Clinical Reference Percentiles for AI-derived Epicardial Adipose Tissue: A Multicenter Study

Authors:
Preprint
Publisher website
Google Scholar
Abstract
Background and Aims: Epicardial adipose tissue (EAT) has emerged as an important cardiovascular biomarker that reflects both inflammatory and cardiometabolic risk. EAT volume and density vary significantly across populations, yet there is a lack of multicenter studies investigating the predictive value of population-specific EAT percentiles. Methods: In this multicenter study, we retrospectively analyzed low-dose computed tomography correction scans from 42,842 patients undergoing myocardial perfusion imaging. A derivation cohort of 15,082 patients was used to establish sex- and age-specific nomograms for EAT density and EAT volume indexed to body surface area. Percentile-based thresholds were tested for outcome prediction in a validation cohort of 27,760 patients. For clinical implementation, we developed an online EAT percentile calculator. Results: Percentile curves demonstrated increased BSA-indexed EAT volume and decreasing EAT density with age. Over a median follow-up of 3.6 years (IQR: 1.83 – 5.14), 4,956 patients experienced a nonfatal myocardial infarction or death. In multivariable Cox models, patients above the 95 th sex- and age-specific percentile had significantly worse outcomes for BSA-indexed EAT volume [adjusted hazard ratio 1.30, 95% CI: 1.14 – 1.49, p < 0.001] and EAT density [adjusted hazard ratio 1.7, 95% CI: 1.51 – 1.92, pth percentile (p<0.001). Conclusion: Age- and sex-specific EAT percentiles provide a clinically interpretable framework for contextualizing automated EAT measurements and identifying patients at increased cardiovascular risk. EAT density was a stronger prognostic marker and identified elevated risk even among patients with normal BMI, supporting its potential to provide information beyond conventional anthropometric assessment. Graphical
Funding Information
  • National Heart, Lung, and Blood Institute (R35HL161195)

This publication has 35 references indexed in Scilit: