AIMS: Obesity in children is heterogeneous, but BMI-based risk stratification does not capture important differences in body composition or cardiometabolic risk factors. We aimed to develop a body composition-based obesity phenotyping framework and examine its associations with cardiometabolic risk factor transitions. MATERIALS AND METHODS: Utilising nine body composition metrics (whole-body and regional fat/muscle mass index, visceral fat area) from 2262 children with obesity in a training cohort, we applied the Discriminative Dimensionality Reduction Tree algorithm to construct a continuous two-dimensional phenotypic manifold. Modified Poisson regression and spatial autocorrelation analyses were used to evaluate associations between the mapped spatial dimensions and 2-year cardiometabolic risk factors transitions (progression and recovery). The topological framework was externally validated in an independent cohort of 330 children with obesity. RESULTS: Obesity phenotypes were mapped onto two principal axes to define three clinical profiles (mixed fat-muscle elevation, adiposity-dominant, and muscle-dominant), with their underlying structure consistently reproduced across external replication cohort. Dimension 1 was positively associated with progression to hypertension (RR = 1.18, 95% CI 1.09-1.28), high LDL-C (1.29, 1.13-1.49), and hyperuricemia (1.22, 1.13-1.30) and inversely with hypertension recovery (0.84, 0.76-0.92). Dimension 2 showed positive association with hypertension progression (1.27, 1.03-1.58) and inverse association with high LDL-C recovery (0.48, 0.23-0.96). DDRTree-derived dimensions showed no clear predictive advantage over BMI or body composition metrics in most analyses, suggesting the DDRTree manifold should be used for exploratory visualisation rather than clinical prediction. CONCLUSIONS: Phenotypic manifold mapping of childhood obesity identifies body composition subtypes with divergent short-term cardiometabolic risk trajectories.
Xiao et al. (Sun,) studied this question.