Association Between Racial Segregation and COVID-19 Vaccination Rates

Abstract

Objective: 

To examine the association between county-level Black-White residential segregation and COVID-19 vaccination rates.

Design: 

Observational cross-sectional study using multivariable generalized linear models with state fixed effects to estimate the average marginal effects of segregation on vaccination rates.

Setting: 

National analysis of county-level vaccination rates.

Main Outcome Measure: 

County-level vaccination rates across the United States.

Results: 

We found an overall positive association between county-level segregation and the proportion population fully vaccinated, with a 6.8, 11.3, and 12.8 percentage point increase in the proportion fully vaccinated by May 3, September 27, and December 6, 2021, respectively. Effects were muted after adjustment for sociodemographic variables. Furthermore, in analyses including an interaction term between the county proportion of Black residents and the county dissimilarity index, the association between segregation and vaccination is positive in counties with a lower proportion of Black residents (ie, 5%) but negative in counties with the highest proportions of Black residents (ie, 70%).

Conclusions: 

Findings highlight the importance of methodological decisions when modeling disparities in COVID-19 vaccinations. Researchers should consider mediating and moderating factors and examine interaction effects and stratified analyses taking racial group distributions into account. Results can inform policies around the prioritization of vaccine distribution and outreach.

Racial disparities in COVID-19 outcomes have been well documented. Studies have shown that communities of color have been disproportionately impacted by COVID-19; people of color experienced a significantly higher burden of prevalence, hospitalization, and mortality than their White counterparts.^1–3 Analyses using data from 2020 have also found that counties with higher COVID-19 death rates had a higher proportion of Black residents and greater levels of adverse social determinants of health.^4,5 Disparities in adverse COVID-19 outcomes are driven by both individual and community-level factors, many of which stem from long-standing systemic inequalities and institutionalized racism.⁶ These factors include an increased risk of exposure, severe illness, and mortality due to reduced health care access, insurance, and higher rates of chronic conditions.^3,7 Research has also demonstrated that residential segregation, a fundamental cause of racial disparities in health,⁸ is associated with worse COVID-19 outcomes (both infections and deaths).^9–12

Given these inequities, many vaccine distribution efforts in the United States have focused on equitable distribution in communities disproportionately affected by COVID-19.¹³ Some examples of these equity-focused distribution efforts include those made by members of the National COVID-19 Resiliency Network (NCRN),¹⁴ a national coalition focused on mitigating the impact of COVID-19 on disproportionately impacted populations led by the National Center for Primary Care at Morehouse School of Medicine (MSM) and funded by the US Department of Health & Human Services Office of Minority Health. In collaboration with NCRN, the University of Texas at El Paso (UTEP) hosted a series of mobile vaccination clinics among migrant workers that resulted in 2100 individuals, including farm workers, dairy workers, meat packing workers, and their families, being vaccinated. The National Association of Community Health Centers (NACHC), another NCRN partner, has had success in strategically reaching underrepresented and uninsured patients, with a collective national effort that vaccinated nearly 23 million people, with 69.59% of vaccinations received by patients of color.¹⁵

However, while vaccinations are a key tool in reducing the morbidity and mortality of COVID-19, racial disparities in vaccine uptake persist.⁶ Research has shown that communities of color are disproportionately impacted by challenges to vaccine access such as lack of information about when and where to receive a vaccine, difficulty scheduling vaccination appointments, insufficient transportation, reduced access to childcare, and reduced flexibility in taking time off work.^6,16,17

Studies show that communities with higher proportions of people of color report significantly lower vaccination rates^7,18–20 and experience higher COVID-19 infections and deaths.^18,21–25 There is also research that explores the association between racial segregation and COVID-19 deaths in the United States.^9,10,21 To the best of our knowledge, however, research specifically examining the association between segregation and vaccination at the national level is lacking. Thus, the aim of this study was to examine the association between county-level Black-White residential segregation and COVID-19 vaccination rates across the United States. In the analysis process, we discovered nuances that underscore the importance of methodological decisions when modeling disparities in COVID-19 vaccinations.

Methods

Data sources

We conducted an observational cross-sectional study using publicly available county-level data sources for calendar year 2021, including the MSM NCRN Database, County Health Rankings (CHR),²⁶ and Area Health Resources Files (AHRF).²⁷ Developed by MSM with support from KPMG LLP, the MSM NCRN Database provides researchers with a centralized data repository that aggregates a variety of publicly available data sources, including the Centers for Disease Control and Prevention (CDC), academic centers, the US Census and American Community Survey, and local public health agencies around the United States. The platform incorporates more than 600 data elements at the zip code tabulation area and county levels and is updated weekly.

