Variation in Risk‐Standardized Acute Admission Rates Among Patients With Heart Failure in Accountable Care Organizations: Implications for Quality Measurement

Abstract

Background

Accountable care organizations (ACOs) aim to improve health care quality and reduce costs, including among patients with heart failure (HF). However, variation across ACOs in admission rates for patients with HF and associated factors are not well described.

Methods and Results

We identified Medicare fee‐for‐service beneficiaries with HF who were assigned to a Medicare Shared Savings Program ACO in 2017 and survived ≥30 days into 2018. We calculated risk‐standardized acute admission rates across ACOs, assigned ACOs to 1 of 3 performance categories, and examined associations between ACO characteristics and performance categories. Among 1 232 222 beneficiaries with HF, 283 795 (mean age, 81 years; 54% women; 86% White; 78% urban) were assigned to 1 of 467 Medicare Shared Savings Program ACOs. Across ACOs, the median risk‐standardized acute admission rate was 87 admissions per 100 people, ranging from 61 (minimum) to 109 (maximum) admissions per 100 beneficiaries. Compared to the overall average, 13% of ACOs performed better on risk‐standardized acute admission rates, 72% were no different, and 14% performed worse. Most ACOs with better performance had fewer Black beneficiaries and were not hospital affiliated. Most ACOs that performed worse than average were large, located in the Northeast, had a hospital affiliation, and had a lower proportion of primary care providers.

Conclusions

Admissions are common among beneficiaries with HF in ACOs, and there is variation in risk‐standardized acute admission rates across ACOs. ACO performance was associated with certain ACO characteristics. Future studies should attempt to elucidate the relationship between ACO structure and characteristics and admission risk.

Nonstandard Abbreviations and Acronyms

ACO

accountable care organization

CMH

Cochran–Mantel–Haenszel test

FFS

fee‐for‐service

MSSP

Medicare Shared Savings Program

RSAAR

risk‐standardized acute admission rate

Research Perspective.

What Is New?

• We identified Medicare fee‐for‐service beneficiaries with heart failure enrolled in Medicare Shared Savings Program acountable care organizations (ACOs) and calculated risk‐standardized acute admission rates across ACOs, assigned ACOs to performance categories, and examined associations between ACO characteristics and performance categories.

• The median risk‐standardized acute admission rate was 87 admissions per 100 people, with variation across ACOs.

• ACO performance was associated with hospital affiliation, geographic region, ACO size, proportion of primary care providers, and proportion of Black beneficiaries.

What Question Should Be Addressed Next?

• Further investigation into drivers of better or worse ACO performance in risk‐standardized acute admission rates among patients with HF is warranted.

Heart failure (HF) affects ≈6 million individuals in the United States, ¹ is the leading cause of hospital admission in older adults, ² is associated with high rates of readmissions ³ and death, ¹ and costs the US health care system ≈$40 billion per year. ⁴ Because of the public health burden of HF, health care delivery and care coordination interventions have often targeted the management of patients with HF. Accountable care organizations (ACOs) encourage provider networks to manage a panel of patients across a continuum of settings, with the goal of improving care and reducing costs. In 2012, the Centers for Medicare and Medicaid Services created an ACO program called the Medicare Shared Savings Program (MSSP), which aims to provide coordinated care for Medicare fee‐for‐service (FFS) beneficiaries. ⁵ From 2012 to 2020, the MSSP nearly doubled in size, and by 2020 it included approximately one‐third of the FFS Medicare population. ⁶

Although there is mixed evidence that ACOs reduce use or improve patient outcomes generally, ⁷ patients with HF may derive particular benefit from an ACO given the high health care use and cost associated with this condition, and ACOs' focus on improving care coordination, medication and appointment adherence, and social support. Data suggest that some ACOs have been able to achieve low admission rates for patients with HF. ⁸ However, few studies have described performance variation across ACOs related to specific diseases or conditions, and none have focused on the population of patients with HF. Therefore, in a population limited to FFS Medicare beneficiaries with a diagnosis of HF, we aimed to (1) explore variation in risk‐standardized acute admission rates (RSAARs) across MSSP ACOs, (2) identify high‐ and low‐performing MSSP ACOs, and (3) determine associations between performance on RSAAR and ACO characteristics.

