Does exposure to certified stroke centers affect all communities equitably? Stroke patient outcomes by community socioeconomic status

Abstract

Objective:

Our aim was to determine how access, treatment, and outcomes change for patients after their community is exposed to a newly certified stroke center based on a community’s disadvantage status.

Methods:

In this retrospective observational study, we included Medicare Fee-for-Service patients from a national claims-based database admitted to hospitals between January 2009 and December 2019 whose primary diagnosis was acute ischemic stroke. We implemented linear probability models with community fixed effects to compare changes in outcomes when communities were exposed to newly certified stroke centers nearby relative to similar communities that did not experience stroke center expansion, controlling for patient demographics and comorbid conditions and secular trends. Outcomes included admission to a certified stroke center, receipt of thrombolytic therapy or mechanical thrombectomy, and 1-year mortality.

Results:

We analyzed 2,807,763 patients with acute ischemic stroke. Only 68% of those in disadvantaged communities had exposure to a newly certified stroke center nearby during the study period, while 92% of those in advantaged communities had the same exposure. In disadvantaged communities, new stroke centers were associated with a 23.1 percentage point (pp) increase in admission to a stroke center, a 0.3 pp increase in “drip-and-ship” thrombolytic therapy, a 0.2 pp decrease in thrombectomy, and no statistically significant changes in “drip-and-stay” therapy or 1-year mortality. In advantaged communities, new stroke centers were associated with a 4.2 pp increase in admission to a stroke center, a 0.6 pp decrease in “drip-and-ship” therapy, a 0.8 pp increase in “drip-and-stay” thrombolytic therapy, a 0.2 pp increase in thrombectomy, and a small reduction in one-year mortality of 0.6 pp.

Conclusions:

Stroke center expansion has been uneven, and its effects on patient care differ by a community’s socioeconomic status. These findings may guide stroke center initiatives to improve care in disadvantaged communities.

Introduction

Background

Despite decades of documented disparities in stroke care for traditionally underserved populations, significant inequities persist across the entire stroke care continuum—from pre-hospital transport to thrombolytic therapy to mortality,¹ particularly affecting racial and ethnic minority groups and individuals with low socioeconomic status. Black, Hispanic, low-income, and rural patients receive lower-quality care,^2–6 and the risk of death from stroke is 44% higher for Black patients compared to White patients.⁷ Furthermore, Black patients are one-fifth as likely, and poorly insured patients one-ninth as likely, to receive thrombolytic therapy compared with their counterparts.⁸ These inequities persist despite rapid improvements in technologies and treatments for stroke and other cerebrovascular diseases.^9–11

Importance

Certification of hospitals that provide specialized stroke care services has proliferated in recent decades, driven by the belief that expanding access to these services would create a “rising tide that lifts all boats” that benefits the broader population.^12,13 Over the past decade, there have been significant improvements in access to stroke centers over time. In 2011, 80% of the U.S. population had access by ground within 60 minutes,¹⁴ which had increased to 91% by 2019.¹⁵ For access within 30 minutes, our work has also shown that in 2009, 34% of stroke patients had no stroke centers within a 30-minute drive time, and by 2019, that number decreased to 17%.¹⁶ However, the theory that increased availability of these services benefits all segments of the population equitably has not been rigorously tested. As there continues to be greater awareness of how the “built environment” (proximity to hospitals, availability of emergency transportation, distribution of healthcare facilities, urban vs. rural infrastructure) contributes to structural discrimination for stroke patients,¹⁷ it is crucial to quantify whether efforts to expand certified stroke centers over the past decade have benefitted patients and communities and how any associated improvements might differ by the location of stroke center expansion.

Previous studies have typically relied on traditional measures of disadvantage, such as race, income, and education. However, more comprehensive measures, such as the Area Deprivation Index (ADI), have been developed in recent years.¹⁸ Employing these broader measures of disadvantage is crucial for comprehensively assessing access to healthcare services, such as stroke, as they enable a multifaceted exploration of the socioeconomic and environmental factors that influence healthcare disparities.

