Disparities in COVID-19 Outcomes by Race, Ethnicity, and Socioeconomic Status

Key Points Question Are race and ethnicity–based COVID-19 outcome disparities in the United States associated with socioeconomic characteristics? Findings In this systematic review and meta-analysis of 4.3 million patients from 68 studies, African American, Hispanic, and Asian American individuals had a higher risk of COVID-19 positivity and ICU admission but lower mortality rates than White individuals. Socioeconomic disparity and clinical care quality were associated with COVID-19 mortality and incidence in racial and ethnic minority groups. Meaning In this study, members of racial and ethnic minority groups had higher rates of COVID-19 positivity and disease severity than White populations; these findings are important for informing public health decisions, particularly for individuals living in socioeconomically deprived communities.

Meta-regression for measures of clinical care quality in the following cohorts: meta-regression for preventable hospital stays in correlation with Asian Americans who tested positive for COVID-19 in cohort studies; meta-regression for primary care physician availability in correlation with Asian Americans and Hispanics who tested positive for COVID-19 (cohort studies) and Whites who are deceased (cross-sectional studies); and meta-regression for the amount of uninsured individuals in correlation with African Americans who tested positive for COVID-19 (cohort studies) and Whites who are deceased (cross sectional studies).
The following keywords were used to search by all fields, which includes full text, author name, journal name, and phrase, in each database: "COVID-19 AND race", "COVID-19 AND ethnicity", "COVID-19 AND Asian patients", "COVID-19 AND Black patients", "COVID-19 AND White patients", "COVID-19 AND Hispanic/Latino patients", "COVID-19 AND American Indian/Alaska Natives patients", "COVID-19 AND Pacific Islander patients", "COVID-19 AND multiracial patients", "income AND COVID-19"; "socioeconomic status AND COVID-19", and "employment AND COVID-19." We used both the keyword and Medical Subject Heading (MeSH) term for the following keywords to increase the scope of our systematic review and meta-analysis: "COVID-19 AND ethnicity (MeSH term: COVID-19 AND ethnic groups)", "COVID-19 AND race (MeSH term: COVID-19 AND race factors)", "socioeconomic status AND COVID-19 (MeSH term: COVID-19 AND social class)". MeSH terms provide controlled vocabulary for searches in databases, such as Pubmed. We chose to use both the MeSH term and the non-MeSH term for these particular keywords, as the non-MeSH term yielded significantly more results than the MeSH term. MeSH terms could not be used for the following keywords, as they were not available on the database: "COVID-19 AND Asian patients", "COVID-19 AND Black patients", "COVID-19 AND White patients", "COVID-19 AND Hispanic/Latino patients", "COVID-19 AND American Indian/Alaska Natives patients", "COVID-19 AND Pacific Islander patients", and "COVID-19 AND multiracial patients". MeSH terms were solely used for the following keywords: "income AND COVID-19" and "employment AND COVID-19".
Our original keyword searches yielded 21,745 total results. Of these articles, 14,519 were unique (eFigure 1). We excluded studies based on Abstract if they met one of the following criteria: (1) The article is irrelevant for the study question or has insufficient data, (2) The article does not discuss an outcome that is of interest, (3) The article is published in a non-standard format and/or in a foreign language. Only studies with original clinical data were included. Following the Abstract review, we screened the full text of the remaining 287 articles. After subsequent fulltext screening using the same 3 exclusion criteria, a total of 68 studies were included for data analysis.
Study and patient characteristics were collected, including the study type, location, mean and median age, total number of patients in the study, and medical comorbidities. Specifically, we extracted data for the following medical comorbidities and conditions which we observed to be commonly reported across various studies: smoking status (both former and current smokers), median body mass index (BMI), BMI over 40, cardiovascular disease (including other heart conditions such as coronary artery disease), hypertension, chronic obstructive pulmonary disease (COPD), diabetes mellitus or diabetes, and occurence of malignancy or cancer. For the purposes of this analysis, we considered Hispanics and Latinos as a single cohort. The studies included did not differentiate between various Asian populations, so many Asian populations were considered as a single cohort.
Following initial data review, we extracted the zip code, geographic location and/or congressional district from each study included in our meta-analysis in order to identify socioeconomic variables for subsequent analyses. In instances where congressional district information was not provided, we determined this information based on the zip code or geographic location of the study. From this extracted information, we obtained the following data for various measures of socioeconomic disparity from external websites for each study: (1) County median income and the percentage of each race in the district where the study was conducted was taken from the US Census Bureau's website at the congressional district level. (  Excluded due to insufficient data: adjusted (eTable 3, eTable 4). Studies in the unadjusted model that did not include information for one of these variables were excluded from the adjustment analysis of that particular variable. Methods to estimate missing data, such as multiple imputation, were not used as the studies were conducted separately (not a randomized trial) and the number of known outcomes would not be sufficient for accurate imputation. No more than two individual measures were adjusted at once in order to minimize the effects of overfitting (see the composite measures mentioned below).
Additionally, fitting was only calculated if the predictor variable(s) had at least 2 more outcomes than the variables being adjusted for. The mixed-effects models were fitted to the median value(s) of the variable(s) being adjusted for in order to reduce the effects of outliers. We calculated a combined measure for both Comorbidities and Clinical Care using a unit-weighted composite function, as several variables were required to appropriately adjust for these factors. The Comorbidity measure was composed using the following comorbidities that were available in the study group: ever smoker, BMI, cardiovascular disease, hypertension, COPD, diabetes, and cancer. The following variables were used to compose an estimate for the quality of Clinical care: Percent of the population under 65 that are uninsured, ratio of the population to primary care physicians, and the rate of hospital stays for ambulatory-care sensitive conditions per 100,000 Medicare enrollees (preventable hospital stays). In order to test for the similarity of variables used for the combined measures, only composed variables with a Chronbach's alpha score > 0.7 were used for adjustment (eTable 3, eTable 4). The clinical-care measure for Hispanic/Latino COVID-19 positive RR/OR was the only unit-weighted composite variable which yielded an alpha score < 0.7. Thus, RR/OR adjustment was not implemented for this cohort.