Methods

Because of the limitations of our data use agreement with the Centers for Medicare and Medicaid Services and the Research Data Assistance Center, we are not able to share any identifiable data used in this study. Requests to access deidentified data from qualified researchers trained in human subject confidentiality protocols may be sent to Dr Penelope Pekow at penny.pekow@baystatehealth.org.

Study Overview and Data Sources

We identified a sample of Medicare beneficiaries with a diagnosis of HF between 2014 and 2017 who were assigned to MSSP ACOs in 2017 and met eligibility in 2018 (outcome year). Using previously described methods ⁹ , ¹⁰ and the 2017 MSSP ACO Provider‐Level Research Identifiable File ¹¹ (which defines each ACO as a collection of provider taxpayer identification numbers), we then attributed beneficiaries to an ACO on the basis of a plurality of their charges for primary care services in 2017. ⁹ We computed ACO‐level, risk‐standardized, acute, all‐cause, unplanned admission rates (RSAARs) for these beneficiaries in 2018. We then described variation in RSAARs across ACOs and identified ACO characteristics associated with higher or lower RSAAR.

To identify ACO‐level characteristics, we used MSSP ACO public use files and data from the claims files themselves. In addition, based on beneficiary zip code, we defined rurality using Rural–Urban Continuum Codes on the basis of population size, degree of urbanization, and adjacency to a metro area. ¹² Hospital affiliation was determined from ACO websites. We categorized the percentage of primary care physicians and percentage of cardiologist and cardiology subspecialist variables at the 50th percentile and 90th percentile due to their skewed distribution. The Baystate Health Institutional Review Board approved this study. We received a waiver of informed consent.

Cohort Definition

To identify our cohort, we used methods similar to those used by the Centers for Medicare and Medicaid Services for the development and validation of ACO quality metrics. ⁸ We first identified a sample of FFS beneficiaries who (1) were >65 years of age and (2) met the diagnostic criteria for HF with at least 1 hospital claim (Part A) with a principal diagnosis code for HF, or 2 claims (Part A or Part B) with codes for HF in any position in the years 2014 to 2017 (Table S1). Then, for each of these years (2014–2017), we identified a random sample of 999 999 Medicare beneficiaries who met diagnostic criteria for HF. Once a beneficiary was in the cohort, we included that beneficiary in each year thereafter through 2018 or until death or loss to follow‐up.

We additionally limited the population to beneficiaries with 12 months of continuous enrollment in Medicare FFS Parts A and B in 2016 to 2017 and 12 months of continuous enrollment in Part A (or until death) during 2018 (to capture all admissions in 2018). We excluded patients who received hospice care in 2017, and those who had left ventricular assist devices (Z95.811) or heart transplant (Z94.1, Z94.3), as these are high‐risk patients clustered among a few ACOs. We excluded patients who had a zip code that lacked information about rurality, and those who had ≤30 days at risk for admission in 2018 (eg, death occurred on January 15). We also excluded patients cared for by ACOs in US territories and in ACOs that had <25 patients with HF in our sample (because we could not calculate reliable RSAARs for very small populations) (Figure 1).

Figure 1. Cohort selection.

*Eligibility is determined on the basis of continuous enrollment in Part A, Parts A and B. ACO indicates accountable care organization; HF, heart failure; and LVAD, left ventricular assist device.

Outcome Definition

We calculated the number of acute, all‐cause, unplanned hospital admissions per 100 beneficiaries with HF in the year 2018 for each ACO, winsorizing at the 99th percentile to account for extreme outliers. We excluded planned admissions, defined on the basis of a planned admission algorithm adapted for the Centers for Medicare and Medicaid Services' hospital readmission measures. ¹³

Risk Adjustment

From the Master Beneficiary Summary Files for 2017, we risk adjusted for age. We identified comorbidities on the basis of individual Charlson Comorbidity groups and risk adjusted as a set of 0, 1 indicators. ¹⁴ We also adjusted for use of ambulatory inotrope therapy and long‐term oxygen use, which was obtained from durable medical equipment files. ¹⁵

