Body composition phenotyping of obesity in children aged 6-18 years: multi-strategy clustering and interpretable...

Area of Science:

• Pediatric Endocrinology
• Obesity Research
• Body Composition Analysis

Background:

• Childhood obesity heterogeneity is not fully captured by Body Mass Index (BMI).
• Non-invasive measures are needed for operational phenotyping of pediatric obesity.
• Existing metrics lack the granularity to describe diverse body compositions in obese youth.

Purpose of the Study:

• To identify body composition-based phenotypes in obese children aged 6-18 years.
• To utilize complementary clustering approaches for phenotype identification.
• To characterize the discriminative structure of these phenotypes using interpretable machine learning.

Main Methods:

• Cross-sectional study of 78 obese children (6-18 years).
• Analysis of 13 Bioelectrical Impedance Analysis (BIA)-derived indices using PCA-K-means, HCPC, and UMAP-DBSCAN.
• Evaluation of cluster quality, stability, and agreement using silhouette scores, Jaccard bootstrap, and Adjusted Rand Index (ARI).
• Assessment of separability with a random forest classifier and SHAP for interpretability.

Main Results:

• Four consistent phenotypes were identified: low-fat/high-muscle, balanced, high-fat/low-muscle, and mixed high-fat/high-muscle.
• The classification model achieved 85.9% accuracy, with macro AUC of 0.953 and micro AUC of 0.965.
• SHAP analysis highlighted fat mass and BMI as key discriminators, with opposing contributions from lean and skeletal muscle mass.

Conclusions:

• This study presents an exploratory, single-center framework for phenotyping childhood obesity based on body composition.
• The identified phenotypes provide a more detailed characterization than BMI alone.
• External validation with outcome-linked markers and imaging is required before clinical application.