Harmonizing multimodality imaging results using Bayesian analysis: the case of CT coronary angiography and CT-derived...

Coronary computed tomographic angiography (CCTA) may provide both anatomic and CT fractional flow reserve data (CTFFR). The objective is to use Bayesian analysis to develop a model wherein the probability of significant coronary artery disease (CAD) by CTFFR can be determined given the prior probability (P) of the combined clinical and CCTA result. 172 patients referred for CCTA and subsequently underwent coronary angiography were automatically referred to CTFFR analysis. A clinical P risk score (CRS) was calculated per patient. CCTA exams were scored using CAD-RADS classification. CTFFR results were generated. CAD was defined as ≥ 3 RAD class for CCTA and ≤ .80 by CTFFR. P was calculated using CCTA and CTFFR accuracy from a prior clinical trial: post-test P for the CCTA result used the CRS as the prior risk, and CTFFR P used the post-test CRS + CCTA P as the prior risk (tri-variable). Patients were classified for each model into low (< 5%), intermediate, (5-70%) and high (> 70%) risk groups. There were 100 patients (58%), who had significant CAD at angiography. 58 patients had discordant CCTA/CTFFR results. The inclusion of the CRS and CRS + CCTA in the prior progressively reduced the intermediate risk cohort from 83 to 41% (p < 0.0001). Correct classifications (low-risk, negative angiogram plus high-risk, positive angiogram) increased by model: CRS = 12%, CRS + CCTA = 25%, CRS + CTFFR = 33%, CRS + CCTA + CTFFR = 44% (p < 0.001). Incorrect classifications were reduced to 15%. The tri-variable model performed better than either CCTA or CTFFR alone for all patients and for the sub-group with discordant imaging results. Discrepant CCTA and CTFFR results are present in one third of patients. The use of both the CRS and CCTA as the prior risk synergistically maximized the accuracy of the accuracy of the CTFFR technique.