Study sample

This national analysis uses county-level data. We excluded counties with a Black population of less than 100, for which the dissimilarity index of Black-White residential segregation is not available (35% of counties). Based on the CDC US county-level vaccination data, Texas and Hawaii did not report county-level vaccination data for most of 2021; therefore, data from these states were not included. A total of 1916 counties were included in the final analysis.

Key measures

Using the MSM NCRN Database, we used county-level vaccination rates (derived from CDC data) as the primary outcome of interest. The rates were expressed as the percentage of the total county population fully vaccinated (having the second dose of a 2-dose vaccine or 1 dose of a single-dose vaccine). We used an index of dissimilarity to measure Black-White residential segregation in this study.²⁸ The dissimilarity index is widely used in research to measure segregation within a population. In addition, the dissimilarity index is a commonly used proxy for structural racism, given that it represents compositional distribution of groups across a population. It quantifies segregation as the degree of deviation from a random residential distribution of 2 groups within a given geographic area. The values range from 0 (total integration) to 100 (total segregation) and can be interpreted as the percentage of one group that would have to be relocated to attain the same spatial dispersion as the second group.²⁸

We categorized covariates in 3 different groupings hypothesized to affect COVID-19 vaccine uptake: (1) sociodemographic, (2) health systems, and (3) COVID-19 outcomes. The sociodemographic variables include area-level factors associated with populations disproportionately impacted by the COVID-19 pandemic (eg, population older than 65 years, household income, geographic region). Health systems factors (eg, county numbers of federally qualified health centers [FQHCs] and hospitals beds) were included to account for differential access to health care resources, which can impact the relationship between residential segregation and vaccine rates. Finally, area-level COVID-19 variables included cumulative deaths, cases, and hospitalizations per capita. COVID-19 outcomes may be associated with community-level vaccination rates—for example, residents of communities with higher rates of infection may be more likely to seek out vaccinations. There is some evidence to support this; those who knew family members or friends affected by COVID-19 have been shown to be less likely to refuse vaccination than those who did not.²⁹ Further details on our model variables are provided in Table 1.

TABLE 1 - List of Variables and Data Sources

Variable (All at County Level)	Definition	Source
Dependent variable
Proportion fully vaccinated	Proportion of county residents that have received either (1) the second dose of a 2-vaccine series or (2) the first dose of a single-vaccine series	MSM NCRN (CDC)
Independent variable of interest
White/Black segregation (dissimilarity index)	Demographic measure of the evenness with which 2 groups are distributed across the component geographic areas (census tracts) that make up a larger area (counties)	County Health Rankings
Covariates
Sociodemographic
Proportion Black	Proportion of Black residents in a county (per census data)	MSM NCRN ACS
Proportion strongly hesitant to vaccination	Proportion of residents reporting strong vaccine hesitancy	MSM NCRN (CDC)
Population density	People per square meter (ZCTA)	MSM NCRN (ACS)
Population age >65 percentile	County percentile ranking of the proportion of the population older than 65 years	MSM NCRN (ACS)
% Households with at least one housing problema	Percentage of households with at least one of 4 housing problems: overcrowding, high housing costs, lack of kitchen facilities, or lack of plumbing facilities	CHR
Gini Index	Measure of income inequality ranging from 0, indicating perfect equality, to 1, indicating perfect inequality Derived from observed cumulative income distribution	MSM NCRN (ACS)
Median household income	County median household income	MSM NCRN (ACS)
Rural-urban classification	Rural Urban Continuum Codes	MSM NCRN (USDA)
Health systems
Population to PCP ratio	Ratio of population to primary care physicians	CHR
# FQHCs	Count of FQHCs by county	AHRF
# Hospital beds	Count of hospital beds by county	AHRF
# SNFs	Count of SNFs by county	AHRF
COVID-19–related variables
Cumulative COVID-19 deaths per capita	County-level cumulative deaths attributed to COVID-19, per capita	MSM NCRN (USAFacts)
Cumulative COVID-19 cases per capita	County-level cumulative cases attributed to COVID-19, per capita	MSM NCRN (USAFacts)
Cumulative COVID-19 hospitalizations per capita	County-level cumulative hospitalizations attributed to COVID-19, per capita	MSM NCRN (USAFacts)

Abbreviations: ACS, American Community Survey; AHRF, Area Health Resources Files; CDC, Centers for Disease Control and Prevention; CHR, County Health Rankings; FQHC, federally qualified health center; MSM NCRN, Morehouse School of Medicine National COVID-19 Resiliency Network; PCP, primary care physician; SNF, skilled nursing facility; USDA, United States Department of Agriculture; ZCTA, zip code tabulation area.

