264: Machine learning model for prediction of neonatal scalp hematomas using first stage of labor data

Abstract

Decision regarding the mode of delivery (either spontaneous, assisted or cesarean) during second stage of labor is one of the major challenges facing modern obstetrics. A model predicting maternal and fetal labor related outcomes, such as subgaleal hemorrhage (SGH), based on maternal and fetal variables collected during first stage of labor, may contribute substantially to the physician's decision-making process. Electronic medical records of labors during a 6-years period in a tertiary medical center were explored. Using gradient boosting, a model incorporating multiple maternal and neonatal variables was created to predict the risk for increased pressure on fetal head during delivery (SGH or cephalohematoma (SGH+CH)). This model can additionally allocate the deliveries with high risk for SGH at the beginning of second stage of labor. Elective repeated cesarean deliveries, multifetal gestations and preterm deliveries were excluded. We identified 659 (1.8%) deliveries with SGH+CH out of 36,607 parturients during a 6-years period. A machine learning based model to predict SGH+CH was developed. The area under the curve of the model was 0.89. Most importantly, the model was able to identify high risk births, using basic maternal, fetal and labor associated parameters, with increased odds ratio (OR) for several major neonatal complications; OR of 11.44 (95% Confidence Interval 7.5-17.5) for SGH, OR of 2.38 (95% CI 1.8-3.2) for neonatal intensive care admission and OR of 3.41 (95% CI 2.5-4.6) for lower than 7 APGAR score in 5 minutes after birth. Our machine learning-based model was able to accurately predict births with higher risk for SGH and CH. The same model was capable of early identification of deliveries with increased risk for several major obstetric and neonatal complications, making it an applicable tool, useful for personalized decision making during the active stage of labor.