Factors Associated With Participation in Cardiac Episode Payments Included in Medicare’s Bundled Payments for Care Improvement Initiative

Key Points

Question

Are hospitals participating in Medicare’s Bundled Payments for Care Improvement initiative for cardiac bundles different from nonparticipating hospitals in ways that could limit the generalizability of program outcomes to all US acute care hospitals?

Findings

In this cross-sectional study, participation in Bundled Payments for Care Improvement model 2 bundled payments for acute myocardial infarction, congestive heart failure, coronary artery bypass graft surgery, and percutaneous coronary intervention was associated with larger hospital size, non–safety net hospital status, and access to cardiac catheterization laboratories.

Meaning

Outcomes of cardiac bundled payments included in Bundled Payments for Care Improvement may have limited external validity, particularly among small and safety net hospitals with more limited cardiac capabilities.

Abstract

Importance

Medicare’s Bundled Payments for Care Improvement (BPCI) is a voluntary pilot program evaluating bundled payments for several common cardiovascular conditions. Evaluating the external validity of this program is important for understanding the effects of bundled payments on cardiovascular care.

Objective

To determine whether participants in BPCI cardiovascular bundles are representative of US acute care hospitals and identify factors associated with participation.

Design, Setting, and Participants

Retrospective cross-sectional study of hospitals participating in BPCI model 2 bundles for acute myocardial infarction (AMI), congestive heart failure (CHF), coronary artery bypass graft, and percutaneous coronary intervention and nonparticipating control hospitals (October 2013 to January 2017). The BPCI participants were identified using data from the Centers for Medicare and Medicaid Services, and controls were identified using the 2013 American Hospital Association’s Survey of US Hospitals. Hospital structural characteristics and clinical performance data were obtained from the American Heart Association survey and Centers for Medicare and Medicaid Services. One hundred fifty-nine hospitals participating in BPCI model 2 cardiac bundles and 1240 nonparticipating control hospitals were compared, and a multivariable logistic regression was estimated to identify predictors of BPCI participation.

Exposures

Bundled payments.

Main Outcomes and Measures

Hospital-level structural characteristics and 30-day risk-adjusted readmission and mortality rates for AMI and CHF.

Results

Compared with nonparticipants, BPCI participants were larger, more likely to be privately owned or teaching hospitals, had lower Medicaid bed day ratios (ratio of Medicaid inpatient days to total inpatient days: 17.0 vs 19.3; P < .001), and were less likely to be safety net hospitals (2.5% vs 12.3%; P < .001). The BPCI participants had higher AMI and CHF discharge volumes, were more likely to have cardiac intensive care units and catheterization laboratories, and had lower risk-standardized 30-day mortality rates for AMI (13.7% vs 16.6%; P = .001) and CHF (11.3 vs 12.4; P = .005). In multivariable analysis, larger hospital size and access to a cardiac catheterization laboratory were positively associated with participation. Being a safety net hospital was negatively associated with participation (odds ratio, 0.3; 95% CI, 0.1-0.7; P = .001).

Conclusions and Relevance

Hospitals participating in BPCI model 2 cardiac bundles differed in significant ways from nonparticipating hospitals. The BPCI outcomes may therefore have limited external validity, particularly among small and safety net hospitals with limited clinical cardiac services.

This study examines whether participants in Bundled Payments for Care Improvement cardiovascular bundles are representative of US acute care hospitals and identifies factors associated with participation.

Introduction

The United States spends more than $200 billion on cardiovascular disease care annually.¹ Bundled payments hold promise for curbing rising health care costs.² In 2013, the Center for Medicare and Medicaid Services (CMS) launched Bundled Payments for Care Improvement (BPCI), a voluntary pilot evaluating the feasibility and efficacy of 4 bundled payment models across 48 clinical conditions.³ Model 2 bundles, the most widely subscribed, subsume the index hospitalization, associated readmissions, and postacute care and include bundles for common cardiovascular conditions and procedures.³

Selective participation represents a potential threat to the external validity of voluntary programs, including BPCI. Prior work suggests that hospitals participating in BPCI may not broadly represent all US acute care hospitals.^4,5 Investigating participation bias in BPCI Model 2 cardiac bundles is critical for understanding the generalizability of cost and quality outcomes from this pilot and may help policymakers mitigate this bias when designing and implementing future programs, such as BPCI Advanced, a voluntary bundled payment pilot CMS plans to launch in October 2018.⁶ Whether participation bias exists in BPCI Model 2 cardiac bundles is unknown.

