(245) Accuracy of deep learning models in interpreting intrapartum fetal monitoring to predict fetal acidemia

Up to 30% of cesareans in the US are performed due to false-positive interpretations of intrapartum electronic fetal monitoring (EFM). EFM interpretation is subjective and vulnerable to bias. Novel deep learning techniques can improve complex data processing and pattern recognition in medicine. We sought to apply deep learning approaches that could interpret EFM data to predict fetal acidemia.

Study Design:

The database was created using intrapartum EFM data from 2006-2020 at a large, multi-site academic health system. We included patients ≥34 weeks with a singleton, vertex fetus with EFM data available for ≥1 hour prior to delivery and an umbilical cord blood pH result available. We excluded those with >30% missingness in EFM data. Data pre-processing removed noise and artifact. Data was divided into training and testing sets with equal distribution of acidemic cases. Several different deep learning architectures were explored, including transformers, convolutional neural networks (CNNs), and long short-term memory (LSTM) networks. The primary outcome was low cord blood pH, investigated at four clinically meaningful thresholds: 7.2, 7.15, 7.1, and 7.05. Receiver operating characteristic (ROC) curves were generated with area under the curve (AUC) assessed to determine the performance of the models.

Results: A total of 124,776 fetal monitoring files were available, 35,604 had a corresponding umbilical cord gas pH result, and the final sample size was 10,176. The prevalence of the outcome in the data was 20.9% with pH < 7.2, 9.1% < 7.15, 3.3% < 7.10, and 1.3% < 7.05. The median AUC values for each deep learning model at each different pH threshold are shown in Figure 1. The best performance was achieved with the CNN multiscale model and a pH threshold of 7.10, with an AUC of 0.82 95% CI [0.82-0.83].

Conclusion:

A novel application of deep learning methods achieves excellent performance in predicting fetal acidemia on umbilical cord blood pH. This technology could improve the accuracy and consistency of EFM interpretation to prevent unnecessary cesarean deliveries and avoidable intrapartum fetal injury.