Deep-learning survival analysis for patients with calcific aortic valve disease undergoing valve replacement

Area of Science:

• Cardiology
• Medical Informatics
• Biostatistics

Background:

• Calcification of the aortic valve (CAVDS) causes aortic stenosis (AS), necessitating valve replacement via surgical aortic valve replacement (SAVR) or transcatheter aortic valve intervention (TAVI).
• High post-intervention mortality rates underscore the clinical importance of accurate risk assessment for SAVR and TAVI procedures.

Purpose of the Study:

• To compare traditional Cox Proportional Hazard (CPH) models with Machine Learning (ML) approaches (DeepSurv, RSF) for predicting one-year mortality risk post-SAVR or TAVI.
• To identify key variables for estimating mortality risk in patients undergoing aortic valve interventions.

Main Methods:

• Comparative analysis of Cox Proportional Hazard (CPH), Deep Learning Survival (DeepSurv), and Random Survival Forest (RSF) models.
• Utilized six variables: albumin, age, BMI, glucose, hypertension, and clonal hemopoiesis of indeterminate potential (CHIP) for risk prediction.
• Evaluated prediction capability using the c-index metric.

Main Results:

• A combination of six variables (albumin, age, BMI, glucose, hypertension, CHIP) accurately predicted one-year mortality across all three methods.
• Machine learning models (DeepSurv, RSF) demonstrated superior prediction capabilities compared to the traditional CPH model.
• ML models effectively capture non-linear relationships, enhancing their utility in medical statistical analysis.

Conclusions:

• Machine learning models offer improved prediction accuracy for post-aortic valve intervention mortality.
• The identified six-variable set provides a robust tool for risk stratification.
• Enhanced early identification of high-risk patients can guide therapeutic interventions and improve outcomes.