Prediction model of preeclampsia using machine learning based methods: a population based cohort study in China

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Maternal-Fetal Medicine

Background:

• Preeclampsia is a significant cause of maternal and perinatal morbidity with unknown pathogenesis.
• Early identification and intervention, such as aspirin therapy, are crucial for preeclampsia prevention.
• Developing effective prediction models is vital for early screening and risk stratification.

Purpose of the Study:

• To develop a robust machine learning-based prediction model for preeclampsia.
• To identify high-risk pregnancies for preeclampsia using early gestational markers.
• To provide an effective tool for early screening and prediction of preeclampsia.

Main Methods:

• A prospective cohort study included 5116 pregnant women.
• Maternal characteristics, biophysical, and biochemical markers (MAP, UtPI, PAPP-A, PLGF) were collected at 11-13+6 weeks' gestation.
• Five machine learning algorithms (Logistic Regression, Extra Trees, Voting, Gaussian Process, Stacking) were applied and validated using cross-validation.

Main Results:

• The Voting Classifier demonstrated superior performance in predicting preterm preeclampsia (AUC=0.884, DR=0.625 at 10% FPR).
• PLGF contribution was higher than PAPP-A in predicting overall preeclampsia, with the opposite trend for preterm preeclampsia.
• Machine learning model performance was comparable to the Fetal Medicine Foundation competing risk model.

Conclusions:

• Developed machine learning models offer an accessible tool for large-scale preeclampsia screening.
• Early prediction can help reduce the disease burden of preeclampsia.
• Improved maternal and fetal outcomes are anticipated through timely intervention based on accurate predictions.