Prediction Model of Late Fetal Growth Restriction with Machine Learning Algorithms

Area of Science:

• Perinatal Medicine
• Clinical Prediction Modeling
• Machine Learning in Healthcare

Background:

• Late-onset fetal growth restriction (FGR) poses significant risks to newborns.
• Accurate prediction of FGR is crucial for timely intervention and improved outcomes.

Purpose of the Study:

• To develop and validate a clinical model for predicting late-onset FGR.
• To assess the utility of machine learning algorithms in FGR prediction.
• To create a simplified model for practical clinical application.

Main Methods:

• Retrospective study involving seven hospitals and over 32,000 patients (January 2009 - December 2020).
• Development of FGR prediction models using first-trimester (E1) and early third-trimester (T1) variables.
• Application of a machine learning algorithm (XGBoost) with embedded feature selection for model simplification and validation.

Main Results:

• The prediction model achieved an Area Under the Curve (AUC) of 0.78 at T1 in the training set and 0.73 in the test set.
• Area Under the Precision-Recall Curve (AUPR) was 0.31 at T1 (training) and 0.24 (test).
• A simplified model demonstrated comparable performance to the original model, indicating its potential for clinical use.

Conclusions:

• A simplified machine learning model effectively predicts late-onset FGR.
• The model shows promise for evaluating individual FGR risk in the early third trimester.
• This tool can support clinical decision-making and enhance perinatal care.