Beyond MVPA minutes: DDRTree-derived activity phenotypes expose metabolic differences in children with obesity

Background: Childhood obesity is linked to early cardiometabolic disturbances. However, the amount of moderate-to-vigorous physical activity (MVPA) alone may not fully capture meaningful behavioral variability. Patterns such as prolonged sedentary time, fragmentation of movement, and the distribution of activity across the day may independently influence metabolic health. Objective: To characterize multidimensional physical activity (PA) profiles in children and adolescents with obesity using discriminative dimensionality reduction via learning a tree (DDRTree) applied to accelerometer data, and to evaluate their relationships with body composition and cardiometabolic risk beyond MVPA duration. Methods: In this cross-sectional study, 91 youths with obesity (39 girls, 43%), aged 6–17 years (mean 11.2 years), wore a thigh-mounted accelerometer for seven consecutive days. Multiple accelerometry-derived variables were extracted, including activity volume, intensity distribution, sedentary accumulation, fragmentation metrics, and transitions between states. DDRTree was used to identify PA phenotypes. Anthropometric measures, estimated body composition, indices of insulin sensitivity, and a continuous metabolic syndrome risk score were compared across phenotypes using ANCOVA adjusted for age and sex. Results: Three distinct PA phenotypes emerged: Sedentary Pattern (24%), Light Activity Pattern (55%), and Higher Activity Pattern (21%). Only three participants (3.3%) achieved current MVPA recommendations. Compared with the Sedentary Pattern, individuals in the Higher Activity Pattern accumulated roughly 122 fewer minutes per day of sedentary time (p<0.0001) and about 16.5 more minutes per day of MVPA (p<0.001). This group also displayed lower fat mass percentage (38.8% vs 41.6%; p=0.045), higher fat-free mass percentage (61.2% vs 58.4%; p=0.037), greater insulin sensitivity (SPISE; mean difference +0.90, 95% CI 0.14–1.66; p=0.016), and a reduced overall metabolic risk score (mean difference −0.42, 95% CI −0.78 to −0.06; p=0.017). Conclusions: Multidimensional activity profiles derived from accelerometry reveal clinically relevant differences in body composition and cardiometabolic risk that are not fully explained by MVPA duration alone. These findings support the use of phenotype-based approaches to improve risk stratification and guide tailored interventions in pediatric obesity.