Antepartum prediction of shoulder dystocia using machine learning

Area of Science:

• Obstetrics and Gynecology
• Machine Learning in Healthcare
• Perinatal Medicine

Background:

• Shoulder dystocia (SD) is a significant obstetric emergency with unpredictable occurrence.
• Accurate prediction of SD is crucial for informed delivery management and reducing neonatal morbidity.
• Current risk assessment methods for SD have limitations.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting shoulder dystocia.
• To identify key predictors of SD using ML algorithms.
• To assess the performance of the ML model in a clinical setting.

Main Methods:

• Retrospective analysis of term singleton vaginal deliveries over 10 years.
• Eleven maternal and fetal features were selected for model development.
• The CatBoost model was trained and validated using a 70:30 data split and cross-validation, with AUC as the primary performance metric.

Main Results:

• Shoulder dystocia occurred in 0.18% of deliveries (94/51,628).
• Predictive factors included higher maternal BMI, shorter stature, diabetes, and increased birthweight.
• The CatBoost model achieved an AUC of 0.83, with sonographic EFW (55.6%) and maternal BMI (20.1%) as the most significant predictors.

Conclusions:

• A machine learning model, particularly CatBoost, demonstrates strong performance in predicting shoulder dystocia.
• The model's high AUC (0.83) suggests its potential utility in clinical decision-making for delivery planning.
• Key features like EFW and maternal BMI are vital for SD prediction, offering a data-driven approach to risk assessment.