Population-Level Disparities in COVID-19: Measuring the Independent Association of the Proportion of Black Population on COVID-19 Cases and Deaths in US Counties

Abstract

Context: 

There is a need to understand population race and ethnicity disparities in the context of sociodemographic risk factors in the US experience of the COVID-19 pandemic.

Objective: 

Determine the association between county-level proportion of non-Hispanic Black (NHB) on county COVID-19 case and death rates and observe how this association was influenced by county sociodemographic and health care infrastructure characteristics.

Design and Setting: 

This was an ecologic analysis of US counties as of September 20, 2020, that employed stepwise construction of linear and negative binomial regression models. The primary independent variable was the proportion of NHB population in the county. Covariates included county demographic composition, proportion uninsured, proportion living in crowded households, proportion living in poverty, population density, state testing rate, Primary Care Health Professional Shortage Area status, and hospital beds per 1000 population.

Main Outcome Measures: 

Outcomes were exponentiated COVID-19 cases per 100 000 population and COVID-19 deaths per 100 000 population. We produced county-level maps of the measures of interest.

Results: 

In total, 3044 of 3142 US counties were included. Bivariate relationships between the proportion of NHB in a county and county COVID-19 case (Exp β = 1.026; 95% confidence interval [CI], 1.024-1.028; P < .001) and death rates (rate ratio [RR] = 1.032; 95% CI, 1.029-1.035; P < .001) were not attenuated in fully adjusted models. The adjusted association between the proportion of NHB population in a county and county COVID-19 case was Exp β = 1.025 (95% CI, 1.023-1.027; P < .001) and the association with county death rates was RR = 1.034 (95% CI, 1.031-1.038; P < .001).

Conclusions: 

The proportion of NHB people in a county was positively associated with county COVID-19 case and death rates and did not change in models that accounted for other socioecologic and health care infrastructure characteristics that have been hypothesized to account for the disproportionate impact of COVID-19 on racial and ethnic minority populations. Results can inform efforts to mitigate the impact of structural racism of COVID-19.

Data that include race and ethnicity in reporting of COVID-19 cases and deaths have been incomplete and inconsistently reported across states. Among places and populations where race and ethnicity COVID-19 data are available, there are stark racial and ethnic disparities.^1–4 In particular, alongside American Indian, Alaska Native, Pacific Islander, and Hispanic/Latino(x) populations, non-Hispanic Black (NHB) populations bear disproportionate burdens of COVID-19 cases and deaths compared with non-Hispanic White populations.^5–7 Discussion of the factors that contribute to these disparities has focused on the presence of other structural inequities in these communities, including disproportionate employment in high-exposure fields, disproportionate burden of underlying chronic health conditions, disproportionate burdens of poverty, inadequate access to health care, and higher rates of uninsured status.^5,8,9 However, there is strong evidence that health disparities experienced by communities of color exist due to structural racism and discrimination, independent of the risk conferred by socioeconomic disadvantage and other social determinants of health.^10–12

Our objective for this ecological analysis was to understand the independent ecologic factors associated with the geographic variation in the burden of COVID-19 cases and deaths in the pandemic. Specifically, we wanted to understand the interplay between racial and ethnic composition, social determinants of health, rurality, and health care access at the ecologic level. This study was designed to identify the extent to which racial composition of a population in a place, independent of other social and health care infrastructure factors, contributes to COVID-19 disparities.¹³ US counties with higher proportions of NHB populations have fewer health-promoting resources.^14–16 We put forth the proportion of NHB persons as a measure of population-level disparities and operationalize this as a continuous measure. Other studies have used this measure but have dichotomized the measure into counties with high versus low NHB populations⁴ or looked at this in relationship to county case rates but not deaths.¹⁷ Our choice to present this as a continuous measure reinforces the linear relationship this measure has on outcomes, which strengthens its predictive potential.

Epidemiologic studies have described racial and ethnic disparities in COVID-19 disease burden in the context of other sociodemographic factors such as poverty, rurality, and health system infrastructure.^18–21 On the basis of exploratory analyses at the state level,^22,23 we hypothesized that the proportion of NHB persons in a county would be associated with higher rates of COVID-19 cases and deaths and that this association would be unchanged after accounting for other high-risk socioecologic factors, demographic composition, and health care system capacity. We conducted this study to test this hypothesis and to put forth the proportion of NHB persons in a county as an independent measure of population-level race disparities in COVID-19.

