Medicare ACOS Are Not Associated with Reductions in the Use Of Low-Value Coronary Revascularization

Abstract

Background

Because specialty care accounts for half of Medicare expenditures, improving its value is critical to the success of Medicare accountable care organizations (ACOs) in curbing spending growth. However, whether ACOs have reduced low-value specialty care without compromising use of high-value services remains unknown.

Methods and Results

Using national Medicare data, we identified two cohorts: beneficiaries for whom the value of coronary revascularization is lower [those with ischemic heart disease without angina, congestive heart failure, or recent admission for acute myocardial infarction (AMI)] and beneficiaries for whom its value is higher (those with recent AMI admission). We then determined the provider groups who cared for the cohorts, distinguishing between those participating (n=298) and those not participating in a Medicare ACO (1,329). After measuring the provider groups’ use of coronary artery bypass grafting and percutaneous coronary intervention among the two cohorts, we fit multivariable models to test the statistical significance of rates of change in low- and high-value revascularization following ACO participation. During the pre-ACO period, participating and non-participating provider groups had similar rates of low- and high-value revascularization. Our multivariable model results show that rates of change for low- and high-value coronary revascularization were not altered by a provider group’s participation in a Medicare ACO {lower value: difference, −0.04 per year [95% confidence interval (CI), −0.11 to 0.03]; higher value: difference, 0.96 per year (95% CI, −0.46 to 2.4)}.

Conclusions

We found no association between provider group participation in a Medicare ACO and use of low- or high-value coronary revascularization.

Given Medicare’s projected increases in spending,¹ many hope that the expansion of accountable care organizations (ACOs) will help curb future spending growth. Accounting for nearly half of all Medicare expenditures,^2,3 attention must be paid to expensive specialty care, particularly cardiovascular services like coronary revascularization, which make an outsize contribution to spending.⁴ Yet when the leaders of Medicare ACOs were recently surveyed about their organizations’ early strategic plans for achieving shared savings, conspicuously missing was any mention of specialty care.⁵

However, Medicare ACOs have the potential to influence specialty care substantially. A core premise of ACOs is shared accountability in patient management and coordination between primary and specialty care. Since participants are held to quality and spending targets for not only the services that they provide but also the care delivered by others, ACO primary care physicians may lead specialists to limit a procedure’s use in situations where its value is low. They might also steer referrals toward specialists who provide higher value care. Further, collective incentives may motivate specialists participating in ACOs to lower their costs by reducing procedures among marginal patients. On the other hand, ACO financial pressures may bluntly discourage use of all procedures, regardless of their value.

To better understand the effects of Medicare ACOs on specialty care, we analyzed national Medicare claims from beneficiaries with ischemic heart disease, who were candidates for coronary revascularization. Coronary revascularization is an ideal procedure to study given that its value to patients varies substantially from scenarios where it can be lifesaving [acute myocardial infarction (AMI)] to those where its value is less certain (asymptomatic ischemia). We determined the provider groups who cared for these beneficiaries, distinguishing between those participating and those not participating in a Medicare ACO. We then compared coronary revascularization use among these groups before and after ACO implementation to assess whether ACOs reduced population-based rates of low-value procedures without affecting high-value specialty care.

Methods

Data source and study population

Our analyses were based on national Medicare claims from a 20% random sample of beneficiaries, including data from the Carrier, Denominator, Medicare Provider Analysis and Review, and Outpatient research identifiable files, allowing us to capture both inpatient and outpatient cardiac services. Because of the sensitive nature of these files, requests to access them from qualified researchers trained in human subject confidentiality protocols may be sent to the Centers for Medicare & Medicaid Services. For each study year from 2008 to 2014, we included fee-for-service beneficiaries if they had continuous enrollment in Parts A and B in that year and the year prior (for purposes of comorbidity assessment). As in other studies,^6,7 we also required that beneficiaries received at least one primary care service (Healthcare Common Procedure Coding System codes 99201 through 99215, 99304 through 99350, G0402, G0438, and G0439) in the study year furnished by an ACO professional.

For each study year, we used International Classification of Diseases, Ninth Revision, codes to identify two cohorts (code lists available in the Appendix). The first cohort included beneficiaries for whom the value of coronary revascularization is widely considered to be lower.⁸ It included beneficiaries with evidence of stable, asymptomatic ischemic heart disease without concomitant diagnoses of angina, congestive heart failure, or recent AMI hospitalization. For the second cohort, we identified beneficiaries for whom the value of coronary revascularization is generally regarded as higher.⁹ This consisted of those admitted in a given year with a primary diagnosis of AMI.

