Explainable machine learning model for predicting cesarean section following induction of labor: Development and...

Area of Science:

• Perinatal medicine
• Machine learning in healthcare
• Predictive modeling for obstetrics

Background:

• Induction of labor (IOL) is a common obstetric procedure with inherent risks, including cesarean section (CS).
• Predicting CS risk following IOL is crucial for informed, individualized patient care.
• Existing prediction tools may lack accuracy or explainability.

Purpose of the Study:

• To develop and validate an explainable machine learning model for predicting CS risk in women undergoing IOL.
• To identify key predictors of CS following IOL using routinely collected data.
• To assess the model's performance and clinical utility.

Main Methods:

• Utilized population-based administrative perinatal datasets from Australian states (NSW, Queensland, Victoria).
• Developed and compared seven machine learning models (XGBoost optimal) with hyperparameter tuning and feature selection.
• Validated model performance using temporal and geographical datasets, assessing AUROC, calibration, and SHAP values for explainability.

Main Results:

• The optimal XGBoost model achieved an AUROC of 0.76 in temporal validation and 0.75 in geographical validation.
• Key predictors included nulliparity, pre-pregnancy BMI, and maternal age; diabetes and hypertension were less influential.
• Higher predicted CS risk correlated with increased inpatient costs and maternal morbidity.

Conclusions:

• An explainable machine learning model using routinely collected maternal factors demonstrates strong predictive performance for CS risk post-IOL.
• The model provides valuable insights into individual CS risk, potentially improving clinical decision-making.
• Further research on co-design and implementation is necessary for clinical adoption.