Provider Care Segregation and Hospital‐Region Racial Disparities in the United States for Acute Ischemic Stroke and Endovascular Therapy Outcomes

Abstract

Background

Reasons for racial disparities in the use and outcomes of endovascular treatment (ET) are not known. We examined patterns in care segregation for acute ischemic stroke (AIS) in the United States, and outcomes of segregation of care after ET.

Methods and Results

We used deidentified Medicare data sets to identify AIS admissions between January 1, 2016 and December 31, 2019, using validated International Classification of Diseases, Tenth Revision (ICD‐10) codes. For AIS, we calculated (1) the proportion of White patients at the hospital, (2) the proportional difference in the proportion of White patients between hospital patients and the county, and (3) provider care segregation by the dissimilarity index for ET cases. Using unadjusted and adjusted multilevel logistic models, we examined associations between measures of segregation and outcomes of discharge home, inpatient mortality, and 30‐day mortality. The mean proportional difference in the proportion of White patients comparing hospitalized patients with AIS to the county race distribution was 0.015 (SD, 0.219) at the hospital level. For ET, the mean proportional difference in the proportion of White patients comparing patients receiving ET to the county race distribution was much higher, at 0.146 (SD, 0.374). The dissimilarity index for ET providers was high, with a mean of 0.48 (SD, 0.29) across all hospitals. Black patients with AIS (compared with White patients) had reduced odds of discharge home, inpatient mortality, and 30‐day mortality.

Conclusions

In this national study with contemporary data in the endovascular era of AIS treatment, we found substantial evidence for segregation of care in the United States, not for only AIS in general but also especially for ET.

Nonstandard Abbreviations and Acronyms

AIS

acute ischemic stroke

ET

endovascular treatment

HRR

health referral region

NIHSS

National Institutes of Health Stroke Scale

Clinical Perspective.

What Is New?

• Using contemporary data in the endovascular era of acute ischemic stroke treatment, we found substantial evidence for segregation of care in the United States, not only for acute stroke in general but also especially for endovascular thrombectomy.

What Are the Clinical Implications?

• More research is needed to describe the extent and outcomes related to segregation of care in acute stroke.

• Clinical implications may include unequal care and inefficient allocation of resources for acute stroke care.

Racial and ethnic minorities have long faced disparate health outcomes in the United States. ¹ In the case of Black individuals, the legacy of slavery and Jim Crow legislation relegated Black Americans to second‐class status and left devastating consequences for future social mobility, economic advancement, and overall health. Although the Civil Rights Movement helped realize considerable advances in racial justice, there still exist persistent health disparities in a multitude of health domains that continue to have tangible life‐altering effects on the population. ¹ The primary driver of these disparities is structural racism, which does not rely on the consequences of individual's actions but instead represents bias that is embedded into institutions, laws, and policies that propagates itself. ²

Significant racial disparities specifically exist in the areas of cerebrovascular disease. These disparities encompass not only a differential burden of cardiac and stroke risk factors but also differences in stroke morbidity and mortality. Compared with White patients, Black Americans more often have hypertension, diabetes, and left ventricular hypertrophy. Black patients have greater disability after an index stroke than their White counterparts and an age‐specific mortality rate 3 times that of White patients. ³ These disparate outcomes are caused by multiple factors, including individual attitudes, beliefs, and health literacy, disparate access to acute stroke treatment and secondary stroke prevention, and differences in the quality of stroke care, with disparities in thrombolytic use, neurology referral, secondary prevention, and surgical interventions. ³

Segregation of care, at the provider and hospital levels, has not been extensively studied and is 1 possible contributor to health disparities. Hospital‐level segregation can be defined as a difference in racial distribution among patients cared for at a hospital compared with the racial distribution in the surrounding community; a greater difference would signify greater care segregation at the hospital. Provider segregation is defined as the situation where Black and White patients are treated by different care teams, potentially causing separate and unequal care. Such segregation may not necessarily arise from conscious and deliberate choices by providers, but rather from institutionalized practices reflecting the effects of structural racism.

