Abstract
OBJECTIVES: We evaluated whether automated CT-based adiposity tools can predict all-cause mortality in a large retrospective adult population. METHODS: This study included 151 177 patients who underwent abdominal CT between 2000 and 2021. An AI-based algorithm measured abdominal visceral adipose tissue (VAT) cross-sectional area and density at the L3. Kaplan-Meier survival curves and hazard ratios assessed VAT and mortality. RESULTS: Among 136 895 patients included, 9059 died within 1 year and 18 829 died within 2 to 20 years post-CT. Higher VAT density predicted 1-year mortality (hazard ratio [HR] up to 3.8) and over 2-20 years (HR up to 2.1). In contrast, VAT area did not significantly predict mortality. High VAT density was associated with the poorest survival, regardless of area. Low VAT density predicted better survival, regardless of area. VAT density consistently predicted mortality across age groups and sexes, whereas BMI did not differentiate risk. CONCLUSIONS: AI-enabled CT measures of VAT density are superior to VAT area for predicting all-cause mortality. Furthermore, we analysed VAT density vs. BMI in our largest age group (40-59) and found BMI was unable to adequately predict risk of mortality. Automated assessment of VAT density may enhance patient risk assessment and management. ADVANCES IN KNOWLEDGE: Assessing visceral fat density using fully automated AI-based CT tools offers a significant advancement in predicting health risk, leading to targeted interventions and improved management strategies. This study is novel due to its large patient population, offering evidence that prognostication with VAT density is broadly generalizable across varying patient populations.