A Multi-Parametric Risk Stratification Model for Early-Pregnancy Prediction of Preeclampsia: Integrating First- and Second-Trimester Biomarkers with Machine Learning

Preeclampsia remains a major contributor to maternal and perinatal morbidity. This study aimed to develop and internally evaluate an interpretable machine-learning framework integrating routinely available first- and second-trimester parameters for preclinical risk stratification. This retrospective case-control study included 524 singleton pregnancies (262 with preeclampsia and 262 normotensive controls). Seven prespecified predictors—12-week mean arterial pressure (MAP), pregnancy-associated plasma protein A (PAPP-A), alpha-fetoprotein (AFP), unconjugated estriol, eosinophil-to-lymphocyte ratio (ELR), neutrophil-to-lymphocyte ratio, and lymphocyte count—were evaluated using logistic regression, random forest, Gaussian naive Bayes, support vector machine, and k-nearest neighbors. Internal validation used stratified 10-fold cross-validation with identical fold assignments across classifiers. A five-variable weighted clinical risk score was also constructed. MAP showed the highest univariable discrimination (area under the receiver operating characteristic curve [AUC] = 0.891). In cross-validation, Gaussian naive Bayes achieved the numerically highest mean AUC (0.948 ± 0.024), accuracy (0.885 ± 0.031), recall (0.893 ± 0.044), and F1 score (0.886 ± 0.032), while random forest had the highest mean precision (0.891 ± 0.042). The weighted clinical risk score achieved an AUC of 0.885 and 81.9% accuracy. The framework showed strong internal discrimination but remains exploratory. Because the score incorporates approximately 16-week AFP and estriol measurements, it should not be interpreted as a purely first-trimester aspirin-selection model. Prospective external validation, calibration, and subtype-specific assessment are required before clinical use.