Predicting Postpartum Depression Risk Using Social Determinants of Health

Postpartum depression (PPD) affects approximately 20% of women after childbirth and has complex etiology. Existing predictive models of PPD lack training on large, national datasets and comprehensive integration of clinical and social determinants. To address this gap, we developed machine learning (ML) models using the Pregnancy Risk Assessment Monitoring System survey (2016-2021) including 51,917 US patients. ML used medical history, pregnancy complications, social factors, and infant outcomes as features, with self-reported PPD as the outcome. We evaluated several ML approaches, including gradient boosting machine, logistic regression, random forest, and support vector machine, and deep significance clustering (DICE). Logistic regression demonstrated the best performance (AUC = 0.726, 95% CI = [0.715, 0.737]).