Assigning beneficiaries to provider groups who cared for them

Next, we used two steps to assign beneficiaries to provider groups (defined as primary care physicians, specialists, and hospitals who care for common patients).¹⁰ First, we linked beneficiaries to their predominant ambulatory provider, defined as the primary care physician who furnished the most primary care services during a given year. For those who received no primary care services by any primary care physician, we assigned them to the specialist or non-physician practitioner responsible for the most primary care services. Second, we linked all predominant ambulatory providers to the acute care hospital where they performed the plurality of their work during the year. Providers who did not bill for any inpatient care were linked to the hospital where most of their patients were referred that year.

Distinguishing provider groups participating in a Medicare ACO

After determining the provider groups who cared for beneficiaries with ischemic heart disease, we used the Leavitt Partners ACO Database to distinguish those participating in a Medicare ACO. This validated database has been in existence since 2010 and contained 839 Medicare, Medicaid, and commercial ACOs at the time of our analysis.¹¹ 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. We considered a provider group to be participating in a Medicare ACO if the hospital where the primary care physicians and specialists in it delivered the plurality of their inpatient care was identified as either being owned by or affiliated with the ACO.

Measuring rates of coronary revascularization

Our unit of analysis was quarterly provider group-level rates of total coronary revascularization for both of our cohorts. Among the cohort with asymptomatic ischemia, the numerator for our rate calculation was the number of times in a study quarter that coronary artery bypass or percutaneous coronary intervention (code lists available in the Appendix) was performed on a given provider group’s assigned beneficiaries. Of note, these procedures are largely done on an inpatient basis. The denominator corresponded to the number of the provider group’s assigned beneficiaries, who had a diagnosis of asymptomatic ischemia that same quarter.

Among the cohort with hospitalization for AMI during the calendar year (for whom revascularization is high-value), the numerator for our rate calculation was the number of times in a study quarter that coronary artery bypass grafting or percutaneous coronary intervention was performed on a given provider group’s assigned beneficiaries during hospitalization and up to 30 days after discharge. The denominator was the number of the provider group’s assigned beneficiaries, who were hospitalized that same quarter for AMI. Using previously described methods,¹² we risk-adjusted all rates using beneficiary-level covariates for age, gender, race/ethnicity, coexisting medical conditions (based on hierarchical condition categories¹³), and socioeconomic status.¹⁴

Statistical analysis

We first examined differences between ACO participating and non-participating provider groups with respect to the beneficiaries for whom they cared, using parametric and non-parametric tests where appropriate. We also linked data from the American Hospital Association Annual Survey to compare groups based on characteristics of the hospitals where they delivered inpatient care, including hospital size, urbanicity, for-profit and teaching status, and region.¹⁵ Since capability and capacity may influence intervention decisions, we limited our analysis to provider groups that performed at least 10 percutaneous coronary interventions and/or coronary artery bypass grafting procedures annually.

For both cohorts, we then plotted adjusted rates of total coronary revascularization in ACO participating and non-participating provider groups by study year. To test the statistical significance of rates of change in revascularization following ACO participation, we fit multivariable linear regression models (see Appendix for full model specification). We accounted for repeated measures by using generalized estimating equations with robust variance estimators,^16,17 weighted by the number of beneficiaries assigned to a group per quarter. Our models included a time-varying indicator for the group (set to 1 the quarter when a group began Medicare ACO participation and 0 otherwise) and covariates for the hospital characteristics described above. To address temporal trends, we introduced quarter and year as fixed effects.

To test the robustness of our findings, we conducted a series of sensitivity analyses. To see if ACO effects differed between early and late adopters, we constructed separate models for groups with contract start dates in 2012, 2013, and 2014. To examine whether a group’s organizational structure, degree of financial risk assumed [based on its participation in the Pioneer ACO or Medicare Shared Savings Program (MSSP)], or its participation in a commercial or Medicaid ACO contract were effect modifiers, we added interaction terms to our primary models. Because practice patterns may change with increased experience, we also fit models that included year lags for Medicare ACO participation. Finally, we carried out an exploratory analysis on provider groups in the MSSP to determine whether specialist participation (defined as the number of specialist physicians per 1000 beneficiaries in the ACO to which a given provider group belonged) modified ACO effects.

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. The Health Sciences Institutional Review Board at our institution deemed this study to be exempt from its oversight.

