Integrating SHAP analysis with machine learning to predict postpartum hemorrhage in vaginal births

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Postpartum hemorrhage (PPH) is a significant cause of maternal morbidity and mortality worldwide.
• Accurate prediction of PPH risk is crucial for timely intervention and improved patient outcomes.
• Existing risk assessment tools often lack precision and interpretability.

Purpose of the Study:

• To develop and validate a machine learning (ML) model integrated with SHapley Additive exPlanations (SHAP) for predicting PPH risk.
• To provide an interpretable tool for personalized risk assessment and prevention strategies in women undergoing vaginal delivery.
• To identify key predictive features for PPH using SHAP analysis.

Main Methods:

• A retrospective multicenter cohort study was conducted in Northeast China.
• An XGBoost ML model was developed and validated using electronic medical record data from vaginal deliveries.
• SHAP analysis was employed for feature selection, interpretation, and model explanation.

Main Results:

• The XGBoost model achieved high predictive accuracy for PPH, with an Area Under the Curve (AUC) of 0.997 in the training set.
• SHAP analysis identified 15 key features contributing to PPH prediction, offering intuitive insights into their impact.
• The model demonstrated strong performance in internal (AUC=0.894) and external (AUC=0.880) validation datasets.

Conclusions:

• An interpretable ML model with high accuracy for predicting PPH following vaginal delivery was successfully developed.
• The SHAP-integrated model offers a promising tool for personalized PPH risk assessment and clinical decision-making.
• Further validation with larger, diverse datasets is recommended to enhance generalizability across different populations and healthcare settings.