Prediction of pre-eclampsia with machine learning approaches: Leveraging important information from routinely...

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Public Health

Background:

• Pre-eclampsia (PE) is a major global cause of maternal and perinatal mortality.
• Routine data-based PE prediction offers wide applicability, especially in low-resource settings.
• Early intervention for high-risk pregnancies can mitigate PE complications.

Purpose of the Study:

• To predict PE using routinely collected maternal health data.
• To identify the optimal machine learning (ML) model for PE prediction.
• To compare ML model performance against logistic regression.

Main Methods:

• Utilized data from 48,250 singleton pregnancies (2016-2021).
• Employed supervised ML models with maternal characteristics as predictors.
• Assessed performance using AUC, calibration plots, and Shapley value analysis.

Main Results:

• Random forest achieved an AUC of 0.84, but with calibration limitations.
• Extreme gradient boosting showed an AUC of 0.77 with good calibration.
• Logistic regression yielded an AUC of 0.75 with perfect calibration.
• Key predictors included nulliparity, BMI, prior PE, maternal age, and hypertension/diabetes history.

Conclusions:

• Two ML models demonstrated high performance in predicting PE risk from routine data.
• External validation is crucial to confirm generalizability and refine prediction.
• Further research should utilize standardized prognostic factors for broader applicability.