Abstract: Objective  To construct a nomogram to predict the risk of hospitalization for heart failure (HF) in maintenance hemodialysis (MHD) patients.  Methods  MHD patients from four centers in northeast Sichuan during 2017 to 2023 were included in this study. Their basic information, clinical data, laboratory and imaging results were collected. Patients in the two centers were used as the training set (n=386), and those in the other two centers were used as the external validation set (n=116). Least absolute shrinkage and selection operator (LASSO) and Cox regression analysis were used to analyze the related risk factors. A nomogram model for the risk of HF hospitalization was established. The prediction efficiency of the model was evaluated by the area under the receiver operating characteristic (ROC) curve, and the accuracy and practicability of the model were analyzed and evaluated by the calibration curve and the decision curve.  Results  The median follow-up periods of the training set and external validation set were 15 months (9, 24) and 14 months (10, 21), respectively. HF hospitalization occurred in 140 cases (36.27%) and 28 cases (24.14%) in training set and external validation set, respectively. Cox regression showed that the N-terminal pro-brain natriuretic peptide (HR=1.532, 95% CI:1.244～1.886, P<0.001), percentage of lymphocytes (HR=0.975, 95% CI:0.952～0.999, P=0.038), right atrium diameter (HR=1.060, 95% CI:1.017～1.105, P=0.005)/right ventricle diameter (HR=1.033, 95% CI:0.998～1.062, P=0.064), weekly dialysis duration (HR=0.667, 95% CI:0.459～0.968, P=0.033), HF score (HR=1.778, 95% CI:1.130～2.798, P=0.013), use of angiotensin-converting enzyme inhibitor/angiotensin receptor blocker medications (HR=0.569, 95% CI: 0.353～0.917, P=0.020), and history of coronary heart disease/diabetes (HR=1.582, 95% CI:1.002～2.500, P=0.049) were independently associated with HF hospitalization. The C-statistic for internal and external validation were 0.836 (95% CI:0.802～0.870) and 0.819 (95% CI:0.786～0.853), respectively. The calibration curve showed that there was a good consistency between actual probability and predicted probability of HF hospitalization (the median of the 2-year calibration slope was 1.018). The decision curve showed that the clinical net income was higher.  Conclusion  The prediction model of this study can accurately and individually evaluate the risk of HF hospitalization in MHD patients, and is of clinical practice value.

Key words: Maintenance hemodialysis, Heart failure, Prediction model, External verification, Nomogram