Identifying First-Trimester Risk Factors for SGA-LGA Using Weighted Inheritance Voting Ensemble Learning

Area of Science:

• Obstetrics and Gynecology
• Neonatal Health
• Machine Learning in Healthcare

Background:

• Accurate classification of fetal growth (Small for Gestational Age - SGA, Large for Gestational Age - LGA) is vital for neonatal assessment.
• Identifying intrauterine growth restriction (SGA) and excessive fetal growth (LGA) early can optimize infant and maternal outcomes.
• Leveraging first-trimester data for prediction offers a window for timely medical intervention.

Purpose of the Study:

• To develop and validate a novel algorithm for predicting SGA, LGA, and Appropriate for Gestational Age (AGA) fetuses using first-trimester data.
• To identify significant latent maternal risk factors associated with SGA and LGA.
• To determine the importance of specific biochemical and maternal factors in predicting fetal growth classifications.

Main Methods:

• Analysis of data from 7943 pregnant women (424 SGA, 928 LGA, 6591 AGA) from 2015-2021.
• Development of the Weighted Inheritance Voting Ensemble Learning Algorithm (WIVELA) for classification.
• Application of relevance determination algorithms to identify key predictive features.

Main Results:

• The WIVELA algorithm achieved an average accuracy of 92.12% in 10-fold cross-validation, outperforming five other machine learning algorithms.
• Identified key maternal risk factors including first-trimester weight change, prepregnancy weight, height, age, and obstetric history.
• Highlighted the significance of first-trimester biomarkers such as HDL, TG, OGTT, TC, FPG, and LDL in reflecting SGA/LGA status.

Conclusions:

• The proposed WIVELA algorithm provides an accurate and reliable method for early classification of fetal growth conditions.
• Identification of latent maternal risk factors and key biomarkers enables proactive clinical monitoring and intervention.
• This approach supports enhanced prenatal healthcare strategies for managing fetal growth abnormalities and improving maternal-fetal outcomes.