular Medicine, Henry Ford Hospital, Detroit, Michigan
- eDepartment of Cardiovascular Diseases, Mayo Clinic, Rochester, Minnesota
- fMeasurement Science Quality Enhancement Research Initiative, Department of Veterans Affairs, San Francisco, California
- ↵∗Address for correspondence:
Dr. Justin M. Bachmann, Division of Cardiovascular Medicine, Department of Medicine, Vanderbilt University Medical Center, 2525 West End Avenue, Cardiovascular Research, Nashville, Tennessee 37205.
Objectives This study characterized cardiac rehabilitation (CR) use in ventricular assist device (VAD) recipients in the United States and the association of CR with 1-year hospitalization and mortality by using the 2013 to 2015 Medicare files.
Background Exercise-based CR is indicated in patients with heart failure with reduced ejection fraction, but no data exist regarding CR participation after VAD implantation.
Methods The study included Medicare beneficiaries enrolled for disability or age >65 years. The investigators identified VAD recipients by diagnosis codes and cumulated CR sessions occurring within 1 year after VAD implantation. Multivariable-adjusted Andersen-Gill models were used to evaluate the association of CR with 1-year hospitalization risk, and Cox regression was used to evaluate the association of CR with 1-year mortality.
Results There were 1,164 VADs implanted in Medicare beneficiaries in the United States in 2014. CR use was low, with 348 patients (30%) participating in CR programs. The Midwest had the highest proportion of VAD recipients who began CR (38%), whereas the Northeast had the lowest proportion of CR participants (25%). Each 5-year increase in age was associated with attending an additional 1.6 CR sessions (95% confidence interval [CI]: 0.7 to 2.5; p < 0.001). CR participation was associated with a 23% lower 1-year hospitalization risk (95% CI: 11% to 33%; p < 0.001) and a 47% lower 1-year mortality risk (95% CI: 18% to 66%; p < 0.01) after multivariable adjustment.
Conclusions Approximately one-third of VAD recipients attend CR. Although it is not possible to account fully for unmeasured confounding, VAD recipients who participate in CR appear to have lower risks for hospitalization and mortality.
Cardiac rehabilitation (CR), a systematic, multidisciplinary program of prescribed exercise, nutritional counseling, psychosocial support, and cardiovascular risk factor control, is indicated in patients with stable heart failure with reduced ejection fraction (HFrEF), as well as after heart transplantation (1). CR decreases mortality rates and improves quality of life in patients with ischemic heart disease (2). Despite the known benefits of CR, <20% of eligible patients participate in CR programs (3–7). Data on CR participation after ventricular assist device (VAD) implantation have been lacking.
Exercise training in VAD recipients is feasible and safe, and it improves self-reported health status, peak oxygen uptake, and skeletal muscle function (8–11). Moreover, exercise capacity is significantly diminished after VAD implantation, thus increasing the relative benefit of improvements in exercise tolerance (12). VAD implantation is not currently 1 of the indications for CR covered by Medicare. However, many VAD recipients are eligible for CR under the HFrEF indication, which covers patients with stable, chronic heart failure and a left ventricular ejection fraction of ≤35% (5). The Centers for Medicare & Medicaid Services (CMS) define patients with stable chronic heart failure as those who have not had cardiovascular hospitalizations within the previous 6 weeks (13). Almost all patients who are candidates for VAD implantation meet medical criteria for disability benefits (left ventricular ejection fraction ≤30% with symptoms affecting activities of daily living) and are thus eligible for Medicare (14).
Using CMS data, we evaluated CR use after VAD implantation in the United States. We also characterized the association of CR with 1-year hospitalization and mortality risks in VAD recipients. We hypothesized that CR is associated with a decreased risk of hospitalization and death in these patients.
We obtained data regarding CR use in VAD recipients in the United States from the 2014 to 2015 Medicare 100% Limited Data Set files from the CMS. These files contain all inpatient and institutional outpatient claims for fee-for-service Medicare beneficiaries. The Institutional Review Board of Vanderbilt University Medical Center in Nashville, Tennessee approved the study, which was carried out under the auspices of a data use agreement with the CMS.
