Racial Disparity in Mechanical Thrombectomy Utilization: Multicenter Registry Results From 2016 to 2020

Abstract

Background

Previous studies on racial disparity in mechanical thrombectomy (MT) treatment of acute large vessel occlusion stroke lack individual patient data that influence treatment decision‐making. We assessed patient‐level data in a large US health care system from 2016 to 2020 for racial disparities in MT utilization and eligibility.

Methods and Results

A retrospective study was performed of 34 596 patients admitted to 43 hospitals from January 2016 to September 2020. Data included patient age, sex, race, residential zip code median income and population density, presenting hospital stroke certification, baseline ambulation, and National Institutes of Health stroke scale. The cohort included 26 640 White, non‐Hispanic (77.0%), and 7956 African American/Black (23.0%) patients. In multivariable logistic regression, Black patients were less likely to undergo MT (adjusted odds ratio [OR], 0.65; 95% CI, 0.54–0.76), arrive within 5 hours of “last known well” (adjusted OR, 0.73; 95% CI, 0.69–0.78), and have documented anterior circulation large vessel occlusion (adjusted OR, 0.78; 95% CI, 0.64–0.96). Race was not associated with MT rate among patients arriving within 5 hours of last known well with documented acute large vessel occlusion.

Conclusions

Black patients with stroke underwent MT less frequently than White patients, likely in part because of longer times from last known well to hospital arrival and a lower rate of documented acute large vessel occlusion. Further studies are needed to assess whether extending the MT time window and more aggressive large vessel occlusion screening protocols mitigate this disparity.

Racial disparity in the utilization of mechanical thrombectomy (MT) for treatment of acute ischemic stroke caused by large vessel occlusion (LVO) has been previously documented. ^{1 , 2} However, there is a paucity of data regarding the status of this disparity in recent years, during which health care infrastructure expansion has increased access to MT. In addition, previous studies utilized data from national databases that lacked the individual patent data necessary to determine eligibility for the procedure. The present study analyzed patient‐level data from one of the largest health care systems in the United States to investigate the extent of, and potential reasons for, persistent racial disparity in MT utilization from 2016 to 2020.

Methods

Institutional review board approval was obtained for this study, including waiver of informed consent. The data that support the findings of this study are available from the corresponding author on reasonable request, pending approval by the Ascension Data Science Institute.

A system of 43 hospitals in 12 states (AL, FL, IL, IN, KS, MD, MI, NY, OK, TN, TX, WI) contributed to a collective stroke database (Table S1), from which data were extracted for all patients presenting between January 1, 2016, and September 30, 2020, with a diagnosis of ischemic stroke or transient ischemic attack. Table 1 lists the demographic and clinical variables of interest. Demographic data were obtained in accordance with hospital‐specific registration protocols. The median income of a patient’s residential zip code was derived from publicly available government data and used as a proxy for socioeconomic status. ³ Urban zip codes were defined in accordance with the US Census Bureau as a densely developed territory with a population of at least 50 000. High‐volume stroke centers were defined as hospitals averaging >10 stroke admissions per month during the study period.

Table 1.

Comparison of Demographics, Clinical Characteristics, and Outcomes of White and Black Patients

	Total cohort (N=34 596)	White patients (n=26 640)	Black patients (n=7956)	P value
Demographics
Age (mean±SD), y	71.3±14.3	73.1±13.9	65.0±14.0	<0.01
Sex				<0.01
Male	16 669	12 983 (48.7)	3686 (46.3)
Female	17 927	13 657 (51.3)	4270 (53.7)
Residential zip code median income <$50 000	24 023	17 098 (64.2)	6925 (87.0)	<0.01
Urban residential zip code	8064	4160 (15.6)	3904 (49.1)	<0.01
Presenting hospital, CSC/TSC	20 329	15 369 (57.7)	4960 (62.3)	<0.01
Presenting hospital, high‐volume	30 784	23 792 (89.3)	6992 (87.9)	<0.01
Region				<0.01
Midwest	21 114	15 843 (59.5)	5271 (66.3)
South	9312	8000 (30.0)	1312 (16.5)
East	4170	2797 (10.5)	1373 (17.3)
Clinical characteristics
Baseline ambulation				<0.01
Independent	31 302	24 086 (90.4)	7216 (90.7)
With assistance	2169	1725 (6.5)	444 (5.6)
Unable	1125	829 (3.1)	296 (3.7)
NIHSS, median (IQR)	3.0 (1.0–8.0)	3.0 (1.0–7.0)	3.0 (1.0–8.0)	0.22
Outcomes
Treated with mechanical thrombectomy	1190 (3.4)	957 (3.6)	233 (2.9)	<0.01
Early arrival*	11 475 (33.2)	9289 (34.9)	2186 (27.5)	<0.01
Early‐arrival patients with documented aLVO	1112 (9.7)	924 (9.9)	188 (8.6)	0.06
Early‐arrival patients with documented aLVO treated with mechanical thrombectomy	726 (65.3)	602 (65.2)	124 (66.0)	0.87

aLVO indicates anterior circulation large vessel occlusion; CSC, comprehensive stroke center; IQR, interquartile range; NIHSS, National Institute of Health Stroke Scale; and TSC, thrombectomy‐capable stroke center.