Results

We identified 298 provider groups that participated in a Medicare ACO contract at some point during the study interval and 1,329 non-participating groups. As shown in Table 1, 40% were early ACO model adopters (i.e., they had a contract start date in 2012). The overwhelming majority (91%) participated in the Medicare Shared Savings Program (as opposed to the Pioneer ACO Program). Most participating provider groups (53.7%) included hospital partners. Sixty-two percent were concurrently in a commercial ACO, but only 7.1% also participated in a Medicaid ACO contract.

Table 1.

Characteristics of ACO participating and non-participating provider groups.

	Participating (n=298)	Non-Participating (n=1,329)	P-Value
Characteristics of the provider group
Contract start date
2012	119 (39.9)	NA	NA
2013	99 (33.2)	NA	NA
2014	80 (26.9)	NA	NA
Organizational type
Physician-Hospital Partnership	160 (53.7)	NA	NA
Hospital-led	103 (34.6)	NA	NA
Physician-led	35 (11.7)	NA	NA
Program type
Medicare Shared Savings	272 (91.3)	NA	NA
Pioneer ACO Program	26 (8.7)	NA	NA
Commercial ACO contract
Yes	186 (62.4)	NA	NA
No	112 (37.6)	NA	NA
Medicaid ACO contract
Yes	21 (7.1)	NA	NA
No	277 (93.0)	NA	NA
Characteristics of the hospitals where provider groups deliver inpatient care
Hospital size			0.0265
Small	11 (3.7)	94 (7.1)
Medium	63 (21.1)	328 (24.7)
Large	224 (75.2)	907 (68.2)
For-profit hospital			<0.001
Yes	30 (10.1)	327 (24.6)
No	268 (89.9)	1,002 (75.4)
Teaching hospital			0.0155
Yes	53 (17.8)	166 (12.5)
No	245 (82.2)	1,163 (87.5)
Urban hospital			0.1543
Yes	298 (100.0)	1,320 (99.3)
No	0 (0.0)	9 (0.7)
Geographic region			<0.001
Northeast	55 (18.5)	178 (13.4)
Midwest	112 (37.6)	305 (22.9)
South	72 (24.2)	573 (43.1)
West	59 (19.8)	273 (20.5)

Abbreviations: ACO, accountable care organization; NA, not applicable.

Note: Parentheses indicate percent.

Table 2 compares beneficiaries cared for by ACO participating and non-participating provider groups during the pre-contract period. The distribution of asymptomatic ischemia was similar between groups, and differences in their beneficiaries’ age, gender, race/ethnicity, and level of comorbid illness were not clinically meaningful. However, beneficiaries cared for by non-participating groups tended to come from lower socioeconomic strata (P<0.01). Participating groups tended to practice at larger, not-for-profit, and teaching hospitals (Table 1). Further, their hospitals were concentrated disproportionately in the Northeast and Midwest.

Table 2.

Characteristics of the beneficiaries with ischemic heart disease cared for by ACO participating and non-participating groups during the pre-contract period.

Beneficiary Characteristic	Pre-Contract Period (2008 to 2011)		Difference between Groups during Pre-Contract Period	P-Value
	Participating (n=444,851)	Non-Participating (n=1,715,606)
IHD Cohort (%)				0.23
Stable ischemia	96.8	96.9	0.0
Recent AMI	3.2	3.1	0.0
Mean age ± SD (year)	77.2 ± 7.2	77.1 ± 7.2	−0.1	<0.01
Female (%)	46.2	45.6	0.6	<0.01
Race/ethnicity (%)				<0.01
White	89.4	89.9	−0.5
Black	6.6	5.8	0.8
Hispanic	1.1	1.5	−0.4
Other	2.9	2.9	0.0
Mean no. of HCCs ± SD	1.9 ± 1.7	1.9 ± 1.7	0.0	0.81
SE stratum (%)				<0.01
Low	25.0	33.6	−8.6
Medium	32.7	33.6	−0.9
High	42.3	32.8	9.5

Abbreviations: AMI, acute myocardial infarction; HCCs, hierarchical condition categories; IHD, ischemic heart disease; no., number; SD, standard deviation; SE, socioeconomic.

Figure 1 displays adjusted rates (per 100 beneficiaries per year) of total coronary revascularization across participating and non-participating provider groups (the former are stratified by their Medicare ACO contract start date). During the pre-contract period, rates of revascularization among beneficiaries with asymptomatic ischemia (panel A) and those among beneficiaries hospitalized for AMI (B) were similar between participating and non-participating provider groups (lower value: 1.0 versus 1.1, respectively; higher value: 53.9 versus 54.2, respectively). Over the study interval, while rates of lower value revascularization remained relatively flat (from 1.0 in 2008 to 1.0 in 2014), rates of higher value revascularization rose, on average, by 13.5% (from 51.5 in 2008 to 58.5 in 2014). For both cohorts, there was no noticeable separation of rates in participating provider groups following the start of an ACO.

