The Effect of Medicare Accountable Care Organizations on Early and Late Payments for Cardiovascular Disease Episodes

Abstract

Background:

Early evaluations of the Pioneer and Shared Savings programs have shown substantial heterogeneity and modest savings associated with receipt of care in a Medicare accountable care organization (ACO). Whether savings are limited to specific types of conditions remains unknown. We examined the association between implementation of Medicare ACOs and episode spending for two diverse cardiovascular conditions.

Methods and Results:

We analyzed a 20% sample of national Medicare data, identifying fee-for-service beneficiaries ages 65 and older admitted for acute myocardial infarction (AMI) or congestive heart failure (CHF) between January 2010 and October 2014. We distinguished admissions to hospitals participating in a Medicare ACO from those that were not. We calculated 365-day, price-standardized episode spending made on behalf of these beneficiaries, differentiating between early (index admission to 90 days post-discharge) and late payments (91+ days). We used an interrupted time series design to fit longitudinal multivariable models to estimate the association between hospital ACO participation and episode spending. Our study included 153,476 beneficiaries admitted for AMI to 401 ACO participating and 2,597 non-participating hospitals, and 260,420 beneficiaries admitted for CHF to 412 ACO participating and 2,796 non-participating hospitals. On multivariable analysis, admission to an ACO participating hospital was not associated with changes in early episode spending {AMI: $95 per beneficiary [95% confidence interval (CI), -$481 to $671]; CHF: $158 (95% CI, -$290 to $605)}. However, it was associated with significant reductions in late episode spending for both cohorts [AMI: -$680 (95% CI, -$1,348 to -$11); CHF: -$889 (95% CI, -$1,465 to -$313)].

Conclusions:

For beneficiaries with AMI or CHF, admission to ACO participating hospitals was not associated with changes in early episode spending, but it was associated with significant savings during the late episode. ACO effects on late episode spending may complement other value-based payment reforms that target the early episode.

Introduction

Cardiovascular disease is both highly prevalent and expensive, with severe events such as acute myocardial infarction (AMI) and hospitalization for congestive heart failure (CHF) exacerbation accounting for over $300 billion in Medicare spending annually.¹ Prior empirical work demonstrates that payments for AMI and CHF episodes vary widely across hospitals for reasons not completely explained by case-mix differences.^{2, 3} Many interpret such variation as evidence of healthcare waste due to a fragmented delivery system.^4–7 Thus, health policy reforms designed to reduce care fragmentation, like Medicare’s rollout of accountable care organizations (ACOs), may help contain spending related to cardiovascular disease.⁸

Early evaluations of the Pioneer and Shared Savings programs have shown substantial heterogeneity and modest savings associated with receipt of care in an ACO.^9–13 Whether savings are limited to specific types of conditions and the mechanisms by which they occur remain unknown. To answer these questions, we studied two diverse cardiovascular conditions—one acute (AMI) and the other chronic (CHF), whose management requires multiple care transitions. Such diversity is important since spending increases for cardiovascular hospitalizations are driven largely by payments for services rendered during the late (versus early) post-discharge period.¹⁴ This time period is when ACOs, through their efforts to enhance care coordination, are likely to exert their greatest effect.¹⁵ It is also when other recent reforms (e.g., bundled payments, readmission penalties) are expected to have little impact.

In particular, we analyzed national Medicare claims from beneficiaries admitted for CHF or AMI. After measuring total payments made on their behalf, we assessed for differences in early and late episode spending between beneficiaries who received care at ACO participating and non-participating hospitals. We had two hypotheses. First, receipt of care in an ACO participating hospital would have larger effects on spending more than 90 days following discharge from the index admission. Second, savings from receipt of care at an ACO participating hospital would differ across conditions as diverse as CHF and AMI.

Methods

Study Population

We analyzed claims and enrollment data from the Medicare Provider and Analysis Review, Outpatient, and Carrier research identifiable files for a random 20% sample of Medicare fee-for-service beneficiaries. Our study population consisted of beneficiaries 65 years of age and older who were admitted to an acute care hospital for a primary diagnosis of AMI or CHF between January 1, 2010 and October 1, 2014 (eTable 1). For inclusion, we required beneficiaries to have continuous enrollment in both Parts A and B of Medicare for 12 months prior and 12 months after their index admission (or, in the case of those who subsequently died, during the time that they were alive). In the event that a beneficiary experienced multiple AMI and/or CHF episodes within the same calendar year, we only considered the first episode and excluded the others. We excluded beneficiaries who were discharged against medical advice and those from Maryland, which is not part of Medicare’s Inpatient Prospective Payment System. We also excluded hospitals caring for fewer than two patients with AMI or CHF over the study period. Because of the sensitive nature of Medicare Research Identifiable Files, requests to access them from qualified researchers trained in human subject confidentiality protocols may be sent to the Centers for Medicare & Medicaid Services.

