What is the optimal number of readers needed to achieve high diagnostic accuracy in coronary computed tomographic angiography? A comparison of alternate reader combinations

Background: Coronary computed tomographic angiography (CCTA) possesses high accuracy to detect coronary artery disease (CAD), although studies have reported differences in diagnostic performance. Prior trials used different numbers of interpreters, and the optimal number to detect CAD is unknown. Objective: We compared the diagnostic performance of 1, 2, 3, and 5 randomly selected interpreters for CCTA. Methods: We evaluated 50 patients randomly selected from 2 multicenter studies with both 64-detector CCTA and invasive quantitative coronary angiography (QCA). Five blinded, experienced readers independently interpreted CCTA and assessed for obstructive CAD (≥50% stenosis) and high-risk CAD (left main, proximal left anterior descending, or 3-vessel stenoses). A core laboratory performed QCA. For each patient, different random combinations of readers were selected; the accuracy of 1, 2, and 5 readers was compared with 3 readers. Results: Obstructive and high-risk CAD were observed in 20 of 50 (40%) and 6 of 50 (12%) patients, respectively. With combinations of 1, 2, 3, or 5 readers, there was a range of per-patient diagnostic performance (sensitivity, 100% each; specificity, 67%-90%; accuracy, 80%-94%; P = NS), per-segment diagnostic performance (sensitivity, 67%-83%; specificity, 87%-93%; accuracy, 87%-92%; P < .001 for 1 vs 3 and 2 vs 3 readers), and detection of high-risk CAD (sensitivity, 83%-100%; specificity, 73%-80%; accuracy, 76%-82%; P = NS). The highest diagnostic accuracy was observed with 3 readers for each comparison. Conclusion: The diagnostic performance of CCTA to detect obstructive or high-risk CAD is generally high irrespective of the number of readers. Consensus interpretation by ≥3 readers provides the highest diagnostic accuracy.