| Objective: To construct a risk prediction model for sarcopenia in the elderly using advanced machine learning algorithms, and to reversely map and explore its Traditional Chinese Medicine (TCM) etiology and pathogenesis based on feature importance. This study aims to provide an objective basis for the early prevention and treatment of sarcopenia through an integrative approach combining Chinese and Western medicine. Methods: Clinical data of 1,253 elderly individuals admitted or undergoing physical examinations at Tongji Hospital, Tongji Medical College of Huazhong University of Science and Technology, and other institutions from January 2021 to June 2024, were retrospectively analyzed. The dataset was divided into a training set and a testing set at an 8:2 ratio. The Bagged Trees algorithm was used for the single imputation of missing values, and LASSO regression was applied to screen for core features. Three models were constructed: traditional Logistic regression, Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). Model performance was evaluated by calculating the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. Furthermore, feature importance was extracted using the XGBoost model to analyze the etiology and pathogenesis in conjunction with classical TCM theories. Results: The prevalence of sarcopenia in the study population was 24.58%. LASSO regression identified 10 core predictive variables from 24 initial variables. A comparison of model performance showed that Logistic regression (AUC=0.766, sensitivity=0.328) had limitations in handling class imbalance and non-linear features; in contrast, the RF model (AUC=0.846, sensitivity=0.820) and XGBoost model (AUC=0.837, sensitivity=0.771) demonstrated superior performance. After optimizing the threshold using the Youden index, the XGBoost model achieved a sensitivity of 0.885 and a specificity of 0.630, indicating excellent overall classification efficacy. The XGBoost feature importance analysis revealed that age, frailty status, history of falls, risk of anxiety/depression, eating status and weight loss, as well as empty-nest and marital status, are the core predictive factors for sarcopenia in the elderly. Conclusion: The machine learning model based on XGBoost can accurately predict the risk of sarcopenia in the elderly. The high-risk features identified by the model highly align with TCM pathogenic changes such as "deficiency of both spleen and kidney," "deficiency of Yangming leading to laxity of the ancestral sinews," and "internal emotional injury." This suggests that in the clinical prevention and treatment of sarcopenia, alongside conventional nutritional and exercise interventions, practitioners can apply the classic principle of "treating flaccidity syndrome by solely targeting the Yangming meridian," supplemented by invigorating the spleen, warming the kidney, soothing the liver, and relieving depression, to construct a more precise, integrative intervention strategy. |