Association of radiomic score with cardiovascular risk in primary prevention: insights from the CAC-CV PREVITAL study

Background Coronary artery calcium (CAC) score is an established marker of subclinical atherosclerosis and is a predictor of cardiovascular events in primary prevention. Radiomics enables high-throughput extraction of quantitative image features which may refine cardiovascular risk assessment. Purpose To assess the value of artificial intelligence (AI)-based radiomic score for cardiovascular risk prediction. Methods This is a substudy of the randomized CAC-CV PREVITAL study. Participants who underwent ECG-gated computed tomography (CT) scan with a CAC score > zero were included. Cardiovascular risk was quantified using SCORE2/SCORE-OP and individuals were classified according to the ESC risk categories as very-high, high, or low-to-moderate risk. CAC score was calculated as Agatston score. AI-based automated calcium segmentation was performed using a deep-learning model. Radiomic features were extracted per calcification. Elastic net regression was used to select radiomic features and generate a radiomic score. Univariable logistic regression assessed the association between the standardized radiomic score and the very-high risk category per ESC definition (model 1) or a SCORE2/SCORE2-OP ≥ 10% (model 2). Multivariable logistic regression models including both the radiomic score and the Agatston score were also assessed. Results A total of 259 individuals, mean age 59 (8.3) years, 36% women were included. Twenty percent subjects were at very-high risk, and 17% had a SCORE2/SCORE2-OP ≥ 10%. A total of 259 CT scans, one per individual, were analysed, and 851 radiomic features were extracted per calcification. Mean total CAC score was 230 (447). A CAC score ≥100 was found in 45% of participants, and a CAC score ≥ 300 was present in 17% of subjects. The radiomic score comprised eight features in model 1 and six features in model 2. By univariable analysis, each 1-SD increase of radiomic score was associated with a significantly higher odd of being in the very-high risk category (odds ratio [OR] 3.17, 95% confidence interval [CI] 1.97-5.10, p <0.001) and of having a SCORE2/SCORE2-OP ≥ 10% (OR 2.00, 95% CI 1.39-2.87, p<0.001). Similarly, each 1-SD increase of CAC score was associated with a higher odd of very-high risk category (OR 1.53, 95% CI: 1.12-2.08, p=0.007) or of SCORE2/SCORE2-OP ≥ 10% (OR 1.68, 95% CI: 1.16-2.42, p=0.006). By multivariable analysis, the radiomic score was independently associated with very-high risk category (OR 2.95, 95% CI 1.82–4.79, p<0.001) and with a SCORE2/SCORE2-OP ≥ 10% (OR 2.95, 95% CI 1.82–4.79, p=0.002), while CAC score did not emerge as a significant predictor (very-high risk category: OR 1.25, 95% CI 0.90-1.73, p=0.18; SCORE2/SCORE2-OP ≥ 10%: OR 1.37, 95% CI: 0.97-1.93, p=0.08). Conclusions AI-based radiomic score is a significant predictor of higher cardiovascular risk. Prospective validation of radiomic score in studies with long term clinical outcome data is warranted.