Clinical utility of the AITIS model for test-free identification of sarcopenia in patients with stage IV–V non-dialysis-dependent chronic kidney disease

Background Identifying sarcopenia in resource-limited settings presents a significant challenge. The objective of this study was to validate the clinical usefulness of the Artificial Intelligence to Identify Sarcopenia (AITIS) model, a test-free artificial intelligence model we previously proposed for identifying sarcopenia, in patients with chronic kidney disease (CKD). Methods This observational cross-sectional study enrolled 236 patients with stage IV–V CKD. Sarcopenia was diagnosed using the Asian Working Group for Sarcopenia 2019 criteria, which are based on handgrip strength, physical performance, and appendicular skeletal muscle mass measured by bioelectrical impedance analysis. Patient data, including age, sex, height, weight, and 20 functional measures, were used as predictors. The AITIS model was applied to predict sarcopenia, and its performance, explainability, and clinical usefulness were comprehensively analyzed. Results The study included 129 men and 107 women (median age = 54.5 years). Sarcopenia was diagnosed in 62 patients (26.3%). The three most common functional limitations reported were jogging 1 km ( n = 77, 32.6%), climbing stairs ( n = 56, 23.7%), and walking 1 km ( n = 13, 5.5%). In contrast, no difficulty was reported in getting in and out of bed, using the toilet, or controlling urination and defecation. The AITIS model demonstrated favorable performance in predicting sarcopenia within the study population [AUC (95% CI) = 0.792 (0.729–0.856), area under the precision-recall curve = 0.610, kappa = 0.398, accuracy (95% CI) = 0.758 (0.699–0.812), sensitivity = 0.597, specificity = 0.816]. Calibration curve analysis revealed good agreement between model predictions and the ground truth. Model performance improved in patients with less physical activity (AUC = 0.843, 95% CI = 0.764–0.922) and in those with stage IV CKD (AUC = 0.824, 95% CI = 0.739–0.909). The SHAP summary plot indicated that age, walking 1 km, and lifting 5 kg were the top three contributors to the model’s predictive ability. Decision curve analysis supported the model’s clinical usefulness. Conclusion The AITIS model demonstrates strong generalizability and performance in predicting sarcopenia in patients with stage IV–V CKD. These findings may enhance clinical decision-making and facilitate the development of novel strategies for managing sarcopenia in CKD patients.