Goals of This Investigation

Thus, we identified where new stroke center certifications occurred in the US from January 2009 to December 2019 to address the following questions: (1) Among Medicare Fee-For-Service patients, does the expansion of certified stroke centers occur at similar rates across communities with different socioeconomic statuses, as defined by the ADI? (2) Is having a newly certified stroke center available near a community associated with similar improvements in patient outcomes (likelihood of admission to a stroke center, receipt of intravenous thrombolytics, receipt of mechanical thrombectomy, home at 90 days, and one-year mortality) for communities with different socioeconomic statuses?

Methods

Study Design and Setting

This study was a retrospective observational study that was deemed exempt by the National Bureau of Economic Research Institutional Review Board (IRB). Our study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cohort studies.

Selection of Participants

Our study population included all Medicare Fee-for-Service patients from a national claims-based database in the United States with a primary discharge diagnosis of acute ischemic stroke who were admitted to hospitals between January 1, 2009, and December 31, 2019. We identified our stroke patient cohort using the following International Classification of Diseases (ICD) codes: 433.x1, 434.x1, or 436 (ICD version 9), or I63 (ICD version 10), in accordance with prior literature.^17,19,20 We excluded patients whose mailing ZIP code was more than 100 miles away from their admitted hospital, since this likely implied that their residential address differed from their mailing address or that they were treated while away from home.

Database Details

Our analysis combined patient-, hospital-, and community-level data. We obtained patient-level data from the 100% Medicare Provider and Analysis Review (MedPAR) between 2009 and 2019, which included admission dates, demographics, comorbid conditions, procedures received, and mailing ZIP codes. These care episode records were then linked to vital statistics to capture each patient’s death status. We linked the MedPAR records with Medicare Beneficiary Summary Files to obtain vital statistics. We collected hospital stroke certification status from national accrediting bodies as well as from states (further details in the Supplemental Methods and elsewhere²¹), and we supplemented this stroke center certification data with data from the American Hospital Association and the Healthcare Cost Report Information System to capture additional hospital characteristics, including geographic coordinates. We used 2010 US Census data and American Community Surveys from 2011 to 2019 to identify geographical coordinates and demographic information for each ZIP code community, and the Neighborhood Atlas to identify the socioeconomic status of each patient’s community.^18,22 Finally, we derived a driving-time database using web-based queries between each patient’s ZIP code and nearby stroke centers based on each location’s geographic coordinates (longitude and latitude).²³

Key Measurements

The Neighborhood Atlas^22,24 ranks each neighborhood’s socioeconomic well-being based on multiple dimensions, where a neighborhood is defined as a Census Block Group (CBG).^24,25 Specifically, each neighborhood is assigned a percentile ranking based on the ADI. This validated marker combines 17 measures of employment, income, housing, and education from the American Community Survey to create a score for each CBG in the US.^22,24–29 These ADI values are based on the entire US and range from 1 to 100, with 1 being the least deprived neighborhood and 100 the most deprived.

While ADI is defined at the neighborhood level, our patient community was defined at the ZIP code level, as this was the smallest geographic unit available from our patient data. Following a prior study³⁰ that mapped neighborhood populations to ZIP code communities (described in more detail in the Supplemental Methods), we determined the percentage of each community’s population that lived in a high ADI neighborhood (i.e., if the ADI value was above 80) and a low ADI neighborhood (if ADI value was 20 or below). We then categorized communities into three groups based on the share of the population in high- or low-ADI neighborhoods. Specifically, advantaged communities were defined as ZIP codes with at least 25% of the population living in low ADI neighborhoods; likewise, disadvantaged communities were defined as ZIP codes with at least 25% of the population in high ADI neighborhoods. The remaining ZIP codes (those with less than 25% of the population in high and low ADIs) were defined as mixed communities. Under this definition, approximately 25% of the population was each in advantaged and disadvantaged communities, while the remaining 50% were in mixed communities. To track communities consistently over time, these community ADI categories were made time-invariant based on ADI values from 2015 (the earliest year available on the Neighborhood Atlas website). Our past experience working with these types of community-level measures showed that they are highly correlated over time.³¹ In other words, we assume that communities’ movement in and out of each advantage category is minimal over time.

