Development and validation of a multivariate model for predicting heart failure hospitalization and mortality in patients receiving maintenance hemodialysis

Abstract

AbstractBackgroundHeart failure (HF) is a common cardiovascular disease in patients receiving maintenance hemodialysis (MHD). Once these patients on MHD exhibit HF, their hospitalization rate, mortality, and economic burden will be significantly increased. Early identification and prediction of hospitalization and death are of great importance for reducing hospitalization and mortality. This study used multicenter clinical data to develop and externally validate clinical risk models to predict expected mortality and HF hospitalization rates in MHD patients with HF.Materials and MethodsFrom January 2017 to October 2022, 348 patients receiving MHD from four participating centers were enrolled. Demographic data, MHD treatment modalities, laboratory tests, and echocardiography data were collected when the initial event occurred. Three centers were randomly assigned to the modeling dataset (n=258), and one center was assigned to the external validation set (n=90). Considering a composite outcome of HF hospitalization and death as the primary endpoint and hospitalization due to HF or death as the secondary endpoint, a COX clinical prediction model was constructed and verified using internal and external datasets.ResultsThe median age of patients in the modeling cohort was 63 years old, 41.5% of patients were women; 165 (61%) had a history of HF; 81 (31.4%) were hospitalized for HF; and 39 (15.1%) patients had died. The c-statistic values for composite outcome, hospitalization for HF, and mortality were 0.812, 0.808, and 0.811, respectively. The predictors of death and hospitalization outcomes caused by HF are significantly different. The strongest predictors of HF hospitalization outcomes were advanced age, multiple HF hospitalizations, hyponatremia, high levels of NT-proBNP and hs-cTnT, and larger MVe values. The strongest predictors of mortality were longer dialysis age, combined atrial fibrillation, calcification of the aortic or mitral valve (especially calcification, and in particular aortic valve calcification), pleural effusion, low serum sodium, and higher levels of hs-cTnT. The median age of the patients in the external validation cohort was 63 years old; 28.8% were female; 35 (38.1%) had a history of HF; 11 (12.2%) were hospitalized for HF; and 5 (5.6%) died. The c-statistic of the predictive models for composite outcome, hospitalisation for HF, and mortality was comparable to that of the modelling cohort.ConclusionThe model established in this study is stable and reliable and the included variables are easily obtained from the routine clinical environment. The model can provide useful risk factors and prognostic information for patients with MHD combined with HF. Keywords: heart failure, MHD patients, mortality, predictive model, external validation.