Figure 1.

Rates of total coronary revascularization among ACO participating and non-participating provider groups over the study interval for low- (A) and high-value scenarios.

Note: Participating provider groups are stratified by contract start date.

As displayed in Table 3, our multivariable model results show that rates of change for low- and high-value coronary revascularization were not altered by a provider group’s participation in a Medicare ACO {lower value: difference, −0.04 per year [95% confidence interval (CI), −0.11 to 0.03]; higher value: difference, 0.96 per year (95% CI, −0.46 to 2.4)}. When considered separately, we noted similar findings when rates for coronary artery bypass grafting [lower value: difference, −0.01 per year (95% CI, −0.04 to 0.01); higher value: difference, 0.01 per year (95% CI, −0.9 to 0.9)] and percutaneous coronary intervention [lower value: difference, −0.02 per year (95% CI, −0.08 to 0.03); higher value: difference, 0.7 per year (95% CI, −0.8 to 2.2)].

Table 3.

Multivariable model examining the relationship between provider group participation in a Medicare ACO and the rate of coronary revascularization, stratified by the value of intervention.

	Parameter Estimates for Percentage Change of Coronary Revascularization
	Lower Value		Higher Value
Variable	Coeff.	95% CI	Coeff.	95% CI
Year (referent 2008)
2009	0.03	(−0.01, 0.07)	3.10	(2.14, 4.06)
2010	0.01	(−0.03, 0.05)	4.01	(3.06, 4.96)
2011	−0.11	(−0.15, −0.07)	4.55	(3.59, 5.52)
2012	−0.16	(−0.21, −0.12)	5.17	(4.19, 6.15)
2013	−0.18	(−0.23, −0.14)	6.41	(5.40, 7.42)
2014	−0.03	(−0.08, 0.02)	6.30	(5.23, 7.37)
Quarter (referent first)
Second	−0.09	(−0.11, −0.07)	−0.38	(−1.09, 0.34)
Third	0.14	(0.10, 0.17)	−1.14	(−1.85, −0.42)
Fourth	0.15	(0.11, 0.19)	−4.05	(−4.78, −3.32)
Urban hospital (referent rural)
Urban	0.31	(0.06, 0.56)	0.83	(−2.16, 3.83)
Teaching hospital (referent non-teaching)
Teaching	−0.15	(−0.23, −0.08)	0.89	(−0.18, 1.95)
Non-profit hospital (referent for-profit)
Non-profit	−0.02	(−0.10, 0.06)	0.06	(−1.06, 1.18)
Hospital region (referent Northeast)
Midwest	0.45	(0.35, 0.54)	5.20	(3.96, 6.45)
South	0.29	(0.22, 0.37)	3.01	(1.77, 4.24)
West	0.30	(0.21, 0.39)	4.13	(2.69, 5.56)
Hospital size (referent small)
Medium	−0.17	(−0.33, −0.01)	−1.11	(−3.76, 1.54)
Large	−0.21	(−0.36, −0.06)	−0.07	(−2.63, 2.49)
1st year participating in ACO (referent non-participating)
ACO	−0.04	(−0.11, 0.03)	0.96	(−0.46, 2.38)

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

Figure 2 illustrates findings from our sensitivity analyses. We observed no association between ACO participation and total coronary revascularization rates when we analyzed early and late model adopters separately. Group organizational structure, degree of financial risk, and commercial or Medicaid ACO participation did not modify Medicare ACO effects on rates of lower or higher value revascularization. When we examined organizational learning effects by including year lags for ACO participation in our models, we noted decreases in the rates of lower value total coronary revascularization and percutaneous coronary intervention in year two [total: difference, −0.1 (95% CI, −0.2 to 0.0); PCI: difference, −0.1 (95% CI, −0.2 to −0.0)], but this effect was not maintained in year three. Finally, on our exploratory analysis, we found evidence suggesting that provider groups with high specialist participation may decrease rates of lower value revascularization [difference, −0.10 per year (95% CI, −0.27 to 0.07)] and increase rates of higher value revascularization [difference, 1.62 per year (95% CI, −1.70 to 4.93)], although these effects were not statistically significant.

Figure 2.

