Distribution of Muscle Mass and Fat Mass to Identify the Risk of Sarcopenia and Sarcopenic Obesity in Adults via Machine Learning.

Objective: This study aimed to evaluate a sample of adults to define phenotypes on the basis of the association between adiposity and muscle mass via an unsupervised machine learning model. Methods: A cross-sectional study was conducted with 2710 adults, comprising women (n = 1907) and men (n = 803). The variables analyzed were absolute and relative muscle mass, the muscle mass index, appendicular skeletal muscle mass, body fat percentage, total fat mass, the trunk-to-limb fat ratio, and the fat-to-muscle ratio. K-means clustering was performed by sex. The clusters were validated via inertia and silhouette measures, and comparisons were performed via post hoc tests based on variance and Tukey's test. Results: Different phenotypes were identified by sex. In women, a phenotype characterized by compensatory muscle, normal/lean, and sarcopenic obesity is observed, whereas in men, phenotypes characterized by metabolic obesity, athletic/protector, and sarcopenia risk/lean are identified. For women, the sarcopenic obesity phenotype is characterized by the highest adiposity burden and a higher fat-to-muscle ratio (FMR), indicating a disproportion between fat mass and muscle quality. In men, the sarcopenia risk/lean phenotype is defined as having limited muscle mass and an unfavorable FMR, despite having body mass index (BMI) values comparable to those of nonobese individuals. Across the whole sample, women presented higher FMR values and greater variability than men did. All between-phenotype differences were statistically significant (p < 0.001). Conclusions: Unsupervised machine learning identified biologically distinct body composition phenotypes that were not adequately captured by BMI alone.