Abstract 208: Reliability of Cardiac CT for the Detection of Significant Luminal Narrowing: Results From the Study of Myocardial Perfusion and Coronary Anatomy Imaging Roles in CAD (SPARC) Trial
Background: Agreement between site and expert core lab reads may serve as an indicator of the quality of data interpretation in registries. We aimed to determine the agreement for significant luminal narrowing between expert readers within a core lab setting as compared to clinically performed reads of coronary CT angiography (CCTA).
Methods: We evaluated 283 patients undergoing CCTA at 25 academic and non-academic sites across the United States and Canada participating in the SPARC trial, a prospective registry assessing post-cardiac imaging outcomes. Two expert core lab readers adjudicated the luminal narrowing of each coronary segment as non-significant (<50% stenosis), significant, or non-evaluable. Disagreement within these categories was adjudicated by a third expert reader on a segmental level. Interobserver agreement between core lab readers and between the final core lab reading and that of the sites was assessed on a per segment, per vessel, and per patient basis (defined as normal intermediate or high risk). Agreements were expressed as weighted Kappa.
Results: The weighted Kappa between core lab readers was 0.87 (0.86, 0.88) on a per segment basis, 0.72 on a per vessel basis, and 0.61 on a per patient basis using a CTA70% stenosis criteria for normal, non-obstructive, 1, 2, or 3VD. The weighted Kappa between the final core lab reading and that of the sites was 0.51 (0.44, 0.59) on a per patient basis using a CTA70% stenosis criteria for normal, non-obstructive, 1, 2, or 3VD. Overall, core lab readers adjudicated more segments as having significant luminal narrowing (17% vs. 26%, p= <0.001) and as unevaluable (9% vs. 11%, p= <0.001), as compared to the site readers.
Conclusions: Consistent with accuracy studies, which demonstrated excellent segmental agreement but poor classification into 1, 2, or 3VD, agreement between expert readers and between core lab reads and clinical sites are limited for advanced CAD classification, with core lab readers detecting more disease.