Acute ischemic stroke (AIS) is one of the most common inpatient neurologic disorders. Since 2015, endovascular treatment (ET) has been a widely accepted treatment option for patients with AIS with proximal large‐vessel occlusion, but the provision of ET requires specialized care teams and equipment that are not equally distributed in the United States. ⁴ , ⁵ There are racial disparities in use and outcomes of ET, but little research exists on the impact of hospital or provider segregation stroke care and outcomes. ⁶ We examined patterns in care segregation for AIS in the United States, and the impact of segregation of care on outcomes of patients with AIS who underwent ET.

METHODS

Medicare provides federal health insurance to ≈44 million people in the United States aged ≥65 years. We used complete, deidentified Medicare inpatient data sets from January 1, 2016, to December 31, 2019. These provided data on demographics, date of encounter, and claims information for all Medicare beneficiaries, as well as date of death (if applicable). The requirement for institutional review board approval was waived for this study because it is a retrospective analysis of deidentified data with reporting of aggregate results; this study was exempted from the need for consent by the Mount Sinai Institutional Review Board. The data sets used in this study are publicly available directly from Medicare; the authors are restricted from directly sharing the data with a data use agreement.

Index AIS admissions were identified using the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD‐10‐CM) code of I63.x in the primary diagnosis position, which has been previously validated with a positive predictive value of ≥82%. ⁷ All AIS admissions were included, excluding those with a rehabilitation revenue code (xx3025–xx3099). ET was identified using ICD‐10‐CM procedure codes 03CG3ZZ, 03CH3ZZ, 03CJ3ZZ, 03CK3ZZ, 03CL3ZZ, 03CP3ZZ, and 03CQ3ZZ. In cases of interhospital acute transfers, ET cases were attributed to the hospital where the procedure was performed. Medicare provides the National Provider Identifier number identifying the operating physician for inpatient procedures performed, and we used this unique physician identifier to identify interventionalists performing ET for AIS at each hospital.

Demographic information abstracted from Medicare included age group, sex, and race and ethnicity, which we grouped into White and Black patients (excluding other race and ethnic categories for this analysis). Age group, reported in 5‐year categories, was categorized into ≥75 and <75 years. The Charlson Comorbidity Index, a validated assessment of comorbidity in administrative data analysis, was calculated for each admission. ⁸ Urban versus rural hospital location was assessed and grouped using National Center for Health Statistics urban‐rural location classification definitions into urban (large metropolitan areas with at least 1 million residents) and nonurban (small metropolitan areas with <1 million residents, micropolitan areas, and not metropolitan or micropolitan). Teaching hospital status was determined by receipt of Medicare graduate medical education payment. ⁹ Total 4‐year stroke volume was calculated as the total number of AIS admissions at the treating hospital during the study period. Total 4‐year ET volume was calculated as the total number ET procedures at the treating hospital during the study period. Socioeconomic status of the hospital zip code was determined by the Multidimensional Deprivation Index, which estimates poverty with data from the US Census American Community Survey and accounts for standard of living, education, health, economic security, housing quality, and neighborhood quality. ¹⁰ National Institutes of Health Stroke Scale (NIHSS) score, when recorded, is coded using ICD‐10‐CM codes.

To examine geographic patterns, we grouped hospitals using the Dartmouth health referral regions (HRRs). ¹¹ HRRs are geographic areas defined not by state or city boundaries, but by patient travel patterns for specialty care. HRRs are an established rubric to analyze geographic patterns of health care use that account for referral patterns.

We used county‐level race and ethnic data from the US Census (2010–2019 annual population estimates) to calculate the proportion of White patients living within a county, as follows: [number of White patients]/[number of White and Black patients]. ¹² When we restricted to those aged ≥65 years, the results were overall similar, so we present the data for all age groups in this article. We chose the county level because of the reliability of available census data on racial composition at the county level, the average size of the county when examined nationally, and the ability to validly link the county to the individual hospital without missing data.

