Predictive model for sarcopenia in chronic kidney disease: a nomogram and machine learning approach using CHARLS data.

Background: Sarcopenia frequently occurs as a complication among individuals with chronic kidney disease (CKD), contributing to poorer clinical outcomes. This research aimed to create and assess a predictive model for the risk of sarcopenia in CKD patients, utilizing data obtained from the China Health and Retirement Longitudinal Study (CHARLS). Methods: Sarcopenia was diagnosed based on the Asian Working Group for Sarcopenia (AWGS 2019) criteria, including low muscle strength, reduced physical performance, and low muscle mass. The 2015 CHARLS data were split randomly into a training set (70%) and a testing set (30%). Forty-nine variables encompassing socio-demographic, behavioral, health status, and biochemical factors were analyzed. LASSO regression identified the most relevant predictors, and a logistic regression model was used to explore factors associated with sarcopenia. A nomogram was developed for risk prediction. Model accuracy was evaluated using calibration curves, while predictive performance was assessed through receiver operating characteristic (ROC) and decision curve analysis (DCA). Four machine learning algorithms were utilized, with the optimal model undergoing hyperparameter optimization to evaluate the significance of predictive factors. Results: A total of 1,092 CKD patients were included, with 231 (21.2%) diagnosed with sarcopenia. Multivariate logistic regression revealed that age, waist circumference, LDL-C, HDL-C, triglycerides, and diastolic blood pressure are significant predictors. These factors were used to construct the nomogram. The predictive model achieved an AUC of 0.886 (95% CI: 0.858-0.912) in the training set and 0.859 (95% CI: 0.811-0.908) in the validation set. Calibration curves showed good agreement between predicted and actual outcomes. ROC and DCA analyses confirmed the model's strong predictive performance. The Gradient Boosting Machine (GBM) outperformed other machine learning models. Applying Bayesian optimization to the GBM achieved an AUC of 0.933 (95% CI: 0.913-0.953) on the training set and 0.932 (95% CI: 0.905-0.960) on the validation set. SHAP values identified age and waist circumference as the most influential factors. Conclusion: The nomogram provides a reliable tool for predicting sarcopenia in CKD patients. The GBM model exhibits strong predictive accuracy, positioning it as a valuable tool for clinical risk assessment and management of sarcopenia in this population.