TCT-504 Comparison of Computed Tomography derived Fractional Flow Reserve to invasive Fractional Flow Reserve in Diagnosis of Functional Coronary Stenosis: A Meta-Analysis

METHODS CTA using a 320-row multidetector CT scanner was performed in 321 consecutive patients with suspected CAD. Plain abdominal CT scanning at the umbilical level was also performed to measure visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT) areas. To assess visceral fat accumulation, VAT/SAT (V/S) ratios were calculated. The severity of coronary artery stenosis was assessed on the basis of CTA. Significant stenosis was defined as 50% diameter stenosis. The CT-SXscore was calculated in patients with 1 significant stenoses. Patients were also divided into four groups based on SAT and VAT area median values.


Results
Eligible studies. Our search yielded 559 articles with relevant publications. Most publications were not relevant for the meta-analysis given the use of a broad search strategy. After exclusion of duplicates, 452 articles remained and further 366 articles were excluded based on the review of title and the abstract. After final review and application of inclusion and exclusion criteria, seventeen studies were included in the analysis. Of the seventeen studies that were included, four were multicenter prospective studies 14,[18][19][20] , six were single center prospective studies 24,[26][27][28]30,33 and seven were single-center retrospective studies [21][22][23]25,29,31,32 . Eight studies reported diagnostic accuracy of CTFFR on a per-patient basis and 17 studies reported on an individual vessel level basis. Study and baseline characteristics are detailed in Tables 1 and S1. Individual study estimates of diagnostic accuracy of CTFFR to identify ischemia-causing lesions are detailed in Supplementary Table S2  for per-patient level, while Supplementary Table S3 details per-vessel diagnostic accuracy results. The overall quality of included studies per quality assessment of diagnostic accuracy studies tool (QUADAS-2) was good (Supplementary Figs S1-S3). Data Synthesis. Our study included a total of 17 studies with 1,294 patients. Among the 1,294 patients, a total of 2,191 individual vessels were assessed using FFR and CTFFR. The FFR was chosen as the gold standard and the results of CTFFR as expressed in reference to the FFR.
Further, posterior estimation using Bayes Nomogram was performed. At the patient level analysis using a pre-test probability of disease of 25%, a positive result on CTFFR would increase the likelihood of disease to 53%, whereas a negative test would decrease the likelihood of disease to 5% ( Supplementary Fig. S4). When individual vessel level analysis was repeated, the results were similar ( Supplementary Fig. S5). This therefore suggests a very high negative predictive value for exclusion of hemodynamically significant stenosis with CTFFR.    Publication bias. Assessment for publication bias was performed using Deek's funnel plot asymmetry test.
There was no evidence of publication bias at both per patient (p = 0.48) and per vessel level (p = 0.86) analyses. However, given the low sensitivity of the Deek's funnel plot as well as the small number of studies included in this meta-analysis, publication bias cannot be absolutely eliminated.

Meta-regression.
Multivariate meta-regression analysis was performed at per-patient level and per-vessel level for the covariates including study design, sample size (less or more than 100), year of publication, region of study, quality of study, method of CTFFR appraisal, prevalence of CAD, proportion of hypertensive patients, proportion of diabetic patients, proportion of smokers and proportion of patients with dyslipidemia to identify potential sources of heterogeneity (Supplementary Table S4). These exploratory variables were not found to have a significant impact on the heterogeneity of the results at both patient level as well as individual vessel level (Supplementary Table S5).

