Identifying High-Risk Pre-Term Pregnancies Using the Fetal Heart Rate and Machine Learning

Fetal heart rate (FHR) monitoring is ubiquitous in antenatal care, yet human visual interpretation poorly predicts adverse pregnancy outcomes. Meanwhile, preterm gestations carry a high burden of stillbirth and severe fetal compromise, where earlier identification of high-risk pregnancies may justify iatrogenic preterm delivery to prevent avoidable fetal death. We analyzed 4867 antepartum FHR recordings from pre-term pregnancies meeting at least one of ten adverse outcome criteria alongside 4014 term uncomplicated controls. Seven clinically validated FHR features were extracted from each trace, and six machine-learning classifiers were trained on 80% of the data (7105 samples) using k-fold cross-validation; the remaining 20% (1776 samples) formed an internal validation cohort. The random forest demonstrated the best performance, achieving an area under the receiver-operating characteristic curve (AUC) of 0.88 (95% confidence interval [CI] 0.87–0.88) during training and 0.88 (95% CI 0.86–0.90) on validation, with good calibration (Brier score 0.14). Median AUC across individual adverse outcomes was 0.85 (interquartile range [IQR] 0.81–0.89) and exceeded 0.80 at all gestational ages assessed; sensitivity and specificity at the Youden threshold were 76.2% and 87.5%, respectively. Decision-curve analysis demonstrated net benefit across a range of clinically relevant probability thresholds. These findings indicate that data-driven interpretation of antepartum FHR can stratify risk in pre-term pregnancies with high accuracy and may support earlier, evidence-based clinical decision-making, particularly in resource-limited settings where specialist expertise is limited.