Related Experiment Videos
Development and validation of a postpartum hemorrhage risk prediction model for advanced maternal age women
Yun Zhou1, Xiao Yao1, Jue Zhou1
1Shanghai Key Laboratory of Maternal Fetal Medicine, Shanghai Institute of Maternal Fetal Medicine and Gynecologic Oncology, Shanghai First Maternity and Infant Hospital, School of Medicine, Tongji University, Shanghai, China.
Background:
Postpartum hemorrhage (PPH) is the leading cause of maternal mortality in women of advanced maternal age (AMA). Accurate antenatal prediction remains challenging. This study aimed to develop and compare the performance of multiple machine learning (ML) algorithms for predicting PPH risk following vaginal delivery in AMA parturients.
Methods:
A retrospective cohort study was conducted using data from 5,369 women who delivered vaginally at Shanghai First Maternity and Infant Hospital. LASSO regression was used to identify key predictors. Seven machine learning algorithms-logistic regression (LR), decision tree (DT), random forest (RF), XGBoost, LightGBM, support vector machine (SVM), and artificial neural network (ANN)-were employed to build prediction models. Model performance was evaluated on an independent validation set using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and F1 score. The optimal model was selected based on a composite criterion prioritizing the highest sensitivity and best F1 score, while requiring a competitive AUC.
Results:
Eight predictors were selected for model construction. Among the seven models, the SVM model achieved the highest sensitivity (0.482) and the highest F1 score (0.504), and was therefore selected as the clinically preferred model, while maintaining a competitive AUC of 0.782 (95% CI: 0.758-0.806). SVM achieved a favorable balance between sensitivity and specificity, indicating better clinical utility for early screening and risk stratification of postpartum hemorrhage.
Conclusion:
Among seven machine learning models, the SVM model was selected as the clinically preferred predictor due to its highest sensitivity and F1-score. Its good discriminative ability and balanced performance make it suitable for clinical application as a screening tool. The integration of SHAP analysis can further enhance model interpretability and facilitate individualized risk assessment.