Artificial Intelligence Applications in Obstetric Risk Prediction: A Systematic Review of Machine Learning Models for Preeclampsia

Abstract

Preeclampsia remains a leading cause of maternal and perinatal morbidity and mortality worldwide. While traditional prediction models have shown limited accuracy, machine learning (ML) approaches offer promising alternatives by handling complex, non-linear relationships in multidimensional datasets. This systematic review evaluates the performance, methodological quality, and clinical applicability of ML models for preeclampsia prediction. Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, we searched five databases (PubMed, Embase, Scopus, Web of Science, and Cochrane Library) for studies published until April 15, 2025, and included studies that developed or validated ML models predicting preeclampsia. Risk of bias was assessed using the Prediction Model Risk-of-Bias Assessment Tool (PROBAST). Eleven studies (n = 11, comprising 116,253 pregnancies) were included. Ensemble methods (XGBoost, Random Forest) demonstrated superior performance, with area under the curve (AUCs) ranging from 0.84 to 0.973. Key predictors included mean arterial pressure, prior preeclampsia, and the biomarkers placental growth factor (PlGF) and pregnancy-associated plasma protein A (PAPP-A). Seven studies (63.6%) showed low overall risk of bias, while three (27.3%) had high risk due to analytical limitations. Only three studies (27.3%) conducted external validation. ML models, particularly ensemble methods, show excellent discriminative ability for preeclampsia prediction. However, heterogeneity in predictors and limited external validation constrain clinical translation. Future research should prioritize prospective validation studies with standardized outcome definitions and predictor sets.

Topics & Concepts