We compared hospitals participating in BPCI Model 2 bundled payments for acute myocardial infarction (AMI), congestive heart failure (CHF), coronary artery bypass graft surgery (CABG), and percutaneous coronary intervention (PCI) with nonparticipants in their health care markets to identify structural and performance characteristics associated with BPCI participation.

Methods

This study was approved by the institutional review board at the Harvard School of Public Health. Medicare claims data had been deidentified prior to use, and informed consent was therefore waived. We used publicly available data from CMS to identify all hospitals that enrolled in BPCI Model 2 bundles for AMI, CHF, CABG, and PCI between October 1, 2013, and January 31, 2017. Using data from the American Hospital Association’s (AHA) Survey of US Hospitals from 2010 to 2013, we identified nonparticipating acute care hospitals in hospital referral regions with at least 1 BPCI participant. Critical access hospitals were excluded. We obtained data on hospital structural characteristics and clinical performance from the AHA survey and CMS’ Hospital Compare website including 30-day risk-standardized mortality and readmission rates and clinical volumes for AMI and CHF (list of covariates provided in eAppendix in the Supplement). Data on each hospital’s disproportionate share payment were obtained from 2014 Medicare claims. We defined safety net hospitals (SNHs) as having disproportionate sharepayments in the top 10% of hospitals nationally. We compared the structural characteristics and clinical outcomes of BPCI participants and control hospitals using t tests, Wilcoxon tests, and χ² tests. We considered a 2-tailed P value less than .05 to be statistically significant. We constructed a multivariable logistic regression model to identify predictors of participation in BPCI Model 2 cardiac bundles. For categorical variables with missing data, we included a missing category alongside other categories in the regression. We excluded patients missing data for one or more continuous variables.

Results

We identified 159 different hospitals participating in BPCI Model 2 cardiac bundles and 1240 control hospitals. Thirty hospitals participated in CABG bundles, 34 in PCI bundles, 52 in AMI bundles, and 114 in CHF bundles; 52 hospitals participated in at least 2 of these bundles.

Compared with nonparticipants, BPCI participants were larger and more likely to be privately owned and teaching hospitals. Participants had lower Medicaid bed day proportions (17.0% vs 19.3%; P = .004) and were less likely to be SNHs (n = 4 [2.5%] vs n = 121 [9.8%]; P < .001) (Table 1). The BPCI participants had higher mean annual emergency department visits (n = 59 565 vs n = 35 323; P < .001) and higher AMI and CHF discharge volumes and were more likely to have cardiac intensive care units (n = 72 [45.3%] vs n = 315 [25.4%]; P < .001), cardiac catheterization laboratories (n = 140 [88.1%] vs n = 643 [51.9%]; P < .001), and cardiac surgery on site (n = 100 [62.9%] vs n = 371 [29.9%]; P < .001). In addition, BPCI participants had lower risk-standardized 30-day mortality rates for both AMI (13.7% vs 16.6%; P = .001) and CHF (11.3% vs 12.4%; P = .005).

Table 1. Characteristics of BPCI Participants and Nonparticipants.

Characteristic	No. (%)
	BPCI Participants (n = 159)	Nonparticipants (n = 1240)	P Valuea
Hospital characteristics
Hospital profit status
Government, public or federal	7 (4.4)	209 (16.9)	<.001
Private, nonprofit	123 (77.4)	761 (61.4)
Private, for-profit	29 (18.2)	270 (21.8)
Bed size
<100	15 (9.4)	504 (40.7)	<.001
101-299	72 (45.3)	485 (39.1)
300-499	45 (28.3)	154 (12.4)
>500	27 (17.0)	97 (7.8)
Region
Midwest	27 (17.0)	298 (24.0)	.008
Northeast	50 (31.5)	251 (20.2)
South	57 (35.9)	492 (39.7)
West	25 (15.7)	199 (16.1)
Safety net hospital
Yes	4 (2.5)	121 (9.8)	<.001
No	154 (96.9)	867 (69.9)
Missing	1 (0.6)	252 (20.3)
Teaching hospital statusb
Major	25 (15.7)	93 (7.5)	<.001
Minor	56 (35.2)	262 (21.1)
Nonteaching	78 (49.1)	873 (70.4)
Missing	0 (0.0)	12 (1.0)
Presence of CICU
Yes	72 (45.3)	315 (25.4)	<.001
No	87 (54.7)	925 (74.6)
Cardiac surgery onsitec
Yes	100 (62.9)	371 (29.9)	<.001
No	59 (37.1)	869 (70.1)
Cardiac catheterization laboratoryb,d
Yes	140 (88.1)	643 (51.9)	<.001
No	17 (10.7)	521 (42.0)
Missing	2 (1.3)	76 (6.1)
Heart transplant centerb
Yes	11 (6.9)	39 (3.1)	.02
No	143 (89.9)	1106 (89.2)
Missing	5 (3.1)	95 (7.7)
Annual emergency department volume, mean (SD)	59 565 (37 639)	35 323 (36 518)	<.001
Ratio of Medicaid inpatient days to total inpatient days, % (SD)e	17.0 (8.5)	19.3 (14.1)	.004
Ratio of Medicare inpatient days to total inpatient days, % (SD)f	51.8 (11.1)	52.0 (16.6)	.84
Medicare AMI volume, 2013, mean (SD)g	1874 (1437)	1241 (1251)	<.001
Medicare CHF volume, 2013, mean (SD)h	4474 (3256)	2723 (2826)	<.001
Hospital outcomes, mean (SD)
Risk-standardized 30-d mortality rate, % (SD)
AMIi	13.7 (6.0)	16.6 (22.0)	.001
CHFj	11.3 (3.6)	12.4 (9.3)	.005
Risk-standardized 30-d readmission rate, % (SD)
AMIk	16.8 (7.9)	15.9 (13.6)	.27
CHFl	20.5 (5.2)	21.0 (11.9)	.32
All-causem	16.1 (1.1)	16.2 (1.1)	.42