Risk Adjustment and Measurement Calculation

We calculated the RSAAR for each ACO on the basis of published methods by Spatz and colleagues, ⁸ who developed and tested a model of RSAAR for patients with HF in ACOs. We developed a 2‐level, hierarchical, negative binomial model with a log offset for the count outcome to estimate RSAARs that accounts for the clustering of patients within ACOs and accommodates the varying sizes of different ACOs. ⁸ Relevant patient factors with clinical significance were selected a priori for inclusion in the model on the basis of prior literature. The first level of the model adjusted for patient factors (covariates described above). The second level of the model estimated a random intercept that reflects the ACO's contribution to admission risk. The RSAAR for each ACO was calculated as the ratio of the number of predicted to the number of expected admissions, multiplied by the crude rate of admissions per 100 patients with HF. The expected number of admissions was calculated on the basis of the ACO's case mix and national average intercept. The predicted number of admissions was calculated on the basis of the ACO's case mix and the estimated ACO‐specific intercept term.

Bootstrapping to Calculate CIs on Estimates

To estimate 95% CIs on RSAAR for each ACO, we estimated RSAAR in 500 bootstrapped samples. Based on the 95% interval estimates, ACOs were assigned to 1 of 3 performance categories. ACOs were rated “better than average” if the 95% interval estimate was completely below the crude overall mean admission rate and “worse than average” if the 95% interval estimate was above the overall mean admission rate. ACOs with 95% interval estimate overlapping the mean were rated as “average.”

ACO Characteristics and Association With RSAAR

We examined associations between ACO characteristics and the “better than average,” “average,” and “worse than average” RSAAR performance categories. We tested for trend across ordered RSAAR performance categories using the Cochran–Mantel–Haenszel test (CMH). We then used a multiple linear regression model to examine the association of ACO characteristics with RSAAR after adjustment for other factors. We considered ACO characteristics on the basis of literature review: percentage of ACO clinicians who are primary care physicians, percentage of ACO clinicians who are cardiologists or cardiology subspecialists, percentage of HF beneficiaries who are Black, percentage of patients with HF who are dual eligible (Medicare and Medicaid benefits), the number of years in existence, ACO region (Northeast, Midwest, South, West, multiple regions), presence or absence of a hospital affiliation (yes, no, unknown), rurality, and ACO size. The most parsimonious model was selected on the basis of adjusted R‐squared, Akaike information criterion, and Bayesian information criterion.

Results

We identified a total of 1 232 222 eligible Medicare beneficiaries with HF in 2017. Of these, 353 031 (29%) patients were assigned to 472 ACOs, with 292 221 of these beneficiaries surviving 2017 and meeting eligibility in 2018. After the exclusion of patients who received hospice care in 2017 (n=6587) or left ventricular assist device/transplant (n=602) in 2017, those with ≤30 days at risk for admission in 2018 (n=5757), those with missing data on rurality (n=243), and patients from ACOs with ≤25 patients with HF in our sample, 283 795 HF beneficiaries in 467 ACOs comprised the final cohort (Figure 1). The majority of included beneficiaries were women (54%) and White (86%), from urban locations (78%) (Table 1). The average (SD) age was 81 (7.4) years. Seventeen percent of patients required long‐term oxygen therapy in 2017. The average Charlson Comorbidity Index score was 4.2 (SD, 2.9).

Table 1.

Baseline Characteristics of Heart Failure Beneficiaries Attributed to an ACO in 2017

