Body Composition–Driven Phenotyping Reveals Obesity and Cardiometabolic Heterogeneity in Children Using a Tree-Like Representation

Diabetes Journal · 4 weeks ago · 48 views

Traditional measures such as BMI do not fully capture obesity-related metabolic risks. Here, we examined DXA-derived body composition patterns in children (n = 11,238) using the discriminative dimensionality reduction tree algorithm, which was validated in an independent cohort (n = 2,001). The tree structure revealed a continuous landscape of body composition phenotypes, with distinct spatial gradients for fat and lean mass. Within the phenotypic tree, fat-dominant phenotypes clustered in the region of low dimension 1 and low dimension 2 (corresponding to the lower left branches), where they demonstrated significant spatial overlap with high-risk clusters for hypertension, elevated LDL cholesterol (LDL-C), total cholesterol, and triglycerides (TG) (Moran I >0.3, all P < 0.001). Meanwhile, phenotypes characterized by elevated lean mass coupled with increased adiposity were localized in the region of low dimension 1 and high dimension 2 (upper left branches), demonstrating an increased risk of low HDL-C and high TG (Moran I >0.6, all P < 0.001). In contrast, lean-dominant phenotypes clustered in the region of high dimension 1 and high dimension 2 (upper right branches), which were associated with a relatively low risk of hyperglycemia, insulin resistance, and high LDL-C (Moran I >0.1, all P < 0.001). For practical utility, we developed a publicly available web application to map individual data on the reference tree architecture, allowing for assessment of cardiometabolic risk. Our findings highlight the value of more precise body composition measures for early identification and prevention of obesity-related health problems.Article HighlightsChildhood obesity shows large differences in body composition and health risk that are not well captured by BMI or simple metabolic classifications, prompting the need for more precise characterization.This study aimed to determine whether a data-driven framework integrating detailed body composition measures could better describe obesity-related phenotypic heterogeneity and its relationship with cardiometabolic risk in children.We delineated a continuous body composition manifold encompassing fat-dominant, lean-dominant, and concomitant high-mass phenotypes, which captured diverging cardiometabolic risk trajectories and yielded modest incremental improvements in risk prediction.These findings support more precise risk stratification and provide a practical tool to improve early identification and prevention of obesity-related health complications in children.