Racial disparities in avoidable hospitalizations in traditional Medicare and Medicare Advantage

Abstract

Importance:

Compared to traditional Medicare (TM), Medicare Advantage (MA) has the potential to reduce racial disparities in hospitalizations for ambulatory care sensitive conditions (ACSC). As racial disparities may be partly attributable to unequal treatment based on where people live, this suggests the need of examining geographic variations in racial disparities.

Objectives:

To examine differences in ACSC hospitalizations between White and Black beneficiaries in TM and MA and examine geographic variations in racial differences in ACSC hospitalizations in TM and MA.

Methods:

We analyzed the 2015–2016 Medicare Provider Analysis and Review files. We used propensity score matching to account for differences in characteristics between TM and MA beneficiaries. Then, we conducted linear regression and estimated adjusted outcomes for TM and MA beneficiaries by race. Also, we estimated racial differences in adjusted outcomes by insurance and hospital referral region (HRR).

Results:

While White beneficiaries in TM and MA had similar rates of ACSC hospitalizations (163.7 versus 162.2 per 10,000 beneficiaries), Black beneficiaries in MA had higher rates of ACSC hospitalizations than Black beneficiaries in TM (221.2 versus 209.3 per 10,000 beneficiaries). However, the racial differences were greater in MA than TM (59.0 versus 45.6 per 10,000 beneficiaries). Racial differences in ACSC hospitalizations in MA were prevalent across almost all HRRs. 95.5% of HRRs had higher rates of ACSC hospitalizations among Black beneficiaries than White beneficiaries in MA relative to just 54.2% of HRRs in TM.

Conclusions:

Our findings provide evidence of racial disparities in access to high-quality primary care, especially in MA.

Ensuring that all Americans receive high-quality health care is a key policy priority. Hospitalizations for ambulatory care sensitive conditions (ACSC) are of particular interest. ACSC hospitalizations could be potentially avoidable with timely access to appropriate primary care¹ and thus are widely recognized as a key measure of access to high-quality primary care. Several programs (e.g., Medicare Accountable Care Organization and the Physician Value-Based Payment Modifier) have included ACSC hospitalizations as a performance metric^2,3 and other programs (e.g., Medicare Advantage [MA] and the Merit-based Incentive Payment) are under consideration for use. ACSC hospitalizations are shown to be associated with the availability of primary care resources^4,5 and factors related to limited access to care such as race/ethnicity, income, and neighborhood conditions.^6–8

Prior studies have documented substantial differences in ACSC hospitalizations between White and Black beneficiaries in traditional fee-for-service Medicare (TM).^9–11 One study found that while ACSC hospitalizations for White beneficiaries declined significantly from 2003 to 2009, there were no changes in ACSC hospitalizations for Black beneficiaries.¹⁰ Recent research showed that ACSC hospitalizations decreased from 2011 to 2015 in both White and Black beneficiaries, but racial differences in ACSC hospitalizations still persist.⁹ Findings from these prior studies suggest that there may exist racial disparities in the accessibility of high-quality primary care. There is evidence showing that Black beneficiaries were less likely to establish primary care as their usual source of care and instead were more likely to use emergency department than White beneficiaries.¹²

Although the majority of Medicare beneficiaries receive care through TM, 36% of Medicare beneficiaries—equivalent to 24.1 million beneficiaries—enrolled in MA plans in 2020.¹³ MA plans are paid on a capitated basis and thus are incentivized to provide services in a way that maintains beneficiary health and reduces the cost of care. Among the strategies MA plans use is increased access to high-quality primary care in order to avoid expensive ACSC hospitalizations. Particularly, MA plans have the potential to decrease racial disparities in ACSC hospitalizations because improved access to primary care in MA plans may disproportionately benefit racial minority beneficiaries who were more likely to have trouble getting needed care.^6–8 However, there is limited evidence. First, several studies found that MA plans have encouraged use of primary care by expanding networks of primary care providers¹⁴ and providing relatively low cost-sharing for primary care¹⁵, but others found that MA plans have provided narrow networks,¹⁶ particularly for primary care services.¹⁷ Furthermore, little is known about racial disparities in ACSC hospitalizations in MA. Evidence suggests that racial disparities in health care use and quality of care were prevalent in both TM and MA.^18–20 Also, there is mixed evidence that MA plans were better at reducing racial differences than TM.^19,21