Differential changes in rates of coronary revascularization in low- and high-value scenarios based on ACO contract start date, organizational type, concurrent commercial or Medicaid contract, and experience. Abbreviations: ACO, accountable care organization; No., number. Note: Error bars represent 95% confidence interval

Discussion

For both low- and high-value scenarios, we found that rates of change for total coronary revascularization, coronary artery bypass grafting, and percutaneous coronary intervention among beneficiaries treated by provider groups participating in a Medicare ACO contract were similar to those of beneficiaries treated by non-participating groups. These findings indicate that the incentives of MSSP and Pioneer may be too weak to limit the growth in spending on cardiac specialty care and that refinements of current alternative payment models and risk-bearing contracts are needed to influence provider behavior.¹⁸

Data on the early experience of ACOs with specialty care are sparse; only two studies have examined ACO effects on specialty care delivery. The first was of an ACO pilot, comparing use of cardiovascular imaging and procedures among participating groups before and after implementation.¹⁹ Consistent with our results, the study’s authors found no evidence that the pilot had any effect on the use of lower value care.¹⁹ More recently, Schwartz and colleagues assessed the performance of Pioneer ACOs.²⁰ In contrast to our results, they demonstrated marginally significant but modest reductions in some lower value procedural care (e.g., carotid endarterectomy for asymptomatic patients) during the program’s first year.²⁰

One reason why we observed no effects of Medicare ACOs on lower value coronary revascularization procedures is that these procedures are already infrequently performed. As such, ACO leaders may have chosen instead to focus on higher prevalence, lower value activities where more room for improvement exists. An alternative explanation has to do with the fact that participating provider groups, and particularly specialists, have too little “skin in the game.” Specifically, most in our study were in one-sided risk models that required no penalty for losses. Thus, the specialists participating in these ACOs may have limited incentive to alter their practice patterns. A second explanation pertains to a lack of specialist involvement. Half of Medicare ACOs have no formal arrangements with specialty practices.²¹ Given that ACO primary care physicians, by law, cannot restrict their beneficiaries’ care choices,²² weakly connected specialists may fall outside the reach of ACOs.

Our study has limitations that merit discussion. First, working with medical claims, we do not know beneficiaries’ burden of symptoms. Therefore, we cannot comment on whether a given coronary revascularization procedure was appropriate or note. Nevertheless, we can comment on the value of intervention. Indeed, other investigators have used frameworks similar to ours for estimating lower value PCI use with administrative data. Second, participation in a Medicare ACO is non-random and may be endogenous with organizational factors. Insofar as participating provider groups already had structures and processes in place before ACO implication to limit low-value procedure utilization among specialists, our results would be biased. That said, we would expect the direction of this bias to be positive towards finding an effect of ACOs.

Third, we must acknowledge the possibility of misclassification bias from two sources. For one, we determined a provider group’s participation in a Medicare ACO based on where its primary care physicians and specialists delivered inpatient services. In addition, our method for categorizing the value of coronary revascularization depends, in part, on the accuracy of providers’ diagnosis coding. However, we have reasons to believe that the risk of this bias is low. When we compared results from our group ACO participation algorithm with data from the Shared Saving Program ACO Provider-level research identifiable file, we observed that our assignment approached 90% accuracy. What is more, the rates of lower value PCI, which we observed are similar to those published elsewhere.²³

Fourth, we concentrated on lower value procedures for ischemic heart disease only. We did so because these procedures are costly, accounting for $6 billion annually in Medicare spending and involve a diverse group of specialists; but we acknowledge that findings on them may not be generalizable to lower value care for other conditions. Finally, it is unclear whether our findings apply to alternative ACO models, emphasizing greater financial risk for providers.

Notwithstanding these limitations, our findings have important implications for clinician leaders and policymakers. The focus of the initial iteration of Medicare ACOs has been on enhanced primary care for beneficiaries with multiple chronic medical conditions and complex medical needs. This narrow focus has produced, at best, slight savings for Medicare. As suggested by our exploratory analysis, new more comprehensive designs that better incorporate specialists may be necessary if ACOs are to reach their full potential. Without such changes, the current Medicare ACO programs may struggle to control specialist costs.

What is Known

• Data on the early experience of Medicare accountable care organizations (ACOs) with specialty care are sparse.

• Studies on a commercial ACO pilot and the first-year performance of the Pioneer ACO Model Program suggest that they may decrease utilization lower value procedural care.

What the Study Adds

• Regardless of a procedure’s value, its use was not altered by provider participation in a Medicare ACO.

• This suggest that the incentives of current Medicare ACO programs may be too weak to influence specialist physician behavior.