Methods

Inclusion criteria and data sources

Counties were included in this study if they met the following criteria: (1) 50 US states or District of Columbia, and (2) had at least 10 COVID-19 cases as of September 20, 2020. We obtained county-level COVID-19 cases and deaths on September 20, 2020, from the USAFacts COVID-19 Web site²⁴ and 2019 county-level population estimates from US Census Bureau.²⁵ Primary Care Health Professional Shortage Area (PCHPSA) designation in 2020 and the number of hospital beds in 2017 were obtained from the Area Health Resource File of the Health Resources and Services Administration.²⁶ PCHPSA is a measure defined by the Health Resources and Services Administration that incorporates the population to provider ratio, the number of people living below the federal poverty level, measures of infant mortality and low birth weight, and the nearest health care location outside of the designated area.²⁷ State-level COVID-19 testing rates by September 20, 2020, were obtained from Johns Hopkins Coronavirus Resource Center.²⁸ Sociodemographic characteristics were obtained from 2018 estimates of 2014-2018 American Community Survey Data.²⁹

Outcome variables

There were 2 main outcomes in this study. COVID-19 case rate was calculated and presented as the number of cases per 100 000 people, and COVID-19 death rate was calculated and presented as the number of deaths per 100 000 people.

County-level characteristics

The primary independent variable was the percentage of NHB people residing in a county. Sociodemographic covariates included the percentage of Asian/Pacific Islander (A/PI) people, the percentage of Hispanic people, the percentage of American Indian/Alaska Natives (AI/AN), the percentage of crowded households (defined as the proportion of households in the county with more than 1 person per room, excluding bathrooms and kitchens), the percentage of uninsured people, the percentage of people living under federal poverty level, the percentage of people 65 years and older, population density presented as the number of people per square mile in the counties, hospital beds per 1000 people, and state-level testing rate, which are all continuous variables, and PCHPSA designation, which is the only categorical variable. Independent variable and covariates were explored on the basis of findings from previously published literature.

Statistical analysis

Descriptive statistics were calculated for outcome variables and independent variables with mean and standard deviation or number and proportion reported. Since distribution of COVID-19 case rate was highly right-skewed, we log-transformed it before fitting the linear models. Assumptions of multiple linear regression were checked, and no violations were found. Then, we performed unadjusted and adjusted linear regression using log-transformed COVID-19 case rate as outcome. We reported exponentiated coefficients, exponentiated 95% confidence intervals (CIs), and P values for each independent variable. Using COVID-19 death numbers as outcome variable and total population in the county as offset, we conducted unadjusted and adjusted negative binomial regression (a negative binomial model was selected over a Poisson model due to the overdispersion of death data), with rate ratios, 95% CIs, and P values reported for each independent variable. With either case rate or death rate as the outcome, 3 regression models were constructed: model 1 has only percentage of NHB people as the predictor; model 2 was based on model 1 with sociodemographic measures added; model 3 was based on model 2 with hospital beds per 1000 people, PCHPSA designation, and state-level testing rate added. We conducted sensitivity analyses to examine the impact of the proportion of NHB in high and low poverty counties and high and low population density counties for both outcome variables. The association between the proportion of NHB and the outcome variables was similar in both high and low poverty and high and low population density counties. We checked for multicollinearity by examining variance inflation factors (VIFs); no variables had a VIF greater than 3, ensuring there was not considerable multicollinearity. All P values were 2-sided, and a P value less than .05 was considered statistically significant. SAS, version 9.4, was used to perform all analyses. We also used ArcGIS Pro (Esri, Redlands, California) to create choropleth maps for the 2 outcomes and proportion of NHB population. The categories on the maps for COVID-19 case rate and death rate were based on quintile groups; the categories on the map for proportion of Black population were based on Jenks Natural Breaks Classification. This study was a secondary data analysis of publicly available data.