Discussion
The results of our meta-analysis suggest that CTFFR has good diagnostic accuracy in detecting functionally significant coronary stenosis when compared to the FFR. These findings remained consistent at both per-patient and per-vessel level analysis.  CCTA is useful for noninvasive assessment of CAD in stable ischemic heart disease, as well as in patients with acute onset chest pain and acute coronary syndromes [34][35][36] . Studies have demonstrated its clinical utility and prognostic value in patients with suspected CAD with a very high negative predictive value 37 . However, CCTA overestimates the severity of stenotic lesions with only a minority of identified lesions being functionally ischemic therefore leading to increased utilization of ICA 4,38,39 . Thus, the management of obstructive lesions identified on CCTA remains challenging 40 , resulting in low positive predictive value. In the cardiac catheterization lab, the functional significance of a stenotic lesion in the coronary arteries has been evaluated using instantaneous wave free ratio (iFR) and or FFR. The FFR technique requires induction of coronary hyperemia with adenosine, but the iFR technique does not need induction of coronary hyperemia 41 . These tests have been well validated and the new hybrid algorithm using iFR and FFR has been adapted in several catheterization laboratories across the United States 41 . Depending on the functional significance of the lesion, immediate revascularization of the stenotic lesion or deferred revascularization with medical treatment strategies could be adopted 13,42,43 . However, these techniques require an invasive approach and the peri-procedural complications associated with such an approach are not trivial. Also, ICA could result in unnecessary hospitalization and additional costs to the patient.
The ideal alternative would therefore be a non-imaging test that could assess the functional significance of a coronary stenotic lesion. The CTFFR was therefore proposed as a test to determine the functional significance of coronary lesions by use of computational fluid dynamics to the CCTA images 15,16 . Using multiplanar imaging and  computer software, virtual hyperemia and computation modeling could be performed to estimate the functional significance of the lesions using CTFFR approach 17 . Calculation of CTFFR involves three main elements: physiological modeling of blood flow along with anatomical modeling of coronary arteries, and solution of governing equations of blood flow using numerical methods 16 . Since CCTA images are obtained during resting conditions, reducing resting microcirculatory resistance in the computational model simulates hyperemic response leading to computation of CTFFR 17 . As diseased and healthy vessels adapt to the amount of blood flow they carry, microcirculatory resistance is modeled. The luminal area of the feeding coronary artery is inversely proportional to the resistance of the vessel 17,44,45 . CTFFR therefore offers several advantages in terms of obviating the need for vasodilator administration, altering of the CCTA protocols or the need for additional radiation beyond what is normally expected during acquiring of CCTA images 46 .
In our meta-analysis, we observed that the CTFFR performed very well compared to the FFR in diagnosing functional significance of coronary stenosis and more importantly affords a very high negative predictive value to the test as seen on the Bayes nomogram. However, the studies did not report outcomes data and therefore at this time it is unclear if CTFFR strategy could result in improved prognosis by deferring coronary angiography and potentially revascularization. But it is clear from the results of our analysis that CTFFR offers more information than the traditional CCTA alone.
More recently, post hoc analysis based on hypothetical simulations of outcomes to assess clinicoeconomical effectiveness of using CTFFR to guide clinical decision making were reported 47,48 . Hlatky et al., projected that the patient stratified by CTFFR to undergo ICA and PCI had 30% lower costs and 12% fewer events at 1 year compared to the patients undergoing ICA with visually guided PCI 47 . In the other study by Kimura et al., a projected cost savings of 32% and 19% fewer adverse cardiac events were observed using a similar strategy as Hlatky et al. 48 . Further, Kim et al., demonstrated the improvement in diagnostic accuracy of CTFFR in identifying ischemic lesions by virtual coronary stenting using advanced computational models 20 . In the Prospective Longitudinal Trial of FFRCT: Outcomes and Resource Impacts (PLATFORM) study, the patients in the CTFFR arm had 32% lower costs of utilization but similar QOL scores in comparison to patients in ICA arm at 90 days 49 .
There are limitations to both the CTFFR and FFR techniques. In the CTFFR technique, artifacts induced from patient motion, blooming artifact from coronary calcification and insignificant contrast enhancement may limit the validity of the results 50 . In the case of FFR, procedural technique, the dose of vasodilator agent used to induce maximal hyperemia, distal microvascular resistance of the coronary arteries and high left ventricular end diastolic pressures can affect the results. Therefore, adequate attention to the patient's clinical presentation and symptoms should also be factored in prior to proceeding with revascularization in addition to the above techniques.
Our study has several limitations. The studies included were few and had a small number of patients which decreases the robustness of the findings. Since, this is a meta-analysis all the weaknesses inherent to the individual studies will be inherited to our study. The included studies were published over a period of 5 years and there could have been alterations to the protocols in the study as well as improvements to the imaging and software algorithms that could contribute to the heterogeneity among the studies. Further, there could be inter-observed and intra-observer variability in interpreting the findings on these respective tests. Also, the dose of adenosine infused to induce coronary hyperemia could be variable among different centers across which these studies were performed contributing to varying findings among the studies. All the studies excluded patients with bypass grafts and therefore our results are not applicable to these patients. Lastly, publication bias although was tested and excluded could contribute to the findings of our study.