Beneficiary characteristics (N=283 795)	n (%)
Age, y,* mean (SD)	81 (7.4)
65–74	64 459 (23)
75–84	123 385 (43)
≥85	95 951 (34)
Female sex	152 310 (54)
Race or ethnicity
White	243 615 (86)
Black	23 612 (8)
Hispanic	9794 (3)
Other	6774 (2)
Medicare/Medicaid dual eligibility (at least 1 mo)	47 399 (17)
Rurality
Urban	222 656 (78)
Rural adjacent to metro area	39 020 (14)
Rural not adjacent to metro area	22 119 (8)
Any hospital admission during 2017	121 224 (43)
Admissions for principal diagnosis of HF in 2017	28 870 (10)
Any SNF stays in 2017	40 193 (14)
Long‐term O2 use* (from 2017 DME file)	47 977 (17)
Received inotropic therapy* in 2017	1271 (0.45)
Palliative care in 2017	2841 (1)
Charlson Comorbidity Index score, mean (SD)	4.2 (2.9)
Charlson Comorbidity groups
Myocardial infarction*	49 998 (18)
Peripheral vascular disease*	136 699 (48)
Cerebrovascular disease*	76 186 (27)
Dementia*	36 083 (13)
Chronic pulmonary disease*	125 787 (44)
Connective tissue disease–rheumatic disease*	17 512 (6)
Peptic ulcer disease*	7857 (3)
Mild liver disease*	16 726 (6)
Diabetes with or without complications*	136 746 (48)
Paraplegia and hemiplegia*	6284 (2)
Renal disease*	130 020 (46)
Cancer*	48 794 (17)
Moderate or severe liver disease*	2368 (1)
Metastatic carcinoma*	5817 (2)
HIV/AIDS	257 (0.09)

Data are presented as frequency (proportions) unless stated otherwise. ACO indicates accountable care organization; DME, durable medical equipment; HF, heart failure; and SNF, skilled nursing facility.

^*Used for risk adjustment for risk‐standardized acute admission rate. Proportions may not sum to 100% due to rounding.

The characteristics of 467 included ACOs are described in Table 2. The majority (266; 57%) of ACOs were composed of <35% primary care physicians, while 10% of ACOs were composed of >60% primary care physicians. Fifty percent (235) of ACOs were composed of <5% cardiologists and cardiology subspecialists, while 45 (10%) had >10% cardiologists. The majority of ACOs were concentrated in the South (39%), with 21% in the Midwest, 19% in the Northeast, and 11% in the West. Forty‐two percent (n=195) of ACOs were affiliated with a hospital.

Table 2.

Characteristics of ACOs

ACO characteristics (N=467)	Overall, n (%)
ACO characteristics from 2017 PUF
PCP,* %
<35	266 (57)
35–60	152 (33)
>60	49 (10)
Cardiologists and cardiology subspecialists,§ %
<5	235 (50)
5–10	187 (40)
>10	45 (10)
ACO region,* %
Northeast	19
Midwest	21
South	39
West	11
Multiple regions	10
ACO size, (tertile)*
Small, <9170 beneficiaries	155 (33)
Medium, 9170–17 017 beneficiaries	156 (33)
Large, >17 017 beneficiaries	156 (33)
Characteristics from ACO websites †
ACO hospital affiliation†
Yes	188 (40)
No	195 (42)
Unknown	84 (18)
ACO characteristics of included HF beneficiaries ‡
Black,‡ %
<5	218 (47)
5–12	125 (27)
>12	124 (27)
Medicare/Medicaid dual eligibility,‡%
<15	215 (46)
15–20	105 (22)
>20	147 (31)
Rural,|| %
<10	244 (52)
10–30	104 (22)
>30	119 (25)

ACO indicates accountable care organization; HF, heart failure; PCP, primary care physician; and PUF, public use file.

^*Data from ACO PUF; proportions may not sum to 100% due to rounding.

^† Data collected from ACO websites in 2019.

^‡ All data taken from 2017 claims files is taken from the limited cohort of patients with a diagnosis of HF.

^§ From 2017 ACO provider file.

^|| Based on Rural–Urban Continuum Codes code definition http://rtc.ruralinstitute.umt.edu/resources/defining‐rural/.

Admission in 2018 was common: 46% of patients had ≥1 hospital admission in 2018 (24% had 1 admission, 22% had ≥2 admissions), and 12% of patients had an admission with the principal diagnosis of HF in 2018. The top 5 principal diagnoses included sepsis (A41.9), acute kidney failure (N17.9), pneumonia (J18.9), chronic obstructive pulmonary disease with acute exacerbation (J44.1), and urinary tract infection (N39.0). Among ACOs, the mean crude rate of acute, unplanned admission was 87 admissions per 100 people (SD, 13). After risk adjustment, the mean RSAAR was 87 admissions per 100 people (SD, 7.5) (Figure 2). The median RSAAR after risk adjustment was also 87 admissions per 100 people. We observed variation in RSAARs across MSSP ACOs, ranging from 61 (minimum) to 109 (maximum) admissions per 100 beneficiaries, with the 25th and 75th percentile RSAARs at 82 and 92 admissions per 100 beneficiaries.

