Determinants of Rejection Rate for Coronary CT Angiography Fractional Flow Reserve Analysis

Background: Coronary artery fractional flow reserve (FFR) derived from CT angiography (FFT CT ) enables functional assessment of coronary stenosis. Prior clinical trials showed 13%–33% of coronary CT angiography studies had insufficient quality for quantitative analysis with FFR CT. To determine the rejection rate of FFR CT analysis and to determine factors associated with technically unsuccessful calcu- lation of FFR Materials and Methods: Prospectively acquired coronary CT angiography scans submitted as part of the Assessing Diagnostic Value of Noninvasive FFR CT in Coronary Care (ADVANCE) registry ( https://ClinicalTrials.gov: NCT02499679) and coronary CT an- giography series submitted for clinical analysis were included. The primary outcome was the FFR CT rejection rate (defined as an inability to perform quantitative analysis with FFR CT ). Factors that were associated with FFR CT rejection rate were assessed with multiple linear regression. Results: FFR clinical cohorts, At multivariable section (odds 1.04; 95% P clinical (OR, 95% CI: 1.02, 1.04; P , .001) cohorts and heart rate in the ADVANCE (OR, 1.05; 95% CI: 1.02, 1.08; P , .001) and clinical (OR, 1.06; 95% CI: 1.05, 1.07; P , .001) cohorts were independent predictors of rejection. flow cohort respectively. CT and lower patient heart rate increase rates of completion of CT fractional flow reserve analysis.

FFR CT analysis could be performed in 87%-88% of coronary CT angiography data sets (7,8); however, a more recent substudy of the Prospective Multicenter Imaging Study for Evaluation of Chest Pain, or PROMISE, trial found that because of low-quality images, only 67% of coronary CT angiography data sets could be used for FFR CT analysis (9). Previous studies have examined the factors that affect the accuracy of FFR CT (for instance, both nitrate use and b-blocker use improve diagnostic accuracy) (10,11). However, factors that affect the ability to perform FFR CT analysis of coronary CT angiography data remain poorly understood. Identification of such factors would allow for targeted intervention to increase the likelihood of using this additional analysis in clinical care.
Thus, the aim of our study was to examine the rejection rate of coronary CT angiography for FFR CT analysis and the factors associated with an inability to perform FFR CT in the controlled Assessing Diagnostic Value of Noninvasive FFR CT in Coronary Care (ADVANCE) registry (12) and in an uncontrolled cohort of clinical cases.

Materials and Methods
This study was conducted in accordance with the Declaration of Helsinki. Participants in the ADVANCE registry all provided written informed consent. The clinical cohort analysis was performed with anonymized metadata. Institutional review board approval and patient consent were obtained at all sites and from all participants in the ADVANCE registry. For the commercial cases, all sites provided local institutional review board approval to publish any metadata available from the FFR CT submission. HeartFlow served as a sponsor of the AD-VANCE registry (https://ClinicalTrials.gov: NCT02499679).
All data in the study were held by the authors, and data analysis, reporting, and submission were performed independent of the sponsor.

