ESPE Yearbook of Paediatric Endocrinology

Digit Health. 2025;11:20552076251331879. doi: 10.1177/20552076251331879

Brief Summary: This cross-sectional study provides a machine learning model using LightGBM and explainable AI techniques to estimate height in 54,374 children and adolescents aged 6–18 years based on body composition data from over 278,000 measurements. The model achieved high accuracy and identified soft lean mass (SLM), body fat mass percentage (BFMP), and skeletal muscle mass as key predictors. Explainable AI tools revealed interpretable relationships between body composition and height, offering insights into pediatric growth patterns.

This study underscores the clinical relevance of lean and fat mass in pediatric growth assessment, demonstrating their significant associations with height projections. While bone age-based height prediction remains the standard in routine clinical practice, this work introduces a novel approach by integrating multiple body composition variables for height estimation in children and adolescents.

However, the model’s applicability is limited by the exclusion of other influential factors such as genetic background, lifestyle behaviors, socioeconomic status, and health conditions. To enhance generalizability and robustness, further studies are warranted to validate the model across diverse ethnic and socioeconomic populations and to incorporate longitudinal data.

Finally, comparative evaluations against existing growth prediction models will also be essential to establish its clinical utility.