Abstract 4366874: AI-Based Cardiac Chamber Volumetry from CAC CT Enhances Heart Failure Prediction Beyond PREVENT-HF

Abstract

Background: Coronary artery calcium (CAC) CT scans are widely used for the assessment of atherosclerosis, but their application for cardiac chamber volumetry in heart failure (HF) risk prediction remains underexplored. Research Question: This study aims to evaluate whether AI-derived cardiac chamber volumetry, obtained from CAC CT scans in asymptomatic patients, improves HF risk prediction beyond the American Heart Association’s PREVENT-HF clinical risk calculator. Methods: This retrospective cohort study included asymptomatic patients aged 45-75 years, without known cardiac disease, who underwent diastolic phase CAC CT imaging between 2010-2023 with at least one year of follow up. Chamber volumes were derived from CT images using a previously validated convolutional neural network based autoencoder model. The AHA base PREVENT-HF score was calculated for each patient. Time-dependent AUCs at 3, 5, and 10 years modeled HF predictive performance for chamber volumetry, PREVENT-HF, and their combination. Cox proportional hazard regression was used to assess the associations between chamber volumes and PREVENT-HF with incident HF. Results: A total of 2,966 patients were included with a mean age of 56.3±9.3 years (42% women; 78% White). Over a mean follow-up period of 4.3±2.6 years, 7.2% (n=215) developed heart failure within 10 years. Higher chamber volumes of the left atrium (LA), left ventricle (LV), right atrium (RA), and LV myocardium were significantly associated with increased risk for incident HF at each follow-up interval (p-value < 0.001). LA volume was the most predictive with time dependent AUC values of 0.711, 0.719, and 0.693 at 3, 5, and 10 years, respectively (p<0.001 compared to PREVENT-HF alone). Time dependent AUC of all chamber volumes alone outperformed PREVENT-HF [3yr: 0.765 vs. 0.629 (p=0.002), 5yr: 0.764 vs 0.673 (p=0.047), 10yr: 0.737 vs. 0.775 (p=0.430)]. Combining chamber volumes and PREVENT-HF (3yr: 0.746, 5yr: 0.776, 10yr: 0.775) resulted in significantly higher performance than PREVENT-HF alone (p<0.001). Conclusion: AI-derived cardiac chamber volumetry derived from CAC CT enhances heart failure risk prediction when integrated with the PREVENT-HF risk calculator. Left atrial volume, in particular, serves as a strong independent predictor of heart failure.