Advancing machine learning tools for early prediction and clinical diagnosis of pre-eclampsia

The global burden of pre-eclampsia is rising, posing significant challenges to women's health, particularly in cases of multiple organ failure. It continues to contribute to 14 % of maternal deaths during pregnancy, with co-morbidities further complicating maternal and fetal outcomes. Risk assessment in pre-eclampsia (early and late) is critical. It remains a challenge due to the limitations of non-specific biomarkers and the discomfort associated with invasive diagnostic tests, making them unsuitable for remote settings. This study explores the integration of advanced machine learning (ML) models for analyzing urine- and serum-based biomarkers to enhance predictive accuracy. ML algorithms, including CatBoost, LightGBM, XGBoost, and Random Forest (RF), were evaluated using clinical datasets of 11,006 pregnant women from Yonsei University Healthcare Centre. The study findings indicate that CatBoost achieved the highest accuracy (∼92.12 % with 10-fold cross-validation), followed closely by RF, LightGBM, and XGBoost. Feature importance plots demonstrated the relevance of selected features in enhancing diagnostic precision. These results underscore the potential of ML-driven analysis for risk assessment and associated diagnosis in variable real-time resource settings, improving maternal and fetal health outcomes in pre-eclampsia.