Study Cohorts
ADVANCE registry.-The ADVANCE registry is a multicenter prospective controlled registry; patients were recruited for inclusion from July 15, 2015, to October 20, 2017. The ADVANCE registry was designed to evaluate clinical utility, outcomes, and resource use after FFR CT -guided treatment in clinically stable symptomatic patients with a coronary artery disease diagnosis at coronary CT angiography (12,13). The inclusion and exclusion criteria have been previously published and are available in Figures E1-E3 (online) (14). Analyses performed using FFR CT analysis software (version 2.0 or higher) were included in the current analysis; 1450 analyses performed using software versions older than version 2.0 were excluded. Version 2.0 or higher software was used because these are the versions that are currently used in clinical care; thus, they are the most generalizable to current practice.
Clinical submission cohort.-All patients referred for clinical consecutive coronary CT angiography between July 2016 and March 2018 and who had FFR CT analysis performed with version 2.0 or higher FFR CT analysis software were considered for inclusion. Before this point, 5062 cases in the clinical cohort were analyzed with software versions older than 2.0 and were not included in the current study. There was no overlap of participants in the clinical cohort with those in the ADVANCE cohort. All coronary CT angiography examinations were performed according to the local policy of the referring center. When a scan was rejected for analysis, a reason was included in the rejection notification to the referring center. If this could be modified, the case could then be resubmitted. Only the final accepted scans were included. The rationale for this was that if a scan could be amended and subsequently processed, it was simply a postprocessing issue, whereas the rejections that were not amenable to such simple steps were of interest because they represented a lost opportunity.
During the initial commercial set-up, there is an intensive process for the first few weeks, during which structured education and feedback are provided to ensure optimization of the data submitted for analysis; this is referred to as onboarding. Centers that were still onboarding at the end of the study period and all scans performed during the onboarding process (n = 739) were excluded from this study.
FFR CT analysis.-The decision to refer a patient for FFR CT analysis was made by the physician interpreting the coronary CT angiography images. All FFR CT analysis was performed at a single center (HeartFlow). Analysis was performed as previously described (15). (c) Image quality rejections included cases in which image quality was inadequate to allow for generation of the anatomic coronary model necessary for FFR CT calculation. This decision was made based on two parallel assessments. The first was an image quality scoring system of the coronary arteries in which the length of the artifact, diameter of the vessel in which the artifact is present, whether a stenosis is present within the region of the artifact, and the type of artifact were combined and weighted according to the location of the artifact within the vessel, with more proximal artifacts given greater weighting. Scans with an inadequate image quality score were rejected. The second related decision process was based on the presence of any regions of uninterpretability (ROUs). Cases were not processed if one vessel system contained two ROUs longer than 5 mm, all three vessel systems contained one ROU longer than 5 mm, or one vessel system contained one ROU longer than 30 mm. If a scan failed because of the image quality or the ROU criteria, it was rejected. (d) For these patients, the reason for poor image quality was recorded as blooming artifact, motion artifact, or image noise. Examples of coronary CT angiography studies rejected for inadequate image quality are shown in Figures E1-E3 (online).
These criteria have remained stable since software was updated from version 1.6 through version 2.0 or higher. DICOM data from technical rejections are not stored and are not available for analysis.

Statistical Analysis
Statistical analysis was performed with statistical software (SPSS, version 23, SPSS, Chicago, Ill; R, version 2.15.2, R Foundation for Statistical Computing, Vienna, Austria). Continuous variables were expressed as mean 6 standard deviation or median and 25th to 75th percentile, as appropriate. Discrete variables were expressed as absolute numbers and percentages. Student independent t tests or Mann-Whitney tests were used, as appropriate, to compare continuous variables among pa-and technical factors was extracted. The patient factors obtained were heart rate (HR) and HR variability. The technical factors obtained included scanner type, number of sources, number of sections, section thickness, pixel spacing (distance between the center of the pixel in the x-y plane, which in turn is determined by the reconstructed field of view), temporal resolution, tube voltage, aortic attenuation, aortic noise, and effective radiation dose measured as the product of dose-length product times for conversion coefficient for the chest (K = 0.026 mSv/mGy·cm) (16). From the DICOM imaging data, aortic contrast material opacification was calculated as the mean attenuation (in Hounsfield units) within the aortic root volume extracted for FFR CT analysis, while image noise was measured as the standard deviation of Hounsfield units within the same volume of the aortic root. The signal-to-noise ratio was calculated as the mean attenuation (in Hounsfield units) in the aorta divided by the standard deviation of attenuation in the aorta. The pixel spacing contained within the DICOM metadata was used to back-calculate the reconstructed field of view, as this is the more intuitive and clinically appreciable input metric. This was done by multiplying the pixel spacing by 512 to coincide with the size of the CT output matrix.
End point.-Rate of coronary CT angiography rejection for FFR CT analysis was defined as the inability of technologists at HeartFlow, who were blinded to the clinical data, to perform FFR CT analysis. The reason for rejection was recorded and classified as follows: (a) Technical rejection included all automatic rejections due to the data failing to meet minimum technical specifications for FFR CT analysis due to data format. Reasons for technical rejection are as follows: section thickness or spacing Flow diagram of the study participants in both cohorts. * The number of cases is greater than the total number of rejections, as cases could be rejected for more than one reason. Technical rejection means all automatic rejections were due to the data failing to meet minimum technical specifications for fraction flow reserve derived from coronary CT angiography (FFR CT ) analysis due to data format. Image quality rejection means image quality was inadequate to allow for generation of the anatomic coronary model necessary for FFR CT (Table 3). Rejection of coronary CT angiography scans for FFR CT analysis was greater in patients with a history of smoking, diabetes, or angina (Table 1). Technical factors associated with coronary CT angiography rejection for FFR CT analysis were lower use of dual-source CT scanning, lower aorta contrast, wider reconstructed field of view, and greater section thickness (Table 1). Finally, HR was higher in the rejected cases than in the accepted cases (63 beats per minute 6 14 vs 60 beats per minute 6 9, P = .03), with no difference in HR variability between groups. Tables 4 and 5 show uni-and multivariable analysis of factors associated with rejection for analysis in the ADVANCE registry. In the fully adjusted model, smoking history (smokers vs tients in whom FFR CT analysis was performed versus those who were rejected for FFR CT analysis. Comparisons among groups of discrete variables were performed by using the x 2 or Fisher exact test if the expected cell count was less than five. Binary logistic regression was used to assess the association between baseline covariates and FFR CT rejection (results are presented as odds ratio [OR] and 95% confidence interval [CI]). Variables with P , .1 at univariable analysis were then included as covariates in multivariable analysis. All tests were two tailed, and P , .05 was considered to indicate a significant difference.