Figure 2. 2018 ACO‐level observed and risk‐standardized rates of acute, all‐cause unplanned hospital admissions among Medicare beneficiaries with HF assigned to an ACO in 2017.

ACO indicates accountable care organization; and RSAAR, ACO‐level risk‐standardized acute all‐cause admissions per 100 beneficiaries. The measure score represents the predicted acute admission rate divided by the expected acute admission rate. This result is multiplied by an average acute admission rate (across all ACOs).

Using the 3 categories of ACO performance noted above (better than average, average, worse than average), we then examined associations between ACO characteristics and RSAAR (Table 3). We found that ACOs rated as better than average had a smaller proportion of Black beneficiaries with HF (70% of the better‐than‐average group had <5% Black patients, versus 45% of ACOs in the worse‐than‐average group; CMH P<0.001), were less likely to have a hospital affiliation (48% of ACOs in the better‐than‐average category were not affiliated with a hospital versus 22% of ACOs in the worse‐than‐average group; CMH P=0.004) and were more likely to be located in the West region (35% of ACOs in the better‐than‐average category were in the West, versus 3% of ACOs in the worse‐than‐average group; CMH P<0.001). ACOs rated worse than average were more likely to have more Black patients (31% of ACOs in the worse‐than‐average category had >12% Black patients versus 8% of the ACOs in the better‐than‐average group; CMH P<0.001); to have fewer primary care physicians (76% of ACOs in the worse‐than‐average group had <35% primary care physicians versus 44% of ACOs in the better‐than‐average group; CMH P=0.005); to be large (61% of ACOs in the worse‐than‐average group were large versus 37% in the better‐than‐average group); and were more likely to be located in the Northeast (42% of ACOs in the worse‐than‐average group were located in the Northeast versus 14% in the better‐than‐average group; CMH P<0.001). When we examined associations between RSAAR and ACO characteristics using multiple linear regression, our findings were similar (Table S2), with the largest and most significant effect being related to region. The following ACO characteristics were not associated with RSAAR: percentage of Medicare/Medicaid dual‐eligible patients, percentage of rural patients, percentage of cardiologists and cardiology subspecialists in the ACO, or number of years in existence for the ACO.

Table 3.

ACO Characteristics by ACO Performance on Risk Standardized Unplanned Acute Admission Rates

ACO characteristics (N=467)	ACO performance on RSAAR
	Better than ACO average (n=63)	No different from ACO average (n=337)	Worse than ACO average (n=67)
	n (%)	n (%)	n (%)	CMH||
Black,* %				<0.001
<5	44 (70)	144 (43)	30 (45)
5–12	14 (22)	95 (28)	16 (24)
>12	5 (8)	98 (29)	21 (31)
Medicare/Medicaid dual eligibility,* %				0.28
<15	37 (59)	148 (44)	30 (45)
15–20	12 (19)	79 (23)	14 (21)
>20	14 (22)	110 (33)	23 (34)
Rural,† %				0.11
<10	32 (51)	168 (50)	44 (66)
10–30	11 (17)	82 (24)	11 (16)
>30	20 (32)	87 (26)	12 (18)
ACO hospital affiliation‡				0.004
No	30 (48)	143 (42)	15 (22)
Yes	21 (33)	132 (39)	42 (63)
Unknown	12 (19)	62 (18)	10 (15)
PCP, %				0.005
<35	28 (44)	187 (55)	51 (76)
35–60%	28 (44)	112 (33)	12 (18)
>60	7 (11)	38 (11)	4 (6)
Cardiologists and cardiology subspecialists,§ %				0.15
<5	31 (49)	177 (53)	27 (40)
5–10	26 (41)	125 (37)	36 (54)
>10	6 (10)	35 (10)	4 (6)
Number of years ACO in existence				0.22
1	13 (21)	75 (22)	10 (15)
2	17 (27)	67 (20)	10 (15)
3+	33 (52)	195 (58)	47 (70)
ACO region, %				<0.001
Northeast	14	15	42
Midwest	8	22	30
South	30	45	16
West	35	8	3
Multiple regions	13	10	9
ACO size (tertile), %				<0.001
Small (<9170 beneficiaries)	29	37	16
Medium (9170–17 017 beneficiaries)	35	35	22
Large (>17 017 beneficiaries)	37	27	61

