Development of Machine Learning-Based Risk Prediction Models to Predict Rapid Weight Gain in Infants: Analysis of...

Area of Science:

• Pediatric Health
• Machine Learning Applications
• Obesity Prevention

Background:

• Rapid weight gain (RWG) in infancy is a significant predictor of later obesity.
• Early identification of infant RWG can enable timely obesity risk assessment.
• RWG is defined as an upward shift across centile lines on infant weight growth charts.

Purpose of the Study:

• To develop and validate machine learning (ML) risk prediction models for identifying infant RWG by age 1 year.
• To leverage routinely collected prenatal and early postnatal data for risk prediction.
• To assess the performance of various ML algorithms in predicting infant RWG.

Main Methods:

• Pooled data from 7 Australian and New Zealand cohorts (n=5233) were used for model development and validation.
• Eight ML algorithms were trained to predict RWG using factors like maternal pre-pregnancy weight, smoking, gestational age, parity, infant sex, birth weight, breastfeeding, and solids introduction timing.
• Data were split into training (70%) and testing (30%) sets, with 5-fold cross-validation used for model consistency evaluation.

Main Results:

• The average prevalence of infant RWG was 27%.
• ML models demonstrated acceptable to excellent discrimination, with Area Under the ROC Curve (AUC) ranging from 0.75 to 0.86 in the training set.
• The Gradient Boosting model showed the best predictive accuracy, with validation in the test set showing good prediction ability for true positives (accuracy and sensitivity >0.75).

Conclusions:

• The study successfully developed the first ML-based risk prediction models for infant RWG with acceptable accuracy.
• These models can be integrated into routine child growth monitoring systems.
• The models hold potential for facilitating population-wide early obesity risk assessment in primary healthcare settings.