Machine Learning Models to Predict Risk of Maternal Morbidity and Mortality From Electronic Medical Record Data:...

Area of Science:

• Medical Informatics
• Public Health
• Machine Learning in Healthcare

Background:

• Over 80% of maternal deaths in the US are preventable, highlighting the need for proactive interventions.
• Machine learning (ML) models utilizing electronic medical records (EMRs) offer a potential strategy for predicting adverse maternal outcomes.
• Existing reviews often lack a comprehensive scope, focusing narrowly on specific outcomes or excluding the full development-to-implementation pipeline of ML models.

Purpose of the Study:

• To systematically review the evidence on ML models for predicting maternal morbidity and mortality risk (RO1).
• To assess the translation of these ML models into clinical applications for healthcare providers (RO2).
• To identify factors influencing the implementation of ML clinical applications in practice (RO3).

Main Methods:

• A comprehensive search was conducted across major databases (PubMed, CINAHL Plus, Scopus, Embase, IEEE Xplore) on February 20, 2023.
• Studies were limited to those using EMR data in healthcare settings, with rigorous title, abstract, and full-text screening by independent reviewers.
• Data extraction followed a structured template, and findings were summarized descriptively.

Main Results:

• Out of 480 identified studies, 39 were included, with over half published in 2022, predominantly from the US, China, and Israel.
• The majority of studies focused on predicting pregnancy and delivery outcomes, particularly cardiovascular risks, hypertensive disorders, gestational diabetes, and postpartum hemorrhage.
• While 30 studies used computable phenotypes and boosting methods were common, no studies reported on clinical applications (RO2) or implementation factors (RO3).

Conclusions:

• Future research should prioritize postpartum outcomes and enhance study transparency and reproducibility using reporting checklists.
• Significant gaps exist in the translation and implementation of ML models into clinical practice for maternal care.
• Expanded efforts are crucial to bridge the gap between ML model development and real-world clinical application to prevent maternal mortality.