^*Early arrival was defined as hospital presentation within 5 h of “last known well.”

Demographic and clinical variables were dichotomized and subgroups were compared with respect to each of the following dependent variables: (1) overall rate of treatment with MT; (2) proportion of patients arriving at the hospital within 5 hours of “last known well” (LKW; “early arrival”); (3) proportion of early‐arrival patients with documented anterior circulation LVO (aLVO), defined as internal carotid artery terminus, M1 or M2 occlusion on vascular imaging (computed tomographic angiography or magnetic resonance angiography, depending on hospital‐specific stroke triage protocols); and (4) rate of MT among early‐arrival patients with documented aLVO. An early‐arrival threshold of 5 hours from LKW was used to allow for groin puncture within the MT “early time window” of 6 hours. Patients were selected for MT according to hospital‐specific protocols. A mixed effects logistic regression model was then constructed for each dependent variable. Demographic and clinical variables with P<0.20 in univariable analysis were included as fixed effects, and presenting hospital was included as a random effect. A P value <0.05 was considered statistically significant.

Statistical methods for addressing missing data are detailed in Data S1. In particular, patients who initially presented to our hospital network and were then transferred to a hospital outside our network (out‐of‐network transfers) were excluded from the study cohort because it is unknown whether these patients underwent MT, and sensitivity analysis confirmed that exclusion of these patients did not qualitatively affect our results with respect to race. All statistical analyses were performed using R version 4.0.3 (The R Foundation).

Results

Cohort characteristics stratified by race are summarized in Table 1. The study included 34 596 patients with a mean age of 71.2 years (interquartile range, 62–82 years); 16 669 were men (48.2%) and 17 927 were women (51.8%); 26 640 were White, non‐Hispanic (77.0%) and 7956 were African American or Black (23.0%). Black patients were more likely than White patients to be women (53.7 versus 51.3, P<0.01), reside in a zip code with an annual median income <$50 000 (87.0% versus 64.2%, P<0.01), reside in an urban zip code (49.1% versus 15.6%, P<0.01), present to a comprehensive or thrombectomy‐capable stroke center (62.4% versus 57.7%, P<0.01), present to hospitals in the Midwest and East Coast (66.3% versus 59.5% and 17.3% versus 10.5%, respectively), ambulate independently (90.7% versus 90.4%), or be unable to ambulate (3.7% versus 3.1%) at baseline. Black patients were less likely to present to a high‐volume stroke center (87.9% versus 89.3%, P<0.01), present to hospitals in the South (16.5% versus 30.0%), and ambulate with assistance at baseline (5.6% versus 6.5%).

The overall rate of MT was 3.4% (1190 of 34 596). Overall, MT utilization rates for demographic and clinical subgroups are listed in Table S2. In multivariable analysis, significantly lower rates of MT were seen among Black patients compared with White patients (adjusted odds ratio [OR], 0.65; 95% CI, 0.54–0.76) (Table 2). The cohort included 11 475 patients (33.2%) who arrived within 5 hours of LKW (Table S3). In multivariable analysis, Black patients were less likely than White patients to arrive at the hospital within 5 hours of LKW (adjusted OR, 0.73; 95% CI, 0.69–0.78) (Table 2). Among early‐arrival patients, 1112 (9.7%) had vascular imaging that showed an aLVO (Table S4). In multivariable analysis, aLVO was less commonly documented in Black patients compared with White patients (adjusted OR, 0.78; 95% CI, 0.64–0.96) (Table 2). Among early‐arrival patients with documented aLVO, 726 (65.3%) underwent MT (Table S5). In univariable analysis, rates of MT among White and Black patients were comparable (65.2% versus 66.0%, respectively; P=0.83).

Table 2.

Adjusted ORs of Study Outcomes for Black Patients Relative to White Patients

Outcome	Adjusted OR (95% CI)	P value
Treatment with mechanical thrombectomy*	0.65 (0.54–0.76)	<0.01
Early arrival (within 5 h of “last known well”) †	0.73 (0.69–0.78)	<0.01
Documented anterior circulation large vessel occlusion among early‐arrival patients ‡	0.78 (0.64–0.96)	0.02

OR indicates odds ratio.

^*Fixed effects included in the model: race, median income of residential zip code, presentation to a comprehensive stroke center/thrombectomy‐capable stroke center (CSC/TSC), presenting hospital region, presentation to a high‐volume stroke center, baseline ambulatory function, and stroke severity.

^† Fixed effects included in the model: median income and population density of residential zip code, presenting hospital region, baseline ambulatory function, and stroke severity.

^‡ Fixed effects included in the model: age, sex, median income and population density of residential zip code, presentation to a CSC/TSC, presenting hospital region, presentation to a high‐volume stroke center, baseline ambulatory function, and stroke severity.