Cohort Characteristics
This study included 2778 participants from the ADVANCE registry (1845 [66%] were men; mean age of the total cohort, 66.2 years 6 10.3; mean age of men, 65.1 years 6 10.2; mean age of women, 68.5 years 6 10.0; P , .001 for between-sex age difference) and 10 621 participants from the clinical cohort. A flow diagram of the study participants can be found in the Figure. The characteristics of the cohorts are listed in Tables 1 and 2.

ADVANCE Registry
Scans of the 2778 participants were submitted from 38 sites in Europe, North America, and Japan. Coronary CT angiography was performed using 16 scanner models produced by four dif- CI: 6.23%, 7.18%) were rejected for inadequate image quality; the remainder were rejected for technical limitations of the submitted data or because of the presence of stents, bypass grafts, or other cardiac hardware ( Table 3). The coronary CT angiography rejection rate was higher in the clinical cohort than in the AD-VANCE registry (P , .001). Participants in whom the coronary CT angiography scan was rejected for FFR CT analysis showed a lower use of dualsource and wide-coverage CT scanners, lower aorta contrast, wider reconstructed fields of view, and greater section thickness as compared with participants in whom coronary CT angiography scans were accepted (P , .001 for all) ( Table 2). Participants in whom the coronary CT angiography scan was rejected had a higher HR (mean, 65 beats per minute 6 12 vs 60 beats per minute 6 9; P , .001) and a higher HR variability (mean, 24 beats per minute 6 42 vs 17 beats per minute 6 29; P = .001) compared with those in whom the coronary CT angiography scan was accepted ( Table 2). When considering only wide-detector and dual-source CT scanners, dual-source CT scanners were associated with lower rates of coronary CT angiography rejection for FFR CT

Discussion
The main findings of our study are that the rate of rejection of coronary CT angiography scans for fractional flow reserve derived from coronary CT angiography (FFR CT ) analysis due to inadequate image quality is low, and it is lower in the Assessing Diagnostic Value of Noninvasive FFR CT in Coronary Care (ADVANCE) cohort than in the clinical cohort. We also found temporal resolution, section thickness, and heart rate (HR) are independent predictors of coronary CT angiography scan rejection for FFR CT analysis.