The study population included Medicare beneficiaries enrolled in 2014 for disability or age ≥65 years who resided in the United States, had uninterrupted fee-for-service coverage until their death or for 1 year following discharge, and did not attend any CR sessions in the year before VAD implantation. The basis for inclusion in the study was the discharge diagnosis code (International Classification of Diseases-9th Revision [ICD-9] codes 37.60, 37.63, 37.65, 37.66) or procedure code (Current Procedural Terminology [CPT] codes 33975, 33976, 33979, 33981, 33982, 33983 0051T, 0052T, 0053T) for VAD implantation, replacement, or repair.
Participation in CR programs, defined as a binary variable (yes/no), was the primary outcome. We searched the Medicare outpatient Limited Data Set files for CR claims (CPT codes 93797, 93798, G0422, G0423, or S9472) occurring within 1 year after the VAD hospitalization discharge date. Secondary outcomes included the following: 1) CR as a continuous variable (number of sessions attended); 2) the number of hospitalizations that occurred in the 1-year period after patients underwent VAD implantation, as determined from the inpatient file; and 3) all-cause mortality, as determined from death dates in the Medicare denominator file.
Patients receiving heart transplants within 1 year of discharge from the VAD hospitalization were identified by CPT code 33945 or ICD-9 code 37.51. We obtained demographic characteristics, including age, sex, race (black, white, or other), and geographic census division (Midwest, Northeast, West, and South) from the denominator file. We characterized the burden of comorbidities with Elixhauser comorbidity groups present during the hospitalization for VAD implantation and the preceding 12 months by using ICD-9 codes as described previously (15). We determined whether the VAD hospital had a CR program from the American Hospital Association Annual Survey of Hospitals (16). We characterized socioeconomic status with median income from the patient’s county of residence, obtained from the United States Census Bureau Small Area Income and Poverty Estimates for 2014 (17).
Baseline demographic and geographic characteristics of VAD recipients participating in CR were compared with characteristic of recipients who did not participate in CR programs by using chi-square tests for categorical variables and Wilcoxon rank sum tests for continuous variables. We used multivariable-adjusted logistic regression to evaluate the effect of individual covariates on CR initiation rates. We used linear regression to analyze predictors of the number of CR sessions attended. The Andersen-Gill model with a robust sandwich covariance estimator (also known as a proportional means model), a technique for the analysis of recurring events, was used to model the effect of participating in CR on 1-year hospitalization risk after multivariable adjustment (18,19). Patients receiving heart transplants were censored at the time of the transplant admission. CR participation was used as a time-updated covariate. All individuals in the sample were considered non-CR participants at baseline and until they began CR. For example, if a VAD recipient had no hospitalizations in the year following VAD implantation and did not initiate CR until 4 months after discharge, they would contribute 4 months of nonhospitalization time in the non-CR participant group, and the remaining 8 months in the follow-up period would contribute to the CR participant group. This analysis strategy was chosen to minimize immortal person-time bias (20,21). Known as the Mantel-Byar method, this approach has been shown to yield unbiased estimates even when event hazards change over time (20).
To evaluate the association of CR with 1-year mortality risk, we constructed a Cox regression model adjusting for clinical characteristics and comorbidities and again used CR participation as a time-varying covariate to minimize immortal person-time bias. As a sensitivity analysis to address potential healthy cohort bias, we created a marginal structural model with inverse probability of treatment weighting (22). For the marginal structural model, the follow-up period was broken into 1-week blocks starting at 3 weeks post-discharge to accumulate enough CR participants, and the sample was reweighted at the beginning of each 1-week interval. This weighting scheme allowed us to estimate the average treatment effect in the CR participants. Thus the CR participants at each time point served as our reference population to which the sample was standardized.
We conducted an additional analysis to measure the sensitivity of the effect of CR on mortality to residual confounding from unmeasured variables, specifically frailty, after adjusting for observed confounders. This method makes statistical inferences about the true exposure effect of CR by specifying distributions of unmeasured confounders in CR participants and nonparticipants along with the effects of these confounders on the outcome (i.e., mortality) (23). Frailty, defined as a score of >0.25 on the Rockwood Frailty Index (24), was found in a study of 99 VAD recipients to have a prevalence of 61.6% before VAD implantation and a hazard ratio (HR) of 2.31 (95% confidence interval [CI]: 1.18 to 4.98; p < 0.05) for 1-year mortality (25). These point estimates were used as the basis for the sensitivity analysis.
A total of 1,647 Medicare beneficiaries received VADs in 2014. We excluded 95 patients who attended CR in the year before hospital admission for VAD implantation, 179 patients who did not have uninterrupted fee-for-service Medicare coverage, and 209 patients who died in the hospital or on the day of discharge, for a final sample size of 1,164 VAD recipients.