Measuring episode payments

The primary outcome of our study was total price-standardized spending per episode. To measure this, we constructed a claims window encompassing the index admission and extending 365 days post-discharge and summed all payments made on a beneficiary’s behalf during this period. We examined payments for inpatient, skilled nursing facility, rehabilitation, outpatient department, and professional services. Payments for durable medical equipment, home health, and medication were not captured in our analysis. We distinguished between early (index admission to 90 days post-discharge) and late (91 to 365 days post-discharge) episode spending. Our secondary outcomes were the major components of episode payments: 1) index admission, 2) readmissions, 3) professional services, and 4) post-acute care.

We price-standardized all payments to account for geographic payment differences and add-on payments for indirect medical education and disproportionate share hospitals, using methods employed by the Medicare Payment Advisory Commission.¹⁶ In addition, we inflation-adjusted all payments to 2014 U.S. dollars.

Distinguishing between ACO Participating and Non-Participating Hospitals

Our independent variable of interest was admission to an ACO participating hospital. Of note, a beneficiary aligned with a Medicare ACO could be admitted to a non-participating hospital and vice versa. To distinguish between receipt of care at a Medicare ACO (MSSP or Pioneer) participating versus non-participating hospital, we used the Leavitt Partners ACO Database. This previously validated database contained 839 Medicare, Medicaid, and commercial ACOs at the time of our analysis.¹⁷ Information on ACOs in the database is updated regularly from press releases, news articles, government announcements, conferences, personal and industry interviews, and other public records. Working with these data, we were also able to determine the organizational structure of the Medicare ACO in which the hospital participated (hospital- or physician-led or hospital-physician partnership).

Statistical Analysis

We first assessed for differences between patients admitted to ACO participating versus non-participating hospitals in the pre-ACO contract period. We compared beneficiary age on admission, gender, race/ethnicity, and level of comorbidity [defined by number of hierarchical condition categories (HCCs)¹⁸]. The pre-contract period for a hospital varied according to the contract start date for the ACO in which it participated (January 2012, April 2012, July 2012, January 2013, or January 2014). Next, we assessed for differences in hospital characteristics between ACO participating and non-participating hospitals, using 2014 American Hospital Association Annual Survey data on hospital size, teaching status, and for-profit status.¹⁹

Using an interrupted time series research design, we then tested the effect of hospital ACO participation on both early and late episode spending for our AMI and CHF cohorts. To do so, we used an interrupted time series design to fit longitudinal multivariable fixed-effects regression models. Our main exposure was a binary, time-varying indicator for a hospital’s ACO participation status [set to 1 during the quarter when the ACO in which the hospital participated began its Medicare contract and 0 otherwise]. This approach allowed us to adjust for baseline trends and rule out maturation effects. It also allowed us to account for rolling ACO contract start dates. We included in our models controls for beneficiary age, gender, race/ethnicity, and level of comorbidity (using 79 HCCs), as well as hospital fixed effects, year fixed effects, and quarter fixed effects. We fit these models separately for early and late spending, as well as for each component payment category. To account for serially correlated outcomes, we estimated robust standard errors.

Finally, we conducted a series of heterogeneity analyses. To determine whether the association between hospital participation and total episode spending differed between early and late adopters of the ACO model, we constructed separate models for hospitals whose ACOs had contract start dates in 2012, 2013, and 2014. We next examined if the effects of hospital participation were modified by ACO organizational structure (i.e., whether the ACO was hospital- or physician-led or a hospital-physician partnership) by adding interaction terms for these characteristics to our primary models. Because practice patterns may change with increased experience, we fit models that included year lags for Medicare ACO participation.

We performed all analyses using SAS Version 9.4 (Cary, NC). Tests were two-tailed, and we set the probability of Type 1 error at 0.05. Our institution’s Health Sciences Institutional Review Board deemed this study to be exempt from its oversight.

