Automated EAT radiomic profiling predicted incident heart failure with C-statistic 0.84; each 10% risk increase raised adjusted HF risk by 31% (HR 1.31).
Does automated radiomic phenotyping of epicardial adipose tissue from CCTA improve the prediction of incident heart failure risk in adult patients without prior HF?
Automated radiomic phenotyping of epicardial adipose tissue from CCTA provides a robust biomarker that significantly improves incident heart failure risk prediction beyond traditional risk factors.
Absolute Event Rate: 0% vs 0%
Abstract Background Adverse epicardial adipose tissue (EAT) remodeling mediates the effects of visceral obesity on cardiovascular disease. Coronary computed tomography angiography (CCTA) provides a scalable diagnostic strategy for high-resolution, high-throughput radiomic phenotyping of EAT, but its role in predicting incident heart failure (HF) risk remains unknown. Purpose To develop and externally validate an automated pipeline for the high-throughput radiomic phenotyping of EAT and develop a robust biomarker for future HF risk. Methods This was a nationwide, multi-centre study of 77,268 adult patients undergoing CCTA across 9 centers in the U.K. (2007-2022), split into development (n=62,877 from 7 centers; 80-20% split for training and internal validation) and external testing sets (n=14,391 from 2 independent centers). Participants had no prior history of HF, and at least 30 days of event-free follow-up post-CCTA. We developed a novel, automated pipeline that enables robust EAT segmentation followed by quantification of 1,655 EAT radiomic features, spanning first-order and texture-related parameters. These features were combined with age and sex and processed through sequential steps of batch-level harmonization, a variational autoencoder (VAE), and a survival classification head to predict incident HF (Fig 1). The model was evaluated in the external testing set using multivariable Cox regression models adjusted for age, sex, as well as history of diabetes mellitus, ischaemic heart disease, myocardial infarction, peripheral arterial disease, and chronic kidney disease. Results The derivation cohort included 62,877 unique participants (age 57±13 years, 29,435 46.8% women), with a median follow-up of 5.1 (IQR: 3.4-7.6) years and 2,108 (3.4%) new HF events, whereas the external testing set included 14,391 unique participants (age 58±12 years, 7,107 49.4% women), with a median follow-up of 4.0 (IQR: 2.8-7.0) years and 1,457 (3.1%) new HF events. The EAT radiomic pipeline predicted incident HF with a C-statistic of 0.84 across both the internal validation and external testing sets. In the external testing set, each 10% increment in the EAT radiomic risk percentile was associated with a ~31% higher adjusted risk of HF (adj. HR 1.31 95%CI: 1.22-1.40) (Fig 2a-b), with those in the ≥90th percentile exhibiting a ~20-fold higher risk of HF compared with those in the 10th percentile (Fig 2b). EAT radiomic phenotyping improved HF discrimination at 5-years beyond traditional risk factors (ΔAUROC: 0.053, p0.001) (Fig 2c), and showed consistent performance across key age, sex and racial subgroups (Fig 2d). Conclusion In the largest nationwide CCTA-based radiomic analysis to date, we developed and validated an automated pipeline for high throughput radiomic phenotyping of EAT, demonstrating reproducible links to incident HF. These findings establish a new visceral adiposity biomarker of HF risk.Figure 1. Figure 2.
Oikonomou et al. (Sat,) reported a other. Automated EAT radiomic profiling predicted incident heart failure with C-statistic 0.84; each 10% risk increase raised adjusted HF risk by 31% (HR 1.31).