Development and validation of a machine learning-based early prediction model for fetal acidosis in vaginal deliveries.

UNLABELLED: Fetal acidosis (FA) is a prevalent clinical symptom in newborns during the perinatal period, which can lead to higher risks of hypoxic ischemic encephalopathy, seizures, intraventricular hemorrhage and delayed developmental milestones. However, there is no effective tool for the risk assessment and prediction of FA in neonates. Therefore, our objective is to develop a FA prediction model in newborns by using machine-learning algorithms. Parturients and their neonates who delivered at the Fuzhou University Affiliated Provincial Hospital between January 2018 and December 2023 were included to identify optimal predictors and develop a prediction model for FA. To prospectively validate the model’s predictive performance, an independent cohort of newborns delivered at the Fujian Maternal and Child Health Hospital between June 2023 and March 2024 was enrolled. Model interpretability was assessed using the Shapley Additive Explanations (SHAP) method. Eventually, a total of 1,587 neonates at the Fuzhou University Affiliated Provincial Hospital constituted the development dataset (1110 cases were assigned to the training cohort, and 477 were included in the internal validation cohort), while 333 neonates from the Fujian Maternal and Child Health Hospital formed the prospective validation cohort. Among various machine learning algorithms evaluated, the Gaussian Naive Bayes (GNB) algorithm demonstrated the best performance across all validation types. The model achieved an area under the receiver operating characteristic curve (AUC) of 0.749 [95% CI: 0.679–0.818] in internal validation, and 0.763 [95% CI: 0.719–0.807] in external validation. This algorithm will be instrumental for identification of high-risk perinatal populations and it will facilitate targeted interventions in neonates. Trial registration: Chinese Clinical Trial Registry Identifier ChiCTR2300074180.

SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1038/s41598-026-45246-1.