Machine Learning Algorithms Versus Classical Regression Models in Pre-Eclampsia Prediction: A Systematic Review.

Area of Science:

• Medical Informatics
• Computational Biology
• Obstetrics and Gynecology

Background:

• Healthcare risk prediction is increasingly utilizing machine learning (ML) approaches.
• Pre-eclampsia prediction remains a critical area for improving maternal and infant outcomes.
• Classical regression models have been the standard for prognostic factor analysis.

Purpose of the Study:

• To synthesize existing literature comparing machine learning (ML) and classical regression models for pre-eclampsia risk prediction.
• To identify key prognostic factors for pre-eclampsia identified in both ML and regression studies.
• To compare the predictive performance of ML algorithms against classical regression models.

Main Methods:

• Systematic literature review of 9382 retrieved studies, including 82 selected publications.
• Analysis of 84 classical regression models and 6 purely ML algorithms.
• Comparative analysis of 10 publications reporting both ML and classical regression models on the same datasets.

Main Results:

• Machine learning (ML) algorithms demonstrated superior prediction performance compared to classical regression models for pre-eclampsia.
• Random forest (AUC=0.94) and extreme gradient boosting (AUC=0.92) were top-performing ML algorithms.
• Frequent prognostic factors included age, BMI, medical conditions, parity, and specific biomarkers (e.g., placental growth factor).

Conclusions:

• ML algorithms, particularly random forest and boosting methods, offer enhanced prediction accuracy for pre-eclampsia.
• Future research should standardize evaluation metrics and datasets for direct comparison between ML and classical regression.
• External validation of ML algorithms is crucial to assess their generalizability in diverse clinical settings.