A total of 348 (30%) of Medicare beneficiaries receiving VADs initiated CR (Table 1). The average age of the cohort was 61 years, and 20% of VAD recipients were female. Most patients undergoing VAD implantation were white (72%), whereas 23% were black, and 5% were in another racial category, including Asian and non-white Hispanic. Almost all patients in the study (96%) received VADs at hospitals that reported having CR programs. A total of 69 patients (6%) underwent heart transplantation within 1 year of VAD implantation, with a greater proportion receiving transplants in CR participants as compared with nonparticipants (8% vs. 5%; p < 0.05). The 1-year mortality rate was 22% in CR nonparticipants (179 deaths) compared with 7% in CR participants (25 deaths; p < 0.0001). After VAD implantation, 31% of patients (n = 363) were discharged to inpatient rehabilitation facilities (IRFs) or skilled nursing facilities (SNFs).
Cardiac rehabilitation use
The only significant predictor of CR initiation in VAD recipients was census region (Table 2). VAD recipients in the Midwest had higher odds of initiating CR than did recipients in the South (odds ratio: 1.59; 95% CI: 1.16 to 2.18; p < 0.01). None of the comorbidities or age was associated with CR initiation.
Those patients who did initiate CR attended a mean of 24.5 ± 15.0 sessions, fewer than the generally recommended program of 36 sessions (Table 2). Less than one-third of CR attendees participated in the full course of 36 sessions. Older patients attended more CR sessions, with a 1.6 session increase (95% CI: 0.7 to 2.5; p < 0.001) per 5-year increase in age. VAD recipients with renal failure attended an average of 7.3 fewer sessions (95% CI: 3.5 to 11.0; p < 0.001). There was a small but statistically significant inverse association between length of stay during the VAD hospitalization and the number of CR sessions attended (−0.5 sessions per 5-day increase in length of stay; 95% CI: −0.9 to −0.1; p < 0.05). In CR participants, the average time between discharge and the first CR session was 109 ± 84 days, with a median of 83 days (interquartile range [IQR]: 44 to 155 days).
Cardiac rehabilitation and hospitalizations
The median number of total hospitalizations within 1 year of VAD implantation in the cohort was 2 (IQR: 1 to 3), with 914 patients (79%) hospitalized at least once during this time. After multivariable adjustment, participation in a CR program was associated with a 23% (95% CI: 11% to 33%; p < 0.001) decrease in 1-year hospitalizations (Table 3). Multivariable-adjusted cumulative hospitalizations over 1 year, stratified by CR participation, are displayed in Figure 1. Patients with chronic pulmonary disease before VAD implantation were more likely to be readmitted (HR: 1.42; 95% CI: 1.20 to 1.68; p < 0.0001), as were patients with renal failure (HR: 1.28; 95% CI: 1.13 to 1.45; p < 0.001). Patients discharged to an IRF or SNF were also more likely to be readmitted (HR: 1.15; 95% CI: 1.02 to 1.29; p < 0.05). In contrast, patients with a prior history of pulmonary circulation disorders (including pulmonary hypertension) had lower odds of being readmitted.
Cardiac rehabilitation and mortality
After adjusting for demographics, clinical factors, and comorbidities, CR was associated with a decreased risk of mortality in the year after VAD implantation (Table 4) (HR: 0.53; 95% CI: 0.34 to 0.82; p < 0.01). Factors associated with increased 1-year mortality included age (HR: 1.09 per 5-year increase; 95% CI: 1.01 to 1.19; p < 0.05), discharge to an IRF or SNF (HR: 1.57; 95% CI: 1.17 to 2.10; p < 0.01), peripheral vascular disease (HR: 1.45; 95% CI: 1.04 to 2.02; p < 0.05), and weight loss (HR: 1.42; 95% CI: 1.05 to 1.90; p < 0.05). A sensitivity analysis using a marginal structural model demonstrated a similar association between CR and 1-year mortality (HR: 0.47; 95% CI: 0.30 to 0.74; p = 0.001). Plots of standardized mean differences for individual covariates after using inverse probability of treatment weighting are displayed in Online Figures 1 and 2, thereby demonstrating that the sample was well balanced on covariates at 1 and 12 months post-discharge.
