Leveraging Machine Learning to Uncover Ethnic-Specific Predictors of Maternal Postpartum Depression

Maternal postpartum depression (PPD) is a significant mental health issue that affects over 15% of mothers, with adverse effects on their children. Ethnic disparities in PPD represent a pressing public health concern, and minority mothers are found to have a higher likelihood of being affected by PPD. Identifying specific risk factors for these mothers is crucial for developing targeted preventive strategies that can help mitigate the disparities in PPD. This study used CDC data (N = 39,637) from the Pregnancy Risk Assessment Monitoring System to examine how risk factors for PPD vary among mothers from different ethnic groups. We applied Random Forest machine learning (ML) algorithms, and the final model achieved a mean ROC AUC of 0.65 and PR AUC of 0.66 during cross-validation on balanced samples, though performance declined on the held-out test set (PR AUC = 0.28; two times higher than the baseline of 0.11). The top predictors include prepregnancy and prenatal depression, family income, prior PPD-related visits, WIC participation, employment-based insurance, breastfeeding, pregnancy intention, and parental education. Subgroup analyses revealed both shared and unique predictors across racial and ethnic groups. While the results for White mothers closely aligned with overall trends, smoking and pregnancy termination ranked high among Black mothers. For Hispanic/Latina mothers, unintended pregnancy and smoking were more prominent, whereas, for Asian mothers, infant sleep arrangements, prepregnancy health behaviors, and infant sex emerged as key predictors. This study demonstrates the value of using ML to identify data-driven, population-specific predictors of PPD and highlights the need for culturally sensitive, early preventative approaches in perinatal mental health care.