Dynamic machine learning models for predicting cesarean delivery risk in women with no prior cesarean delivery: A retrospective nationwide cohort analysis.

Abstract

Objective: To develop and validate advanced machine learning (ML) models for predicting unplanned intrapartum cesarean deliveries in women with no previous cesarean delivery, using both static and dynamic clinical data.

Methods: A retrospective cohort study was conducted using nationwide data from a large integrated healthcare provider, including 262 632 women whose labor had started. Two ML models, logistic regression and decision tree algorithms, were employed to predict unplanned cesarean delivery. The models incorporated demographic, medical, and obstetric variables collected at multiple time points during labor. Model performance was evaluated based on accuracy, sensitivity, specificity, and the area under the receiver operating characteristics curve (AUC-ROC).

Results: The logistic regression model demonstrated an accuracy of 95% with an AUC-ROC of 0.92. The decision tree model showed adaptability in highly variable labor conditions, achieving an F1 score of 0.91 and excelling in real-time prediction. Key predictors included maternal age, gestational age, body mass index, fetal heart rate patterns, and labor dynamics. Model performance remained robust across various demographic subgroups but was slightly reduced in nulliparous women.

Conclusion: These ML models provide an innovative approach to predicting unplanned cesarean delivery by integrating diverse clinical parameters, enhancing decision making, and optimizing labor management. Prospective validation and seamless integration into clinical workflows are required to establish their utility in broader obstetric practice.