Abbreviations: AHA, American Heart Association; AMI, acute myocardial infarction; BPCI, Bundled Payment for Care Improvement; CHF, congestive heart failure; CICU, cardiac intensive care unit; COTH, Council of Teaching Hospitals.

^aP values calculated using t test, χ² test, or analysis of variance as appropriate.

^bTeaching hospital status was defined using response to the AHA annual survey of US hospitals. Consistent with previous work, major teaching hospitals were defined as being members of the COTH; minor teaching hospitals were defined as non-COTH members, which indicated an affiliation with a medical school on the AHA survey; and all other hospitals were defined as nonteaching hospitals.

^cPresence of cardiac surgery capabilities on-site was defined using data from the AHA annual surveys from 2010 to 2013 (preferentially using the most recent available survey data).

^dRepresents access to a cardiac catheterization laboratory through the hospital, system, or venture.

^eMissing observations from 20 BPCI nonparticipants.

^fMissing observations from 5 BPCI nonparticipants.

^gMissing observations from 5 BPCI participants and 445 BPCI nonparticipants.

^hMissing observations from 1 BPCI participant and 187 BPCI nonparticipants.

ⁱMissing observations from 1 BPCI participant and 168 BPCI nonparticipants.

^jMissing observations from 1 BPCI participant and 85 BPCI nonparticipants.

^kMissing observations from 2 BPCI participants and 214 BPCI nonparticipants.

^lMissing observations from 1 BPCI participant and 85 BPCI nonparticipants.

^mMissing observations from 72 BPCI nonparticipants.

In multivariable analysis, larger hospital size and having a cardiac catheterization laboratory were positively associated with BPCI participation. Being an SNH was strongly and negatively associated with participation (odds ratio, 0.3; 95% CI, 0.1-0.8; P = .01) (Table 2).

Table 2. Multivariable Analysis of Factors Associated With Hospital Participation in BPCI Model 2 Cardiac Bundles^a.

Characteristic	Odds Ratio (95% CI)	P Value
Hospital characteristics
Hospital profit status
Government, federal	1 [Reference]	NA
Private, nonprofit	2.3 (0.9-5.5)	.07
Private, for-profit	2.1 (0.8-5.4)	.14
Bed size
<100	1 [Reference]	NA
101-299	1.6 (0.8-3.2)	.15
300-499	2.5 (1.1-5.7)	.03
>500	1.8 (0.6-5.3)	.27
Region
Midwest	1 [Reference]	NA
Northeast	1.7 (0.99-3.0)	.06
South	1.1 (0.6-2.0)	.62
West	1.6 (0.8-3.1)	.15
Safety net hospital	0.3 (0.1-0.8)	.01
Teaching status
Nonteaching	1 [Reference]	NA
Minor	1.1 (0.7-1.7)	.76
Major	1.1 (0.5-2.4)	.73
Presence of CICU	0.8 (0.5-1.2)	.29
Presence of cardiac surgery	1.5 (0.9-2.5)	.09
Cardiac catheterization laboratory	2.0 (1.1-3.9)	.03
Heart transplant center	0.9 (0.4-2.1)	.77
Ratio of Medicaid inpatient days to total inpatient days	0.98 (0.96-1.0)	.13
Ratio of Medicare inpatient days to total inpatient days	1.00 (0.98-1.00)	.91
Emergency department volume	1.0 (0.99-1.1)	.14
Hospital outcomes
Risk-standardized 30-d mortality rate
AMI	1.0 (0.8-1.3)	.74
CHF	0.9 (0.6-1.4)	.77
Risk-standardized 30-d readmission rate
AMI	1.1 (0.9-1.3)	.44
CHF	0.8 (0.5-1.1)	.10

