Machine Learning for Fetal Growth Prediction.

Area of Science:

• Obstetrics
• Perinatal Medicine
• Biostatistics

Background:

• Birthweight is commonly used as a proxy for estimated fetal weight (EFW).
• This practice has limitations, potentially leading to misinterpretations of fetal growth and health.
• Accurate EFW is crucial for assessing fetal well-being and identifying growth abnormalities.

Purpose of the Study:

• To evaluate machine learning and regression models for predicting EFW using data available at birth.
• To compare the performance of different predictive models, including generalized boosted models and quantile regression.
• To re-evaluate the association between maternal smoking and small-for-gestational-age (SGA) birth using accurate EFW standards.

Main Methods:

• Utilized two datasets: 18,517 pregnancies (31,948 visits) and 240 high-risk pregnancies.
• Applied linear regression, quantile regression, random forests, Bayesian additive regression trees, and generalized boosted models.
• Assessed model performance using mean squared error and median absolute deviation criteria.

Main Results:

• Generalized boosted models demonstrated the best overall performance in predicting EFW.
• Quantile regression was the top-performing regression-based approach.
• The association between smoking and SGA was significantly altered when using EFW standards compared to birthweight standards, with the risk ratio decreasing substantially.

Conclusions:

• Machine learning algorithms show promise for estimating EFW from birth data, addressing limitations of using birthweight as a proxy.
• Accurate EFW standards are essential for correctly assessing the impact of factors like smoking on fetal growth.
• The study highlights the need for improved methods to estimate fetal weight for better clinical and research outcomes.