An understudied but potentially insightful question is whether racial differences in ACSC hospitalizations are prevalent across geography in TM and MA. While racial disparity is driven by unequal treatment within a hospital or by a provider, it may be also attributable to unequal treatment based on where people live.²² Prior research has found that heterogeneity exists in the extent of racial disparities across regions.²³ In some areas, no or negligible racial disparities were observed.²² These findings indicate that overall national racial differences mask sizable variations across regions. Ignoring geography may yield misleading policy prescriptions because the policy prescriptions may differ depending on whether the racial disparities are attributable to geographic variations instead of differences in treatment within hospitals or providers. If racial minority beneficiaries tend to live in areas where quality of care is lower for all beneficiaries, then reducing geographic disparities in quality of care could significantly reduce racial disparities in treatment and health outcomes.

To address the gap in the literature on racial disparities in ACSC hospitalizations in TM and MA, we conducted two main analyses. First, we examined differences in ACSC hospitalizations between White and Black beneficiaries in TM and MA. Second, we examined geographic variations in differences in ACSC hospitalizations between White and Black beneficiaries in TM and MA.

METHODS

Data and Sample

Our primary data was the 100% Medicare Provider Analysis and Review (MedPAR) files between 2015 and 2016, which include hospitalization-level data on all TM and MA beneficiaries who were admitted to a MedPAR reporting hospital. Since 2008, the Centers for Medicare and Medicaid Services tied a hospital’s Medicaid Disproportionate Share Hospital payments to submission of shadow bills for MA beneficiaries staying at the hospital. Consequently, the completeness of MedPAR hospital records for MA beneficiaries has increased and thus MedPAR currently accounts for about 90% of all MA hospitalizations.²⁴ The data has been frequently used to measure hospitalization-related outcomes for MA beneficiaries.^24–27 Patient characteristics were derived from the Master Beneficiary Summary File.

We identified White and Black Medicare beneficiaries 65 years or older with 12-month continuous enrollment in MA or TM (both Parts A and B benefits). We excluded those whose original Medicare eligibility was attributable to end-stage renal disease which is carved out of the MA program. We also excluded those who switched between MA and TM at any time as these beneficiaries may result in carry-over bias.

Measures

Our primary outcome was a binary indicator of any ACSC hospitalization in a given year. We also included two secondary outcomes: number of ACSC hospitalizations and length of stay (LOS) for ACSC hospitalizations among those with any ACSC hospitalization. We identified ACSC hospitalizations using the Prevention Quality Indicators (PQI) developed by the Agency for Healthcare Research and Quality.²⁸ Specifically, we identified inpatient visits associated with ACSCs by applying algorithms of PQIs onto International Classification of Diseases (ICD) diagnosis and procedure codes in the MedPAR data. Algorithms of PQIs are available for the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) as well as the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM). Since the transition from ICD-9-CM codes to ICD-10-CM codes occurred on October 2015, we used both ICD-9-CM codes and ICD-10-CM codes for the 2015 data. For the 2016 data, we used ICD-10-CM codes.

Our primary independent variables were race, insurance, and geographic area. We used the Research Triangle Institute (RTI) Race Code to more accurately identify non-Hispanic White and non-Hispanic Black. This variable was developed to improve classification of Hispanics and Asians/Pacific Islanders by utilizing lists of Hispanic and Asian/Pacific Islander names from the US Census and simple geography. Although Hispanic were not included in our study, distinguishing non-Hispanic White from Hispanic White allows us to more appropriately estimate racial disparities between non-Hispanic White and Black enrollees.²⁹ Prior research assessed the sensitivity and specificity of the RTI race variable compared with self-reported race/ethnicity data and found that the variable has high sensitivity and specificity for non-Hispanic White and Black.³⁰ Particularly, the RTI race variable is less likely to misclassify Hispanic as non-Hispanic White. We assigned Medicare beneficiaries to either TM or MA based on a full 12-month enrollment for each calendar year. We used the hospital referral region (HRR) as the geographic unit of analysis. HRRs have been defined by the Dartmouth Atlas project based on referral patterns for tertiary care and designed to approximate health care markets, in which practice patterns may be similar. We assigned beneficiaries to HRRs by their zip code of residence. To control for differences in characteristics between TM and MA beneficiaries, we included age, sex, Medicare-Medicaid dual eligibility, and (unstandardized) Hierarchical Condition Category (HCC) risk scores.