Our key determination in this analysis was whether a patient’s community was exposed to a newly certified stroke center within a 30-minute driving time. Broadly speaking, hospitals can be certified as stroke centers at four levels: Acute Stroke Ready Hospital (ASRH), Primary Stroke Center (PSC), Thrombectomy-Capable Stroke Center (TSC), or Comprehensive Stroke Center (CSC). ASRHs are the most basic type of stroke center certification, primarily focusing on patient stabilization and initiating emergency stroke treatment. Typically, patients initially treated at an ASRH are transferred to a more advanced facility for further care. PSCs offer structured inpatient stroke care, while TSCs and CSCs are notable for their enhanced capabilities, including performing mechanical thrombectomy. Following previous literature,¹⁶ we grouped TSCs and CSCs together since TSCs constitute a very small proportion of hospitals, and because they are the only two levels of stroke-certified hospitals that are required to be able to perform mechanical embolectomy for acute ischemic stroke. CSCs are also required to be able to treat hemorrhagic stroke and perform procedures to treat aneurysms, which are not the focus of our analysis, which is limited to ischemic stroke. Based on the stroke center certification data collected, we defined a hospital as a stroke center on and after the year and quarter (year-quarter) that it achieved any level of stroke center certification. Then, together with our driving time database, we identified all hospitals within a 30-minute driving time (defined as nearby) of the geographic center (as listed in the US Census) of a ZIP code community for each quarter. We selected a 30-minute driving time threshold due to the time-sensitive nature of efficacy and risks associated with acute stroke interventions, as well as evidence from prior studies on travel times for medical treatments, as well as prior work on travel times for medical treatments, including emergency cesarean delivery,³² cardiac care,^33,34 and primary care.³⁵ We computed the actual driving time between each community–hospital pair using automation software from Stata that interfaced with HERE Technologies’ web-based platform^36,37 and then evaluated quarter-to-quarter changes in hospital stroke certification status within the set. A community was said to experience a stroke center expansion on and after the quarter that any hospital within a 30-minute drive acquired stroke center certification. We conducted a sensitivity analysis with the threshold changed to 15 minutes.

Outcomes

We examined whether nearby stroke center certification was associated with changes in stroke care access, acute stroke treatment, and health outcomes that differed based on a community’s socioeconomic disadvantage status. Specifically, we examined admission to stroke centers, receipt of thrombolytic therapy, receipt of mechanical thrombectomy, home (not in an inpatient acute or rehabilitation hospital) at 90 days, and finally, one-year mortality rates. We further differentiated two models of thrombolytic therapy: “drip-and-ship” cases, where patients receive intravenous thrombolytics at one center but are then transferred to the current hospital for admission, and “drip-and-stay” cases, where stroke patients receive thrombolytics and then are admitted to that initial hospital.

Main Analysis

Our empirical model followed a difference-in-differences framework, and our goal was to compare changes in outcomes when a community experienced an expansion in stroke center access (treatment communities) relative to the same type of community (advantaged, mixed, or disadvantaged) that did not experience a change in stroke center access (reference communities).

Since all outcomes were binary, we implemented a linear probability model with community fixed effects. The key independent variable is a binary indicator that took on a value of 1 on and after the time that a community was exposed to a newly certified stroke center within a 30-minute drive. We estimate this coefficient separately for communities classified as advantaged, mixed, and disadvantaged. Using access outcomes as an example, the coefficient on this crucial variable for the advantaged community has the following interpretation: it captured changes in the probability of admission to a certified stroke center among stroke patients in advantaged communities when their community experienced a stroke center expansion relative to changes in this same probability for patients who also lived in advantaged communities but did not experience a stroke center expansion. Likewise, analyses were performed for mixed and disadvantaged communities.

We also included year indicators to capture temporal trends common to all communities. The community fixed effects were critical to our identification strategy because they controlled for unobserved time-invariant heterogeneity across communities, including any baseline differences in underlying patient health and socioeconomic conditions. Under this model specification, each community’s socioeconomic status (advantaged, mixed, or disadvantaged) was subsumed entirely by the community fixed effects because disadvantage status was time-invariant. Lastly, we controlled for patient demographic covariates (sex, 5-year age groups, race, ethnicity) and 23 disease-related risk adjustments (based on comorbid conditions recorded on the MedPAR records), following prior work.^38,39

Sensitivity Analyses

To assess the robustness of our findings and to explore potential mechanisms behind our main findings, we conducted a series of sensitivity analyses.