We defined 3 primary independent variables at the level of the hospital: (1) the proportion of White patients at the hospital, (2) the proportional difference in the proportion of White patients at the hospital compared with the proportion of White patients in the county, and (3) provider care segregation. For (1), the proportion of White patients with AIS at the hospital was calculated as follows: [number of White patients with AIS treated at the hospital]/[number of White and Black patients with AIS treated at the hospital]. Similarly, the proportion of White patients who underwent ET at the hospital was calculated as follows: [number of White patients who underwent ET treated at the hospital]/[number of White and Black patients who underwent ET treated at the hospital]. This variable summarizes the balance of White and Black patients with AIS or who underwent ET who received care at a hospital. For (2), the proportional difference in the proportion of White patients between hospital patients with AIS and the county was defined as follows: (proportion of White patients with AIS at the hospital)—(proportion of White patients in the county where the hospital is located)/(proportion of White patients in the county). Similarly, the proportional difference in the proportion of White patients between hospital patients who underwent ET and the county was defined as follows: [proportion of White patients who underwent ET at the hospital]—[proportion of White patients in the county where the hospital is located]/[proportion of White patients in the county]. This variable provides a summary of the racial distribution of patients with AIS or patients who underwent ET treated at a hospital, compared with the distribution of race in the surrounding county, compared with the distribution of race in the county.

For (3), provider care segregation was defined by calculating the dissimilarity index for ET cases. ¹³ , ¹⁴ , ¹⁵ , ¹⁶ For each individual provider, we calculated the dissimilarity index as follows. We calculated the relative frequency of Black patients for each ET provider, b _i , as follows: (number of Black patients who a given provider treated with ET)/(number of Black patients who a hospital treated with ET). The relative frequency of White patients, w _i , was calculated similarly. The dissimilarity index for each hospital was then calculated as: $\frac{1}{2}\sum \left|b_i-w_i\right|$. This index has a range of 0 to 1, where 0 means that, for each provider at a hospital, the ratio between the numbers of Black and White patients treated by that provider is the same. A value of 1 signifies that the providers are 100% segregated, and each provider sees only Black or only White patients. Because we were focused on provider team segregation for procedures, we calculated the dissimilarity index only for ET cases, and not AIS. Index values have the following interpretation: low segregation (between 0 and 0.3), moderate segregation (between 0.31 and 0.6), and high segregation (between 0.61 and 1). ¹⁷

We examined the following outcomes: discharge home, inpatient mortality, and 30‐day mortality. Medicare data sets include discharge disposition, and we defined discharge home as home/self‐care discharge, versus all other discharge dispositions. We defined inpatient mortality as death during the index hospitalization, and we defined 30‐day mortality as death within 30 days of index admission.

Statistical Analysis

We summarized the number of AIS admissions and ET procedures performed across the 4‐year period of the study. Among hospitals that treated AIS, we calculated the proportion of hospitals that provided ET at least once during the 4‐year period. If a county included at least 1 hospital that provided at least 1 ET, it was an ET‐providing county. We summarized the distribution (mean and SD) of the proportion of White patients in the county, comparing ET‐providing counties with counties that did not provide ET using the t test.

We summarized the distribution (mean, SD, and interquartile range [IQR]) of the proportional difference in the proportion of White patients between hospital patients and the county across all hospitals for AIS and ET separately. We also summarized this distribution at the HRR level. For ET, we calculated the distribution of provider care segregation across hospitals and HRRs separately.

We calculated the proportion of AIS‐treating hospitals that treated no Black patients with AIS over the study period. Of these, we calculated the mean proportional difference in the proportion of White patients with AIS. We performed a similar analysis at the HRR level, and a similar analysis at the hospital and HRR levels for ET cases.

We then analyzed outcomes using unadjusted and adjusted multilevel logistic models, with clustering of individuals within individual hospitals. Individual‐level adjusting variables were sex, age ≥75 years, Black (versus White) race, and Charlson Comorbidity Index score; hospital‐level variables were proportion of White patients with AIS or White patients who underwent ET at hospital, AIS or ET hospital volumes, Multidimensional Deprivation Index, urban (versus nonurban) hospital location, and teaching status of hospital. There were separate models for AIS and ET. Model 1 included adjustment variables as above and the primary independent variables of Black (versus White) race at the patient level, and the proportion of White patients at the hospital at the hospital level. Model 2 included the adjustment variables listed above and tested the primary independent variables of Black (versus White) race at the patient level, and the proportional difference in the proportion of White patients between hospital patients and county at the hospital level. The reason we included primary independent variables at both the individual and hospital levels was to assess an individual‐level effect of race and a hospital‐level effect related to segregation of care. All models estimated odds ratios (ORs) and 95% CIs. The primary independent variables were centered at the variable mean and estimated for an interval of 0.05. In secondary analysis, we additionally adjusted for NIHSS score, among the minority of cases with recorded NIHSS score. All analyses were performed in SAS, version 9.4.

