Noninvasive PPROM risk stratification with explainable AI using routine antenatal CRP and albumin
摘要
Premature rupture of membranes (PROM) and preterm PROM (PPROM) are significant obstetric complications, affecting 10–20% and 3% of pregnancies globally and constituting a major cause of preterm birth and neonatal morbidity. While intrauterine infection is a key underlying driver, timely identification of high-risk cases remains a clinical challenge. We developed a noninvasive artificial intelligence (AI) predictive framework using a large-scale dataset of 114,601 antenatal laboratory and ultrasound measurements. Our multimodal model integrated longitudinal clinical measures, including maternal demographics, laboratory results, and ultrasound data. The optimized XGBoost algorithm achieved high performance (AUC: 0.952) and identified elevated C-reactive protein (CRP) and decreased albumin levels in late pregnancy as top features, underscoring the critical roles of inflammation and oxidative stress. SHAP analysis ensured interpretability, unveiling dynamic risk patterns across gestation. This AI-driven approach provides a precise, noninvasive tool for late-gestation PPROM risk stratification, paving the way for translation into precision obstetric care.