An additional analysis was conducted to measure the sensitivity of the effect of CR on mortality to residual confounding from unmeasured variables (23), specifically frailty. Assuming that frailty has an HR of 2.31 for 1-year mortality and a prevalence of 61.6% in the VAD population on the basis of earlier work (25), CR participants would need to have a frailty prevalence of 36.8% or less to make the observed effect of CR on 1-year mortality nonsignificant (i.e., a frailty prevalence of 36.8% would make the effect of CR nonsignificant exactly at p = 0.05).
This study reported CR use rates in patients undergoing VAD implantation in the United States. Approximately one-third of VAD recipients participated in CR programs. There was geographic variation in CR after VAD implantation, with the Midwest having the highest CR initiation rates. VAD recipients participating in CR programs began an average of 3 months after discharge and attended two-thirds of the recommended course of 36 sessions. Younger CR participants attended significantly fewer CR sessions than did older patients. Although it is not possible to account fully for all confounding variables, VAD recipients who participate in CR appear to have lower risk for hospitalization and all-cause death.
Cardiac rehabilitation by indication
CR use varies by indication, with reported initiation rates ranging from <10% in patients with systolic heart failure (5), to 10% to 20% in patients with acute myocardial infarction (AMI) and percutaneous coronary intervention (6,28), and up to 50% in patients receiving heart transplants (29). Approximately one-third of patients undergoing coronary artery bypass grafting (CABG) (3,4,7,30) participate in CR programs, a proportion similar to that seen in VAD recipients in the current study. Unlike the aforementioned indications, Medicare does not specifically cover CR after VAD implantation, and these patients are often referred to CR programs under the auspices of other conditions. VAD recipients could potentially be eligible for CR Medicare coverage under the HFrEF indication (13), stable angina pectoris (1) (which covers most patients with ischemic heart disease), and/or AMI (1) (which covers patients experiencing an AMI within the previous year).
Cardiac rehabilitation initiation
The only significant predictor of CR initiation in VAD recipients was census region. The Midwest census region had a significantly higher proportion of VAD recipients initiating CR than the other regions. This geographic variation in CR use is consistent with prior studies of CR use after AMI and CABG (7). The finding that geographic location is more strongly associated with the odds of initiating CR than any of the clinical characteristics or comorbidities in this population underscores the importance of further research to characterize variation in CR referral patterns and access. For those patients initiating CR, the time between discharge and the first CR appointment was much longer in VAD recipients (median 83 days) as compared with a recent study in patients with ischemic heart disease (median 42 days) (31). This delay is likely attributable to the significant post-operative recovery period after VAD implantation, as well as the Medicare requirement that patients referred to CR programs for systolic heart failure be stable for 6 weeks (e.g., no cardiovascular hospitalizations) before attending the first session (13).
Cardiac rehabilitation dose
A dose-dependent relationship has been identified between the number of sessions attended and death in patients with ischemic heart disease (3,4). Interestingly, older CR participants were more likely to attend more sessions than younger participants. One could expect that older VAD recipients would be inclined to participate in fewer CR sessions because of a higher burden of comorbidities and frailty (24,25). It is possible that younger VAD recipients may be more likely to return to work, and work responsibilities are a significant barrier to attending CR programs (32,33). Renal failure was associated with a significant decrease in the number of sessions attended in CR participants, but not with the odds of initiating CR. This finding likely represents that the time demands of hemodialysis are a major barrier to attending CR sessions 3 times weekly.
Cardiac rehabilitation and 1-year outcomes after VAD implantation
CR was associated with fewer hospitalizations in the year following VAD implantation. The magnitude of this association in our analyses (a 23% decrease; 95% CI: 0.67 to 0.89) is similar to that in other studies (19). A recent meta-analysis of the effect of exercise-based CR versus usual care on hospitalizations demonstrated an 18% decrease in hospitalization risk (95% CI: 4% to 30%) (2). The etiology of this association is likely multifactorial. Beyond CR’s known beneficial effects on skeletal muscle function, peak oxygen uptake, and health status in VAD recipients (8–10), CR offers an opportunity for health care professionals to monitor these patients serially, thus potentially averting unplanned hospitalizations.
In our adjusted analyses, chronic obstructive pulmonary disease and renal failure were the only comorbidities associated with an increased hospitalization risk in VAD recipients. Curiously, pulmonary circulation disorders (including pulmonary hypertension) before VAD implantation were associated with a decreased risk of hospitalization. It is possible that such patients experience disproportionate benefit from a VAD because these devices significantly improve pulmonary arterial pressures (34), although this conclusion would be speculative with the available data and warrants further study.
