Malnutrition Severity Drives Mortality in Geriatric Heart Failure: A Multicenter Extreme Gradient Boosting Analysis

Background: Heart failure (HF) and malnutrition frequently coexist in older patients (≥65 years) and are major determinants of in-hospital mortality. However, predictive models specifically addressing this high-risk population remain limited. Therefore, this study aimed to develop and validate a personalized machine learning model to assess key risk factors. Methods: This study was a multicenter retrospective investigation that collected clinical data from older patients with HF and malnutrition admitted to two Chinese tertiary hospitals. Key predictors were selected using least absolute shrinkage and selection operator (LASSO) regression, followed by development of an extreme gradient boosting (XGBoost) model. Model performance was assessed using receiver operating characteristic (ROC) curve analysis, accuracy, sensitivity, specificity, and F1 score. Shapley additive explanation (SHAP) analysis was applied to provide interpretable feature importance. Moreover, the robustness of the model was externally validated in an independent cohort. Results: The final analysis included 1080 older patients with HF and malnutrition, among whom 244 experienced in-hospital mortality, yielding an in-hospital mortality rate of 22.6%. The XGBoost model achieved high area under the curve (AUC) values (training: 0.979, 95% confidence interval (CI): 0.969-0.990; validation: 0.890, 95% CI: 0.844-0.937; test: 0.936, 95% CI: 0.899-0.974). SHAP analysis highlighted the Geriatric Nutritional Risk Index (GNRI) as the primary predictive factor, with secondary contributions from inflammatory profiles and traditional cardiorenal and electrolyte markers. Conclusions: The constructed XGBoost model demonstrated robust predictive performance. The SHAP analysis provided a clear visualization of key risk factors, thereby providing a valuable reference for clinical risk assessment.