Predictive Models Using Machine Learning to Identify Fetal Growth Restriction in Patients With Preeclampsia:...

Area of Science:

• Obstetrics and Gynecology
• Machine Learning in Healthcare
• Perinatal Medicine

Background:

• Fetal growth restriction (FGR) is a significant complication of preeclampsia, increasing risks for newborns.
• Existing FGR prediction methods lack clinical explainability and are often class-biased, delaying timely interventions.

Purpose of the Study:

• To develop an accurate and explainable machine learning (ML) model for predicting FGR in preeclampsia patients.
• To improve early detection and intervention for FGR, mitigating neonatal morbidity and mortality.

Main Methods:

• A retrospective case-control study analyzed 38 features from preeclampsia patients with and without FGR.
• Multiple ML algorithms were evaluated, with Random Forest (RF) selected for its performance.
• The Shapley Additive Explanations (SHAP) method was used for feature importance ranking and model interpretability.

Main Results:

• An interpretable RF model with 9 key features accurately predicted FGR in both training and external datasets (AUC 0.83 and 0.82, respectively).
• Urinary protein quantification, gestational week of delivery, and umbilical artery systolic-to-diastolic ratio were the most significant predictors.
• The model was translated into a web tool for practical clinical use.

Conclusions:

• An accurate and explainable ML model for predicting FGR in preeclampsia has been successfully developed.
• SHAP analysis effectively identified critical risk factors, addressing the 'black box' issue of ML in clinical applications.
• The tool facilitates clinical decision-making, potentially improving perinatal outcomes.