Results

There were 3044 of 3143 US counties that had at least 10 confirmed COVID-19 cases included. The average number of confirmed COVID-19 cases per 100 000 population was 1785.3 (SD = 1392.2). The mean COVID-19 deaths per 100 000 population was 41.0 (SD = 51.5). Descriptive statistics for the sample of counties are presented in Table 1.

TABLE 1 - Description of Sociodemographic Characteristics, Health Care Accessibility, and Confirmed COVID-19 Cases and Deaths in US Counties (N = 3044)

	Mean	SD
COVID-19 case rate, n/100 000	1 785.3	1 392.2
log (COVID-19 case rate)	7.2	0.8
Deaths, n	64.9	326.6
Deaths per 100 000	41.0	51.5
Non-Hispanic Black, %	9.3	14.7
Asian/Pacific Islander, %	1.5	2.9
Hispanic, %	9.3	13.7
American Indian/Alaska Native, %	1.9	7.6
Crowded households, %	2.4	2.4
Uninsured, %	10.1	5.1
Poverty, %	15.7	6.5
65 y or older, %	18.2	4.4
Population density, no. per square mile	281.6	1 814.2
Hospital beds per 1000	3.0	4.9
State-level testing rate, n/100 000	28 706.6	11 027.5
	n	%
PCHPSA Yes	709	23.3

Abbreviation: PCHPSA, Primary Care Health Professional Shortage Area.

County proportion of NHB people was significantly and positively associated with confirmed COVID-19 case rates, where a 1% increase in the proportion of NHB was associated with a 2.6% increase in the case rate (Table 2). The unadjusted model association (Exp β = 1.026; 95% CI, 1.024-1.028; P < .001) did not substantially change after adjustment for sociodemographic factors (Exp β = 1.022; 95% CI, 1.021-1.024; P < .001) and the addition of health care system factors (Exp β = 1.022). In model 2, where sociodemographic characteristics were added, percent Hispanic (Exp β = 1.008; 95% CI, 1.006-1.010; P < .001), proportion of crowded households (Exp β = 1.017; 95% CI, 1.001-1.033; P = .04), proportion of people 65 years or older (Exp β = .965; 95% CI, 0.959-0.970; P < .001), and proportion uninsured (Exp β = 1.024; 95% CI, 1.018-1.030; P < .001) were associated with county case rate. In model 2, percent A/PI (Exp β = .985; 95% CI, 0.977-0.994; P = .001), and percent AI/AN (Exp β = .995; 95% CI, 0.991-0.999; P = .01) were associated with a lower county case rate. These factors remained significant and minimally changed in model 3, which added health care system factors. In the final model 3, a 1% increase in the proportion of Hispanic people was associated with 0.8% increase in case rate (Exp β = 1.008; 95% CI, 1.006-1.010; P < .001), a 1% increase in A/PI people was associated with a 1.5% decrease in the case rate (Exp β = .985; 95% CI, 0.977-0.994; P = .001), a 1% increase in AI/AN people was associated with a 0.7% decrease in case rate (Exp β = .993; 95% CI, 0.989-0.997; P = .002), a 1% increase in proportion living in crowded households was associated with a 1.6% increase in case rate (Exp β = 1.016; 95% CI, 1.001-1.032; P = .048), a 1% increase in the proportion of people 65 years or older was associated with a 3.6% decrease in case rate (Exp β = .964; 95% CI, 0.959-0.970; P < .001), and a 1% increase in the proportion of uninsured was associated with a 2.7% increase in case rate (Exp β = 1.027; 95% CI, 1.020-1.033; P < .001). Model 3 also had a significant association for state-level testing rate (Exp β = 1.042; 95% CI, 0.020-1.064; P < .001).

