Machine learning-enabled maternal risk assessment for women with pre-eclampsia (the PIERS-ML model): a modelling...

Area of Science:

• Maternal Health
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Pre-eclampsia affects 2-4% of pregnancies, posing a significant global risk for maternal mortality and morbidity.
• Current methods for assessing pre-eclampsia risk may not be sufficiently accurate or responsive for clinical decision-making.

Purpose of the Study:

• To develop and validate a novel machine learning (ML)-based model to predict adverse maternal outcomes in women with pre-eclampsia.
• To create a clinical setting-responsive tool for ruling out and ruling in severe maternal complications.

Main Methods:

• Utilized health system, demographic, and clinical data from 8843 patients across 11 countries for model development.
• Employed random forest ML methods and ten-fold cross-validation on a development dataset (75%) and validation on the remaining 25%.
• External validation was performed on 2901 inpatient women in England; predictive accuracy assessed using AUROC and likelihood ratios.

Main Results:

• The PIERS-ML model demonstrated high accuracy (AUROC 0.80) compared to the logistic regression model (AUROC 0.68).
• PIERS-ML effectively categorized women into risk groups: very low (0% adverse events), low (2%), moderate (5%), high (26%), and very high (91%) risk within 48 hours.
• External validation confirmed accurate risk classification, with 0% adverse events in the very low-risk group.

Conclusions:

• The PIERS-ML model significantly improves the identification of women with pre-eclampsia at the lowest and greatest risk of severe adverse maternal outcomes.
• This tool can support accurate clinical guidance for patients, families, and healthcare providers, potentially improving maternal care.
• The model's performance in both internal and external validation underscores its potential for widespread clinical application.