Machine learning-based predictive model for postpartum post-traumatic stress disorder: A prospective cohort study.

Abstract

Background: Postpartum Post-Traumatic Stress Disorder (PTSD) is a public health issue affecting both mothers and infants. Early identification of high-risk women for PTSD could mitigate its impacts. This study aimed to develop and validate a machine learning (ML)-based model for predicting PTSD risk in the early postpartum period.

Methods: A prospective cohort study collected sociodemographic and clinical, adverse childhood experiences, and biochemical data at 3 days postpartum, with PTSD symptoms assessed at 42 days postpartum. Five ML models were developed using Logistic Regression (LR), Decision Tree, Random Forest, Support Vector Machine, and eXtreme Gradient Boosting. Model performance was assessed using discrimination, calibration, and clinical application in an independent validation cohort, with Area Under the Receiver Operating Characteristic Curves (AUC). The best-performing model was deployed as a web application.

Results: A total of 900 postpartum women were included in model development and 300 in the validation cohort. AUCs for the five models ranged from 0.768 to 0.850, with the LR model performing best (AUC of 0.850; 95%CI: 0.776-0.923). The LR model achieved a Brier score of 0.069, sensitivity of 0.844, specificity of 0.724, and F1 score of 0.406. A web-based calculator incorporating 8 predictors was developed for clinical use.

Conclusion: This study demonstrates the effectiveness of a ML-based model in predicting postpartum PTSD risk. The developed web-based risk calculator enables early identification of high-risk women, supporting timely and targeted interventions. These findings highlight the potential of ML tools to improve maternal mental health care, though further calibration in independent and diverse cohorts is needed.