Abstract
Hypoproteinemia is a common complication across patients receiving maintenance hemodialysis (MHD). Moreover, it is associated with increased risks of cardiovascular events, infection risk, and mortality. This study aimed to construct a classification model for identifying the risk of hypoproteinemia in MHD patients to support precise clinical interventions. To this end, a retrospective analysis was conducted on 288 MHD patients at the Affiliated Hospital of Jining Medical University (January-December 2023). Hypoproteinemia was defined as the primary outcome. Four machine learning (ML) models-Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost)-were trained and validated using 3-fold cross-validation. Feature selection was performed using the Least Absolute Shrinkage and Selection Operator (LASSO). Model performance was evaluated using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, balanced F-score, and Brier score. SHapley Additive exPlanations (SHAP) was used to interpret the model. The SVM model demonstrated the highest identification performance with an AUC of 0.937. Thirteen key features were identified, with globulin, hemoglobin, β2 microglobulin, prealbumin, and hemodialysis mode being the most significant. In conclusion, the classification model based on ML can accurately identify the risk of hypoproteinemia in MHD patients and provide targeted guidance for early clinical intervention. The combination of SVM and SHAP not only enhances model interpretability but also establishes a scientific foundation for personalized risk classification and the development of precision diagnosis and treatment strategies by visualizing the dynamic effects of key features.