deepsurvcompeting risk: cause-specific cox deep neural network for predicting heart failure patient's survival

Predicting the survival of patients with heart failure (hf) is crucial for improving the management of cvd. this study aimed to model a cause-specific cox (csc) deep neural network for predicting the survival of patients with heart failure using a competing risk approach. our retrospective study included 435 patients treated for heart failure at rajaie cardiovascular medical and research center in iran from 2018 to 2023. patient survival data were analyzed based on the cause of death. in this study, instead of feature selection, which is targeted by most classical methods, we introduce a combined approach to provide a flexible and general framework for survival analysis and interpretation. in this approach, the random survival forest (rsf) model first selects features, and then the deep survival model is fitted to the significant variables. finally, the hazard ratio (hr) of the variables was calculated using the multivariable csc model. the performance of the models was evaluated based on the c-index of the training and test sets. the deepsurv model showed the best performance, with c-index values of 0.58 and 0.66 for the training set and the test set, respectively, for the risk of mortality due to hf. for the risk of mortality due to other causes, the rsf had a c-index of 0.61/0.66. finally, for both causes of death, the csc model demonstrated high accuracy, indicating its usefulness in predicting these outcomes. these results emphasize the importance of accurately predicting hf patient survival and identifying risk factors to inform treatment decisions and improve patient outcomes and suggest that survival prediction becomes more accurate when rsf and deepsurv models are used together.