Development and validation of an integrated prognostic model for all-cause mortality in heart failure: a comprehensive analysis combining clinical, electrocardiographic, and echocardiographic parameters

Abstract

Abstract Background Accurate risk prediction in heart failure remains challenging due to its complex pathophysiology. We aimed to develop and validate a comprehensive prognostic model integrating demographic, electrocardiographic, echocardiographic, and biochemical parameters. Methods We conducted a retrospective cohort study of 445 heart failure patients. The cohort was randomly divided into training (n = 312) and validation (n = 133) sets. Feature selection was performed using LASSO regression followed by backward stepwise Cox regression. A nomogram was constructed based on independent predictors. Model performance was assessed through discrimination, calibration, and decision curve analyses. Random survival forest analysis was conducted to validate variable importance. Results During a median follow-up of 4.14 years, 142 deaths (31.91%) occurred. Our model development followed a systematic approach: initial feature selection using LASSO regression identified 15 potential predictors, which were further refined to nine independent predictors through backward stepwise Cox regression. The final predictors included age, NYHA class, left ventricular systolic dysfunction, atrial septal defect, aortic valve annulus calcification, tricuspid regurgitation severity, QRS duration, T wave offset, and NT-proBNP. The integrated model demonstrated good discrimination for 2-, 3-, and 5-year mortality prediction in both training (AUCs: 0.726, 0.755, 0.809) and validation cohorts (AUCs: 0.686, 0.678, 0.706). Calibration plots and decision curve analyses confirmed the model’s reliability and clinical utility across different time horizons. A nomogram was constructed for individualized risk prediction. Kaplan-Meier analyses of individual predictors revealed significant stratification of survival outcomes, while restricted cubic spline analyses demonstrated non-linear relationships between continuous variables and mortality risk. Random survival forest analysis identified the top five predictors (age, NT-proBNP, QRS duration, tricuspid regurgitation severity, NYHA), which were compared with our nine-variable model, confirming the superior performance of the integrated model across all time points. Conclusions Our integrated prognostic model showed robust performance in predicting all-cause mortality in heart failure patients. The model’s ability to provide individualized risk estimates through a nomogram may facilitate clinical decision-making and patient stratification. Clinical trial number Not applicable.

Keywords

• Heart failure
• Prognosis
• Risk prediction
• Machine learning
• Nomogram