RESULTS

Over the 4‐year period of the study, there were 937 535 AIS admissions and 34 469 ET procedures performed. Among hospitals that treated AIS, only 26.7% (1186/4445) provided ET at least once. The mean proportion of White patients in counties that included hospitals that provided ET was 0.822 (SD, 0.142), which is lower than the mean in non‐ET counties (0.889; SD, 0.145; P<0.0001). Despite the higher proportion of Black patients in counties with hospitals that provide ET, relatively lower proportions of Black patients received ET. Among AIS cases at the hospital level, the mean proportion of White patients was 0.886 (SD, 0.192), and 0.886 (SD, 0.119) at the HRR level. As summarized in Table 1, the mean proportional difference in the proportion of White patients comparing hospitalized patients with AIS to the county race distribution was 0.015 (SD, 0.219) at the hospital level and 0.020 (SD, 0.07) at the HRR level. The Figure shows the proportional difference in the proportion of White patients by county, showing a clustering of the highest differences in the south and southeast United States. For ET, the mean proportion of White patients was 0.916 (SD, 0.173) at the hospital level and 0.923 (SD, 0.100) at the HRR level. The mean proportional difference in the proportion of White patients comparing patients who underwent ET to the county race distribution was much higher than for patients with AIS overall, at 0.146 (SD, 0.374) at the hospital level and 0.106 (SD, 0.234) at the HRR level. The dissimilarity index for ET providers, signifying provider care segregation, was high, with a mean of 0.48 (SD, 0.29) averaged across all hospitals and 0.52 (SD, 0.23) averaged across HRRs.

Table 1.

Distribution of Variables at Hospital and HRR Levels

Variable	Proportional difference in the proportion of White patients between hospital patients and county			Provider care segregation (dissimilarity index)
	No.	Mean (SD)	IQR	No.	Mean (SD)	IQR
At hospital level
Acute ischemic stroke	4412	0.015 (0.219)	0.0004–0.054	…	…	…
Endovascular treatment	1173	0.146 (0.374)	0.026–0.202	467	0.48 (0.29)	0.27–0.67
At HRR level
Acute ischemic stroke	304	0.020 (0.07)	0.0006–0.0362	…	…	…
Endovascular treatment	289	0.106 (0.234)	0.021–0.123	197	0.52 (0.23)	0.35–0.66

HRR indicates health referral region; and IQR, interquartile range.

Figure . Proportional difference in the proportion of White patients, comparing hospital patients with acute ischemic stroke (AIS) to the county.

Among hospitals treating AIS, 1757 (39.5%) treated no Black patients with AIS. Of these, the mean proportional difference in the proportion of White patients with AIS was 0.053 (SD, 0.153; IQR, 0.008–0.0387). At the HRR level, there were 7 HRRs (2.3%) that treated no Black patients with AIS. Of these, the mean proportional difference in the proportion of White patients with AIS was 0.019 (SD, 0.010; IQR, 0.015–0.022).

Among hospitals providing ET, 694 (58.5%) treated no Black patients undergoing ET. Of these, the mean proportional difference in the proportion of White patients undergoing ET was 0.218 (SD, 0.386; IQR, 0.044–0.229). At the HRR level, there were 116 HRRs (37.9%) that treated no Black patients undergoing ET. Of these, the mean proportional difference in the proportion of White patients undergoing ET was 0.124 (SD, 0.326; IQR, 0.018–0.098).

As summarized in Table 2, Black patients with AIS (compared with White patients) had reduced odds of discharge home, inpatient mortality, and 30‐day mortality, in adjusted multilevel models. The magnitude of reduced odds was 26% for discharge home (OR, 0.74 [95% CI, 0.72–0.75]), 26% for inpatient mortality (OR, 0.74 [95% CI, 0.72–0.77]), and 33% for 30‐day mortality (OR, 0.67 [95% CI, 0.66–0.69]). There was no significant association with the proportion of White patients with AIS at the hospital (model 1) for any outcome, and a significant association for the proportional difference in proportion of White patients between hospital patients and the county (model 2) for inpatient mortality only (OR, 0.985 per 0.05 increase [95% CI, 0.977–0.992]).