First, because stroke certification is granted only after hospitals meet the required criteria and participate in a review process, it is likely that hospitals may have improved their stroke care prior to the actual certification date. Hence, our difference-in-differences estimates may be conservative. We investigated this possibility by expanding the binary post-exposure indicator into the following periods: 3 or more years prior to exposure (reference period), 1–2 years prior to exposure (the pre-certification period), 0–2 years after exposure (the initial period), and 3+ years after exposure (the steady state period). Second, we investigated whether differential treatment rates were driven by the level of stroke certification obtained. Third, we examined whether our findings were dependent on the 30-minute driving time definition by applying a more conservative 15-minute threshold.

Lastly, aggregating the Neighborhood Atlas ADI to the ZIP code level introduces noise in our classification of the community. As an internal validity test, we used an alternative ADI index to classify community socioeconomic status, constructed using socioeconomic measures directly at the zip code level based on previously validated measures.²⁵ A community is classified as advantaged if its deprivation index is at the lowest quartile, average community if the index is in the inter-quartiles, and disadvantaged if the index is in the highest quartile of the ADI index value. (More details are provided in the technical supplement.) All analyses were done using Stata 18.⁴⁰

Results

Characteristics of Study Subjects

We identified 2,807,763 patients with acute ischemic stroke who were hospitalized from January 2009 to December 2019 in the US. Twenty-two percent of these patients lived in advantaged communities, 48% in mixed communities, and 29% in disadvantaged communities (Table 1). Among those living in advantaged communities, 92% had exposure to any tier of newly certified stroke centers during the study period, while only 68% of those in disadvantaged communities had the same exposure. Demographically, there were more White patients (79% vs. 75%) and fewer Black patients (10% vs. 21%) in advantaged communities versus disadvantaged communities. Advantaged communities also had more elderly patients (37% vs. 27% for patients 85 years and older) and a much smaller rural population (2% vs. 42%).

Table 1.

Descriptive Statistics of Patient Characteristics

	Advantaged Communities		Mixed Communities		Disadvantaged Communities
	N	%	N	%	N	%
N	631,351	22%	1,352,365	48%	824,047	29%
Lives in community where a hospital within a 30-minute drive received stroke certification during study period (2009–2019)
Overall	583,090	92%	1,085,120	80%	557,358	68%
By highest certification level
ASRH	3,130	0%	58,346	4%	74,167	9%
PSC	61,359	10%	302,074	22%	185,321	22%
TSC or CSC	518,601	82%	724,700	54%	297,870	36%
Lives in rural area	12,846	2%	231,297	17%	348,223	42%
Patient demographics
White	497,318	79%	1,161,523	86%	621,540	75%
Black	61,380	10%	127,570	9%	172,523	21%
Hispanic	18,151	3%	24,067	2%	12,662	2%
Other non-white races	24,453	4%	22,328	2%	8,630	1%
Female	353,417	56%	753,578	56%	468,690	57%
Male	277,934	44%	598,787	44%	355,357	43%
Age distribution at time of admission
65–69 years	80,634	13%	210,791	16%	151,525	18%
70–74 years	92,990	15%	229,003	17%	150,173	18%
75–79 years	105,895	17%	244,292	18%	152,302	18%
80–84 years	120,918	19%	255,896	19%	150,312	18%
85+ years	230,914	37%	412,383	30%	219,735	27%
Patient clinical conditions
Recurrent stroke	53,336	8%	115,204	9%	71,498	9%
Transfer	30,066	5%	97,564	7%	84,369	10%
Peripheral vascular disease	67,054	11%	132,473	10%	78,970	10%
Pulmonary circulation disorders	22,597	4%	47,889	4%	26,373	3%
Diabetes	184,765	29%	437,260	32%	297,943	36%
Kidney failure	115,392	18%	251,229	19%	156,846	19%
Liver	6,929	1%	12,955	1%	7,836	1%
Cancer	28,454	5%	53,466	4%	30,240	4%
Dementia	71,982	11%	139,881	10%	84,711	10%
Valvular disease	67,922	11%	132,465	10%	69,287	8%
Hypertension	528,570	84%	1,148,151	85%	707,009	86%
Chronic pulmonary disease	85,441	14%	219,049	16%	144,724	18%
Rheumatoid arthritis/collagen vascular	17,484	3%	38,718	3%	21,113	3%
Coagulation deficiency	26,085	4%	50,298	4%	28,429	3%
Obesity	41,810	7%	114,552	8%	72,646	9%
Substance use	13,199	2%	28,546	2%	19,163	2%
Depression	55,816	9%	136,199	10%	78,195	9%
Psychosis	43,132	7%	92,360	7%	52,961	6%
Hypothyroidism	108,419	17%	242,396	18%	135,209	16%
Paralysis and other neurological disorder	364,641	58%	773,845	57%	475,845	58%
Ulcer	2,232	0%	4,516	0%	2,842	0%
Weight loss	27,657	4%	56,688	4%	39,747	5%
Fluid and electrolyte disorders	140,896	22%	305,356	23%	195,023	24%
Anemia (blood loss and deficiency)	79,772	13%	164,125	12%	107,687	13%
Access, Treatment Received, and Mortality
Admitted to stroke hospital	527,333	84%	1,050,828	78%	570,097	69%
Received thrombolytic therapy during hospitalization	70,381	11%	136,884	10%	73,884	9%
Received thrombectomy	45,769	7%	94,560	7%	54,173	7%
At home by 90-days	440,598	70%	943,785	70%	559,155	68%
1-year mortality	178,119	28%	378,244	28%	237,375	29%