TABLE 2 - Linear Regression Models of the Association Between Sociodemographic Characteristics, Health Care Accessibility, and Confirmed COVID-19 Case Rates in US Counties (N = 3044)^a

	Model 1			Model 2			Model 3
	Exp β	Exp 95% CI	P	Exp β	Exp 95% CI	P	Exp β	Exp 95% CI	P
Non-Hispanic Black, %	1.026	1.024-1.028	<.001	1.022	1.021-1.024	<.001	1.022	1.020-1.024	<.001
Asian/Pacific Islander, %				0.985	0.977-0.994	.001	0.985	0.977-0.994	.001
Hispanic, %				1.008	1.006-1.010	<.001	1.008	1.006-1.010	<.001
AI/AN, %				0.995	0.991-0.999	.01	0.993	0.989-0.997	.002
Crowded households, %				1.017	1.001-1.033	.04	1.016	1.001-1.032	.05
Uninsured, %				1.024	1.018-1.030	<.001	1.027	1.020-1.033	<.001
Poverty, %				1.000	0.996-1.005	.87	1.000	0.996-1.005	.98
65 y or older, %				0.965	0.959-0.970	<.001	0.964	0.959-0.970	<.001
Population density (unit = 100 per square mile)				1.001	0.999-1.002	.55	1.000	0.999-1.001	.78
Hospital beds per 1000							1.006	1.002-1.011	.007
PCHPSA Yes vs No							1.006	0.952-1.063	.83
State-level testing rate, n/10							1.042	1.020-1.064	<.001

Abbreviations: AI/AN, American Indian/Alaska Native; CI, confidence interval; PCHPSA, Primary Care Health Professional Shortage Area.

^aCoefficients and 95% CI of linear models were exponentiated for ease of interpretation.

The proportion of NHB people was significantly and positively associated with COVID-19 death rates in the unadjusted model (RR = 1.032; 95% CI, 1.029-1.035; P < .001) (Table 3). Models that included sociodemographic and health system factors did not substantially change the association of NHB % with death rates. In the final model, a 1% change in NHB % was associated with a 3.5% change (RR = 1.035; 95%CI, 1.031-1.038; P < .001) in death rates. The proportion of Hispanic people (RR = 1.018; 95% CI, 1.014-1.022; P < .001), proportion of AI/AN people (RR = 1.011; 95% CI, 1.004-1.018; P = .002), and population density (RR = 1.005; 95% CI, 1.002-1.009; P = .003) were found to be significant when added in model 2. In the final model, these factors remained significant, with a 1% increase in the proportion of Hispanic people (odds ratio [OR] = 1.018; 95% CI, 1.014-1.022; P < .001) associated with a 1.8% increase in death rate, a 1% increase in the proportion of AI/AN people (OR = 1.010; 95% CI, 1.003-1.016; P = .006) associated with a 1.0% increase in death rate, and a 100 people per square mile increase in population density (RR = 1.005; 95% CI, 1.001-1.008; P = .007) associated with a 0.5% increase in death rate. State testing rate (RR = 1.080; 95% CI, 1.040-1.122; P < .001) was also associated with an increase in death rate.

TABLE 3 - Negative Binomial Regression Models of the Association Between Sociodemographic Characteristics, Health Care Accessibility, and Confirmed COVID-19 Death Rates in US Counties (N = 3044)

	Model 1			Model 2			Model 3
	Rate ratio	95% CI	P	Rate Ratio	95% CI	P	Rate Ratio	95% CI	P
Non-Hispanic Black, %	1.032	1.029-1.035	<.001	1.035	1.032-1.039	<.001	1.035	1.031-1.038	<.001
Asian/Pacific Islander, %				1.005	0.988-1.022	.59	1.003	0.986-1.020	.74
Hispanic, %				1.018	1.014-1.022	<.001	1.018	1.014-1.022	<.001
AI/AN, %				1.011	1.004-1.018	.002	1.010	1.003-1.016	.006
Crowded households, %				1.002	0.971-1.035	.88	0.999	0.968-1.032	.97
Uninsured, %				1.003	0.993-1.014	.54	1.009	0.998-1.020	.12
Poverty, %				1.004	0.996-1.012	.35	1.003	0.995-1.011	.46
65 y or older, %				1.008	0.998-1.018	.13	1.009	0.999-1.019	.09
Population density (unit = 100 per square mile)				1.005	1.002-1.009	.003	1.005	1.001-1.008	.007
Hospital beds per 1000							0.997	0.987-1.006	.44
PCHPSA Yes vs No							1.004	0.910-1.106	.94
State-level testing rate, n/10							1.080	1.040-1.122	<.001

Abbreviations: AI/AN, American Indian/Alaska Native; CI, confidence interval; PCHPSA, Primary Care Health Professional Shortage Area.

Maps of the county COVID-19 case and death rates are presented side by side in Figure 1. The spatial pattern of counties with high rates of COVID-19 deaths is similar to counties that have high rates of COVID-19 cases. Figure 2 shows the spatial distribution of the county proportion of NHB people. The spatial pattern of counties with higher COVID-19 case and death rate counties overlaps the geographic areas of the country with large populations of NHB people. COVID-19 case and death rate patterns are visually similar to patterns of high rates of uninsured status and poverty; these overlapping patterns were concentrated in the South and Southeast parts of the country (see Supplemental Digital Content Appendix A, available at https://links.lww.com/JPHMP/A784, for maps of covariates).

FIGURE 1: Maps of Confirmed COVID-19 Case Rates and Death Rates per 100 000 persons in US Counties with 10 or more Confirmed Cases as of September 20, 2020; n = 3044

FIGURE 2: County Map of Proportion Non-Hispanic Black Population in US Counties, 2018 American Community Survey Estimates

Discussion

This national ecologic analysis examined the association for the proportion of NHB people in a county, as a measure of population-level disparities, on county COVID-19 case and death rates. We found that this measure was associated with county COVID-19 case and death rates in a sample that included 97% of US counties (n = 3044). The association between county proportion of NHB people and county COVID-19 case and death rates did not change in models that accounted for other socioecologic characteristics and health care infrastructure characteristics that have been hypothesized to partially account for the disproportionate impact of COVID-19 on minority populations. These results can be used to quantify population disparities in COVID-19 on NHB populations and attempt to measure the impact of structural racism on county COVID-19 case and death rates after accounting for other social and environmental factors that also contribute to the geographic variation in COVID-19 burden in the United States. We found that county access to primary care and hospital care was not significantly associated with county COVID-19 case and death rates; state-level testing rate was the only health system variable positively associated with county COVID-19 case and death rates. In fact, testing rate was found to be the strongest of all the factors included in our study. We found the proportion of Hispanic population in a county, household crowding, and the proportion of uninsured were positively and significantly associated with county COVID-19 case rates. The proportion of Hispanic population in a county, the proportion of AI/AN population in a county, and population density were positively and significantly associated with county COVID-19 death rates.

These results also attempt to quantify the potential impact of structural racism and discrimination, independent of other social and environmental inequities, associated with high COVID-19 burden from a population health perspective. Structural racism encompasses discrimination embedded and reinforced across all sectors, including housing, residential segregation, health care, education, employment, banking and credit, and the justice system.³⁰ Ecologic studies are limited in their ability to measure structural racism, but this study attempts to tease apart the effect of common social, geographic, and health care factors associated with health inequities from demographic composition as an independent contributor to these inequities.¹³ These results are important because they can guide targeted efforts to combat structural racism and provide a baseline comparison to measure the impact of such efforts that take place in the future.¹² In addition, this is the only national county-level study that has examined the proportion of NHB people in a county as a continuous variable. The positive independent effects of the proportion of NHB population in a county after models that include demographic, socioecologic, and health care infrastructure variables on COVID-19 disease burden strengthen the argument that there are circumstances outside of socioeconomic and health care infrastructure factors that may be contributing to the observed geographic disparities in COVID-19 in the United States. Communities of color have been chronically underresourced with respect to health-promoting resources^14–16; structural inequities in the distribution of health-promoting resources in counties with higher proportions of NHB people may be contributing to these results. These findings demonstrate the need for nuanced, population-level analyses to guide resource allocation and decision making to mitigate the impacts of COVID-19 on racial and ethnic minority communities, especially in the midst of a national COVID-19 vaccine rollout. The racial, ethnic, and geographic disparities observed here and by other studies may be attributable to structural racism and historic and current underinvestment in minority and rural communities. The COVID-19 pandemic did not create these disparate outcomes but is magnifying the disparities in health outcomes historically experienced by these communities.

Finding state testing rates to be associated strongly with the outcomes was not surprising in that one would expect in counties within states with higher testing rates, more cases as well as COVID-19 deaths would be identified. In other words, counties within states with low testing rates may be missing cases and misclassifying deaths, causing them to have lower than actual rates for both outcomes represented in the data. Neither access to primary care nor access to hospital care was associated with rates of COVID-19 cases and deaths. More access to hospital care could result in more testing and higher rates of COVID-19, while better access to primary care could result in better management of underlying chronic conditions associated with higher risk for COVID-19 death. However, these hypotheses were not supported by the findings. While health care access measures represent what services are available to a population, they do not necessarily reflect care that is accessible, welcoming, and culturally competent. In addition, there were disruptions in access to care during the pandemic due to closures or changes in hours and procedures. The measure of access to primary care, PCHPSA designation, may not be accurate, given potential disruption in access to care due to practice closure and strain on practices that have occurred as a result of the pandemic after these designations were assigned. The magnitude of financial stress inflicted by the COVID-19 pandemic on small- and medium-sized primary care practices has been particularly severe. Basu et al³¹ used microsimulation modeling to estimate the financial impact of COVID-19 on a practice and found that practices could lose approximately $67 774 per full-time physician in anticipated 2020 revenue. They conclude that this level of financial strain may precipitate practice closure and worsen access to health care across the country.³¹ In addition, the Larry Green Center at the University of Colorado and the Primary Care Collaborative have been conducting weekly surveys of primary care physicians across the country that also show that primary care practices are experiencing high levels of financial stress.³² Further work is needed to better understand the ways in which strain on primary care practices impacted access to care for communities during the pandemic.

These findings should drive public health decision making and actions to mitigate the impact of COVID-19 on communities of color. Importantly, targeted efforts to address structural racism within the context of COVID-19 should be driven by communities.^33,34 State and local health departments should actively engage NHB, Hispanic, and AI/AN communities to understand their needs and priorities. Community engagement may involve inclusion of community members in advisory roles, on health equity working groups, and in the public health workforce. Based on these findings, state and local health departments could prioritize essential workers in fields with dispro-portionate minority representation for early vaccination. Public health authorities may find that mobile and/or work-site vaccination programs could be more effective in reaching minority communities. However, there may be work needed to address vaccine hesitancy and mistrust within these communities first, which will come to light through community engagement. Policies are also critical to reducing COVID-19 racial and ethnic disparities.³⁵ Evidence is showing that mask mandates, paid sick leave policies, and eviction moratoria slow the spread.^36–39 Public policies directly affect some of the factors included in this study, such as crowded households and uninsured status, which were found to be positively associated with COVID-19 cases. However, further research is needed to understand the role of policy-driven outcomes such as residential segregation on COVID-19 disparities.

This study confirms several studies that have found relationships between COVID-19 burden and race inequity. The Centers for Disease Control and Prevention's Morbidity and Mortality Weekly Report (MMWR) found that NHB populations may be disproportionately impacted by COVID-19, with NHB patients making up 18% of the catchment area population but 33.1% of COVID-19 hospitalizations.¹ Another MMWR report found that 80% of patients hospitalized with COVID-19 in metropolitan Atlanta were Black.⁴⁰ A Commonwealth Fund report early in the pandemic found that, nationally, counties with higher than average proportion of Black population also had higher rates of COVID-19 deaths.^4,41 Gross and colleagues⁴² found an 18-fold higher risk of COVID-19 death among Black populations than among White populations in Wisconsin. Similarly, they found an 88% higher risk of death in Hispanic/Latino(x) populations than in White populations. Another national, county-level ecologic study by Millett et al⁴ compared COVID-19 cases and deaths among counties with greater than and less than 13% NHB populations and found that high NHB population counties had higher COVID-19 case rates and death rates after accounting for other sociodemographic, chronic disease burden, and environmental health variables. Yang et al¹⁷ noted the proportion of NHB was associated with increased COVID-19 case rates in US counties and that this relationship was moderated by racial segregation. A recent article also found that the potential years of life lost among people who have died of COVID-19 is much greater in NHB populations than in White populations.⁴³ Our study builds upon these findings by showing that the association between this measure and both COVID-19 case and death rates in US counties is not attenuated by other social and health care access factors that are also influenced by structural inequity. In addition, our study operationalized percent NHB as a continuous variable and is the first study to account for testing rates, primary care access, and hospital access in the same model.