Clinical Cohort
A total of 10 621 cases were submitted for FFR CT analysis between July 2016 and November 2017. A total of 205 submissions were automatically rejected due to the DICOM files not meeting the minimum technical data specifications. Therefore, the final population comprised 10 416 patients. The cases were submitted by 76 centers (median, 78 scans per center; range, one to 884 scans) covering 10 countries in North America, Europe, and Asia. These were performed by using 22  ). Low-tube potential (100 kVp) was used in 587 participants (5.6%) ( Table 2). The mean HR and HR variability were 60 beats per minute 6 9 and 17 beats per minute 6 30, respectively, with a mean dose of 9.9 mSv 6 7.8 ( Table 2).
The rejection rate was 892 of 10 416 cases submitted (8.60%; 95% CI: 7.89%, 8.93%). Of the 10 416 cases, 711 (6.8%; 95% * Values are greater than the total number of rejections because scans can be rejected for more than one reason. † Inadequate image quality is based on (a) length of the artifact, diameter of the vessel in which the artifact is present, whether a stenosis is present within the region of artifact, and the type of artifact is weighted according to the location of the artifact and (b) the presence of any regions of uninterpretability (ROUs). Cases are not processed if one vessel system contains two ROUs longer than 5 mm, if all three vessel systems contain one ROU longer than 5 mm, or if one vessel system contains one ROU longer than 30 mm.
ADVANCE registry and 79% of the clinical cases submitted for analysis were performed with these types of scanners. An additional explanation for the difference observed between our study and the PROMISE FFR CT substudy is that only 67% of CT angiography data sets met Society of Cardiovascular Computed Tomography, or SCCT, guidelines, yielding a higher rejected rate.
The main reason for FFR CT rejection in our study was motion artifacts. In the Diagnostic Accuracy of Fractional Flow Reserve from Anatomic CT Angiography, or DEFACTO, trial, misalignment artifacts resulted in impaired sensitivity and single-source scanners. Previous studies with dual-source technology show that image quality is less dependent on HR as compared to 64-section single-source coronary CT (17,18) and that wide-coverage scanner technology allows for single-beat whole-heart acquisition, reducing the occurrence of motion or step artifacts (19,20). Despite these advances in CT hardware over the past 2 decades, HR continues to be closely associated with image quality at coronary CT angiography (21). Coronary CT angiography performed using these scanners accounted for only 17% of scans in the PROMISE FFR CT substudy, which had a 33% rejection rate (9). In our study, 85% of cases in the The reconstructed field of view was associated with rejection in the clinical cohort but not in ADVANCE cohort. The reconstructed field of view was 10% smaller in the ADVANCE cohort (18.5 cm) than in the clinical cohort (20.7 cm), suggesting a reticence of the clinical referral centers to optimally crop the images for cardiac analysis, presumably because they feared cropping out the coronary arteries or cardiac chambers. Our study had some limitations. A referral bias is likely present, as referring cardiac imagers were unlikely to submit visually nondiagnostic coronary CT angiography scans for consideration for FFR CT . Thus, our study most likely represents the percentage of visually diagnostic coronary CT angiography scans amenable to subsequent FFR CT analysis rather than the percentage of all coronary CT angiography scans that are amenable to FFR CT analysis. Moreover, there will be differences between the ADVANCE cohort and the clinical cohort, and the results of the two populations cannot be considered interchangeable, nor can the conclusions for one cohort be applied to the other.
In conclusion, in both a clinical registry and in cases undergoing clinical analysis, the rate of rejection of coronary CT angiography for fractional flow reserve derived from coronary CT angiography (FFR CT ) due to inadequate image quality is significantly lower than that reported in previous studies. Optimization of scans through heart rate control, timing of contrast material administration, minimization of section thickness and reconstructed field of view, and greater use of dual-source technology or a wide-coverage scanner holds the potential to enable FFR CT analysis in most patients. overall accuracy of FFR CT , while use of b-blockers yielded higher FFR CT specificity (10). These findings support close adherence to the current guidelines for coronary CT angiography (22,23). Unfortunately, previous studies showed that adherence to these guidelines is variable and that the ideal HR for coronary CT angiography is achieved in only 55% of patients (17,24).
For both cohorts in our study, section thickness and pixel size showed a clear association with rejection of FFR CT analysis. Given that greater section thickness and pixel size have reduced image noise (25), it may be that the cardiac imagers are selecting settings to improve image quality for visual analysis. However, peak aortic enhancement is a more important contributor than image noise for the ability to perform FFR CT analysis, as evidenced by our observation that lower contrast opacification but not greater image noise was associated with rejection. Thus, it may be beneficial to reconstruct studies undergoing FFR CT analysis with a small field of view and the smallest possible section thickness.
Our result showing that the ADVANCE cohort had a lower rate of coronary CT angiography rejection than the clinical cohort warrants consideration. Aortic contrast opacification was a significant predictor of rejection in the clinical cohort but not in the ADVANCE registry. Tube potential is an easily amendable scanning factor that boosts intravascular contrast opacification as it moves the kilovoltage peak closer to the k-edge of iodine; however, low-kilovoltage-peak scanning was implemented in only 6% of the clinical cases, whereas it was implemented in 50% of the ADVANCE cases, suggesting that optimizing scanner parameters eliminates suboptimal aortic contrast opacification. Note.-Data in parentheses are 95% confidence intervals. ADVANCE = Assessing Diagnostic Value of Noninvasive FFR CT in Coronary Care, FFR CT = fractional flow reserve calculated from CT. * Heart rate variability and dose were excluded from multivariable analysis in the clinical cohort, as one or the other were not routinely collected with the Digital Imaging and Communications software of GE Healthcare, Philips, and Toshiba scanners, and this would have introduced a systematic bias in the results. A total of 7867 cases were included in the model due to missing variables in some cases.