Abbreviations: ASRH, Acute Stroke Ready Hospital; PSC, Primary Stroke Center; TSC, Thrombectomy Capable Stroke Center; CSC, Comprehensive Stroke Center.

Main Results

Figure 1 shows how the gap in geographic access to different levels of stroke centers across the three community types changed over time. Specifically, disparities in access to TSCs or CSCs grew over time (Figure 1A). In 2010, the gap between advantaged and disadvantaged communities in access to TSCs or CSCs was 3.3 percentage points (pp) (3.5% vs. 0.2%), but by the end of 2019, it had grown to 45 pp (78% vs. 33%). In contrast, the observed disparity in access to PSCs within 30 minutes shrank over time between advantaged and disadvantaged communities, from 39 pp in 2009 to 30 pp in 2019 (Figure 1B), as more PSCs were upgraded to TSCs or CSCs in advantaged communities in the later period. Finally, although there were more ASRHs in advantaged communities in 2009 (7.4% in advantaged vs. 0.2% in disadvantaged), ASRHs increased in number to a greater extent in disadvantaged communities versus advantaged communities during this period, such that by 2019, the gap was less than 1 pp (Figure 1C).

Figure 1. Percent of patients who had geographic access to a certified stroke center within a 30-minute drive between 2009 and 2019.

Abbreviations: Abbreviations: ASRH, Acute Stroke Ready Hospital; PSC, Primary Stroke Center; TSC, Thrombectomy Capable Stroke Center; CSC, Comprehensive Stroke Center.

Patient Outcomes

The remaining results focused on patient outcomes in a community when a nearby stroke center became certified. Figure 2 displays the coefficient estimates of the critical variables, and Table E1 provides the full results. Disadvantaged communities saw the greatest increase in the likelihood of being admitted to a stroke center after nearby stroke center certification, with a 23.1 pp (95% CI: 21.5, 24.7) increase in this outcome (Figure 2A), representing a 51.3% relative increase (baseline rate of 45.0%). Mixed and advantaged communities also experienced increases in admission to any tier of stroke center after certification of a stroke center nearby, but of a lesser magnitude. Specifically, mixed communities saw a 16.4 pp (CI: 15.4, 17.5) increase, and advantaged communities a 4.2 pp (CI: 2.9, 5.4) increase, representing 29.2% and 5.9% relative increases from baseline, respectively.

Figure 2. Changes in patient access, treatment, and health outcomes when a hospital within a 30-minute drive gained stroke certification.

Note: X-axis represents absolute percentage point changes in outcomes relative to changes in the reference community during the same period (community within the same ADI category where no hospital within a 30-minute drive had gained stroke certification level). Error bars represent a 95% confidence interval.

Figure 2B1 shows that after nearby stroke center certification, “drip-and-ship” thrombolytic cases increased only in disadvantaged communities, by about 0.3 pp (CI: 0.1, 0.4), or a 38.0% relative increase, and decreased in advantaged and mixed communities. On the other hand, Figure 2B2 shows that “drip-and-stay” cases increased after nearby stroke center certification in advantaged and mixed communities, by 0.8 pp (CI: 0.6, 1.0) and 0.2 pp (CI: 0.1, 0.4), respectively, a 14.6% relative increase.