Materials and Methods
Data sources and searches. An extensive electronic search of Cochrane, SCOPUS and PubMed databases was performed for relevant articles using the following search terms: "computed tomography", "CT angiography", "computed tomography angiography", "fractional flow reserve", "FFR", "fractional flow reserve, myocardial". The search was restricted to publications in English and the final search was performed on April 2017. Study Selection. The following inclusion criteria were applied: (1) Age of the study population >18 yrs.; (2) CTFFR being the index test; (3) Design of the study being a diagnostic accuracy study; (4) FFR was chosen as the reference standard; (5) Data must allow construction of two-by-two contingency table. In case of multiple studies published from the same institution, the study reporting the highest number of subjects was included and the remaining studies were excluded from the final analysis.
Data extraction and quality assessment. The initial search was performed independently by two reviewers and studies were selected for inclusion by mutual consensus. In case of disagreement, a third reviewer resolved disagreements between reviewers through discussion to achieve a consensus. Once the studies were selected for inclusion, the data was screened to meet the inclusion criteria and thereafter the following data points were extracted: true positive (TP), true negative (TN), false positive (FP) and false negative (FN). Additionally, cumulative demographic variables of patients included in the study were also extracted. We then performed meta-analysis at both the patient and the individual vessel levels. The quality of included studies was assessed using QUADAS-2 tool 51 .
Data synthesis and Analysis. Continuous variables are presented as mean values and categorical data as percentages. The analysis of diagnostic performance of CTFFR was performed at both patient and individual vessel levels. We estimated the pooled sensitivity and specificity. In addition, negative predictive value (NPV), positive predictive value (PPV), LR+, LR− and DOR with respective 95% confidence intervals were performed. A weighted average of the pooled data on per-patient or per-vessel level was performed using bivariate random effects model 52  transformation of the mean logit sensitivity and specificity and respective standard errors. Statistical heterogeneity was defined as I 2 statistic value greater than 50% 53 . The summary receiver operating characteristic curve (SROC) was derived using the logit estimates of sensitivity, specificity and their respective variances. This was then used to construct a hierarchical SROC curve and the area under the curve (AUC) estimated. An AUC range of 0.75-0.92 was considered to have good degree of diagnostic accuracy and an AUC range of 0.93-0.96 was considered to be a very good diagnostic test 54 . Publication bias was visually assessed using Deek's funnel plot. The Deek's funnel plot utilizes regression of diagnostic log odd's ratio against 1/square root (effective sample size) and further weighting by effective sample size. A p-value of the slope coefficient of <0.10 indicated asymmetry and thereby publication bias 55 .
Meta-regression analysis was performed to identify potential sources of heterogeneity. We performed meta-regression for covariates including study design, sample size (less or more than 100), year of publication, region of study, quality of study, method of CTFFR appraisal, prevalence of coronary artery disease, proportion of hypertensive patients, proportion of diabetic patients, proportion of smokers and proportion of patients with dyslipidemia to identify potential sources of heterogeneity. Further, post-estimation models using Bayes methods were performed to compute the post-test likelihood of presence or absence of disease using a pre-test likelihood of having disease of 25%. This meta-analysis was performed in compliance with the Meta-analysis of Observational Studies in Epidemiology and the quality of reporting for meta-analysis and Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines 56,57 . All the above analyses were performed in R, version 3.4.3, using "mada" package. The calculations for AUC and its confidence interval were performed using MetaDiSc version 1.4.

Conclusion
The results of our study suggest that CTFFR has a good diagnostic accuracy and performs well in comparison to the FFR in the assessment of functional significance of the coronary stenotic lesions. The CTFFR technique is still evolving and needs further evaluation with assessment of clinical outcomes prior to widespread clinical adaptation.