Postpartum depression risk prediction using explainable machine learning algorithms.

Area of Science:

• Reproductive Medicine
• Psychiatry
• Machine Learning

Background:

• Postpartum depression (PPD) is a significant mental health issue affecting mothers and infants.
• Early identification and intervention are crucial for managing PPD.

Purpose of the Study:

• To develop an explainable machine learning model for predicting PPD risk.
• To identify key predictive factors for PPD.

Main Methods:

• Retrospective analysis of 1,065 women's postpartum data.
• Feature selection using LASSO regression and Boruta algorithm.
• XGBoost model development and evaluation using AUC, accuracy, precision, and specificity.
• SHAP for model interpretability.

Main Results:

• An 11-variable XGBoost model demonstrated excellent predictive performance (AUC 0.955, accuracy 0.95).
• Identified key predictors: weight gain, mother-in-law relationship, sleep quality, marital status, planned pregnancy, fetal sex preference, pregnancy anxiety, pelvic-floor endurance, cervix status, prenatal education, and postpartum care satisfaction.
• SHAP analysis provided insights into individual predictions.

Conclusions:

• The XGBoost model effectively predicts PPD risk, aiding clinical decision-making.
• Explainable AI (SHAP) enhances understanding of PPD causes and prevention strategies.
• Improved identification of high-risk individuals can lead to better patient outcomes.