Proportions may not sum to 100% due to rounding. ACO indicates accountable care organization; HF, heart failure; PCP, primary care physician; and RSAAR, risk standardized unplanned acute admission rate.

^*From 2017 claims files (HF cohort); all other characteristics from 2017 ACO public use file unless otherwise specified.

^† Based on Rural–Urban Continuum Codes code definition http://rtc.ruralinstitute.umt.edu/resources/defining‐rural/.

^‡ Data collected from ACO websites.

^§ From 2017 ACO provider file.

^|| Cochran–Mantel–Haenszel test for trend.

Discussion

In a large population of FFS Medicare beneficiaries with a diagnosis of HF enrolled in MSSP ACOs, we observed variation in RSAARs across MSSP ACOs, ranging from 61 (minimum) to 109 (maximum) admissions per 100 beneficiaries, with the 25th and 75th percentile RSAARs at 82 and 92 admissions per 100 beneficiaries, respectively. We identified ACO characteristics that were associated with both better (eg, greater proportion of primary care physicians) and worse (eg, a higher percentage of Black patients) performance, although these effects tended to be small. The strongest association was related to region. ACOs in the Northeast had higher RSAARs, and ACOs in the West had lower RSAARs when compared with ACOs in the South. Also of note, we found that admissions were common, with 46% patients admitted at least once during the outcome year. Once patients were admitted, readmission was common (48% of patients with 1 admission experienced readmission). Importantly, our cohort resembles a “real‐world” population of older patients with HF with a high comorbidity burden and the majority of hospitalizations for non‐HF diagnoses. ¹⁶

HF is a complex condition that increases the risk for hospital admission related to both HF and all causes. ¹⁷ , ¹⁸ , ¹⁹ This hospital admission risk is higher among patients with comorbid conditions ²⁰ and who are older, ²¹ with Medicare beneficiaries representing 70% to 80% of all patients hospitalized with HF each year. ²² Compared with a similar study conducted by Spatz and colleagues in 2016, ⁸ we found higher risk‐adjusted all‐cause admission rates among patients with HF in ACOs: with median 81.5 (interquartile range, 73.6–88.8) admissions per 100 person‐years (Spatz et al) versus median 87 (interquartile range, 82–92) admissions per 100 people (current study). These findings may mimic trends that have been observed in readmissions among patients with HF in non‐ACO populations. In a retrospective study of Medicare hospitalizations with a principal or secondary diagnosis of HF between 2008 and 2015, Blecker and colleagues ¹⁶ found that readmissions for patients with a principal diagnosis of HF and patients with a secondary diagnosis of HF significantly decreased after the passage of the Affordable Care Act in 2010, but plateaued following the introduction of the Medicare Hospital Readmissions Reduction Program in 2012. In a more recent analysis of data from the Nationwide Readmission Database, crude rates of overall and unique hospitalizations with a primary diagnosis of HF declined from 2010 to 2014, followed by an increase from 2014 to 2017, with similar trends in 30‐day readmissions after HF index hospitalizations. ³ To our knowledge, trends in acute, all‐cause hospitalizations among patients with HF have not been examined in more contemporary data sets. Further studies are needed to examine trends in HF admissions in recent years and are also needed to examine competing risk between admission and mortality among patients with HF ²³ , ²⁴ to contextualize the high admission rate we found for this patient population (eg, declines in death rates might explain the high admission rate we observed).

