Machine Learning Prediction of Obesity Development in Children With Overweight Using Longitudinal Body Composition Data.

Obesity developed in approximately one-third of male and one-quarter of female overweight children during follow-up.

• 32.3% of males developed obesity (BMI ≥ 95th percentile) during follow-up
• 25.4% of females developed obesity during follow-up
• Study population consisted of 2801 overweight Korean children aged 7-9 years
• Children with overweight were defined as having BMI between the 85th and 95th percentiles

XGBoost machine learning models achieved moderate predictive performance for obesity development with sex-stratified results.

• Male model achieved AUROC of 0.671 (95% CI: 0.619-0.721)
• Female model achieved AUROC of 0.652 (95% CI: 0.589-0.700)
• Models were evaluated using bootstrap validation
• Sex-stratified models were developed separately for males and females

Key predictors of obesity development identified by SHAP analysis included standardised weight and adiposity measures, height-adjusted skeletal muscle mass, and growth velocity parameters.

• Shapley Additive exPlanations (SHAP) was used to identify key predictive features
• Predictors included standardised deviation scores in addition to raw anthropometric values
• Growth velocity parameters were identified as important predictors beyond static anthropometric data
• Bioelectrical impedance analysis-derived body composition measures, including height-adjusted skeletal muscle mass, were among key features

Children with overweight have substantially elevated rates of obesity progression compared to normal-weight peers.

• Obesity progression rates in overweight children are 10- to 20-fold higher than normal-weight peers
• This elevated risk identifies children with overweight as a critical target for obesity prevention
• The study population was drawn from Korean children aged 7-9 years with BMI between the 85th and 95th percentiles

The study integrated both anthropometric measurements and bioelectrical impedance analysis parameters in longitudinal XGBoost models to predict obesity.

• Longitudinal data from 2801 overweight Korean children were analysed
• Models incorporated anthropometric measurements, bioelectrical impedance-derived body composition, standardised deviation scores, and growth velocity parameters
• XGBoost was the machine learning algorithm used
• The outcome was defined as BMI ≥ 95th percentile