Objective To analyze the risk factors of sarcopenia in hemodialysis patients, construct a prediction model for sarcopenia in hemodialysis patients based on machine learning algorithms to provide a reference for the formulation of clinical prevention and treatment strategies.

Method A total of 211 hemodialysis patients admitted from January 2024 to May 2025 were selected as the research subjects. The patients were divided into the sarcopenia group and the non-sarcopenia group based on whether sarcopenia occurred as the outcome variable. The clinical data of the patients were collected. In addition, thee prediction models were constructed based on decision tree, random forest and SMOTE algorithm. Finally, the predictive value of the prediction models established by the three methods for the occurrence of sarcopenia bewteen two groups were compared.

Results Among the 211 patients, 49 cases developed sarcopenia, with an incidence rate of 23.22%. There were statistically significant differences in the age, BMI, diabetes, dialysis duration, nutritional risk and low physical activity level between two groups (P ＜ 0.05). The results of logistic regression analysis showed that the age (OR = 2.566, P = 0.02), BMI (OR = 0.800, P = 0.02), diabetes (OR = 2.362, P = 0.04), dialysis duration (OR = 3.600, P ＜ 0.01), nutritional risk (OR = 3.684, P ＜ 0.01), low physical activity level (OR = 4.931, P ＜ 0.01) were the risk factors of sarcopenia in hemodialysis patients. The resultsn of ROC analysis showed that the AUC values of the decision tree algorithm, random forest algorithm and SMOTE algorithm for predicting the risk of sarcopenia in hemodialysis patients were 0.811, 0.829 and 0.817, respectively, with good predictive efficacy. Among them, the AUC value of the random forest algorithm was the highest. However, the Delong test showed that there was no statistical significance in the AUC values among the three models (P ＞ 0.05)

Conclusions The risk prediction models for sarcopenia in hemodialysis patients constructed based on decision tree, random forest algorithm and SMOTE algorithm in this study all have good predictive efficacy. Among them, the random forest model has more advantages, and can effectively identify the risk of sarcopenia, which is conducive to clinical prediction and identification of high-risk patients with sarcopenia.