Discussion

Previous studies have shown that Black patients were less likely than White patients to be treated with MT before the landmark MT clinical trials. ^{1 , 2} For example, using diagnosis and procedure codes in the Nationwide Inpatient Sample from 2006 to 2014, Esenwa et al ² found that MT rates were a third lower in Black patients compared with White patients (OR, 0.67; 95% CI, 0.58–0.76). A more recent analysis of a national database that included billing records and diagnosis codes from 2016 to 2018 similarly found that even in the thrombectomy era, Black race was independently associated with lower institutional utilization of MT. ¹ Our study shows that despite recognition of this racial disparity and increased access to MT in recent years, this problem has persisted, with Black patients still ≈35% less likely than White patients to be treated with MT. In addition, while prior studies could not control for clinical factors because of a lack of patient‐level data, our multivariable analysis shows that this racial disparity is independent of baseline ambulatory status and stroke severity.

Our study identifies two potential targets for interventions to minimize racial disparity in MT utilization. First, Black patients were almost 30% less likely than White patients to arrive at the hospital within 5 hours of LKW. Longer LKW‐to‐arrival times among Black patients, primarily attributed to lower stroke literacy and lack of trust in the health care system owing to historic inequalities, is hypothesized to be a driver of racial disparity in intravenous tissue plasminogen activator administration ^{4 , 5} and likely also contributes to racial disparity in MT eligibility. Randomized controlled trials have shown that education initiatives such as the HipHop Stroke intervention and Black beautician stroke education effectively reduce racial disparities in timely hospital arrival after stroke onset. ⁶ Moreover, studies are needed to assess whether advances that extend the MT time window, such as perfusion imaging, impact racial disparity in MT utilization.

Second, our study also shows that Black patients less frequently underwent MT because of a lower incidence of documented aLVO (adjusted OR, 0.78; 95% CI, 0.64–0.96). The rate of aLVO may have been lower among Black patients because of racial differences in stroke cause, including a higher prevalence of cardioembolic stroke in White patients and a higher prevalence of intracranial atherosclerosis in Blacks. ^{7 , 8} However, studies have also shown that Black patients with stroke experience longer times between hospital arrival and head computed tomography scan ⁹ and are less likely to undergo noninvasive cerebrovascular testing. ¹⁰ It is unknown whether emergent vascular imaging was obtained at comparable rates in Black and White patients because our data set lacked consistent documentation of the timing of vascular imaging. Well‐documented racial disparities in stroke prevention, treatment, and recovery have been attributed to implicit bias, as well as structural factors such as long‐standing inequities in health care access. ¹¹ Therefore, it is entirely possible that systemic racial bias contributes to lower rates of LVO screening of Black patients with resultant underreporting of MT eligibility. Further studies are needed to assess for racial disparities in LVO screening among patients arriving within the MT time window, in which case racial disparity may be mitigated by more aggressive LVO screening protocols.

This study has limitations to acknowledge. First, as with any large multicenter registry, our data are subject to biases and recording error, including racial designations. While self‐reporting of race and ethnicity is recommended, some patients’ races may have been documented by observation of admitting or registration staff. We also acknowledge that the broad racial categories utilized in this and other publications on demographic disparities in health care poses a potential limitation. Similarly, our results may be confounded by changes in hospital‐specific protocols during the study period, as is always the case with temporal data. Second, our study focused on patients with aLVO presenting during the early time window because American Heart Association guidelines regarding MT utilization in this population are supported by level I evidence. Still, we acknowledge that variation in MT utilization as a result of the discretion of treating physicians, unrelated to implicit racial bias, is a potential study limitation. Third, while our study included data from 43 hospitals in 12 states, our results may not be generalizable to states not represented. Fourth, 11.6% of baseline ambulatory data and 9.2% of presenting National Institute of Health Stroke Scale scores were imputed. However, a key strength of this study is the overall availability of clinical data, which unlike prior studies, allowed us to control for these variables when analyzing demographic disparities. Last, patients transferred out of our network were excluded from this analysis because their subsequent treatment is unknown. As detailed in Data S1, exclusion of these patients did not qualitatively affect our results regarding racial disparities. However, out‐of‐network transfers were disproportionately from low‐income, nonurban residential zip codes. Therefore, our data cannot be used to draw independent conclusions regarding the impact of patient residential zip code median income and population density on MT rates.

Conclusions

Recent data from a large multicenter cohort show Black patients with stroke are still treated with MT less frequently than White patients. Contributing factors likely include longer times from LKW to hospital arrival and a lower rate of documented aLVO among Black patients. Further studies are needed to assess whether racial disparity in MT treatment may be mitigated by advances in stroke care that extend the MT time window or more aggressive LVO screening protocols.

Sources of Funding

None.

Disclosures

D.H.S. is a consultant for Medtronic, Stryker, Microvention, and Phenox, and a speaker and proctor for Medtronic. D.P.G. is a consultant for iSchemaView, Medtronic, and Siemens Healthineers A.G. The remaining authors have no disclosures to report.

Supporting information

Data S1. Supplemental Methods

Tables S1–S5

References 12, 13, 14, 15