^aOut of 4 potential problems: overcrowding; high housing costs; lack of kitchen facilities; lack of plumbing facilities.

Statistical approach

We ran multivariable generalized linear models (GLMs) with a binomial family and logit link and estimated the average marginal effects of segregation on vaccination rates. We included analytic weights in all models to account for differences in county population size. In the primary models, we included state-level fixed effects to account for differences across states in all models (eg, timing of stay-at-home orders, mask mandates, vaccine rollouts) and adjusted for various county-level sociodemographic, health system, and COVID-19–related covariates. We ran these primary analyses using the following: (1) the full pooled sample; (2) stratified subsample analyses of counties with high versus low proportions of Black residents; and (3) stratified subsample analyses of counties with high versus low dissimilarity indices to examine potential effect heterogeneity. In subsequent sensitivity analyses, we tested for effect heterogeneity across the county proportion of Black residents by adding its interaction with the dissimilarity index to the primary model specification. The inclusion of this interaction effect allowed us to assess whether the relationship between segregation and vaccination rates varies on the basis of the proportion of Black residents in the county.

To account for changes over time, our analyses were conducted using data from 3 time points spanning the initial rollout of COVID-19 vaccines in the United States: May 3, 2021; September 27, 2021; and December 6, 2021. On April 19, 2021, all adults nationwide were eligible for COVID-19 vaccines; on September 22, 2021, the US Food and Drug Administration started allowing booster doses of the vaccine; on November 21, 2021, the US Department of Health & Human Services Secretary issued a directive to expand eligibility for the vaccine for all adults 18 years and older to receive a booster dose of COVID-19 vaccines.³⁰ These dates indicated major milestones in the rollout of COVID-19 vaccines; the availability of booster doses signaled a new phase in the vaccination process. The dates in the analysis were chosen to reflect these milestones.

Results

The proportion of the population that was fully vaccinated increased over time, from 25% in May 2021 to 47% in December 2021 (Table 2). Notably, the proportion fully vaccinated was significantly higher in counties with above-median segregation than in counties with below-median segregation at all time points, that is, 50% versus 43%, respectively, by December 6 (Table 2). Counties with above-median segregation had a significantly lower proportion of Black residents (8% vs 19%), a significantly higher median household income ($56 825 vs $52 150), were significantly more likely to be urban (46% vs 36%), and had significantly more health care resources (eg, FQHCs, hospital beds), compared with counties with below-median segregation (Table 2).

TABLE 2 - Descriptive Statistics, Overall and by Low Versus High Segregation

Variable	All Counties, Mean (SD) or n (%)	Low Segregation (N = 945), Mean (SD) or n (%)	High Segregationa (N = 971), Mean (SD) or n (%)	P
Proportion fully vaccinated, mean (SD)
May 3	0.25 (0.1)	0.22 (0.1)	0.28 (0.08)	<.001
Sep 27	0.43 (0.13)	0.39 (0.13)	0.46 (0.12)	<.001
Dec 6	0.47 (0.13)	0.43 (0.13)	0.50 (0.12)	<.001
Dissimilarity index	0.46 (0.17)	0.32 (0.1)	0.59 (0.09)	<.001
Proportion Black residents	0.13 (0.17)	0.19 (0.18)	0.08 (0.13)	<.001
Proportion strongly hesitant to vaccines	0.09 (0.03)	0.1 (0.03)	0.09 (0.03)	<.001
Population density	0.0002 (0.0009)	0.0001 (0.0005)	0.0002 (0.0012)	.004
Population age >65 percentile	0.49 (0.16)	0.49 (0.17)	0.49 (0.16)	.81
Proportion with at least one housing problem	0.14 (0.04)	0.14 (0.04)	0.14 (0.04)	.53
Gini coefficient	0.45 (0.04)	0.45 (0.04)	0.45 (0.03)	.009
Median household income	54 520.25 (15 791.72)	52 150.42 (16 682.74)	56 826.62 (14 514.81)	<.001
Rural-urban classification
Rural	330 (17%)	196 (21%)	134 (14%)	<.001
Suburban	796 (42%)	409 (43%)	387 (40%)
Urban	790 (41%)	340 (36%)	450 (46%)
Cumulative COVID-19 deaths per capita	0.002 (0.002)	0.002 (0.002)	0.002 (0.001)	<.001
Cumulative COVID-19 cases per capita	0.102 (0.08)	0.105 (0.107)	0.1 (0.04)	.18
Cumulative COVID-19 hospitalizations per capita	0.005 (0.007)	0.005 (0.008)	0.005 (0.006)	.97
Population to PCP ratio	2 533.98 (2 106.27)	3 029.26 (2 547.56)	2 068.34 (1 436.11)	<.001
# Federally qualified health centers	4.41 (11.07)	2.85 (5.81)	5.93 (14.3)	<.001
# Hospital beds	4 21.4 (1 133.94)	240.76 (655.06)	597.2 (1 434.63)	<.001
# Skilled nursing facilities	6.46 (13.48)	3.97 (6.49)	8.89 (17.48)	<.001

