Diagnostic performance of a novel AI algorithm for determining coronary ischemia (AI-QCT ISCHEMIA) according to patient comorbidities

Abstract Background This study evaluated the diagnostic accuracy of a novel AI algorithm (AI-QCT ISCHEMIA) for the prediction of coronary ischemia as assessed using fractional flow reserve (FFR) according to the presence of diabetes, hypertension, and dyslipidemia. Methods This is a post-hoc pooled analysis of CREDENCE (validation cohort; 305 patients, mean age 64.4±9.8 years, 69% men, 868 vessels) and PACIFIC-1 (208 patients, mean age 58.1±8.7 years; 64% men, 612 vessels) studies. Symptomatic patients with suspected coronary artery disease underwent coronary computed tomography angiography (CCTA), myocardial perfusion imaging (MPI), FFR-CT, and invasive FFR. The diagnostic accuracy of AI-QCT ISCHEMIA in predicting myocardial ischemia assessed using invasive FFR was compared to MPI and FFR-CT across different subgroups using vessel-level AUC analysis. Results AI-QCT ISCHEMIA had an AUC of 0.82, compared to 0.75 for FFR-CT and 0.64 for MPI in patients with diabetes, while in those without diabetes the AUCs were 0.87, 0.78, and 0.67. In patients with hypertension, the AUCs were 0.87, 0.77, and 0.67, while in those without hypertension, the AUCs were 0.85, 0.78, and 0.65. In patients with dyslipidemia, the AUCs were 0.84, 0.78, and 0.63, while in patients without dyslipidemia, the AUCs were 0.88, 0.76, and 0.69. Conclusions AI-QCT ISCHEMIA demonstrated high diagnostic accuracy to detect myocardial ischemia regardless of diabetes, hypertension and dyslipidemia status, with consistent superiority to MPI and FFR-CT.