Application of the DeepSurv Model to Predict Survival in Patients with Kidney Failure Undergoing Hemodialysis

Abstrak

This study aims to improve survival prediction in patients with kidney failure undergoing hemodialysis, given their high mortality risk. Traditional models such as Cox Proportional Hazards (Cox PH) have limitations in capturing complex and nonlinear relationships in clinical data. Therefore, this study applies DeepSurv, a deep learning–based survival model, and compares its performance with Cox PH and Cox PH Spline. A total of 300 patients were included, with 165 events and 135 censored observations. The data were split into training and testing sets. DeepSurv was implemented using two hidden layers (64 and 32 neurons), a dropout rate of 0.2, and a learning rate of 1e-3. The model was trained for up to 1000 epochs with early stopping at epoch 435. Performance was evaluated using the concordance index (C-index) and time-dependent AUC at 365, 544, and 730 days. Patients were stratified into low-, medium-, and high-risk groups based on predicted scores. Results showed that Cox PH achieved a C-index of 0.913 and average AUC of 0.964, while Cox PH Spline reached 0.917 and 0.971. DeepSurv achieved a C-index of 0.920 and average AUC of 0.969. Performance differences were small, but DeepSurv provided consistent individual risk estimates. In conclusion, DeepSurv is a flexible approach with performance comparable to Cox-based models. Further external validation and clinical evaluation are needed before wider application