Statistical analyses

Evidence suggests that healthy beneficiaries are more likely to enroll in MA than TM, making a direct comparison between TM and MA beneficiaries potentially biased.³¹ Several studies have used propensity score matching to account for differences in characteristics between TM and MA beneficiaries.^32,33 However, there are some concerns that MA enrollment may be related to health status not captured by commonly used variables to measure health status (e.g., HCC risk scores based on diagnosis codes).³⁴ Thus, we followed prior research and estimated selection on health status not captured by HCC risk scores using observed mortality differences between TM and MA beneficiaries.³⁵ The idea behind this approach relies on the fact that mortality is lower for MA beneficiaries than TM beneficiaries assuming that mortality is not affected by enrollment in MA. Indeed, analysis without risk adjustment showed that MA spending per enroll-month was 30% lower than TM spending per enroll-month. The spending difference between TM and MA beneficiaries decreased by 25% when county fixed effects and risk scores were accounted for in the analysis. The magnitude of the spending difference was just 9% when mortality differences between TM and MA beneficiaries were additionally accounted for.³⁵ We adapted this approach and conducted a linear probability model on mortality (death within the year) after controlling for age, sex, Medicare-Medicaid dual eligibility, HCC risk scores, and year and county fixed effects. Then, we estimated predicted mortality rates for each beneficiary. Next, we computed the inverse probability of treatment weighting (IPTW) as a propensity for enrolling in MA based on all variables described above as well as the predicted mortality rates. After estimating the predicted mortality rates, we excluded those who died within each year from the analysis because end-of-life health care use may result in bias, and then limited to those with 12-month continuous enrollment in TM or MA. We checked whether the IPTW-weighted samples were balanced on individual-level demographic, socioeconomic, and health status variables between TM and MA beneficiaries. We also checked whether the propensity score has an adequate overlap between TM and MA beneficiaries.

We then estimated IPTW-adjusted outcomes between TM and MA beneficiaries by race. Specifically, we estimated adjusted outcomes using linear regression models while controlling for variables described above as well as race, MA enrollment, and the interaction term between race and MA enrollment and applied the IPTW. We conducted linear probability models for binary outcomes and linear regression models for continuous outcomes. To make findings more interpretable, we then estimated the adjusted mean values of the outcome for TM and MA beneficiaries by race. Next, we estimated the differences in the adjusted outcome among Black beneficiaries relative to White beneficiaries by insurance coverage. We conducted several sensitivity analyses to examine robustness of our findings. First, the MedPAR files do not provide claims data for all MA beneficiaries. Thus, we conducted the analysis described above by limiting to hospitals with 100% reporting (i.e., those receiving Medicaid Disproportionate Share Hospital payments or medical education credits). Second, our estimates may be biased due to model misspecification. Thus, we conducted sensitivity analysis by using logistic regression. Finally, accounting for differences in the predicted mortality rates between TM and MA beneficiaries in the IPTW estimation allows us to control for differences in health status not captured by HCC risk scores. However, this may have the potential to increase bias because mortality may be also related to MA enrollment. Thus, we conducted the analysis described above without including the predicted mortality rates in the IPTW estimation.

We conducted a similar analysis as described above and estimated the adjusted mean values of the outcome for TM and MA beneficiaries by race and HRR. Specifically, we estimated IPTW-adjusted outcomes using a linear probability model while controlling for variables described above as well as race, MA enrollment, HRR, and the interaction terms among race, MA enrollment, and HRR. We estimated the adjusted mean values of the outcome for each group by race, insurance coverage, and HRR. Then, we estimated the differences in the adjusted outcome among Black beneficiaries relative to White beneficiaries by insurance coverage and HRR. Due to concerns about the stability of our estimates, we did not report values for HRRs with less than 3,000 white or Black beneficiaries and HRRs with less than 300 Black beneficiaries. As secondary analysis, we also estimated the HRR-level correlations of ACSC hospitalizations among White and Black beneficiaries in TM and MA to examine racial patterns of HRR-level variations in ACSC hospitalizations in TM and MA.

For all analyses, we clustered the standard errors within individuals. We also included year and county fixed effects.