Table 2.

Multilevel Models of Outcomes Among Patients With AIS

Outcome	Model 1*		Model 2†
	OR (95% CI)	P value	OR (95% CI)	P value
Discharge home
Black (vs White) patients	0.74 (0.72–0.75)	<0.0001	0.74 (0.72–0.75)	<0.0001
Proportion White patients with AIS at hospital‡	0.998 (0.993–1.003)	0.41	…	…
Proportional difference in proportion of White patients between hospital patients and county§	…	…	1.00 (0.99–1.002)	0.26
Inpatient mortality
Black (vs White) patients	0.74 (0.72–0.77)	<0.0001	0.74 (0.71–0.77)	<0.0001
Proportion White patients with AIS at hospital‡	0.99 (0.98–1.002)	0.11	…	…
Proportional difference in proportion of White patients between hospital patients and county§	…	…	0.985 (0.977–0.992)	<0.0001
30‐d Mortality
Black (vs White) patients	0.67 (0.66–0.69)	<0.0001	0.68 (0.66–0.69)	<0.0001
Proportion White patients with AIS at hospital‡	0.996 (0.992–1.000)	0.058	…	…
Proportional difference in proportion of White patients between hospital patients and county§	…	…	1.00 (1.00–1.003)	0.69

AIS indicates acute ischemic stroke; and OR, odds ratio.

^*Model 1 is adjusted for: sex, age ≥75 years, Black (vs White) race, Charlson Comorbidity Index score, proportion White patients with AIS at hospital, AIS volume at hospital, Multidimensional Deprivation Index, urban (vs nonurban) hospital location, and teaching status of hospital.

^† Model 2 is adjusted for: sex, age ≥75 years, Black (vs White) race, Charlson Comorbidity Index score, proportional difference in percentage White patients between hospital patients and county, AIS volume at hospital, Multidimensional Deprivation Index, urban (vs nonurban) hospital location, and teaching status of hospital.

^‡ OR estimate is centered on a value of 0.886, with an interval of 0.05.

^§ OR estimate is centered on a value of 0.015, with an interval of 0.05.

For ET, provider care segregation was not associated with any outcome in any of the models (results not shown). Similar to AIS, Black patients receiving ET (compared with White patients) had reduced odds of discharge home (24%–25% reduced odds), inpatient mortality (16%–18% reduced odds), and 30‐day mortality (32%–34% reduced odds) (Table 3). In addition, better outcomes were independently associated with a higher proportion of White patients receiving ET at the hospital, including higher odds of discharge home (OR, 1.07 per 0.05 increase [95% CI, 1.04–1.10]), lower odds of inpatient mortality (OR, 0.97 per 0.05 increase [95% CI, 0.94–0.99]), and lower odds of 30‐day mortality (OR, 0.98 per 0.05 increase [95% CI, 0.96–0.99]) (Table 3; model 1). Also, better outcomes were independently associated with a higher proportional difference in the percentage of White patients, comparing ET admissions with the corresponding county, including higher odds of discharge home (OR, 1.05 per 0.05 increase [95% CI, 1.04–1.07]) and lower odds of inpatient mortality (OR, 0.99 per 0.05 increase [95% CI, 0.97–0.996]) (Table 3; model 2). The secondary analysis among cases with recorded NIHSS score showed that, for the AIS analysis (Table S1), NIHSS score was recorded in only 32.4% of AIS admissions. For the ET analysis (Table S2), NIHSS score was recorded in only 51.8% of AIS admissions with ET. Because of concerns about substantial missing data and bias in missingness, no definitive conclusions can be drawn from these supplementary models.

Table 3.

