Racial disparity in COVID-19 deaths
One of the many alarming features of COVID-19 is its tendency to magnify inequalities by hitting minorities disproportionately. African Americans, Latino/Hispanic, and Native American groups have suffered mortality rates that are in many cases much more than double the national average (see, for example, Oppel et al. 2020, Gross et al. 2020, Price-Haywood et al. 2020, and Romero and Healy 2020).
Researchers have two separate contributions to make on this issue. The first is to measure the size of these disparities. The second, likely the more important, is to identify their sources – particularly to try to identify socioeconomic sources that can be addressed by policy.1 This column will provide a sampling of work on both of those tasks.
The most complete data set for measuring COVID-19 racial disparities is that obtained by Oppel et al. (2020), through a Freedom of Information Act suit, of all individual COVID-19 data held by the US Centers for Disease Control and Prevention (CDC) up to 28 May 2020. Even this dataset is very limited: a majority of cases lacked either race or county, and the data are all from fewer than a third of US counties, representing 55% of the national population.
The per-capita COVID-19 case rate in the data for African Americans is 2.7 times that for whites, and 3.2 times for Latino/Hispanics. The ratios tend to be even greater within each age category. These ratios also vary greatly across the country. Gross et al. (2020) obtain age-corrected relative mortality ratios by state for the 28 states whose data is broken down by race. They report a range of African-American-to-white ratios from a sky-high 18 for Wisconsin down to 0.44 for Pennsylvania. Their aggregate estimate for the US is that COVID-19 mortality is 3.57-times greater for African Americans relative to whites, and 1.88-times greater for Hispanic/Latinos relative to whites.
We can thus take it that these disparities are well established in the data and large. Conceptually, one may classify their sources into three categories:
(i) factors that lead to higher rates of minorities becoming infected;
(ii) factors that lead minorities with the virus to require hospitalisation at higher rates; and
(iii) factors that lead to a higher death rate for minority patients once hospitalised.
Category (i) could include the concentration of workers in essential services that cannot be done from home, requiring workers to go to a physical workplace where they will be exposed to other workers or customers (Hooper et al. 2020); differential availability of paid sick leave; differences in reliance on public transit (Harris 2020); differences in access to healthcare insurance;2 and differences in residential density.3
Categories (ii) and (iii) could include differential incidence of pre-existing conditions that can make a COVID-19 infection more dangerous4 or discrimination in treatment by healthcare institutions.5 Environmental racism6 may contribute to all three categories.7
The most direct approach to untangling these three categories is employed by Price-Haywood et al. (2020), who obtained confidential data from a large hospital system in Louisiana on 3,481 patients who tested positive for COVID-19 between 1 March and 11 April 2020. For each patient, race, zip code, and medical record were available to the researchers. In total, 70.4% of those who tested positive were Black, while only 31% of the service-area population was Black. About 40% of these patients were hospitalised due to the virus, with 77% of those cases Black; about 9% died in hospital, of which 70.6% were Black.
Since the disparity in death rates is the same as the disparity in infection rates (namely, a Black mortality rate that is 2.28 times the average), the disparity seems to be due to category (i) above, and not (ii) or (iii). Further, the authors also estimate a hazard-rate model for mortality conditional on hospitalisation and find that there is no statistically significant effect of race once comorbidities have been controlled for, which seems to confirm that disparities in care in the hospital are not at work in this case. An earlier but much smaller study in Georgia found similar results (Gold et al. 2020).
Given that the COVID-19 disparities seem to be more rooted in rates of exposure to the virus than in subsequent treatment, a natural focus is to hunt down socioeconomic reasons for the differential exposure. Here data problems are a major obstacle. Ideally, one would have a large sample of patients and full information about their employment, education, occupation, income, and so on, as well as medical outcomes for each individual, but that is not available. Researcher Merlin Chowkwanyun has commented: “When a patient comes to a clinic, they can self-report their race, but how do they self-report socioeconomic status? How do you actually gather that information?” (Wood 2020).
One imperfect workaround can be described as follows. Recall that the mortality data are not broken down by race for most counties; however, population figures are. Suppose that mortality rates for each minority do not vary much across the country. Imagine we regress county mortality rates on county minority-population shares. Then if the population share of minority group i has a positive coefficient, that indicates that group i has a higher-than-average mortality rate, and the ratio of its mortality rate to the average can be backed out with a simple calculation. If, then, the disparity in mortality rates thus measured disappears when we control for additional socioeconomic variables, that can be taken as evidence that those variables are the underlying source of the disparity. This is the approach I take in McLaren (2020).