The probability of thrombectomy after stroke center certification increased in advantaged communities by 0.2 pp (CI: 0.1, 0.3) (Figure 2B3), a 22.8% relative increase. In contrast, thrombectomy cases in disadvantaged communities decreased by 0.2 pp (CI: −0.3, −0.1), or a 35.8% relative decrease, with stroke center certification. While the probability of being home at 90 days had positive coefficients for all community types, none were statistically significant (Figure 2C1). Finally, one-year mortality rates decreased for advantaged and mixed communities, by 0.6 pp (CI: −1.0, −0.3) and 0.3 pp (CI: −0.6, −0.1), respectively (Figure 2C2). For disadvantaged communities, one-year mortality rates slightly decreased by 0.03 pp (CI: −0.4, 0.3), but this result was not statistically significant.

Sensitivity Analyses

In our first sensitivity analysis, we expanded the binary post-exposure indicator into the following periods: 3 or more years prior to exposure (reference period), 1–2 years prior to exposure (the pre-certification period), 0–2 years after exposure (the initial period), and 3+ years after exposure (the steady state period). Figure 3a shows that in advantaged communities, the probability of “drip-and-stay” thrombolysis and thrombectomy began increasing during the pre-certification period. Similarly, Figure 3b shows that the probability of being home at 90 days also increased during the pre-certification period for advantaged communities, while one-year mortality started to show statistically significant improvement during the initial post-exposure period (detailed results of both Figures available in Table E2). Disadvantaged communities exhibited a different pattern from the other communities: these communities began experiencing an increase in receipt of “drip-and-ship” thrombolytics during the pre-certification and initial periods, and a small increase in “drip-and-stay” thrombolytics in the steady state period. There were no statistically significant changes in the probabilities of thrombectomy or health outcomes for disadvantaged communities.

Figure 3a. Changes in receipt of stroke treatments when a hospital within a 30-minute drive gained stroke certification, by exposure period.

Note: X-axis represents absolute percentage point changes compared to >2 years pre-expansion period within the treatment community, relative to changes in the reference community during the same period. Error bars represent a 95% confidence interval.

Figure 3b. Changes in patient health outcomes when a hospital within a 30-minute drive gained stroke certification, by exposure period.

Note: X-axis represents absolute percentage point changes compared to >2 years pre-expansion period within the treatment community, relative to changes in the reference community during the same period. Error bars represent a 95% confidence interval.

Second, we investigated whether differential treatment rates were driven by the level of stroke certification obtained. Figure 4 shows that the probability of “drip-and-stay” thrombolytic therapy and thrombectomy increased across all communities if patients were exposed to newly certified TSCs or CSCs (complete results available under Table E3). Exposure to newly certified TSCs or CSCs was also associated with decreased one-year mortality across all community types (from −1.0 pp [CI: −1.3, −0.6] in advantaged communities to −0.5 pp [CI: −0.9, −0.0] in disadvantaged communities). On the other hand, exposure to an ASRH was associated with an increased probability of “drip-and-ship” thrombolytics in mixed (2.0 pp, CI: 1.5, 2.6) and disadvantaged communities (2.0 pp, CI: 1.4, 2.5).

Figure 4. Changes in treatment and health outcomes when a hospital within a 30-minute drive gained stroke certification, by certification levels.

Note: X-axis represents absolute percentage point changes in outcomes relative to changes in the reference community during the same period (community in the same ADI category where no hospital within a 30-minute drive had gained stroke certification level). Error bars represent a 95% confidence interval.

Third, in our sensitivity analysis, we evaluated the impact of reducing the 30-minute driving time to 15 minutes, and our results remained robust to this specification (Table E4). Finally, we use ADI constructed directly using ZIP code level socioeconomic measures (details in technical supplement eMethods), and our coefficients remain similar to our main results (Table E5).