RESULTS

Our sample included 51,740,797 TM beneficiaries and 24,778,503 MA beneficiaries (Table 1). There were differences in sample characteristics between TM and MA beneficiaries. Compared to TM beneficiaries, MA beneficiaries were more likely to be younger (72.9% vs. 74.0% for those aged between 65 and 79 years), to be male (44.7% vs. 42.1%), to be dual-eligible for Medicare and Medicaid (9.6% vs. 12.2%), and to be healthy (12.3% vs. 9.5 for those in the lowest HCC risk score group). However, these differences decreased after applying the IPTW (Table 2). Also, we found substantial overlap in propensity scores between TM and MA beneficiaries (Appendix Figure).

Table 1.

Descriptive statistics for outcomes and sample characteristics between traditional Medicare (TM) and Medicare Advantage (MA) beneficiaries by race.

	N (%) or mean (SD)
	TM		MA
Variables	White (N=47303628)	Black (N=4437169)	White (N=21632271)	Black (N=3146232)
Outcomes
ACSC hospitalization, N (%)	783199 (1.7)	108655 (2.4)	317655 (1.5)	77650 (2.5)
Number of ACSC hospitalization (if any), mean (SD)	1.3 (0.7)	1.4 (0.9)	1.2 (0.6)	1.4 (0.9)
Length of stay for ACSC hospitalization (if any), mean (SD)	5.6 (5.9)	6.7 (7.6)	5.5 (5.6)	6.5 (7.4)
Covariates
Age, N (%)
65–69	13863685 (29.3)	1565419 (35.3)	5687181 (26.3)	981687 (31.2)
70–74	12011032 (25.4)	1113230 (25.1)	6002816 (27.7)	859213 (27.3)
75–79	8440849 (17.8)	734970 (16.6)	4199981 (19.4)	602233 (19.1)
80–84	6006303 (12.7)	494519 (11.1)	2900117 (13.4)	382536 (12.2)
>=85	6981759 (14.8)	529031 (11.9)	2842176 (13.1)	320563 (10.2)
Female, N (%)	26047400 (55.1)	2546522 (57.4)	12355189 (57.1)	1979553 (62.9)
Medicare-Medicaid dual eligibility, N (%)	3863327 (8.2)	1091640 (24.6)	2033901 (9.4)	985154 (31.3)
HCC risk score†, N (%)
0–0.299	5833049 (12.3)	552453 (12.5)	2103537 (9.7)	243660 (7.7)
0.3–0.499	24104881 (51.0)	2136541 (48.2)	11694950 (54.1)	1558843 (49.5)
0.5–0.999	12760475 (27.0)	1275317 (28.7)	5963781 (27.6)	1023222 (32.5)
1–1.999	2625125 (5.5)	241819 (5.4)	1087872 (5.0)	172111 (5.5)
>=2	1980098 (4.2)	231039 (5.2)	782131 (3.6)	148396 (4.7)
Year, N (%)
2015	23439089 (49.6)	2188452 (49.3)	10616793 (49.1)	1517594 (48.2)
2016	23864539 (50.4)	2248717 (50.7)	11015478 (50.9)	1628638 (51.8)

ACSC indicates ambulatory care sensitive conditions; TM, traditional Medicare; MA, Medicare Advantage; HCC, Hierarchical Condition Category.

^†We estimated (unstandardized) HCC risk scores using the 2015 Centers for Medicare and Medicaid Services (CMS)-HCC model.

Table 2.

Sample characteristics between traditional Medicare (TM) and Medicare Advantage (MA) before and after applying the inverse probability of treatment weighting (IPTW).

	%
	Before applying the IPTW†		After applying the IPTW†
Variables	TM	MA	TM	MA
Age
65–69	29.8	26.9	28.9	29.4
70–74	25.4	27.7	26.1	25.9
75–79	17.7	19.4	18.3	18.1
80–84	12.6	13.2	12.8	12.7
>=85	14.5	12.8	14.0	14.0
Female	55.3	57.9	56.1	55.6
Race
White	91.4	87.3	90.1	90.2
Black	8.6	12.7	9.9	9.8
Medicare-Medicaid dual eligibility	9.6	12.2	10.4	10.4
HCC risk score
0–0.299	12.3	9.5	11.4	12.0
0.3–0.499	50.7	53.5	51.6	51.2
0.5–0.999	27.1	28.2	27.5	27.3
1–1.999	5.5	5.1	5.4	5.4
>=2	4.3	3.8	4.1	4.2
Year
2015	49.5	49.0	49.4	49.4
2016	50.5	51.0	50.6	50.6

IPTW indicates inverse probability of treatment weighting; TM, traditional Medicare; MA, Medicare Advantage; HCC, Hierarchical Condition Category.