Abbreviations: PCP, primary care provider; SD, standard deviation.

^aDefined as having a dissimilarity index below versus above the sample median (0.46).

Our GLMs also suggest an overall positive association between county-level dissimilarity indices and the proportion population fully vaccinated. Specifically, when only controlling for state fixed effects, full segregation (dissimilarity index = 1) was significantly associated with a 6.8 percentage point (95% CI, 2.8-10.7), 11.3 percentage point (95% CI, 4.4-18.1), and 12.8 percentage point (95% CI, 5.6-20) increase in the proportion fully vaccinated compared with full integration (dissimilarity index = 0) by May 3, September 27, and December 6, respectively (Table 3). Effects were muted after adjustment for sociodemographic variables, with minimal change after further adjusting for health systems and COVID-19–related variables. Notably, after adjustment for covariates (Table 3, models (2), (3), (4)), full segregation was only consistently significantly associated with the proportion population fully vaccinated on May 3 compared with full integration. Finally, we note that in fully adjusted models, there was less than 3 percentage point difference in the estimated percent fully vaccinated between counties with dissimilarity indices of 0.25 and 0.75 (Table 3).

TABLE 3 - Effects of County-Level Segregation on the Proportion Population Fully Vaccinated

	(1) State Fixed Effects, AME (95% CI)	(2) Model (1) + Sociodemographic Variables, AME (95% CI)	(3) Model (2) + Health Systems Variables, AME (95% CI)	(4) Model (3) + COVID-19 Variables, AME (95% CI)	(5) Model (4) + Interaction: Dissimilarity Index and Proportion Black Residents
May 3, 2021	0.068a (0.028, 0.107)	0.055a (0.028, 0.082)	0.048a (0.021, 0.076)	0.054a (0.025, 0.082)	0.060a (0.033, 0.088)
Sep 27, 2021	0.113a (0.044, 0.181)	0.048b (0.001, 0.095)	0.042 (−0.008, 0.093)	0.049 (−0.004, 0.102)	0.062b (0.008, 0.116)
Dec 6, 2021	0.128a (0.056, 0.200)	0.049 (−0.000, 0.099)	0.045 (−0.007, 0.096)	0.056b (0.002, 0.109)	0.069b (0.015, 0.122)

Abbreviations: AME, average marginal effect; CI, confidence interval.

^aP < .01.

^bP < .05.

Analyses stratified by the county proportion of Black residents (above vs below median [13.4%]) and stratified by segregation (above vs below median [0.46]) generally did not yield statistically significant marginal effects (see Supplemental Digital Content Table 1, available at https://links.lww.com/JPHMP/B157), likely a result of reduced statistical power. However, in analyses including an interaction term between the county proportion of Black residents and the county dissimilarity index, we find evidence to suggest that the association between segregation and vaccination is positive in counties with a lower proportion of Black residents (ie, 5% as shown in the Figure) but negative in counties with the highest proportions of Black residents (ie, 70% as shown in the Figure).

FIGURE: Predictive Margins for County-Level Vaccination Rates Across Levels of Segregation, May 3 and December 6 (Model 6)This figure is available in color online (www.JPHMP.com).

