HypoxIA: A bicenter study of machine learning-based prediction of fetal hypoxia using fetal heart rate and obstetric features.

Published in American journal of obstetrics & gynecology MFM. Abstract: Cardiotocography (CTG) is the gold standard to monitor fetal well-being during labor, but its clinical interpretation remains limited for detecting fetal hypoxia. Artificial intelligence approaches offer new perspectives to improve CTG-based fetal hypoxia detection. To evaluate the performance of an AI model combining intrapartum CTG signal and routinely collected clinical data to predict neonatal acidemia. This bicenter study included 3,574 singleton deliveries from May 2024 to August 2025. Neonatal acidemia was defined as umbilical arterial pH <7.15, and severe acidemia as pH <7.05. Raw CTG signals were segmented into 10-minute windows during the final hour and final three hours before birth. CTG-derived features were automatically extracted and combined with clinical variables in machine learning models. Model performance was evaluated using areas under the receiver operating characteristic curves (AUCs) with 95% confidence intervals (95% CIs), and compared with clinical intervention for suspected fetal hypoxia. Neonatal acidemia occurred in 380/3,574 newborns (10.6%), including severe acidemia in 65/3,574 (1.8%). Medical intervention for suspected fetal hypoxia occurred in 580/3,574 deliveries (16.2%). Model performance was higher for severe acidemia than for overall acidemia and was best during the final hour before birth. For the final-hour analysis, the AUC was 0.78 (95% CI, 0.71-0.85) for severe acidemia and 0.68 (95% CI, 0.61-0.75) for acidemia. For the final three h...