Abstract
BACKGROUND: Heart failure (HF) in patients undergoing maintenance hemodialysis (MHD) increases their hospitalization rates, mortality, and economic burden significantly. We aimed to develop and validate a predictive model utilizing contemporary deep phenotyping for individual risk assessment of all-cause mortality or HF hospitalization in patients on MHD. MATERIALS AND METHODS: A retrospective review was conducted from January 2017 to October 2022, including 348 patients receiving MHD from four centers. The variables were adjusted by Cox regression analysis, and the clinical prediction model was constructed and verified. RESULTS: The median follow-up durations were 14 months (interquartile range [IQR] 9-21) for the modeling set and 14 months (9-20) for the validation set. The composite outcome occurred in 72 (29.63%) of 243 patients in the modeling set and 39 (37.14%) of 105 patients in the validation set. The model predictors included age, albumin, history of cerebral hemorrhage, use of angiotensin-converting enzyme inhibitors/angiotensin II receptor blockers/"sacubitril/valsartan", left ventricular ejection fraction, urea reduction ratio, N-terminal prohormone of brain natriuretic peptide, and right atrial size. The C-index was 0.834 (95% CI 0.784-0.883) for the modeling set and 0.853 (0.798, 0.908) for the validation set. The model exhibited excellent calibration across the complete risk profile, and the decision curve analysis (DCA) suggested its ability to maximize patient benefits. CONCLUSION: The developed prediction model offered an accurate and personalized assessment of HF hospitalization risk and all-cause mortality in patients with MHD. It can be employed to identify high-risk patients and guide treatment and follow-up.