The magnitude of the association between CR and 1-year mortality risk (HR: 0.53; 95% CI: 0.32 to 0.76) is also similar to that in prior studies. Suaya et al. (4) identified a 56% reduction in 1-year mortality risk in an analysis of more than 600,000 Medicare beneficiaries hospitalized for AMI or CABG. Another study demonstrated a 46% reduction in all-cause mortality in a cohort of 846 CABG recipients (30).
It is important to interpret all of these results in the context of potential confounding because of healthy cohort bias, which could overestimate the effect of CR on mortality as well as hospitalizations. However, our analysis controlled extensively for sociodemographic and clinical factors, and Elixhauser comorbidity groups provide effective comorbidity adjustment in surgical populations (15), including heart transplant recipients (35) and patients receiving VADs (36). We also used multiple statistical techniques, including marginal structural models, to control for observed confounders.
Frailty or functional impairment represents 1 of the most significant unobserved confounders because frailty cannot be well characterized with administrative claims data (24). Using prior work by Dunlay et al. (25), who found that 62% of VAD recipients were frail (as defined by the Rockwood Frailty Index ) before device implantation, we demonstrated that the prevalence of frailty in CR participants would have to be very low (<37%) for the effect of CR on mortality to become nonsignificant. It is unlikely that frailty would be this infrequent in VAD recipients who participated in CR because frailty prevalence was 37% after 10 years of follow-up in a community cohort of myocardial infarction survivors of similar age (37) and was >35% in younger patients in intensive care units (38). Given this context, the association of CR with 1-year mortality appears to be quite robust, even the setting of unobserved confounding.
Clinical and policy implications
VAD recipients necessitate multidisciplinary care and require an enormous amount of resources. Our results suggest that CR is associated with improved outcomes in this population. Further study is needed on the mechanisms by which VAD recipients are being referred to CR (i.e., HFrEF, stable angina pectoris, or AMI) and whether the 6-week interval after discharge required for a patient to be deemed stable under the HFrEF indication is leading to delays in CR initiation.
Our study has limitations in addition to those previously addressed. First, we were able to capture use data only on VAD recipients ≥65 years old or with Medicare disability benefits. The finding that a significant number of patients in our cohort received disability coverage does not indicate that they were less likely to participate in CR than those eligible by age because almost all patients receiving VADs would meet the chronic heart failure medical criteria for disability benefits. Second, our data are obtained from CMS administrative claims, which are not adjudicated. However, CMS data have been used effectively to study many cardiovascular therapies, including CR, in previous work (3,4,7). Third, our analyses were limited to VAD recipients enrolled in fee-for-service Medicare and may not be generalizable to patients enrolled in Medicare private health plans. However, fee-for-service Medicare still accounted for 71% of Medicare beneficiaries in 2014 (39). Finally, the CMS decision memo approving HFrEF as an indication for CR was issued in February 2014, so it is possible that CR uptake in VAD recipients under the HFrEF indication has increased since that time.
Less than one-third of Medicare beneficiaries receiving VADs participate in CR programs in the United States. Although it is not possible to account fully for all confounding variables, VAD recipients who participate in CR appear to have lower risks for hospitalization and mortality. These exploratory results suggest opportunities for further, more definitive studies of the effectiveness of CR in this population, as well as a need to understand factors that drive patient and caregiver decisions regarding CR participation.
COMPETENCY IN MEDICAL KNOWLEDGE: CR is indicated in patients with stable HFrEF, including those receiving VADs. Participation in CR programs is associated with decreased 1-year hospitalizations and mortality in VAD recipients.
TRANSLATIONAL OUTLOOK: Further studies are needed to characterize the barriers to CR participation in patients receiving VADs, along with quality improvement interventions to increase CR uptake in this population.
The authors would like to acknowledge the Million Hearts Initiative co-led by the Centers for Medicaid & Medicare Services and the Centers for Disease Control and Prevention. Million Hearts hosts the Cardiac Rehabilitation Collaborative, within which the authors had many discussions with other cardiac rehabilitation professionals that helped shape this study. The authors would also like to acknowledge Benjamin D. Levine, MD, Director of the Institute for Exercise and Environmental Medicine at Texas Health Presbyterian Hospital Dallas and Professor of Internal Medicine at University of Texas-Southwestern Medical Center, who reviewed the manuscript and provided helpful guidance.