Abbreviations: AMI, acute myocardial infarction; BPCI, bundled payment for care improvement; CHF, congestive heart failure; CICU, cardiac intensive care unit; NA, not applicable.

^aWe estimated a multivariable logistic regression analysis to identify hospital structural characteristics and clinical outcomes independently associated with participation in BPCI Model 2 cardiac bundles for AMI, CHF, coronary artery bypass grafting, and percutaneous coronary intervention. For categorical variables with missing data, a separate “missing” response option was included for the variable in the regression analysis. Two hundred twenty-five patients who were missing data from 1 or more continuous variables were excluded from the regression analysis.

Discussion

The annual costs of treating cardiovascular disease are projected to quadruple to more than $818 billion by 2030.⁷ Curbing this spending without compromising quality is critical to ensuring the financial sustainability of the US health care system. Bundled payments represent a promising approach for improving quality and reducing unnecessary resource use.⁸ The BPCI is the largest study of bundled payments to date.

We found significant differences between BPCI model 2 cardiac bundle participants and comparison hospitals with respect to key cardiac capabilities and mortality rates for common cardiovascular conditions. After multivariable adjustment, participating hospitals were significantly more likely to be larger, non–safety net facilities with cardiac catheterization laboratories.

Despite growing interest in using bundled payments to improve care value, little empirical evidence on the merits and risks of bundled payments for cardiovascular disease care or factors associated with success under this new model exists. Our analysis indicates that selection bias could limit the generalizability of cost and quality outcomes data from BPCI model 2 cardiac bundles.

This study has implications for designing and interpreting outcomes from future alternative payment programs, including BPCI Advanced. Specifically, participation bias represents a threat to future voluntary bundled payment pilots, which may struggle to enroll a broadly representative cohort of hospitals, including small hospitals and SNHs. Smaller hospitals may be less likely to participate in voluntary programs because they treat lower volumes of common cardiovascular diseases, lack the capabilities to treat severe and complex cardiovascular disease, and do not possess the administrative and quality improvement infrastructure to track and rapidly improve performance.^9,10,11 Additionally, SNHs treat outsized proportions of poor and underserved patients, and bundled payment risk adjustment methods do not account for social risk factors.^12,13 Finally, many hospitals contract with private physician groups to staff their facilities rather than directly employing clinicians. This arrangement can complicate efforts to incent high performance within bundles and reduce administrators’ enthusiasm to participate in these pilots. These and other barriers must be mitigated to achieve broader participation in future pilots.

Policymakers could address these barriers in a few ways. First, they could adjust payments to account for patient-level social risk factors affecting quality and cost outcomes, which could alleviate concerns managers of SNHs have about participating in pilots.¹⁴ Second, future pilots, including BPCI Advanced, could offer quality improvement assistance or limits to downside risk for underrepresented hospitals to entice them to participate. Third, future pilots could include separate tracks for larger non-SNHs and smaller SNHs; separating these facilities would ensure that hospitals with fundamentally different missions, clinical capabilities, and patient populations are not forced to compete against each other. Fourth, policymakers could reintroduce bundled payment programs with mandatory participation; however, while mandatory participation could solve the generalizability problem, payment adjustments for social determinants of health and quality improvement assistance may still be needed to help some hospitals succeed.

Limitations

This study has limitations. First, our comparisons of clinical outcomes and patient populations treated were limited to publicly available data. Second, we did not have data necessary to compare baseline costs of care across BPCI participants and nonparticipants and cannot evaluate differences in care value. Third, we cannot rule out confounding owing to unmeasured covariates.

Conclusions

Hospitals participating in bundled payments for AMI, CHF, CABG, and PCI through BPCI model 2 are larger, less likely to be SNHs, and have greater cardiac capabilities. Selection bias may limit efforts to use BPCI outcomes data to understand how bundled payments affect quality and costs of cardiovascular disease care across a diverse cohort of delivery systems.

Supplement.

eAppendix. Covariate Definition and Source