Results

Our study included 153,476 beneficiaries admitted for AMI to 2,998 hospitals (401 ACO participating and 2,597 non-participating hospitals), and 260,420 beneficiaries admitted for CHF to 3,208 hospitals (412 ACO participating and 2,796 non-participating hospitals). Table 1 compares structural differences between ACO participating and non-participating hospitals during the pre-contract period. ACO participating hospitals tended to be larger, were more likely to be teaching and non-profit, and concentrated in the Northeast.

Table 1.

Beneficiary and Hospital Characteristics for the Acute Myocardial Infarction and Congestive Heart Failure Cohorts in the Pre-Contract Period, Stratified by Hospital ACO Participation.

	Acute Myocardial Infarction			Congestive Heart Failure
	Non-Participating	Participating	P-Value	Non-Participating	Participating	P-Value
No. of beneficiaries	54,800	17,984		94,048	28,376
No. of hospitals	2,597	401		2,796	412
Mean early episode payment, in dollars	24,427	25,091	<.001	18,909	19,671	<.001
Mean late episode payment, in dollars	9,906	10,462	0.001	12,420	13,105	<.001
Beneficiaries characteristics
Female (%)	49.9%	49.6%	0.490	55.7%	55.6%	0.671
Age group, in years (%)			0.178			0.164
65 to 69	16.1%	16.3%		11.7%	12.0%
70 to 74	17.2%	17.8%		12.2%	11.6%
75 to 79	18.0%	17.5%		15.6%	14.9%
80 and older	48.7%	48.4%		60.5%	61.5%
Race (%)			<.001			0.001
White	88.4%	87.2%		85.1%	84.8%
Black	7.1%	8.4%		10.3%	10.9%
Other	4.5%	4.4%		4.6%	4.3%
No. of HCCs (%)			0.641			0.296
0	20.6%	21.1%		7.2%	7.2%
1	20.4%	20.4%		10.6%	10.3%
2	17.4%	16.3%		14.2%	13.9%
3 or more	41.6%	42.2%		68.0%	68.6%
Hospital factors
Number of beds (%)			<.001			<.001
Less than 200	30.2%	20.1%		36.3%	23.6%
200 to 349	29.7%	26.1%		28.4%	26.4%
350 to 499	18.7%	19.6%		16.8%	19.0%
500 or more	21.4%	34.2%		18.5%	31.0%
% Teaching hospital	13.7%	23.7%	<.001	12.2%	22.1%	<.001
% Urban	97.6%	99.6%	<.001	95.8%	99.2%	<.001
Profit status (%)			<.001			<.001
For-profit	18.8%	6.7%		19.3%	6.8%
Non-profit	69.8%	89.9%		68.9%	89.3%
Other	11.4%	3.4%		11.8%	3.9%
Region (%)			<.001			<.001
South	44.3%	23.2%		44.2%	22.8%
Northeast	18.1%	28.3%		19.5%	30.0%
West	14.1%	16.7%		13.1%	14.8%
Midwest	23.5%	31.8%		23.2%	32.4%

Abbreviations: ACO, accountable care organization; HCC, hierarchical condition category; No., number.

The characteristics of beneficiaries in the AMI and CHF cohorts during the pre-ACO contract period are shown in Table 1. For both cohorts, beneficiaries admitted at ACO participating and non-participating hospitals had similar age and gender distributions. They also had similar levels of comorbid illness. While ACO participating hospitals treated a higher proportion of black beneficiaries, this difference was not clinically meaningful. In terms of baseline spending, ACO participating hospitals had significantly higher mean early and late episode payments for both AMI and CHF cohorts.

Figure 1 displays early and late payments for AMI and CHF episodes by year across ACO participating (stratified by their ACO contract start date) and non-participating hospitals. Although there was a general trend towards declining early episode spending for AMI (1A), it remained relatively flat for CHF (1C). While late episode spending for AMI (1B) and CHF (1D) also declined over the study interval, there was no noticeable separation of spending trends for ACO participating hospitals following the start of an ACO contract.

Figure 1.

Trends in Early and Late Payments for Acute Myocardial Infarction (AMI) and Congestive Heart Failure (CHF) Episodes across ACO Participating and Non-Participating Hospitals. Note: Trends in early and late episode payments for the AMI cohort are displayed in panels A and B respectively. Trends in early and late episode payments for the CHF cohort are displayed in panels C and D, respectively. Red, green, and purple vertical lines indicate ACO contract start dates for the 2012, 2013, and 2014 Medicare ACO participating hospitals, respectively. Trends are price-standardized, but not risk-adjusted.

