Predictive model of sarcopenia in chronic kidney disease: an integrated approach of bioinformatics, machine learning, and clinical validation.

Background: Sarcopenia represents a prevalent and clinically significant complication of chronic kidney disease (CKD), featuring insidious onset and diagnostic complexity. Elucidating the underlying pathogenic factors of CKD-related sarcopenia and constructing reliable predictive models are critical to optimizing clinical prognosis and patient outcomes. Methods: This study enrolled two independent cohorts to develop and validate a predictive model for sarcopenia in patients with chronic kidney disease (CKD). The primary cohort comprised 2,979 participants from the National Health and Nutrition Examination Survey (NHANES) database (2011-2018), while the independent external validation cohort consisted of 428 CKD patients recruited from the Affiliated Hospital of Southwest Medical University. Predictive variables were selected via least absolute shrinkage and selection operator (Lasso) regression, followed by multivariable logistic regression to identify independent predictors. The NHANES dataset was randomly partitioned into a training set (n = 2,085) and testing set (n = 894) at a 7:3 ratio. Baseline characteristics were compared across the training, testing, and external validation cohorts to evaluate inter-cohort heterogeneity. Furthermore, five machine learning algorithms, including Gradient Boosting (XGBoost), Logistic Regression (LR), Light Gradient Boosting Machine (LightGBM), Random Forest (RF), and Multilayer Perceptron (MLP), were constructed and hyperparameter-optimized using the training dataset. The best-performing model was selected based on predefined performance metrics and further assessed via ten-fold cross-validation to confirm internal robustness. The optimized model was subsequently validated in the external cohort to evaluate its generalizability to real-world CKD populations. Finally, Shapley Additive Explanations (SHAP) were applied to interpret the model at both the global and individual levels, elucidating the contribution of each predictor to sarcopenia risk. Results: The final model included 12 predictors, with XGBoost demonstrating robust superiority across cohorts. It achieved an AUC of 0.890 (95% CI: 0.860-0.909) for internal testing set, 0.818 (95% CI: 0.790-0.885) for external validation, and 0.888 (95% CI: 0.880-0.896) in the internal validation. The external Brier score was 0.156, indicating good calibration and reliable risk prediction. Decision curve analysis (DCA) confirmed superior net clinical benefit over a wide threshold range, supporting clinical applicability. SHAP analysis further identified gender, BMI, and weight as the top three dominant predictors. Conclusion: In this study, we developed an XGBoost-based model to predict sarcopenia in patients with chronic kidney disease (CKD). Importantly, to our knowledge, this is the first study to apply SHAP values to interpret model predictions in the context of CKD-related sarcopenia, thereby significantly enhancing the transparency and interpretability of the predictive model. Critically, independent external validation further confirmed the model's reliability and robustness, offering a practical tool for the early detection and prevention of sarcopenia and supporting the development of personalized treatment strategies for patients with CKD.