Interpretable machine learning to identify important predictors of birth weight: A prospective cohort study

Area of Science:

• Perinatal Medicine
• Biostatistics
• Machine Learning in Healthcare

Background:

• Accurate prediction of infant birth weight is clinically significant.
• Identifying risk factors for low or high birth weight is essential for maternal and child health.
• Interpretable machine learning offers a novel approach to birth weight prediction and risk factor identification.

Purpose of the Study:

• To employ interpretable machine learning models for predicting infant birth weight.
• To identify the most significant predictors influencing infant birth weight.
• To evaluate the performance of various supervised learning models in birth weight prediction.

Main Methods:

• A prospective cohort study involving 4,754 mother-child dyads was conducted.
• Twenty-four features including maternal and paternal biometrics, lifestyle factors, and biomarkers were collected.
• Eight supervised learning models, including linear regression, SVR, and ensemble methods, were utilized to predict birth weight, with accuracy assessed by Root Mean Squared Error (RMSE).

Main Results:

• The voting ensemble, linear regression, and SVR models demonstrated superior performance.
• The top five predictors for infant birth weight were identified as gestational age, fetal sex, preterm birth, mother's height, and pre-pregnancy BMI.
• Mother's height and pre-pregnancy BMI remained significant predictors even after incorporating ultrasound-measured fetal growth indicators.

Conclusions:

• Mother's height and pre-pregnancy BMI are critical factors in predicting infant birth weight.
• Interpretable machine learning presents a valuable and promising methodology for birth weight prediction and understanding its determinants.