^†To account for selection on health status not captured by HCC risk scores, we used mortality differences between TM and MA beneficiaries. Specifically, we conducted a linear probability model on mortality (death within the year) after controlling for age, sex, Medicare-Medicaid dual eligibility, and HCC risk scores. Then, we estimated predicted mortality rates for each enrollee. Next, we computed the IPTW as a propensity for enrolling in MA based on all variables described above as well as the predicted mortality rates.

Our analysis showed that adjusted rates of ACSC hospitalizations were significantly higher among Black beneficiaries than White beneficiaries in both TM and MA, but there were greater racial differences in the adjusted rates in MA than TM (59.0 [95% CI, 57.2 to 60.8] vs. 45.6 [95% CI, 44.3 to 46.9] per 10,000 beneficiaries) (Table 3). Specifically, while White beneficiaries in TM and MA had similar adjusted rates of ACSC hospitalizations (163.7 [95% CI, 163.7 to 164.1] per 10,000 beneficiaries for TM beneficiaries vs.162.2 [95% CI, 161.1 to 163.2] per 10,000 beneficiaries for MA beneficiaries), Black beneficiaries in MA had higher adjusted rates of ACSC hospitalizations than Black beneficiaries in TM (221.2 [95% CI, 219.2 to 223.1] per 10,000 beneficiaries for TM beneficiaries vs. 209.3 [95% CI, 207.9 to 210.7] per 10,000 beneficiaries for MA beneficiaries).

Table 3.

Adjusted racial differences in hospitalizations for ambulatory care sensitive conditions between TM and MA beneficiaries.

	Adjusted mean, 95% (CI)†
	TM			MA
Outcomes	White	Black	Racial difference	White	Black	Racial difference
ACSC hospitalization, % per 10,000 enrollees	163.7 (163.4 to 164.1)	209.3 (207.9 to 210.7)	45.6 (44.3–46.9)	162.2 (161.1 to 163.2)	221.2 (219.2 to 223.1)	59.0 (57.2–60.8)
Number of ACSC hospitalization (if any), N	1.2 (1.2 to 1.2)	1.3 (1.3 to 1.3)	0.1 (0.1 to 0.1)	1.2 (1.2 to 1.2)	1.3 (1.3 to 1.3)	0.1 (0.1 to 0.1)
Length of stay for ACSC hospitalization (if any), day	5.7 (5.7 to 5.8)	6.3 (6.3 to 6.4)	0.8 (0.7 to 0.8)	5.8 (5.8 to 5.8)	6.4 (6.3 to 6.4)	0.8 (0.8 to 0.9)

ACSC indicates ambulatory care sensitive conditions; TM, traditional Medicare; MA, Medicare Advantage.

^†We estimated selection on health status not captured by HCC risk scores using observed mortality differences between TM and MA enrollees. Specifically, we conducted a linear probability model on mortality (death within the year) after controlling for age, sex, Medicare-Medicaid dual eligibility, HCC risk scores, and year and county fixed effects. Then, we estimated predicted mortality rates for each beneficiary. Next, we computed the inverse probability of treatment weighting (IPTW) as a propensity for enrolling in MA based on all variables described above as well as the predicted mortality rates. We then estimated IPIW-adjusted outcomes using linear regression models while controlling for variables described above as well as race, MA enrollment, and the interaction term between race and MA enrollment and applied the IPTW. We conducted a linear probability model for a binary outcome and linear regression models for continuous outcomes. Then, we estimated the adjusted mean values of the outcome for TM and MA enrollees by race. Next, we estimated the differences in the adjusted outcome among Black beneficiaries relative to White beneficiaries by insurance coverage.

