Refined prediction and validation of individual risk using machine learning in transcatheter aortic valve implantation: TAVI Risk Machine (TRIM) scores

Abstract

Abstract Background Given the recent option for treatment using TAVI irrespective of surgical risk, general surgical risk scores have become less relevant, while TAVI-specific scores require refinement. Additionally, post-TAVI risk models are lacking; however, such risk models can support decision between post-TAVI treatment approaches, such as early discharge or close surveillance. Purpose This study aimed to predict 30-day mortality following transcatheter aortic valve implantation (TAVI) based on machine learning (ML) using data from the German Aortic Valve Registry. Methods Mortality risk was determined using a random forest ML model that was condensed in the newly developed TAVI Risk Machine (TRIM) scores, designed to represent clinically meaningful risk modelling before (TRIMpre) and after (TRIMpost) TAVI. Algorithm was trained and cross-validated on data of 24,452 patients and generalisation was examined on data of 5,889 patients. Results TRIMpost demonstrated significantly better performance than traditional scores (C-statistics value, 0.79; 95% confidence interval [CI] [0.74; 0.83]). An abridged TRIMpost score comprising 25 features (calculated using a web interface) exhibited significantly higher performance than traditional scores (C-statistics value, 0.74; 95% CI [0.70; 0.78]). Conclusion TRIM scores have high performance for risk estimation before and after TAVI. Together with clinical judgement, they may support standardised and objective decision-making before and after TAVI. Funding Acknowledgement Type of funding sources: None.