B54-39 Artificial Intelligence-Enabled Coronary Artery Calcium Quantification Predicts All-Cause Mortality

Abstract Rationale Investigate whether artificial intelligence-detected coronary artery calcification (AI-CAC) on non-ECG-gated non-contrast chest CT is associated with increased all-cause mortality. Methods In this retrospective study, non-contrast non-ECG-gated chest CTs were identified within a large multicenter healthcare system occurring from November to December 2013 with follow-up through April 2025. Imaging studies were analyzed using a machine-learning algorithm to detect and grade CAC corresponding to Agatston scoring (none, mild 1-99, moderate 100-399, and severe =400). Study cohort data including demographic, comorbidity, and mortality data were obtained retrospectively by matching imaging study accession identifiers to electronic medical records with the assistance of the institutional Joint Data Analytics Team. Association between AI-CAC and all-cause mortality was measured by log-rank testing in addition to multivariable logistic regression analysis adjusting for sex, race, smoking status, insurance, and comorbid conditions. R (version 4.5.0; R Foundation for Statistical Computing) and R Studio (Version 2025.05.0 + 496; Posit Software, PBC) were utilized to perform Kaplan-Meier survival analysis and multivariable logistic regression analysis for all-cause mortality. Results A total of 1332 subjects were included in the cohort. Of these, 625 (47%) were female, median age 68.1 years (IQR: 58.2, 78.4), 605 (45.4%) expired during the study follow up period from 2013 to 2025 (median follow up 11.3 years, IQR 1.8-11.3). Of these, 450 (33.8%) had no CAC, 283 (21.2%) had mild CAC, 304 (22.8%) had moderate CAC, and 295 (22.1%) had severe CAC. Log-rank testing demonstrated significant differences in overall survival between groups of varying CAC severity (Figure 1, p 0.0001). After adjustment for age, race, sex, smoking status, and comorbid conditions patients with detected moderate (OR = 2.21, 95%CI 1.38-3.24) and severe CAC (OR = 3.40, 95%CI 2.12-5.44) were associated with increased hazard of all-cause mortality compared to those without CAC. Uninsured subjects (OR = 2.52, 95% CI 1.24-5.12) and those with commercial insurance (OR 1.69, 95% CI 1.10-2.59) independently had increased risk of all-cause mortality as compared to subjects with Medicare insurance. Chronic comorbidities independently associated with increased hazard of mortality included metastatic cancer, chronic lung disease, diabetes mellitus, and congestive heart failure (p 0.05). Conclusions Increased severity of AI-detected CAC on routine non-contrast chest CT carries increased risk of all-cause mortality over an 11-year follow-up period. These findings demonstrate that AI can enable scalable coronary calcium scoring which can identify high-risk patients without additional radiation or cost. Future directions include validation measures and clinical workflow integration. AI-driven CAC screening has the potential to transform routine imaging into a predictive tool for population health. This abstract is funded by: None