However, there were no or small racial differences in the number of ACSC hospitalizations and LOS for ACSC hospitalizations (Table 3). Specifically, the adjusted number of ACSC hospitalizations were similar across all groups (1.2 [95% CI, 1.2 to 1.2] for White beneficiaries in TM, 1.3 [95% CI, 1.3 to 1.3] for Black beneficiaries in TM, 1.2 [95% CI, 1.2 to 1.2] for White beneficiaries in MA, and 1.3 [95% CI, 1.3 to 1.3] for Black beneficiaries in MA). There were marginal racial differences in the adjusted number of ACSC hospitalizations (0.1 percentage points [95% CI, 0.1 to 0.1] and 0.1 percentage points [95% CI, 0.1 to 0.1]). On the other hand, Black beneficiaries had slightly longer adjusted LOS for ACSC hospitalizations than White beneficiaries in both TM and MA (6.3 days [95% CI, 6.3 to 6.4] for Black beneficiaries in TM, 6.4 days [95% CI, 6.3 to 6.4] for Black beneficiaries in MA, 5.7 days [95% CI, 5.7 to 5.8] for White beneficiaries in TM, and 5.8 days [95% CI, 5.8 to 5.8] for White beneficiaries in MA). However, there were no racial differences in the adjusted LOS for ACSC hospitalizations (0.8 days [95% CI, 0.7 to 0.8] for TM beneficiaries and 0.8 days [95% CI, 0.8 to 0.9] for MA beneficiaries). Regression results for the IPTW are presented in Appendix Tables A and B. Regression results for the IPTW-adjusted analysis are presented in Appendix Table C. We found that our findings were robust to several sensitivity analyses.

While there were HRR-level variations in racial differences in adjusted rates of ACSC hospitalizations in TM, the racial differences in MA were prevalent across almost all HRRs (Figure). For TM, in 148 out of 273 HRRs (54.2%), Black beneficiaries had higher adjusted rates of ACSC hospitalizations than White beneficiaries. For MA, however, in 234 out of 245 HRRs (95.5%), Black beneficiaries had higher adjusted rates of ACSC hospitalizations than White beneficiaries.

Figure.

Variations in racial differences in hospitalizations for ambulatory care sensitive conditions (ACSC) (per 10,000 enrollees) among Black beneficiaries relative to White beneficiaries in traditional Medicare (TM) and Medicare Advantage (MA) by hospital referral region (HRR).

Due to concerns about the stability of our estimates, we did not report values for HRRs with less than 3,000 white or Black beneficiaries in TM and HRRs with less than 300 Black beneficiaries in TM.

We found suggestive evidence showing geographic concentration in racial disparities in ACSC hospitalizations. The adjusted rates of ACSC hospitalizations are strongly correlated between White beneficiaries in TM and MA at the HRR level (0.758; P<0.001), but this was not observed in other groups (Table 4). The correlation between Black beneficiaries in TM and MA was relatively weak (0.449; P<0.001). Furthermore, there was a more modest correlation between White and Black beneficiaries in MA than between White and Black beneficiaries in TM (0.501; P<0.001 versus 0.687; P<0.001).

Table 4.

Area-level correlations of adjusted rates of hospitalizations for ambulatory care sensitive conditions (ACSC) (per 10,000 enrollees) among white and Black beneficiaries in traditional Medicare (TM) and Medicare Advantage (MA).

	Pearson’s correlation of HRR-level ACSC hospitalizations†	P value
White enrollees in TM versus white enrollees in MA (N=245)	0.758	<0.001
Black enrollees in TM versus Black enrollees in MA (N=245)	0.449	<0.001
White enrollees in TM versus Black enrollees in TM (N=273)	0.687	<0.001
White enrollees in MA versus Black enrollees in MA (N=245)	0.501	<0.001

HRR indicates hospital referral region; ACSC, ambulatory care sensitive conditions TM, traditional Medicare; MA, Medicare Advantage.

^†Using the adjusted mean values of the outcome for each group by race, insurance type, and HRR, we used a Spearman rank-order correlation coefficient to assess the correlations between the HRR rate for white and Black beneficiaries in TM and MA. Due to concerns about the stability of our estimates, we excluded HRRs with less than 3,000 white or Black beneficiaries and HRRs with less than 300 Black beneficiaries.

DISCUSSION

Using inpatient hospital national data for Medicare beneficiaries, we found substantial racial differences in adjusted rates of ACSC hospitalizations in both TM and MA. However, the magnitude of the racial differences was larger in MA than TM. Our finding aligns with finding from prior research that racial differences in readmission rates were greater in MA than TM, suggesting that compared to White beneficiaries, Black beneficiaries may receive lower-quality care in the hospital or receive care in hospitals with lower quality of care.¹⁹

Our finding of geographic variations in racial differences confirmed that racial differences in adjusted rates of ACSC hospitalizations persist after accounting for geography.