Limitations

Our study has several limitations. First, our study population was limited to Medicare Fee-For-Service patients. Given that stroke is a disease of aging, our working assumption is that this nationally representative population captures a substantial share of the stroke patient population and that the differences observed across communities would be similar in a non-Medicare population. However, we acknowledge that younger stroke patients, who may be overrepresented in lower socioeconomic status populations,⁴¹ are not captured in this analysis, and that Medicare Advantage enrollment has grown significantly over the study period,⁴² potentially limiting generalizability. Second, due to the limitations of our administrative dataset, health outcomes were limited to patients being home at 90 days and mortality rates, and we could not study the effects of certification on functional outcomes. For instance, increased use of thrombolytics at newly certified centers would be expected to impact functional outcomes, but not mortality. Third, while stroke center certification requires a certain level of capabilities, some centers certified at a lower tier may have capabilities that are seen at higher certification tiers (e.g., PSCs that perform thrombectomies). Fourth, our analysis did not separately model access to telestroke services, which were available in 45% of US emergency departments in 2019, because stroke certification is often a proxy for having such capacity.¹⁵ Only 36% of non-stroke centers had telestroke services compared with 63% of ASRHs and 58% of PSCs, suggesting that the majority of telestroke expansion has occurred in hospitals with existing stroke center certification.¹⁵ To the extent that there is a general growth of telestroke services across all hospitals that correlates with our outcomes, our time dummies in the model would capture this macro trend. Fifth, our analysis also did not account for fixed-wing or rotary air transport which play a role in stroke care delivery, particularly in critical access settings. These modes of transport could influence access to timely intervention independent of proximity to certified stroke centers. As such, our estimates assume that stroke care access was determined by driving distance alone, an assumption that may not fully reflect all modes of acute stroke care delivery. At the same time, the proportion of strokes that are actually brought into the hospital directly from the field are extremely small (while the exact proportion of strokes that are brought into a hospital directly via helicopter is not well known, only 0.36% of EMS activations for cardiac arrest were transported via helicopter)⁴³, and we therefore highly doubt that this tiny proportion of strokes presenting through aeromedical transport influences our study. Almost all stroke patients who are transported via helicopter must at least first present to a hospital, and we do capture interfacility transports. Sixth, even though our models controlled for comorbid conditions, the lack of detailed clinical information that would only be available through medical charts limited our ability to capture the severity of stroke cases or to formally evaluate whether the proportion of candidates for thrombolysis or thrombectomy may differ in these communities. Seventh, our driving time estimates were measured with errors, given that we employed identical geographical coordinates for all patients within a particular ZIP code and assumed standard traffic conditions, particularly in rural areas. This could bias our results towards zero, making our findings a conservative approximation. Eighth, while prior studies have established external validity of the Neighborhood Atlas ADI, we recognize that summarizing it at the ZIP code level introduces noise. Our sensitivity analysis shows that we arrive at the same conclusion when we replace the ADI measure with one that were directly constructed at the ZIP code level, further confirming the internal validity of using this measure to delineate community’s socioeconomic conditions. Ninth, while our stroke center certification data is collected from national and state certification organizations and are likely the most comprehensive dataset available, we might still be missing some stroke centers.²¹ This would bias results only if the missing data pattern systematically differed by geography, which we do not expect.

Lastly, there is growing evidence that disparities in the timing of when stroke patients present for care exist across socioeconomic groups and relative levels of disadvantage. A recent study⁴⁴ found that patients residing in higher area deprivation index (ADI) communities were significantly more likely to have delayed hospital presentations and were less likely to utilize emergency medical services (EMS), regardless of their distance from a hospital. While our analysis could not account for patient-level time-to-presentation or EMS usage, our difference-in-differences design relies on the assumption that such unmeasured factors were either stable over time or changed similarly across comparison groups. However, if disparities in stroke awareness or EMS activation evolved unevenly during the study period, they could have influenced treatment access independently of geographic access to certified stroke centers.

Discussion

In our study of approximately 2.8 million patients with acute ischemic stroke in the United States, we found that geographic access to any tier of certified stroke center at baseline and over time differed for advantaged, mixed, and disadvantaged communities. Disparities in potential access to the highest tier stroke centers (TSC and CSC) between advantaged and disadvantaged communities grew by 14-fold from 2010 to 2019, while the gap in access to primary stroke centers shrank during this period. Although stroke center expansion improved admission likelihood most in disadvantaged communities up to 51%, gains in treatment and outcomes were concentrated in advantaged communities. When we expanded the binary post-exposure indicator (by making the >2 years prior to exposure as a reference period), we found that only advantaged and mixed communities experienced improved treatment and health outcomes, primarily driven by greater exposure to TSC/CSCs.

