Development and external validation of an interpretable machine learning model for predicting perinatal depression in...

This study aimed to develop a machine learning (ML)-based prediction model for antenatal depression (AND) in Chinese women. Given the significant impact of AND on maternal and infant health, the goal was to create an accurate and interpretable model to identify women at risk early.

Five ML models were compared, including Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM). The models were evaluated on accuracy, sensitivity, specificity, the area under the receiver operating characteristic curve (AUC), and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) were utilized to improve model interpretability and highlight important predictive features. The training set comprised 442 pregnant women during prenatal visits at the obstetrics clinic of a major hospital in Wuhan, China. In comparison, the external validation set comprised 190 pregnant women at the obstetrics clinic of another major hospital in Wuhan, China.

The LightGBM model outperformed the other models in terms of predictive accuracy and generalizability. Key predictors of AND included educational attainment, coping strategies, social support and personality traits. The model demonstrated robust performance in both the training and validation datasets, offering an effective tool for AND prediction.

This study developed a reliable and interpretable ML-based model for early identification of women at risk for AND. The findings suggest that ML techniques, specifically LightGBM, can be integrated into prenatal care to enhance early screening and provide timely interventions for at-risk women, improving maternal mental health outcomes.