Results of our multivariable analysis are displayed in Table 2. Admission to an ACO participating hospital was not associated with changes in early episode spending {AMI: $95 per beneficiary [95% confidence interval (CI), -$481 to $671]; CHF: $158 (95% CI, -$290 to $605)}. However, it was associated with significant reductions in late episode spending for both cohorts [AMI: -$680 (95% CI, -$1,348 to -$11); CHF: -$889 (95% CI, -$1,465 to -$313). These savings were driven by reductions in payments for readmissions during the late episode (eTable 2).

Table 2.

Multivariable Model Examining the Association between Hospital Participation in a Medicare ACO and Early and Late Payments for Acute Myocardial Infarction and Congestive Heart Failure Episodes.

	Acute Myocardial Infarction						Congestive Heart Failure
	Early Episode Spending			Late Episode Spending			Early Episode Spending			Late Episode Spending
Parameter	Estimate	95% CI	P Value	Estimate	95% CI	P Value	Estimate	95% CI	P Value	Estimate	95% CI	P Value
Medicare ACO Status
Participating	$95	(−481, 671)	0.746	-$680	(−1348, −11)	0.046	$158	(−290, 605)	0.490	-$889	(−1465, −313)	0.003
Not Participating	Referent			Referent			Referent			Referent
Year (Continuous)	-$612	(−701, −523)	<.001	-$124	(−223, −25)	0.014	-$92	(−148, −35)	0.001	$70	(−18, 158)	0.117
Quarter
Quarter 1	$1,983	(1675, 2291)	<.001	$650	(352, 949)	<.001	$1,850	(1636, 2063)	<.001	$2,298	(2031, 2566)	<.001
Quarter 2	$1,560	(1250, 1869)	<.001	$825	(530, 1121)	<.001	$1,866	(1638, 2094)	<.001	$2,156	(1877, 2435)	<.001
Quarter 3	$1,219	(896, 1543)	<.001	$68	(−219, 356)	0.641	$1,982	(1748, 2216)	<.001	$908	(612, 1205)	<.001
Quarter 4	Referent			Referent			Referent			Referent
Age (Continuous)	-$277	(−291, −263)	<.001	-$196	(−209, −182)	<.001	-$251	(−263, −239)	<.001	-$358	(−372, −344)	<.001
Sex
Female	-$1,450	(−1687, −1212)	<.001	$406	(189, 622)	<.001	-$878	(−1041, −715)	<.0001	$429	(219, 639)	<.001
Male	Referent			Referent			Referent			Referent
Race
Black	-$746	(−1273, −219)	0.006	$1,689	(1135, 2242)	<.001	-$510	(−853, −166)	0.0036	$2,582	(2137, 3026)	<.001
Other	$297	(−376, 969)	0.387	$13	(−566, 591)	0.967	-$1,112	(−1525, −698)	<.0001	$410	(−209, 1030)	0.195
White	Referent			Referent			Referent			Referent

Abbreviations: ACO, accountable care organization; CI, confidence interval.

Note: Quarter 1 corresponded to January through March; Quarter 2 corresponded to April through June; Quarter 3 corresponded to July through September; and Quarter 4 corresponded to October through December. Estimates for individual HCCs and hospital fixed effects are not listed due to space constraints but are available upon request.

Note: To understand the referent within this multivariable model, consider the following interpretation: When examining the association between gender and late episode spending among beneficiaries admitted for CHF, payments were, on average, $429 higher for women than men.

Figure 2 illustrates findings from our sensitivity analyses. Estimated late episode savings were consistently seen in ACOs with contract start dates in 2012. In addition, for the CHF cohort, hospital-led ACOs and organizations in the second year of their ACO contracts were associated with greater late episode spending reductions. However, contract start date, group organizational structure, and years of program experience did not modify Medicare ACO effects on early episode spending for either the AMI or CHF cohorts.

Figure 2.

Differential Changes in Early and Late Spending for AMI and CHF Episodes Based on ACO Contract Start Date, Organizational Type, and Years of Program Experience. Note: The point estimates represent change in spending (dollars), and the error bars represent 95% confidence intervals.

Discussion

We found that late episode spending decreased by $680 and $889 among beneficiaries admitted with AMI and CHF, respectively, to an ACO participating hospital. These spending reductions were driven by lower payments for readmissions during the late episode. To provide some context for our findings, we estimate that the Medicare program realized approximately $106.4 million in savings from the over 26,000 beneficiaries in our study that received their inpatient AMI and CHF care at an ACO participating hospital. These savings were consistently greater in early ACO model adopters (i.e., organizations with contract start dates in 2012).

