Development of an interpretable machine learning model for predicting sarcopenia in patients undergoing maintenance hemodialysis

Abstract

Background: Sarcopenia has a high incidence among patients undergoing maintenance hemodialysis (MHD), significantly increasing the risk of falls, fractures, and mortality. Traditional diagnostic methods, however, are costly and complex, limiting their widespread clinical application. Therefore, developing an efficient and interpretable sarcopenia prediction model using routine clinical and laboratory data is crucial, with explainability techniques applied to further enhance model transparency. Methods: This study included 256 MHD patients and developed five machine learning models based on clinical and laboratory data: Logistic Regression, Extreme Gradient Boosting, Random Forest, Support Vector Machine, and Gaussian Naive Bayes. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis. Additionally, SHapley Additive exPlanations (SHAP) were employed as an explainability tool to enhance and visualize the interpretability of the optimal model. Results: The Logistic Regression model demonstrated the best performance on the validation set (AUC = 0.828, 95% CI: 0.626–0.989). Key predictive factors included body mass index (BMI), age, gender, creatinine (Cr), 25-hydroxyvitamin D3, left ventricular ejection fraction (LVEF), and estimated glomerular filtration rate (eGFR). SHAP analysis revealed that high BMI and 25-hydroxyvitamin D3 levels were protective factors, while low Cr, LVEF, and eGFR levels, as well as female gender, significantly increased the risk of sarcopenia. Conclusion: This study developed a Logistic Regression model using an interpretable machine learning approach, offering an efficient tool for early screening of sarcopenia risk in MHD patients and facilitating personalized intervention strategies. However, the single-center design limits the model’s external applicability, and further multi-center studies are necessary to validate its generalizability.