The most notable was that whereas half of HRRs demonstrated higher adjusted rates of ACSC hospitalizations among Black beneficiaries relative to White beneficiaries in TM, this was observed across almost all HRRs in MA. The HRR-level rates of ACSC hospitalizations were modestly correlated between Black beneficiaries in TM and MA, suggesting differential impacts of MA enrollment among Black beneficiaries. Also, the HRR-level rates were relatively weakly correlated between White and Black beneficiaries in MA than between White and Black beneficiaries in TM, suggesting larger racial disparities in ACSC hospitalization in MA than TM.

High adjusted rates of ACSC hospitalizations among Black beneficiaries in both TM and MA may be attributable to limited access to high-quality primary care through a myriad of channels. Evidence suggests that Black beneficiaries have limited access to primary care as their usual source of care than White beneficiaries.¹² Furthermore, Black beneficiaries are more likely to enroll in lower-quality MA plans.³⁶ This could be due to which plans that are offered, financial constraints, confusion or lack of knowledge about insurance coverage. Future work should disentangle the underlying mechanisms.

Our findings have important policy implications. There is a need to improve access to high-quality primary care for Black beneficiaries. This is particularly important with significant consequences for Black beneficiaries’ health and financial implications. The costs associated with the disparities in ACSC hospitalizations are substantial, ranging from $38,000 (urinary tract infection) to $8 million (heart failure).¹¹ Excess costs from the disparities are still substantial even after adjusting for demographic and socioeconomic factors. Also, MA enrollment has nearly doubled over the last decade,³⁷ but evidence suggests that MA enrollment was more prevalent among racial/ethnic minority beneficiaries.³⁸ These suggest the need of paying more attention to how to reduce racial disparities in MA. Policymakers need to closely monitor racial disparities in MA and consider how to narrow the gap in ACSC hospitalizations between White and Black beneficiaries.

Our study had several limitations. First, we measured race from administrative data sources. Although prior research found that non-Hispanic White and Black beneficiaries were relatively accurately measured,³⁰ there may be still concerns about misclassification errors. Second, we accounted for differences in sample characteristics between TM and MA beneficiaries, but there is still the potential for residual confounding due to unobservable factors.³⁵ Specifically, our measure of health status relied on inpatient data only, raising some concerns about imperfect case-mix adjustment. Also, we could not include detailed information on socioeconomic status. Third, we used mortality differences between TM and MA beneficiaries to address selection on health status not captured by HCC risk scores, but it is unlikely to completely correct for selective enrollment. This approach may help to decrease bias attributable to selection into MA to some extent, but less is known about how much the bias is still present. Furthermore, this could vary for different types of health care use outcomes. The approach relies on the assumption that mortality is not affected by enrollment in MA,³⁵ but this assumption may be too strong to be met. Fourth, MA plans code health diagnoses more aggressively than TM and thus there may be some over-reporting of co-morbidities in MA.³⁹ Thus, our findings may be subject to differences in coding patterns between providers serving TM and MA beneficiaries. Fifth, the MedPAR files do not provide claims data for all MA beneficiaries, and thus there may be some concerns about validity of our findings. Due to limited data availability, we examined whether our findings were consistent with findings from prior research that used data for all Medicare beneficiaries. We found that our estimates in unadjusted rates of ACSC hospitalizations among TM and MA beneficiaries were close to those from state inpatient database in 13 states (17.2 and 16.0 ACSC hospitalizations per 1,000 beneficiaries for TM and MA beneficiaries from our analysis vs. 17.5 and 16.9 ACSC hospitalizations per 1,000 beneficiaries for TM and MA beneficiaries from prior research using state inpatient database).⁴⁰ Finally, we could not identify the causal mechanisms driving the observed differences in hospitalizations for ACSC between MA and TM beneficiaries by race. Thus, our findings do not necessarily have a causal interpretation.

Conclusions

Adjusted rates of ACSC hospitalizations were significantly higher among Black beneficiaries than White beneficiaries in both TM and MA. However, the racial differences in the adjusted rates were greater in MA than TM. Notably, racial differences in ACSC hospitalizations in MA were prevalent across almost all HRRs. Our findings provide evidence of racial disparities in access to high-quality primary care, especially in MA. Narrowing the racial gap in primary care and ACSC hospitalizations should be a policy priority.