This indirect approach is applied to four census demographic groups: African Americans, Hispanic/Latinos, Asian Americans, and the census category ‘American Indian and Alaska Native’, abbreviated here as ‘First Nations’. The control variables are pre-COVID-19 county features, averaged over 2013–8 and taken from the American Community Survey of the Census Bureau. They include median household income, poverty rate, education, uninsured fraction, the occupational profile of the county, and the fraction who use public transit to get to work.
The main findings are:
(i) This approach confirms very strong racial disparities in mortality rates, not far out of line with the estimates provided by direct measures.
(ii) For Latino/Hispanic and Asian populations, those disparities mostly disappear once education and occupation are controlled for.
(iii) By contrast, for African-American and First Nations populations, the disparity is very robust. Surprisingly, it is barely affected by controlling for occupation, income, poverty rates, or – importantly – even access to healthcare insurance, so those factors do not seem to be an important source of the disparity.
(iv) One factor that does seem to be significant for the African-American disparity is the pre-COVID-19 use of public transport – at least for April, when the partial correlation between mortality and public transit is very strong. Harris (2020) builds a strong case that the subway was a crucial element in the early spread of COVID-19 in New York City, but that the effect dissipated as ridership dropped, and these results are consistent with that (even when New York City is dropped).
These negative results for African Americans and First Nations people can at least help narrow down the search for the source of the disparities. Important candidates that cannot be ruled out at present include:
(i) Disproportional employment in ‘essential’ occupations that cannot be done from home. Despite the insignificant effect of occupations in the regressions above, these occupational effects could still be a reason for the disparity because the county occupational categories in the American Community Survey are very crude;8 this may be the reason occupational variables show no significant results (even though they were very significant for the other two minorities).
(ii) Residential density. Emeruwa et al. (2020) study every woman admitted for labour and delivery at two New York hospitals over several weeks. As required by law, every one of these patients was tested for COVID-19, making this an ideal sample for studying the correlates of positivity. The authors found that the strongest predictor of COVID-19 positivity was whether or not the patient lived in a neighbourhood with a high average number of people per dwelling. This variable mattered far more than neighbourhood income or population density. Unfortunately, the data do not include race, so it is an open question whether this is a source of the disparity or not.
(iii) Environmental racism. Wu et al. (2020) study the effect of county-level particulate matter in the air on COVID-19 death rates and find alarmingly strong effects. Together with the abundant evidence on environmental hazards disproportionately located in minority neighbourhoods, this seems as if it could be a reason for the disparities. However, the study separately controls for the Black population share and finds that it has just as strong an effect as the studies that do not control for pollution.9 This seems to rule out particulate matter as the underlying source of the disparity in death rates, but there are many other forms of environmental harm that could be explored in a similar manner.
The inequities of COVID-19 are grotesque. Nailing down which mechanisms are most at fault can help focus attention on which policies are needed to address them. So far, the exact mechanisms have been elusive.
Berchick, Edward R, Jessica C Barnett and Rachel D Upton (2019), “Health insurance coverage in the US: 2018”, Current Population Reports P60-267, US Census Bureau.
Centers for Disease Control and Prevention (2020), “COVID-19 in racial and ethnic minority groups”.
Chowkwanyun, Merlin, and Adolph L Reed, Jr (2020), “Racial health disparities and COVID-19 – caution and context”, The New England Journal of Medicine, 6 May.
Desmet, Klaus, and Romain Wacziarg (2020), “Understanding spatial variation in COVID-19 across the US”, NBER Working Paper 27329.
Emeruwa, Ukachi N, Samsiya Ona, Jeffrey L Shaman, Amy Turitz, Jason D Wright, Cynthia Gyamfi-Bannerman and Alexander Melamed (2020), “Associations between built environment, neighborhood socioeconomic status, and SARS-CoV-2 infection among pregnant women in New York City”, JAMA, 18 June.
Gold, Jeremy A W, et al. (2020), “Characteristics and clinical outcomes of adult patients hospitalized with COVID-19 — Georgia, March 2020”, Morbidity and Mortality Weekly Report, Centers for Disease Control and Prevention, 69(18).
Gross, Cary P, Utibe R Essien, Saamir Pasha, Jacob R Gross, Shi-yi Wang and Marcella Nunez-Smith (2020), “Racial and ethnic disparities in population level COVID-19 mortality”, Working Paper, Yale School of Medicine, hosted by medRxiv.
Harris, Jeffrey E (2020), “The subways seeded the massive coronavirus epidemic in New York City”, NBER Working Paper 27021.
Hoffman, Kelly M, Sophie Trawalter, Jordan R Axt and M Norman Oliver (2016), “Racial bias in pain assessment”, Proceedings of the National Academy of Sciences, April.
Hooper, Monica Webb, Anna María Nápoles and Eliseo J Pérez-Stable (2020), “COVID-19 and racial/ethnic disparities”, Journal of the American Medical Association 323(24): 2476–7.