This project was supported by Vanderbilt Clinical and Translational Science grant UL1 TR000445 from the National Center for Advancing Translational Sciences at the National Institutes of Health and grant K12HS022990 from the Agency for Healthcare Research and Quality. The content is the responsibility of the authors alone and does not necessarily reflect the views or policies of these agencies. Dr. Lindenfeld has been a consultant for Abbott, Relypsa, Resmed, CVRx, VWave, and Cardionomic; and has received grants from Novartis and AstraZeneca. Dr. Keteyian is a technical advisor to Nimble Heart. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
Barry H. Greenberg, MD, served as Guest Editor for this paper.
- Abbreviations and Acronyms
- acute myocardial infarction
- coronary artery bypass grafting
- confidence interval
- Centers for Medicare & Medicaid Services
- Current Procedural Terminology
- cardiac rehabilitation
- heart failure with reduced ejection fraction
- hazard ratio
- International Classification of Diseases-9th Revision
- inpatient rehabilitation facility
- skilled nursing facility
- ventricular assist device
- Received July 20, 2017.
- Revision received October 10, 2017.
- Accepted November 3, 2017.
- 2018 American College of Cardiology Foundation
- Thomas R.J.,
- King M.,
- Lui K.,
- et al.
- Anderson L.,
- Oldridge N.,
- Thompson D.R.,
- et al.
- Hammill B.G.,
- Curtis L.H.,
- Schulman K.A.,
- Whellan D.J.
- Suaya J.A.,
- Stason W.B.,
- Ades P.A.,
- Normand S.L.,
- Shepard D.S.
- Golwala H.,
- Pandey A.,
- Ju C.,
- et al.
- Doll J.A.,
- Hellkamp A.,
- Ho P.M.,
- et al.
- Suaya J.A.,
- Shepard D.S.,
- Normand S.L.,
- Ades P.A.,
- Prottas J.,
- Stason W.B.
- Kerrigan D.J.,
- Williams C.T.,
- Ehrman J.K.,
- et al.
- Jung M.H.,
- Gustafsson F.
- ↵Centers for Medicare & Medicaid Services. Decision Memo for Cardiac Rehabilitation (CR) Programs - Chronic Heart Failure (CAG-00437N). 2014. Available at: https://www.cms.gov/medicare-coverage-database/details/nca-decision-memo.aspx?NCAId=270. Accessed March 31, 2017.
- ↵Social Security Administration. Disability Evaluation Under Social Security (Blue Book - October 2008). 2008. Available at: https://www.ssa.gov/disability/professionals/bluebook/4.00-Cardiovascular-Adult.htm#4_02. Accessed May 27, 2017.
- ↵American Hospital Association. American Hospital Association Annual Survey of Hospitals. 2016. Available at: https://www.ahadataviewer.com/book-cd-products/AHA-Survey/. Accessed November 2, 2016.
- ↵United States Census Bureau. Small Area Income Estimates. 2014. Available at: https://www.census.gov/did/www/saipe/data/statecounty/data/2015.html. Accessed June 1, 2017.
- Mi X.,
- Hammill B.G.,
- Curtis L.H.,
- Lai E.C.,
- Setoguchi S.
- ↵SAS Institute, Inc. Statistical Analysis Software, version 9.4. 2013. Cary, NC: SAS Institute, Inc. Available at: www.sas.com. Accessed November 17, 2017.
- ↵R Core Team. R: A language and environment for statistical computing. 2015. R Foundation for Statistical Computing, Vienna, Austria. Available at: https://www.R-project.org/. Accessed October 22, 2016.
- Doll J.A.,
- Hellkamp A.,
- Thomas L.,
- et al.
- Bachmann J.M.,
- Shah A.S.,
- Duncan M.S.,
- et al.
- Pack Q.R.,
- Goel K.,
- Lahr B.D.,
- et al.
- Pack Q.R.,
- Mansour M.,
- Barboza J.S.,
- et al.
- Mujib M.,
- Khanna N.,
- Mazumder N.K.,
- et al.
- Muscedere J.,
- Waters B.,
- Varambally A.,
- et al.
- ↵Kaiser Family Foundation. Fact Sheet: Medicare Advantage. 2016. Available at: http://files.kff.org/attachment/Fact-Sheet-Medicare-Advantage. Accessed September 27, 2016.