Abstract
Background: Sarcopenia is a progressive and multifactorial condition linked to aging, malnutrition, and chronic diseases, presenting significant clinical and public health challenges. Current screening tools vary in complexity and diagnostic accuracy, emphasizing the need for simple, evidence-based predictive models suitable for settings with limited resources. Methods: A cross-sectional study was conducted among community-dwelling older adults to develop and internally validate hierarchical predictive models for sarcopenia using readily available primary care variables. Three models were built: (1) a basic clinical model (age, sex, BMI, calf circumference, and SARC-F), (2) a model including nutritional status (Mini Nutritional Assessment, MNA), and (3) an extended model adding bioelectrical impedance parameters (phase angle, PhA). Model performance was assessed using AUC, Brier score, Hosmer-Lemeshow test, and decision curve analysis. Results: The parsimonious model demonstrated excellent discrimination (AUC = 0.91) and good calibration (Hosmer-Lemeshow p = 0.36), while the extended model with MNA and PhA achieved the highest overall performance (AUC = 0.95; Brier = 0.064; p = 0.97). Incorporating MNA and PhA enhanced calibration and clinical utility, especially for risk probabilities between 0.10 and 0.40. Internal validation showed minimal optimism and stable coefficients, with BMI, sex, and PhA as consistent predictors. Conclusions: A model combining anthropometric, nutritional, and bioelectrical variables provides high diagnostic accuracy for sarcopenia while remaining practical for clinical use. Its stepwise design facilitates application at various healthcare levels, supporting early detection and targeted interventions in aging populations.