Abstract
BACKGROUND AND OBJECTIVES: Malnutrition among older hospitalized adults with chronic heart failure (CHF) is associated with adverse clinical outcomes, yet reliable early risk stratification tools remain lacking. This study aimed to develop and validate a machine learning (ML) model for malnutrition risk stratification in this population. METHODS AND STUDY DESIGN: Malnutrition among older hospitalized adults with chronic heart failure (CHF) is associated with adverse clinical outcomes, yet reliable early risk stratification tools remain lacking. This study aimed to develop and validate a machine learning (ML) model for malnutrition risk stratification in this population. RESULTS: Malnutrition prevalence was 44.1% (348/790). In the internal testing, CatBoost (CAT) achieved superior performance with an AUC of 0.901 (95% confidence interval [CI]: 0.858-0.943), accuracy of 0.840, recall of 0.753, and the lowest Brier score of 0.113. This model demonstrated strong calibration, clinical utility, and the highest composite score (62/64). External validation confirmed CAT's generalizability (AUC: 0.916, 95% CI: 0.887-0.945). SHAP analysis identified body mass index (BMI), calf circumference, New York Heart Association (NYHA) classification, age, and diabetes as signifi-cant contributors to malnutrition risk. CONCLUSIONS: The CAT-based model effectively stratifies malnutrition risk in older hospitalized CHF patients, offering a tool for early intervention to improve outcomes. Further multicenter prospective studies are needed to validate its real-world applicability.