Prior studies have also documented variation in outcomes, including RSAARs, across ACOs. ⁸ , ²⁵ Among patients with HF enrolled in ACOs in the first year of the MSSP, Spatz and colleagues ⁸ found similar variation in RSAARs across ACOs, with 53% being “no different” than the national rate, 32% “better than” the national rate, and 14% “worse than” the national rate. Qualitative studies of better‐than‐average ACOs could help to describe the ways that high‐performing ACOs have used strategies such as care coordination to reduce or maintain low admission rates.

Our finding that certain factors (region, percentage of primary care providers, racial composition of patients in the ACO, size of the ACO, and percentage of patients with dual eligibility) may be associated with patterns of health care use in these organizations adds to a growing body of literature on this topic. ACOs differ widely with respect to both physician composition and the distribution of care provided by primary care physicians versus specialist physicians. Prior work has demonstrated that a greater “primary care focus” and cardiologist participation in ACOs may improve outcomes among all patients ²⁶ and patients with general cardiovascular disease, ²⁷ respectively. The association of physician composition with outcomes among patients with HF specifically warrants further study. Finally, our findings demonstrate that ACOs serving populations with higher proportions of Black patients are more likely to have worse performance on RSAARs. Previous data have shown that, while there have been quality gains among patients of racial minority groups who receive care from Medicare ACOs, ACOs that disproportionately care for Black, Hispanic, Native American, and Asian beneficiaries still have worse performance on quality metrics than those that do not. ²⁸ Policymakers must consider how to refine ACO programs and incentive structures to encourage inclusion of racial and ethnic minority patients without compromising quality of care provided.

We found a strong association between geographic region and RSAARs. This result aligns with extensive prior research that shows that geographic variation in health care use, outcomes, and spending appear to be unrelated to patient factors, regional differences in illness, or price differences. ²⁹ , ³⁰ , ³¹ In HF specifically, Northeast regions may have higher risk‐adjusted rates of in‐hospital death and longer length of stay when compared with the Midwest, South, and West, ³² although this has not been extensively studied. Future studies that examine whether this reflects regional variation generally or is a finding specific to ACOs are needed. Additionally, prior studies have observed wide geographic variation in HF deaths, with the highest rates in the South and Midwest. ³³ Studies examining competing risk of death and ACO RSAARs to validate these regional variations in RSAARs are needed.

Our study has some limitations. First, the outcome data from which we calculated RSAARs was limited to 1 year (2018). We also were not able to compare our data to regional rates for patients who are and are not enrolled in ACOs. Future studies should examine trends in RSAARs over time, including potential changes in RSAAR that have occurred in the setting of the COVID‐19 pandemic. Second, we used claims data that do not capture pertinent clinical information (eg, ejection fraction). While the medical therapy of HF does vary somewhat on the basis of phenotype (eg, HF with preserved or reduced ejection fraction), our goal was to evaluate the ability of ACOs to provide coordinated, team‐based care to patients with HF, irrespective of HF type. Future studies may evaluate whether risk‐standardized admission rates among patients in ACOs vary on the basis of the type of HF. Third, while claims data may not fully capture disease severity, previous studies have shown that claims‐derived comorbidity data have good agreement with chart data, ³⁴ , ³⁵ including in HF. ³⁶ Finally, our information regarding ACO‐level characteristics was limited; these data were taken from MSSP ACO public use files, ACO websites, and claims files. Further studies describing the association between RSAARs and other specific ACO characteristics (eg, leadership structure, governance, subspecialist involvement, inclusion of a hospital, care coordination strategies, and other factors) that can be obtained from national ACO surveys or qualitative data could advance understanding of drivers of high and low performance.

In conclusion, there is variation in risk‐standardized rates of acute, unplanned, all‐cause hospital admissions among Medicare beneficiaries with HF across MSSP ACOs. We observed high rates of hospitalization as well as variation in ACO performance, suggesting that there may be patient‐level, ACO‐level, and regional opportunities to improve care. Certain ACO characteristics, namely, region, were associated with RSAARs, warranting further investigation into the drivers of better or worse ACO performance.

Sources of Funding

This study was funded by the National Heart, Lung, and Blood Institute of the National Institutes of Health under Award Number R01 HL139985‐01A1.

Disclosures

None.

Supporting information

Tables S1–S2