Disentangling race and ethnicity in predicting symptoms of depression among young adults: A machine learning approach

Research is needed to understand racial and ethnic differences in symptoms of depression. Unfortunately, most studies examine these differences using ethnically-stratified, mono-racial categories (e.g., non-Hispanic Black), producing inaccurate estimates due to heterogeneity across racial and ethnic identities. In this study, we compare different operationalizations of race and ethnicity in predicting symptoms of depression within a diverse cohort of young adults.

We analyzed cross-sectional data from n = 2340 young adults (mean age: 21 yrs. (SD: 1.6), 59% female, 38% Hispanic, 29% low SES) via the Texas Adolescent Tobacco and Marketing Surveillance (TATAMS) study. Random forest models evaluated prediction and identified sociodemographic features for symptom classification. We modeled eight operationalizations of race and ethnicity, four applying mutually-exclusive categorizations (e.g., non-Hispanic White) and four allowing for overlapping categorizations. Models comprised a) race and ethnicity, alone, and b) included SES, age, sex, and geography.

Models with race and ethnicity, alone, demonstrated poor prediction of symptoms of depression (Sn range: 0.33-0.48). Including other sociodemographic features, prediction remained poor for symptoms of depression (Sn range: 0.15 - 0.62). Prediction decreased upon separation of Hispanic ethnicity and 'Other' race (e.g. non-Hispanic Asian). SES was the most influential feature across all models.

Race and ethnicity poorly predict symptoms of depression, particularly when using standard OMB categories (i.e., ethnically-stratified, mono-racial). Models allowing for overlapping racial and ethnic identities outperformed those using mutually-exclusive categorizations. Results suggest that health equity research should account for racial and ethnic heterogeneity and consider SES in addressing racial and ethnic differences in mental health among young adults at the population level.