Multilevel Models of Outcomes Among Patients With AIS Receiving ET

Outcome	Model 1*		Model 2†
	OR (95% CI)	P value	OR (95% CI)	P value
Discharge home
Black (vs White) patients	0.76 (0.68–0.85)	<0.0001	0.75 (0.67–0.84)	<0.0001
Proportion White patients receiving ET at hospital‡	1.07 (1.04–1.10)	<0.0001	…	…
Proportional difference in proportion of White patients between hospital patients and county§	…	…	1.05 (1.04–1.07)	<0.0001
Inpatient mortality
Black (vs White) patients	0.82 (0.73–0.92)	0.001	0.84 (0.75–0.95)	0.004
Proportion White patients receiving ET at hospital‡	0.97 (0.94–0.99)	0.002	…	…
Proportional difference in proportion of White patients between hospital patients and county§	…	…	0.99 (0.97–0.996)	0.011
30‐d Mortality
Black (vs White) patients	0.66 (0.60–0.72)	<0.0001	0.68 (0.62–0.75)	<0.0001
Proportion White patients receiving ET at hospital‡	0.98 (0.96–0.99)	0.004	…	…
Proportional difference in proportion of White patients between hospital patients and county§	…	…	0.99 (0.98–1.00)	0.064

AIS indicates acute ischemic stroke; ET, endovascular treatment; and OR, odds ratio.

^*Model 1 is adjusted for: sex, age ≥75 years, Black (vs White) race, Charlson Comorbidity Index score, proportion White patients receiving ET at hospital, ET volume at hospital, Multidimensional Deprivation Index, urban (vs nonurban) hospital location, and teaching status of hospital.

^† Model 2 is adjusted for: sex, age ≥75 years, Black (vs White) race, Charlson Comorbidity Index score, proportional difference in percentage White patients between hospital patients and county, ET volume at hospital, Multidimensional Deprivation Index, urban (vs nonurban) hospital location, and teaching status of hospital.

^‡ OR estimate is centered on a value of 0.916, with an interval of 0.05.

^§ OR estimate is centered on a value of 0.15, with an interval of 0.05.

DISCUSSION

In this national study with contemporary data in the endovascular era of AIS treatment, we found substantial evidence for segregation of care in the United States, for not only AIS in general but also especially for ET. This analysis confirmed that ET is widely unavailable in the United States. Despite ET‐providing counties having a higher proportion of Black patients, relatively lower proportions of Black patients received ET in those areas. A substantially higher proportion of White patients received ET compared with the racial composition of the surrounding area (14.6% higher at the county level and 10.6% higher at the geographically larger HRR level). Also, focusing on ET care provision at the hospital level, there was substantial care segregation, shown by a dissimilarity index of 0.48. Strikingly, more than half of hospitals that provided ET (58.5%) treated no Black patients receiving ET during the 4‐year period of the study. In multilevel models accounting for the clustering of individuals within hospitals, Black patients with AIS and Black patients receiving ET had reduced odds of discharge home, inpatient mortality, and 30‐day mortality. For ET, better outcomes were independently associated with a higher proportion of White patients receiving ET at the hospital, as well as with a higher proportional difference in the percentage of White patients, comparing ET admissions with the corresponding county. These findings demonstrate the segregation of ET in the United States, as well as better ET outcomes in care settings where there is greater segregation of care as measured by the difference in racial distribution at the hospital versus the surrounding area.

This study is the first, to our knowledge, to examine the extent of ET segregation at both the county and provider level as well as the effects of care segregation as a contributor to racial disparities in acute stroke management. Prior research examining inequities in the acute setting have found racially disparate receipt of telestroke, ET, and tPA (tissue‐type plasminogen activator) administration. Several retrospective observational studies have used the Nationwide Inpatient Sample (2001–2008, 2004–2010, 2007–2011, and 2016–2018) to examine tPA administration practices. These studies have found that Black patients and other minorities were consistently less likely to receive thrombolytic treatment than their White counterparts. ¹⁸ , ¹⁹ , ²⁰ , ²¹ Similarly designed studies using a local statewide database found similar results in 2014. However, although there exists a near consensus reflecting a racially disparate application of tPA, several studies found no significant discrepancy between Black and White patients. ²² , ²³ , ²⁴ A national study examined factors associated with stroke certification from 2009 to 2019 and found that patients in Black racially segregated communities have easier geographic access to stroke care but may be faced with resource constraints because stroke centers in such areas serve a much larger stroke population. ²⁵