Johnson, Akilah (2020), “On the minds of Black Lives Matter protesters: A racist health system”, ProPublica.org, 5 June.
Mtshali, Marya T (2020), “How medical bias against black people is shaping COVID-19 treatment and care”, Vox.com, 2 June.
Oppel, Richard A, Jr, Robert Gebeloff, K K Rebecca Lai, Will Wright and Mitch Smith (2020), “Racial disparity in cases stretches all across board”, The New York Times, 6 July.
Price-Haywood, Eboni G, Jeffrey Burton, Daniel Fort and Leonardo Seoane (2020), “Hospitalization and mortality among Black patients and white patients with COVID-19”, The New England Journal of Medicine, 27 May.
Romero, Simon, and Jack Healy (2020), “Tribal nations face most severe crisis in decades as the coronavirus closes casinos”, The New York Times, 11 May.
Rubin, Eric J, Lindsey R Baden, Michele K Evans and Stephen Morrissey (2020), “Audio interview: The impact of COVID-19 on minority communities”, New England Journal of Medicine, 11 June.
United Church of Christ Commission for Racial Justice (1987), Toxic wastes and race in the US: A national report on the racial and socio-economic characteristics of communities with hazardous waste sites, New York: United Church of Christ.
Wiemers, Emily E, Scott Abrahams, Marwa AlFakhri, V Joseph Hotz, Robert F Schoeni and Judith A Seltzer (2020), “Disparities in vulnerability to severe complications from COVID-19 in the US”, NBER Working Paper 27294.
Wood, Graeme (2020), “What’s behind the COVID-19 racial disparity?”, The Atlantic, 27 May.
Wu, Xiao, Rachel C Nethery, M Benjamin Sabath, Danielle Braun and Francesca Dominici (2020), “Exposure to air pollution and COVID-19 mortality in the US: A nationwide cross-sectional study”, working paper, Harvard TH Chan School of Public Health, hosted by MedRxiv.
1 Numerous commentators and activists have warned that there is a pernicious tendency to ascribe these disparities to biological differences or irresponsible behaviour by minorities; see, for example, Hopper et al. (2020), Johnson (2020), Mtshali (2020), and Rubin et al. (2020). Research into the causes can help fight this tendency.
2 Black and Latino/Hispanic Americans are much more likely to have no insurance; see Berchick et al. (2019) and Centers for Disease Control (2020). Timothy Freeman, pastor at Trinity African Methodist Episcopal Zion Church in Washington DC, calls insurance coverage crucial for COVID disparities: “I have seen diagnostic tests not performed … and hospitalizations cut extremely short – or not happen at all – because of insurance” (quoted in Johnson 2020).
3 Desmet and Wacziarg (2020) find a strong relationship between COVID-19 cases and two measures of county density. However, Emeruwa et al. (2020) study every woman admitted for labour and delivery at two New York hospitals between 22 March and 21 April, all of whom were tested for COVID-19, and find the strongest predictor of COVID-positivity to be residence in a neighbourhood with many people per dwelling, and not people per square mile.
4 Wiemers et al. (2020) use pre-pandemic retrospective questions from the Panel Study on Income Dynamics to identify socioeconomic correlates of factors that have turned out to be dangerous comorbidities for COVID-19. Several of these comorbidities are strongly correlated with race, education, and income.
5 These factors are emphasised by Mtshali (2020). Hall et al. (2015) survey studies that demonstrate implicit bias in healthcare providers and find some effect on healthcare outcomes. Hoffman et al. (2016) report evidence that a substantial fraction of health professionals undervalue the pain experienced by African-American patients due to racial attitudes, and there are claims that African Americans who have had adverse experiences with healthcare professionals have trouble trusting them and so are slower to seek treatment (Johnson 2020).
6 See United Church of Christ Commission for Racial Justice (1987) for both the origin of the term and the pioneering demonstration of the finding that policy had resulted in toxic environments for racial minorities.
7 Wu et al. (2020) show in county data that a high rate of particulate matter in the air, produced by a wide range of industrial facilities, is a powerful predictor of COVID-19 mortality rates after controlling for numerous variables.
8 For example, one control variable is the county share of the workforce employed in the Department of Labor’s Occupational Category 31: Healthcare Support Occupations, which includes occupations such as Home Health Aides, Nursing Assistants, and Orderlies. This variable is strongly correlated with COVID-19 deaths.
9 Wu et al. (2020) find that a one-standard-deviation increase in the Black population share is associated with a 45% increase in deaths, controlling for particulate matter and other county characteristics. The Hispanic/Latino share is also controlled for but is statistically insignificant.