MACHINE LEARNING–BASED EARLY PREDICTION OF SEVERE PREECLAMPSIA USING ELECTRONIC MATERNAL HEALTH RECORDS IN PAKISTAN

Abstract

Preeclampsia is a major pregnancy-related complication that can rapidly progress to severe maternal and fetal outcomes, particularly where timely risk identification and referral are limited. This study examined the potential of machine learning (ML) to predict severe preeclampsia using routinely available electronic maternal health records in Pakistan. A retrospective quantitative design was proposed using demographic, obstetric, clinical, and laboratory variables, including maternal age, BMI, previous preeclampsia, hypertension, diabetes, gestational age, and blood-pressure measurements. A dataset of 1,000 maternal records was used to demonstrate model development and evaluation. Logistic Regression, Random Forest, Support Vector Machine, and XGBoost were compared using AUC, accuracy, sensitivity, specificity, and F1-score. In the analysis, XGBoost achieved the highest predictive performance (AUC = 0.91), followed by Random Forest (AUC = 0.89), SVM (AUC = 0.86), and Logistic Regression (AUC = 0.82). Blood pressure, previous preeclampsia, BMI, chronic hypertension, and diabetes emerged as important predictors. The findings suggest that ML may support early risk stratification and clinical decision-making for severe preeclampsia. However, the reported results are based on synthetic data; therefore, they do not represent actual clinical outcomes from Pakistani patients. Future research should validate the proposed framework using large, multicenter Pakistani datasets and prospective clinical studies.