Development of a nomogram for predicting incident heart failure and all-cause mortality in patients with chronic kidney disease: a 3-year follow-up study

Background Chronic kidney disease (CKD) confers a substantially elevated risk of cardiovascular (CV) mortality, primarily driven by heart failure (HF) and other CV complications. This study aimed to develop a nomogram for predicting incident HF and all-cause death in patients with CKD, and further identify individuals at high risk of these outcomes. Methods We prospectively recruited 440 patients with CKD stages 3–5 and preserved ejection fraction (LVEF) from a nephropathy center between November 2020 to September 2024. Baseline clinical and demographic characteristics, biochemical and echocardiographic parameters, and data from quality-of-life scales were collected. Key predictors were identified using univariate Cox regression, Lasso, Random Forest, and XGBoost algorithms; these identified predictors were then integrated into a nomogram via multivariate Cox regression. Model performance was assessed using the C-index, ROC analysis, and calibration plots. Results Over 29 months (IQR 24–35), 104(23.6%) patients met the primary composite endpoints, which included 62(14.1%) incident HF and 42(9.5%) all-cause death. After selecting variables through forward stepwise regression, age[HR 1.030, P = 0.016], growth differentiation factor 15(GDF-15) [HR 1.076, P < 0.001], lipoprotein(a)[Lp(a)] [HR = 1.011, P = 0.002], ferritin(FER) [HR = 1.009, P = 0.030], and EuroQol-5D (EQ-5D) score [HR 0.099, P = 0.015] were utilized to construct the nomogram. In the training cohort, the model achieved a C-index of 0.772. For temporal validation, time-dependent area under the ROC curve(AUC) values further confirmed its accuracy: 0.779 (95%CI, 0.696–0.863) at 24 months, 0.816(95%CI, 0.743–0.888) at 27 months, 0.836(95%CI, 0.768–0.903) at 32 months, and 0.850(95%CI, 0.771–0.929) at 36 months. Decision curve analysis (DCA) also demonstrated sustained clinical utility across a broad range of threshold probabilities. Conclusion This study identified specific predictors of HF and all-cause mortality in CKD patients. This model has enhanced predictive ability for early identification of these risks, thereby facilitating timely clinical interventions to improve the prognosis of CKD patients.