These disparities were partly due to differences in the types of centers added. Advantaged communities were more likely to gain TSC/CSC centers, associated with increased “drip-and-stay” thrombolytic therapy, thrombectomy, and reduced mortality. In contrast, disadvantaged areas were more likely to gain ASRHs and saw increases only in “drip-and-ship” thrombolytic cases, but not “drip-and-stay” cases. This may stem from integrating newly certified, lower-tier stroke centers into established stroke networks,⁴⁵ which often have pre-existing protocols for transferring patients to higher-tier stroke centers. Conversely, advantaged and mixed communities saw a decrease in “drip-and-ship” cases and an increase in “drip-and-stay” cases when there was a stroke center expansion nearby. This can be understood in the context of our data, which showed that higher-tier stroke centers were more common in advantaged communities, and lower-tier centers were common in disadvantaged communities.

Advantaged communities also showed a smaller relative increase (6%) in stroke center admissions compared with the 51% relative increase seen in disadvantaged communities. This smaller increase suggests a potential duplication of services and ceiling effect when additional stroke centers are concentrated in one geographic area. For example, exposure to a newly certified stroke center may mean there is now a stroke center in a geographic area where there previously had been none, or it may mean that there is, for example, now a third PSC in an area that previously had two, the latter case illustrating the concept of diminishing returns of adding a newly certified stroke center in a potentially saturated area. Despite the smaller increase in admissions, advantaged communities benefited from a 15% relative increase in the receipt of thrombolytics and a 23% relative increase in thrombectomy. This is likely because higher-level stroke centers were more likely to receive patients rather than transfer them. Disadvantaged communities experienced none of these treatment or outcome improvements. These findings suggest that a community’s clinical needs may not be the driving force for stroke center certification, and that other factors, such as local health care market forces and competition between centers, particularly in advantaged communities, could be important drivers of decisions to seek stroke center certification for some centers.

Finally, our findings of improved health outcomes for patients in advantaged and mixed communities can be interpreted in two ways. The most straightforward explanation is that the quality of care improved with certification of higher-level stroke centers, which predominantly took place in non-disadvantaged communities, though thrombolysis and thrombectomy would not necessarily be expected to improve mortality as opposed to functional outcomes. However, improved health outcomes could also be explained by improvements in diagnosing minor strokes compared to the reference community due to increased awareness, recognition, education, and quality initiatives at certified stroke centers, as suggested by prior literature.^46,47 However, if this were the case, we would expect that the volume of stroke patients would increase, which was not borne out in our analysis. Another possibility would be that stroke centers may be more aggressive in offering life-prolonging care for severe stroke patients, thereby decreasing the observed mortality rate for stroke patients in these communities overall.

Our findings reveal that while the expansion of certified stroke centers has been associated with improvements in stroke care, it has also widened disparities between communities. While we cannot definitively ascertain the mechanisms behind the observed changes, disparities between advantaged and disadvantaged communities were apparent at baseline and have widened as the built environment of stroke centers has impacted treatment rates and outcomes. These results underscore the importance of strategic stroke system planning that considers the geographic accessibility to higher-level stroke centers for disadvantaged communities, where feasible, and that local healthcare market forces may not effectively address the goal of addressing known disparities in stroke care without public policy interventions. If patient volumes are insufficient to warrant certification of higher-tier stroke centers in disadvantaged communities, such as in rural areas, streamlining transfer agreements and protocols across hospitals to facilitate rapid transport will be imperative to improve outcomes. Overall, this improved understanding of the relationship between stroke center certification and stroke care across communities of differing socioeconomic statuses may guide and enhance public policy planning for stroke center certification initiatives that aim to maximize patient benefits and promote more equitable improvements in stroke systems of care.

Non-standard Abbreviations and Acronyms

ADI

Area Deprivation Index

ASRH

Acute Stroke Ready Hospital

PSC

Primary Stroke Center

TSC

Thrombectomy-Capable Stroke Center

CSC

Comprehensive Stroke Center