Abstract
OBJECTIVE: Early-onset preeclampsia (EOPE) represents a particularly severe clinical subtype of preeclampsia (PE) and is frequently complicated by placental abruption, which can result in serious maternal and fetal morbidity or mortality. This study aimed to develop and validate an interpretable machine learning (IML) model for predicting placental abruption in patients with EOPE. METHODS: A retrospective cohort of 580 EOPE patients who delivered between January 2021 and June 2025 was analyzed and randomly divided into training (70%) and validation (30%) sets. Dual-step feature selection combining LASSO regression and the Boruta algorithm identified the most relevant predictors. Six supervised algorithms, including decision tree (DT), k-nearest neighbor (KNN), logistic regression, random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost), were developed and compared. Model performance was evaluated using AUC, F1 score, calibration curve, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) were employed for model interpretation. RESULTS: Eight optimal predictors were selected: urinary protein, placental growth factor (PlGF), diastolic blood pressure (DBP), age, fibrinogen (FIB), prepregnancy BMI, disease severity, and smoking during pregnancy. The RF model achieved the best performance (training AUC = 0.968; validation AUC = 0.894), along with the highest accuracy and F1 score among all algorithms. Calibration curves showed strong consistency between predicted and observed probabilities, and DCA confirmed greater net clinical benefit across a wide range of threshold probabilities. The confusion matrix demonstrated high sensitivity and specificity, indicating stable classification performance. SHAP analysis revealed that urinary protein, PlGF, FIB, and DBP were the dominant predictors, where elevated urinary protein and DBP and reduced FIB and PlGF significantly increased abruption risk. CONCLUSION: The SHAP-based RF model demonstrated high predictive accuracy and interpretability, providing a transparent, data-driven framework for individualized risk assessment of placental abruption in EOPE. This interpretable approach may facilitate early risk identification and personalized management in clinical practice.