Artificial Intelligence–Driven Prediction of Adverse Maternal and Perinatal Outcomes: Development and Validation of a Personalized Risk Assessment Model

Abstract

Background: Adverse maternal and perinatal outcomes are still important causes of morbidity and mortality in the world. Early accurate risk stratification can enhance antenatal care and achieve better pregnancy outcomes. This study set out to create and externally validate an explainable artificial intelligence (AI) based personalised risk assessment tool for predicting adverse maternal and perinatal outcomes.

Materials and methods: Between 2020-25, a multicenter retrospective cohort study of 5,248 pregnant women was carried out. Data regarding demographic, clinical, obstetric, laboratory and fetal assessment parameters were retrieved from electronic medical records. The primary outcome was composite outcome which amounted to preeclampsia, gestational diabetes mellitus, preterm delivery, fetal growth restriction, still birth, admission to the neonatal intensive care unit and perinatal death. Logistic regression, random forest, neural network and XGBoost models were created and evaluated. Both internal and external validation was carried out. SHAP analysis was used to measure the interpretability of models.

Results: There were 1,162 pregnancies with adverse outcomes (22.1%). XGBoost demonstrated the best performance with an AUROC of 0.914 (95% CI: 0.899–0.928), sensitivity of 87.6%, specificity of 86.9%, and F1-score of 0.84. Fetal growth parameters, previous preeclampsia, fasting plasma glucose, body mass index, maternal age and systolic blood pressure were the most influential predictors. External validation demonstrated excellent AUROC (0.908) and slope (0.96) measures of discrimination and calibration, respectively.

Conclusion: The interpretable model built with artificial intelligence was able to successfully forecast adverse maternal and perinatal outcomes, and had high generalizability. It could be combined into the regular antenatal manager to help tailor risk stratification and preventive measures.

Keywords: Artificial intelligence, Machine learning, Maternal outcomes, Perinatal outcomes, Risk prediction, Explainable AI, XGBoost, Obstetrics.