Identifying prenatal risk factors of postpartum depression with machine learning

Postpartum depression (PPD), a common mental illness among mothers, can affect the well-being of both mothers and their children. Early intervention is essential but hindered by difficulties in identifying at-risk women, as it remains unclear how soon PPD can be predicted and which prenatal risk factors are most informative at each stage of pregnancy. We applied machine learning algorithms to (1) determine how early PPD can be predicted and (2) identify trimester-specific risk factors. The analysis leveraged data from 2865 mothers enrolled in the Brabant Study (Netherlands), including 233 psychological, biological, and social variables. A Lasso-regularized linear regression model trained on all trimesters performed best, identifying depressive symptoms during pregnancy, negative affectivity, neuroticism, body mass index, and history of mental health treatment as the most predictive risk factors. Although our predictive performance matched that of similar machine learning studies, high specificity but relatively low sensitivity indicates the model reliably excludes low-risk women but is less effective at detecting those who will develop PPD. Consistent performance across trimesters suggests low-risk status can be identified as early as 12 weeks. Clinicians could therefore use early-pregnancy data to rule out low-risk cases and allocate preventive resources more efficiently.