(543) Predictive Modeling for Antenatal Shoulder Dystocia Risk

Shoulder dystocia (SD) represents a significant complication during delivery, posing substantial risks. Despite its challenging predictability, assessing individual risk is crucial for informed counseling on optimal delivery methods. The objective of this study was to develop and validate a model prediction system for SD using fetal ultrasound and maternal data.

Study Design:

Data was retrospectively obtained from deliveries in Meir Hospital between 2014 and 2023.The inclusion criteria were singleton pregnancies and vaginal deliveries. The features included in the tested models were those that had a small percentage of null values in the SD cases. In cases of null values in normally distributed features, we completed the data with mean values. The final model parameters included maternal age, BMI, obstetric history, maternal height, gestational or pre-gestational diabetes, gestational age at delivery, clinical and sonographic estimated fetal weight, fetal gender, and mode of delivery (vaginal or operative vaginal). To address imbalanced data, we used repeated random sampling of the negative cases. The training and test sets were split in a 70:30 ratio and standardized.
Several classifiers, including logistic regression, decision tree, random forest, support vector machine, XGBoost, and CatBoost, were evaluated using cross-validation and area under the ROC Curve (AUC). We chose the model based on the highest mean AUC. The analysis was conducted in Python using Pandas, Scikit-learn, Numpy, and visualization libraries.

Results: A total of 51,628 deliveries were analyzed, including 94 SD cases. SD patients had higher rates of diabetes (13% vs. 4.8%) and obesity (23% vs. 5%) compared to non-SD patients, and higher mean neonatal birthweight (3751 g vs. 3287 g). The highest mean AUC was achieved using the CatBoost model, which demonstrated high discriminatory ability in predicting SD (AUC = 0.83).

Conclusion: The SD prediction model presented in this study serves as a valuable adjunct for clinical decision-making regarding the appropriate mode of delivery.