Methods/Statistics
Covariate selection and justification in studies of multidimensional indices of structural racism and health: A systematic review Annika Gompers* Annika Gompers Gompers Gompers Gompers Gompers Gompers Department of Epidemiology, Emory University Rollins School of Public Health
Background: There has been a substantial increase in research measuring structural racism, often using multidimensional indices, and its association with health. To date, there is no consensus on best practices for covariate selection for such studies. We aimed to identify and quantify commonly selected covariates and justifications for their use.
Methods: We conducted a MEDLINE search to identify studies meeting inclusion criteria: US-based, empirical studies measuring structural racism using an index comprised of at least three indicators of contemporary racial inequities in resources or opportunities (e.g., education, poverty) or contemporary policies (e.g., voter ID laws) and including a health outcome. Studies of interpersonal or self-reported experiences of racism were excluded. We extracted each study’s exposure, outcome, selected covariates and stated justification, which we graded based on inclusion of references, stated relationship to exposure and outcome—including potential mediation—or mention of a causal diagram. The study protocol is published on PROSPERO.
Results: Fifty-four studies, published in 2020–2025, met inclusion criteria. Six studies (11%) did not adjust for any covariates, 15 (28%) adjusted for area-level variables, 13 (24%) for individual-level variables, and 20 (37%) for both. Seventeen studies (31%) used multiple adjustment sets. Seventeen (31%) did not state any justification for selected covariates, and an additional 16 (30%) did not cite any references in support of their approach. Selected covariates and justification paradigms are shown in Table 1.
Conclusions: Among studies of multidimensional structural racism indices, substantial heterogeneity exists in the covariates selected and the justifications employed, if at all. Given that these indices capture a diffuse construct with unclear etiologic timing, thoughtful approaches to model adjustment are necessary to strengthen inference from this rapidly growing literature.
