The population-level heterogeneity of obesity has yet to be systematically investigated. We aimed to apply the data dimensionality reduction tree (DDRTree) method to obese individuals and to examine how distinct phenotypic patterns align with different outcomes. To characterize the heterogeneity of obesity, a two-dimensional (2D) tree structure based on the DDRTree algorithm was employed. Associations between embedding dimensions and metabolic traits, obesity-related indices, and clinical outcomes were evaluated using multivariable linear, logistic, and Cox proportional hazards regression models. The DDRTree revealed distinct, dimension-specific phenotypic patterns. We found that dimension 1 was strongly associated with insulin resistance, dysglycemia, visceral adiposity, subclinical atherosclerosis, hyperuricemia, and early renal injury, including microalbuminuria (MAU) (all P < 0.001). Dimension 2 was more closely aligned with β-cell function and was associated with all-cause mortality (ACM) in the CHARLS cohort ( P < 0.05). Individuals in the upper-right section of the tree exhibited a higher risk of vascular abnormalities, while the lower-right region clustered obese individuals with pronounced insulin resistance, hyperuricemia, and hyperglycemia. Obesity-related indices demonstrated heterogeneity: waist-based measures (WC, WHTR, ABSI, VAT) consistently aligned with Dimension 1, while lipid- and liver-related indices (LAP, VAI, FLI) showed enrichment along combined phenotypic gradients. Our findings demonstrate that DDRTree can be applied to obese populations to characterize continuous phenotypic heterogeneity and its associations with metabolic and clinical outcomes. This phenotypic mapping framework may support risk-oriented stratification of obese individuals in population-based settings. Not applicable. Not applicable.
Zhao et al. (Sat,) studied this question.