Abstract
Sarcopenia is associated with an elevated burden of depressive symptoms, yet screening tools may have limited accuracy and generalizability in this population. We developed and validated an interpretable machine-learning model to predict depressive symptoms risk among middle-aged and older adults with sarcopenia using National Health and Nutrition Examination Survey (NHANES) 2007-2020 data. In this cross-sectional study, we included 913 participants with sarcopenia aged ≥45 years from NHANES 2007-2020. Candidate predictors were selected using Boruta followed by least absolute shrinkage and selection operator (LASSO). Multiple machine-learning models were developed and internally validated for discrimination, calibration, and clinical utility. Shapley Additive exPlanations (SHAP) were used to support interpretability. Reporting followed the TRIPOD+AI guidance. Nine predictors were retained after Boruta-LASSO selection. In the validation set, the logistic regression model showed the best overall performance (AUC 0.794; Brier score 0.065). SHAP analysis highlighted key contributors including education level, sleep disorder, sex, poverty-income ratio, blood urea nitrogen, osteoarthritis, white blood cell count, absolute lymphocyte count, and body mass index. The final model was presented as a clinically usable nomogram for individualized depressive symptoms risk estimation. We developed a validated, interpretable machine-learning model for predicting depressive symptoms risk in middle-aged and older adults with sarcopenia using NHANES data. The nomogram may facilitate rapid risk stratification and targeted interventions to support risk stratification and targeted supportive care addressing both physical and mental health needs.