Deep neural network base competing risk in predicting heart failure patient's survival

Area of Science:

• Cardiology
• Biostatistics
• Artificial Intelligence

Background:

• Heart failure (HF) presents complex prognostication challenges due to competing risks like HF-specific mortality and other causes of death.
• Accurate prediction of time-to-event outcomes is crucial for effective patient management in HF.

Purpose of the Study:

• To compare the predictive performance of a Deep Neural Network Competing Risks (DNNCR) model against a Random Survival Forest (RSF) model for time-to-event outcomes in HF patients.
• To evaluate the models' ability to handle competing risks in survival analysis for heart failure.

Main Methods:

• Retrospective analysis of 435 heart failure patients with a five-year follow-up.
• Application of Deep Neural Network Competing Risks (DNNCR) and Random Survival Forest (RSF) models to analyze survival data with competing risks.
• Model performance assessed using the C-index and Integrated Brier Score (IBS).

Main Results:

• The DNNCR model demonstrated superior predictive performance compared to RSF, indicated by higher C-index values for both HF-specific death and other causes of death.
• DNNCR achieved a C-index of 0.65 for HF and 0.63 for other causes, outperforming RSF's C-indices of 0.65 and 0.61, respectively.
• Calibration analysis using IBS showed the DNNCR model's superior performance with IBS values of 0.16 for HF and 0.18 for other causes.

Conclusions:

• The DNNCR model significantly outperforms the RSF model in predicting survival outcomes for heart failure patients, especially when dealing with competing risks.
• Enhanced predictive accuracy facilitates better identification of high-risk patients, enabling personalized treatment strategies.
• Future research should focus on diverse datasets to improve DNNCR performance and clinical integration.