Fetal Acidemia Prediction Using a Random Forest Model With Cumulative Fetal Stress Assessment During Labor
Ayumu Ito, Yoko Nagayasu, Rei Mitsuhashi, Mio Kamiya, Eijiro Hayata, Masahiko NakataABSTRACT
Aim
To improve prediction of fetal acidemia during labor using a random forest model incorporating cumulative fetal stress assessment, focusing on an umbilical cord arterial pH < 7.20 as an early clinically relevant threshold.
Methods
This retrospective case‐enriched model development study was conducted at Toho University Omori Medical Center. Data were collected from full‐term vaginal deliveries between January 2017 and June 2021 for cases with umbilical cord arterial pH < 7.20 and between September 2018 and April 2019 for those with pH ≥ 7.20. Fetal acidemia was defined as pH < 7.20 to account for the time gap between fetal heart rate–based prediction and actual pH measurement at birth. A random forest model was developed using fetal heart rate data, including all decelerations within 30 min before delivery. Model performance was evaluated using five‐fold cross‐validation. Two datasets were created: Dataset 1 included fetal heart rate data alone, whereas Dataset 2 additionally included maternal and perinatal factors.
Results
Dataset 1 achieved a test accuracy of 91.0% and an area under the receiver operating characteristic curve (AUC) of 0.95. Dataset 2 showed improved test accuracy of 93.0% with a similar AUC. Variable importance analysis identified variability as the strongest predictor, followed by the duration of mild and severe prolonged decelerations.
Conclusions
In this case‐enriched derivation dataset, a random forest model incorporating cumulative CTG‐derived fetal stress variables showed high discrimination for umbilical artery pH < 7.20. External validation in independent consecutive populations is required before clinical implementation.