Discussion

These results add to the evidence base for informing policies around prioritization of vaccine distribution and outreach. Overall, our analyses suggest a small, but significant positive association between segregation and vaccination rates across counties in the United States. However, models including an interaction term between the county proportion of Black residents and the dissimilarity index suggest that the relationship between these 2 variables differs on the basis of the proportion of Black residents in a county. Specifically, residential segregation was found to be associated with lower levels of vaccination in counties with the highest proportions of Black residents. Conversely, segregation is associated with higher vaccination rates in counties with the lowest proportions of Black residents. This demonstrates that vaccination rates are still relatively lower in areas with a higher proportion of Black residents, although these may classify as “low segregation.” For policy makers and providers designing equitable vaccination interventions, these findings suggest that residential segregation and the proportion of Black residents constitute one area-level indicator that can help guide the distribution of limited resources.

Vaccinations among non-Black residents may be driving the positive association between the dissimilarity index and the proportion fully vaccinated in predominantly White counties. The predominantly White counties may be the “high segregation” counties with relatively higher vaccination rates. Because of lack of data on race-ethnicity–specific vaccination rates at the county level, we are unable to examine the disparity in vaccination rates across Black versus White residents in high dissimilarity index counties. However, in the absence of these data, our results show more segregated counties that are predominantly White have higher vaccination rates and more segregated counties that are predominantly Black are disadvantaged, having lower vaccination rates, even after accounting for differences in the social determinants of health. Therefore, the classification of a “high segregation” or “low segregation” county must be paired with information about the proportion of White and Black residents to understand the true effect on vaccination rates.

Given the institutional and individual practices and policies created to maintain the physical separation of communities of color from White communities, residential segregation is often used as a measure to assess one dimension of structural racism.³¹ This analysis provides a starting point for understanding the association between this aspect of structural racism and COVID-19 vaccination, but these findings also suggest that factors such as education, income inequality, and rurality are important mediators in the relationship between residential segregation and COVID-19 vaccination.

Our study highlights the importance of analyzing differences in county racial composition to assess the true effects in “low segregation” and “high segregation” counties. In addition, while our analyses showed an overall positive association between higher levels of segregation and higher vaccination rates, we find that much of this association is attenuated after controlling for sociodemographic variables. This is consistent with previous findings that the effect of racial residential segregation on COVID-19 outcomes is exacerbated by both economic and educational factors.^12,32 These results emphasize the importance of study design considerations and methodological decisions in analyses when assessing racial equity in COVID-19 vaccinations. Because of the complex and multifaceted relationships between residential segregation and health outcomes such as vaccination rates, the impact of mediating and moderating factors, interaction effects, and stratified analyses can significantly affect findings.

While introducing sociodemographic mediating variables into our models attenuated the relationship between racial segregation and vaccination rates, these findings do not suggest that the effects of segregation can be “explained away” by sociodemographic factors. Rather, our findings regarding the mediating effects of sociodemographic variables align with recent research showing that because structural racism is multidimensional, multidimensional measures may better capture its complex effects.³³ For example, the lack of ability to examine the disparity in vaccination rates across Black versus White residents in high dissimilarity index counties shows that a more multidimensional measure that can capture these complexities and differences in the proportion of residents who make them “low segregation” counties and “high segregation” counties is needed.

This study has a number of limitations. We conducted ecological (county-level) analyses, which introduces the potential for unmeasured confounding variables. Texas and Hawaii are excluded from this analysis because of a lack of vaccination data during the study time frame. The dissimilarity index measures residential segregation at the county level and does not capture more nuanced information about patterns of segregation within counties (eg, at the neighborhood level) and does not account for the presence of other racial/ethnic groups. However, the dissimilarity index is a commonly used measure of residential segregation, publicly available to researchers, that allows us to describe the relationship between county-level segregation and overall vaccination rates. Finally, without race-specific vaccine data, we cannot assess whether higher community-level vaccination rates accrue equally to both White and Black populations. More granular race-ethnicity–specific data are needed, and future research should explore racial and ethnic disparities in vaccination rates. The MSM NCRN Database used for these analyses aggregates a wide variety of data sources that are currently publicly available and can be used to further explore more nuanced relationships between race, ethnicity, and COVID-19 outcomes.

Implications for Policy & Practice

• These findings highlight the importance of considering the proportion of Black residents in segregated areas to continue mitigating disparities; areas with high rates of residential racial segregation and high proportions of Black residents may merit prioritization for vaccine access and outreach campaigns.¹⁶
• The nuance of these results also highlights the importance of methodological decisions when modeling disparities in COVID-19 vaccinations—researchers should consider the impact of mediating and moderating factors and examine interaction effects and stratified analyses taking racial group distributions into account when assessing the relationship between residential segregation and health outcomes such as vaccination rates.