Recently, there have been several studies using individual electronic health record data to examine race and ethnic disparities in COVID-19 hospitalization, intensive care stay, and mortality.^44–46 These studies found that after adjusting for socioeconomic and health conditions, there was not a higher risk of in-hospital mortality or critical illness among Black patients. However, these studies noted that neighborhood environmental factors and delays in care may have resulted in the higher out-of-hospital deaths for Black patients in these places. It is important to note that these studies note overrepresentation of COVID-19 on Black patients in their health systems, which they assert are attributable to structural inequities in exposure to COVID-19 in high-risk fields and the burden of chronic conditions among minority populations compared with White populations.

This work has several limitations. First, this study was limited by a lack of nationally available, consistent set of race and ethnicity fields in both aggregated and individual-level COVID-19 case and death data. Ideally, we would be able to assess county variation in racial and ethnic health disparities in COVID-19 and study the impact of structural racism on those disparities. Granular and complete race ethnicity data would also enable researchers to measure and understand the drivers of equitable COVID-19 case and death rate outcomes across racial and ethnic subgroups in US counties. There may also be subtle variations in how states report cases or deaths across the study period due to a lack of national data reporting standards for COVID-19 cases and deaths. Second, this work was completed cross-sectionally at a single point in time in late September 2020. This analysis should be updated moving forward from this point to understand how the proportion of NHB population in a county impacts COVID-19 outcome at future time points. Third, we use population density as a proxy for rural status and do not examine rural-urban classifications of counties. We did this because population density was a way to measure crowding in public spaces that might promote more opportunities for disease transmission, and we also accounted for household crowding in this analysis as a measure of crowding in private space. We note there are specific susceptibilities to COVID-19 in rural versus urban places that could have warranted using rural-urban categorization in the model but are reassured that Millet et al⁴ utilized these variables and had similar results.⁴⁷ Fourth, we did not account for the variation in the burden of chronic conditions in US counties, but no comprehensive data set of county-level disease burden across chronic conditions for the US population exists; small area estimates that are available already account for many of the covariates included in this model. Fifth, we present the proportion of NHB population in a county as measure of population health race disparities in models that account for other place-based factors, but we acknowledge that inequity has negative consequences for the health of all people.⁴⁸ As with all observational studies, it was not possible to adjust for all potential intermediary or confounding factors. It is possible factors were not unaccounted for that also stem from structural racism and may have attenuated the observed association of the proportion of NHB population with the outcomes if they had been included in models. When and if reliable race-specific COVID-19 data become available, it will be possible to assess how much the proportion of NHB population impacts COVID-19 across racial/ethnic groups. Finally, this model cannot reliably account for the variation in health behaviors that exists across US counties; a recent MMWR found that African Americans were more likely to socially distance than other racial and ethnic groups and rural populations were less likely to wear masks than urban populations.⁴⁹

This national ecologic study that examined the relationship between the proportion of NHB people in a county on COVID-19 case and death rates found that the positive association between this measure and COVID-19 case and death rates remained unchanged after accounting for the independent effects of sociodemographic factors and health care infrastructure characteristics of counties. This work can inform targeted efforts to counter the compounded effects of structural racism. The disproportionate effects of the COVID-19 pandemic on NHB and other minority communities further validate the need for advocacy for policies and programs that tackle structural racism and other contributory factors to health disparities. A community-engaged approach to dismantling structural inequities in health-promoting resources and investment in minority communities is one important consideration for local and national public health leaders to build trust and mitigate the impact of COVID-19 on disproportionately impacted places and populations.

Implications for Policy & Practice

• The population-level racial disparities in COVID-19 disease burden observed here and by other studies may be attributable to structural racism and historic and current underinvestment in minority and rural communities.
• The COVID-19 pandemic did not create these disparate outcomes but is magnifying the disparities in health outcomes historically experienced by these communities.
• To directly measure whether we are achieving equity in COVID-19 disease burden across places and populations, there is a need for high-quality race and ethnicity data not only for cases and deaths but also for other measures including testing and hospitalizations.
• There is a need for advocacy for policies and programs directed at mitigating the impact of structural racism broadly and specifically in public health practices and health systems.