A machine learning based fetal monitoring system to predict and prevent fetal hypoxia.

Project Summary/Abstract: Although EFM is widely deployed in the United States for most deliveries, it has failed to reduce rates for hypoxic injuries such as neonatal encephalopathy, despite an increased rate of cesarean sections. This lack of improvement has been attributed to inconsistent applications of vague guidelines during manual analysis of EFM tracings. Existing automated tools available in the market to augment physician capabilities take the form of low-precision simplistic rule-based alerts, which cause alarm fatigue and also fail to deliver improvements. This project proposes the creation and validation of a machine learning model for prediction of intrapartum fetal hypoxia with high sensitivity and specificity to address this need. Using a multi-site dataset of 50,000 tracings coupled with electronic health records, a combination of clinical knowledge and a variety of machine learning techniques will be used to create a model with leading performance. To clear the high bar set by FDA for patient safety with a de novo device, this proposal aims to validate this model by demonstrating high sensitivity and specificity on a held-out portion of this large multi-site data set, along with a user study to demonstrate improved performance by clinicians with software assistance. After this project demonstrates the safety and efficacy of this model for patient care, a future Phase II will beta test a software solution integrating this model in labor and delivery wards. The research plan outlined in this proposal will give obstetricians a valuable evidence-based tool to help them interpret EFM tracings.