Artificial Intelligence Improves Prediction of Major Adverse Cardiovascular Events in Patients Undergoing Transcatheter Aortic Valve Replacement Planning CT

Rationale and Objectives: Coronary CT angiography (CCTA) is mandatory before transcatheter aortic valve replacement (TAVR). Our objective was to evaluate the efficacy of artificial intelligence (AI)-powered software in automatically analyzing cardiac parameters from pre-procedural CCTA to predict major adverse cardiovascular events (MACE) in TAVR patients. Materials and Methods: Patients undergoing pre-TAVR CCTA were retrospectively included. AI software automatically extracted 34 morphologic and volumetric cardiac parameters characterizing the ventricles, atria, myocardium, and epicardial adipose tissue. Clinical information and outcomes were recorded from institutional database. Cox regression analysis identified predictors of MACE, including non-fatal myocardial infarction, heart failure hospitalization, unstable angina, and cardiac death. Model performance was evaluated with Harrell's C-index, and nested models were compared using the likelihood ratio test. Manual analysis of 170 patients assessed agreement with automated measurements. Results: Among the 648 enrolled patients (77 +/- 9.3 years, 58.9% men), 116 (17.9%) experienced MACE within a median follow-up of 24 months (interquartile range 10-40). After adjusting for clinical parameters, only left ventricle long axis shortening (LV-LAS) was an independent predictor of MACE (hazard ratio [HR], 1.05 [95% confidence interval, 1.05-1.11]; p = 0.04), with significantly improved Cindex (0.620 vs. 0.633; p < 0.001). When adjusted for the Society of Thoracic Surgeons Predicted Risk of Mortality score, LV-LAS was also predictive of MACE (HR, 1.08 [95%CI, 1.03-1.13]; p = 0.002), while improving model performance (C-index: 0.557 vs. 0.598; p < 0.001). All parameters showed good or excellent agreement with manual measurements. Conclusion: Automated AI-based comprehensive cardiac assessment enables pre-TAVR MACE prediction, with LV-LAS outperforming all other parameters.