Fig. 1. Overview of study design and model development.

a The workflow of the study outlines the cohort construction, patient characteristics extraction, dataset splitting into training and testing datasets (including subdivision into antepartum, intrapartum, and postpartum), feature engineering, feature selection, machine learning models (cross-subject validation) and final evaluation. b The proposed eMerge algorithm to identify preeclampsia (PE) patients to construct the binary prediction problem. c The schematic of 19 timeline models including: monthly models, weeks 4–20; biweekly models, weeks 22–34; weekly models, weeks 35–39; intrapartum and postpartum model.