Abstract / Summary
To explore machine learning models for predicting diuretic resistance (DR) in patients with heart failure with preserved ejection fraction (HFpEF) comorbid with metabolic dysfunction-associated steatotic liver disease (MASLD). In this retrospective study of 586 patients, four predictive algorithms were evaluated. The eXtreme Gradient Boosting (XGBoost) model was further interpreted using SHapley Additive exPlanations (SHAP) to identify non-linear feature interactions. XGBoost achieved the highest predictive performance (AUC: 0.850). SHAP analysis indicated that hepatic parameters, particularly the FIB-4 index and serum albumin, were strongly associated with DR, potentially rivaling traditional cardio-renal indices. Interaction analyses suggested that concurrent hepatic and renal impairment synergistically correlated with elevated DR risk. Patients stratified into the AI-derived high-risk phenotype experienced lower freedom from HF re-hospitalization (Log-rank P < 0.001). The XGBoost model, coupled with SHAP analysis, may serve as a valuable tool for DR risk stratification. Assessing multi-organ crosstalk across the cardio-hepato-renal axes could inform personalized volume management in this complex cohort.