Prediction of fetal growth restriction using machine learning algorithms

Objective: To investigate the significant clinical and anamnestic predictors of fetal growth restriction (FGR) and develop effective predictive models using machine learning methods (MLM).

Materials and methods: This retrospective study included 620 pregnant women who were observed and delivered at the V.I. Kulakov NMRC for OG&P, Ministry of Health of Russia. The study group comprised 300 patients with FGR, while the control group included 320 patients with healthy pregnancies. An analysis of the clinical and anamnestic data was conducted to build MLM models, including logistic regression and random forest.

Results: The logistic regression model identified the following predictors: age over 40 years, height less than 1.60 m, chronic arterial hypertension, smoking, a history of FGR, and threatened miscarriage in the first trimester with the formation of retrochorial hematoma and bleeding. This model predicts the development of FGR with a sensitivity of 73% and specificity of 80% (AUC 0.81). An alternative model constructed using random forest demonstrated an increased sensitivity of 78% and a decreased specificity of 74% (AUC 0.79). Within the random forest framework, the most significant contributors to the accuracy of the prognosis were age over 40 years, height less than 1.60 m, chronic arterial hypertension, a history of surgery resulting in a uterine scar, a history of FGR, and threatened miscarriage in the first trimester with retrochorial hematoma without bleeding.

Conclusion: Both models exhibited high predictive value for screening for FGR. Logistic regression offers interpretability, whereas random forest enhances the accuracy by accounting for nonlinear relationships. Implementing these models in clinical practice will optimize the monitoring of pregnant women at risk.