Initial evaluations of Medicare ACO programs have focused primarily on their global effects, demonstrating modest spending reductions related to receipt of care from participating provider groups. For instance, the first full year of Shared Savings Program contracts was associated with a 1.4% savings among 2012 program entrants; however, no savings were achieved by 2013 program entrants.¹¹ Further, spending in Pioneer ACOs fell by 1.2% during the program’s first year, but these reductions were substantially lower by year 2, leading to a net savings of just $67 per beneficiary in 2014.¹⁰ Our study is the first to examine ACO effects on episode spending for two diverse cardiovascular diseases that are common in older adults.

To our surprise, we observed consistent results for AMI and CHF episodes. CHF is a chronic condition marked by acute exacerbations, involving multiple care transitions,^{20, 21} which we posited would lie more within the purview of ACOs than AMI—a sudden and severe event. Hence, our findings suggest positive spillovers from ACOs’ processes for chronic disease management to acute cardiac conditions. Moreover, our findings on the drivers of savings in the late episode indicate possible mechanisms by which ACOs may exert their effects. An AMI or CHF admission at a hospital participating in an ACO could draw the organization’s attention to the patient and trigger care management programs that facilitate care transitions. This long-view approach to improved cardiac management, which separates ACOs from other health policy reforms, results in savings through fewer readmissions in the late episode period.

Our study should be interpreted in the context of several limitations. Participation in an ACO is voluntary. Consequently, groups who chose to participate (as well as the patients whom they serve) may be different from those who do not. While we controlled for observed patient factors and time-invariant hospital differences through our use of hospital fixed effects, we acknowledge that there may be unobserved differences confounding the observed association between ACO participation and episode spending. Our findings on higher baseline episode spending among participating hospitals and greater late episode savings among early ACO adopters lend support to the possibility of self-selection. Second, over the study period, Medicare launched several other payment reforms, including the Hospital Readmissions Reduction Program and the Bundled Payment for Care Initiative, which could also affect episode spending. However, it is important to note that these reforms target early episode payments, while the spending reductions that we observed were concentrated in the late episode.

Third, we acknowledge the possibility of differences in disease severity between beneficiaries admitted to participating and non-participating hospitals, which may confound the observed association. While we used an established claims-based method for measuring levels of comorbidity, ACOs have incentives to code more intensively as a means to increase HCC scores.²² As such, for a given HCC score, a beneficiary admitted to a participating hospital may be, on average, healthier than one admitted to a non-participating hospital.

Lastly, we examined admissions to ACO participating hospitals and episode spending related to those admissions, but Medicare beneficiaries retain choice and can use whatever hospital they want. Therefore, it is unlikely that all AMI and CHF admissions to ACO participating hospitals are made by ACO beneficiaries. To the extent that an ACO’s effect on late episode spending is moderated, in part, by improved outpatient care coordination between cardiology and primary care providers in the ACO, its signal could be muted if a large proportion of admissions to a participating hospital is made by non-ACO beneficiaries. As such, our analysis may underestimate the savings associated with ACOs.

Conclusions

These limitations notwithstanding, our findings have important implications for patients and policymakers. They highlight that the action of ACOs appears to be in the late episode, as opposed to bundled payments and readmission penalties, which almost exclusively target spending in the early episode. Future work is needed to determine if these findings extend to other common medical and surgical conditions, as well as to elucidate the causal pathways by which ACOs affect outcomes.

What is Known:

• -Episode spending for cardiovascular disease like acute myocardial infarction (AMI) and congestive heart failure (CHF) varies widely across hospitals due, in part, to a fragmented delivery system.
• -Thus, health policy reforms designed to reduce care fragmentation, such as Medicare’s accountable care organizations (ACOs), may help reduce the costs of cardiovascular care.

What the Study Adds:

• -While early episode spending (index admission to 90 days post-discharge) was unchanged, late episode spending (91 to 365 days post-discharge) decreased significantly among Medicare beneficiaries admitted with AMI and CHF to an ACO participating hospital.
• -These spending reductions were driven by lower payments for readmissions during the late episode.
• -Taken together, ACOs may complement other health policy reforms, including bundled payments and readmission penalties, that target the early episode.