Studies also show racially disparate use of ET and telestroke. Multiple retrospective observational studies using databases, such as Nationwide Inpatient Sample and Perspective, among others, have shown lower ET rates in Black and Hispanic patients compared with their White counterparts. ²⁶ , ²⁷ , ²⁸ , ²⁹ Only 1 study conflicted with this consensus and found no difference between ET use between White and non‐White patients. ²⁴ Fewer studies have examined potential racial bias in telestroke practices. Of those that are available, a retrospective observational study of the Medical University of South Carolina's telestroke program from 2016 to 2018 found that White patients were more likely to achieve better door‐to‐needle time and had higher odds of receiving tPA. ³⁰ Other studies that examined telestroke use in Texas failed to find any differences in access to care or tPA/ET use between Black and White patients. ³¹ , ³²

Racial disparities in AIS outcomes likely result from the pervasive effects of structural racism, which have caused disparities in disease awareness, control of vascular risk factors, medication prescription and adherence, access to care, and acute and preventative care treatment. This study demonstrates that care segregation is a potential contributing factor to disparate health outcomes in patients with stroke. We found that higher percentages of White patients received ET than Black patients, despite making up less of the racial composition of the surrounding areas. We also found that most hospitals treated no Black patients receiving ET during the period of the study. Although the causes of these phenomena are unclear, they likely represent 1 manifestation of structural racism, taking the form of proximate causes, such as profit maximization at the hospital level, transportation and hospital accessibility in emergencies, health care costs, access to health insurance, or lack of paid sick leave at the patient level, among others.

We also found that a higher proportion of White patients receiving ET at the hospital, and a higher proportional difference in White patients between hospital and corresponding county was associated with higher odds of discharge home and decreased risk of inpatient and 30‐day mortality. This is likely because hospitals with higher White populations are often situated in wealthier neighborhoods and tend be well resourced. Hospitals serving predominantly minority communities are typically situated in low‐socioeconomic neighborhoods and are many times low performing because of lack of adequate funding. This is supported by observational studies that demonstrated that racial disparities could also be driven by the hospitals where minorities seek care. ³³

Last, our study demonstrated that there was significant provider care segregation between Black and White patients. Although there was no significant association with outcomes, it does raise concern about the nature of provider‐patient assignments, which is a process that is largely hidden from significant oversight. External monitoring of such relationships from agencies such as The Joint Commission do not focus on issues of race, leaving such disparities to exist and perpetuate. ¹⁴ The cause of provider segregation is less clear and could be secondary to racial bias, profit‐focused selection of patients, or institutional policies. Additional smaller single‐center studies are necessary to reproduce these findings and further elucidate the related mechanisms.

This study has several strengths. We used a national data set of Medicare data with all beneficiaries, allowing a large sample size and thorough geographic US coverage. Because the study population involved the entire United States, the findings are generalizable to various US geographic settings. This study also had several limitations. We used ICD‐10‐CM codes to define medical conditions and procedures, which do not have perfect validity; however, we relied on validated ICD‐10‐CM codes. Also, because we used a Medicare data set, patients aged >65 years were best represented, and findings are not necessarily generalizable to younger age groups. Also, we could not determine some variables associated with ET use, such as stroke severity at presentation, time to presentation, patient eligibility for mechanical thrombectomy, or the presence of large‐vessel occlusions. We used the racial distribution in the county as the comparison with the hospital distribution, but the county does not always describe the catchment area for hospitals. Also, in the analysis of outcomes, there is the possibility that we were not able to adjust for all relevant variables.

This study highlights directions for future study. Further studies are necessary to clarify the causes of county‐level racial disparities in ET use and determine whether these findings are secondary to racially discriminatory practices, socioeconomic bias, legitimate medical contraindication, or structural barriers. Additional single‐center and multicenter studies are necessary to explore causes and consequences of provider care segregation. Such data would inform best practices to reduce these disparities and improve access to acute stroke treatments. Such interventions may include targeted development of stroke center capability in vulnerable geographic areas, improved prehospital transport triage, directed interhospital transfer protocols, or community education about acute stroke symptoms and the importance of prompt activation of emergency medical services.

Sources of Funding

None.

Disclosures

None.

Supporting information

Tables S1–S2