Machine Learning for Dynamic and Short-Term Prediction of Preeclampsia Using Routine Clinical Data

Area of Science:

• Maternal-fetal medicine
• Clinical informatics
• Machine learning in healthcare

Background:

• Preeclampsia poses significant risks to maternal and perinatal health due to unpredictable onset.
• Current prediction methods often lack generalizability and clinical utility, relying on specialized biomarkers or limited data.

Purpose of the Study:

• To develop and validate machine learning models for dynamic, short-term prediction of preeclampsia onset.
• Utilize longitudinal electronic health record (EHR) data for prediction.

Main Methods:

• Retrospective, multisite cohort study involving over 58,000 pregnancies.
• Developed extreme gradient boosting models using routine EHR data (blood pressure, lab results, demographics).
• Validated models using nested cross-validation and external transfer learning techniques.

Main Results:

• Models achieved strong predictive performance, with AUCs up to 0.863 at training and 0.834 at validation, peaking at 34 weeks gestation.
• Blood pressure was the most significant predictor, with laboratory and demographic factors contributing at different gestational stages.
• Negative predictive values exceeded 0.993, indicating high reliability in ruling out preeclampsia.

Conclusions:

• Dynamic, short-term prediction of preeclampsia is feasible using readily available EHR data.
• This machine learning approach shows promise for earlier clinical intervention and is adaptable to various healthcare settings.