J A C C : C A R D I O V A S C U L A R I M A G I N G VO L . - , N O . - , 2 0 1 8

ª 2 0 1 8 B Y T H E AM E R I C A N C O L L E G E O F C A R D I O L O G Y F O UN DA T I O N

P U B L I S H E D B Y E L S E V I E R

Anatomical and FunctionalComputed Tomography forDiagnosing HemodynamicallySignificant Coronary Artery Disease

A Meta-Analysis

Csilla Celeng, MD, PHD,a Tim Leiner, MD, PHD,a Pál Maurovich-Horvat, MD, PHD, MPH,b Béla Merkely, MD, PHD,b

Pim de Jong, MD, PHD,a Jan W. Dankbaar, MD, PHD,a Hendrik W. van Es, MD, PHD,c Brian B. Ghoshhajra, MD, MBA,d

Udo Hoffmann, MD, MPH,d Richard A.P. Takx, MD, PHD, MSCa,c,d

ABSTRACT

ISS

Fro

Im

St.

Ma

the

Me

au

Ma

OBJECTIVES This meta-analysis determined the diagnostic performance of coronary computed tomography (CT)

angiography (CTA), CT myocardial perfusion (CTP), fractional flow reserve CT (FFRCT), the transluminal attenuation

gradient (TAG), and their combined use with CTA versus FFR as a reference standard for detection of hemodynamically

significant coronary artery disease (CAD).

BACKGROUND CTA provides excellent anatomic, albeit limited functional information for the evaluation of CAD.

Recently, various functional CT techniques emerged to assess the hemodynamic consequences of CAD.

METHODS This meta-analysis was performed in adherence to the PRISMA (Preferred Reporting Items for Systematic

Reviews and Meta-Analyses) guidelines. PubMed, EMBASE, and Web of Science were searched from inception until

September 7, 2017. Bayesian random effects analysis was used to compute pooled sensitivity, specificity, and the

summary receiver-operating characteristic curve of the index tests and compare them with the FFR as a reference

standard. Analyses were performed on vessel and patient levels. Because CTA has excellent sensitivity, specificity was

considered most relevant. Individual FFRCT values were collected.

RESULTS Overall, 54 articles and 5,330 patients were included. At vessel level, pooled specificity of CTP (0.86; 95%

confidence interval [CI]: 0.76 to 0.93), FFRCT (0.78; 95% CI: 0.72 to 0.83) and TAG (0.77; 95% CI: 0.61 to 0.89) were

substantially higher than that of CTA (0.61; 95% CI: 0.54 to 0.68). The addition of FFRCT, CTP, and TAG to CTA resulted

in high to excellent specificities (0.80 to 0.92). The summary receiver-operating characteristic curve at vessel level

yielded superior diagnostic accuracy for CTP, FFRCT, and combined CTA and CTP, compared with CTA. A subanalysis of on-

site versus off-site FFRCT revealed no substantial differences between the sensitivity (0.84 vs. 0.85) and specificity (0.80

vs. 0.73) of the 2 techniques. In a second subanalysis, dynamic CTP showed higher sensitivity (0.85 vs. 0.72), but had a

lower specificity (0.81 vs. 0.90) than static CTP.

CONCLUSIONS CTP and FFRCT demonstrated a substantial improvement in the identification of hemodynamically

significant CAD compared with CTA; therefore, their integration to clinical workflow before revascularization is

recommended. (J Am Coll Cardiol Img 2018;-:-–-) © 2018 by the American College of Cardiology Foundation.

N 1936-878X/$36.00 https://doi.org/10.1016/j.jcmg.2018.07.022

m the aDepartment of Radiology, University Medical Center Utrecht, Utrecht, the Netherlands; bMTA-SE Cardiovascular

aging Research Group Heart and Vascular Center, Semmelweis University, Budapest, Hungary; cDepartment of Radiology,

Antonius Hospital, Nieuwegein, the Netherlands; and the dCardiac MR PET CT Program, Department of Radiology,

ssachusetts General Hospital, Harvard Medical School, Boston, Massachusetts. Dr. Celeng has received grant support from

European Association of Cardiovascular Imaging. Dr. Ghoshhajra has been a consultant for Siemens Healthcare and

dtronic. Dr. Hoffman has received research grants from KOWA, MedImmune, Pfizer, Siemens, and HeartFlow, Inc. All other

thors have reported that they have no relationships relevant to the contents of this paper to disclose.

nuscript received January 17, 2018; revised manuscript received July 8, 2018, accepted July 12, 2018.

ABBR EV I A T I ON S

AND ACRONYMS

CTA = computed tomography

angiography

CAD = coronary artery disease

CTP = computed tomography

myocardial perfusion

FFR = fractional flow reserve

LR = likelihood ratio

MRI = magnetic resonance

imaging

FFRCT = fractional flow reserve

computed tomography

SROCs = summary receiver-

operating curves

TAG = transluminal

attenuation gradient

Celeng et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . - , N O . - , 2 0 1 8

Functional CT for Diagnosing CAD - 2 0 1 8 :- –-

2

C omputed tomography angiography(CTA) has a high sensitivity in theevaluation of coronary artery dis-

ease (CAD) (1–4). However, its specificity forthe detection of hemodynamically significantCAD is still moderate at best, especially incases of intermediate coronary stenosis(30% to 70% reduction in coronary lumendiameter) (5).

Invasive fractional flow reserve (FFR) isthe current reference standard for deter-mining the hemodynamic significance of CAD(6–8). FFR is calculated as the ratio betweenthe maximum achievable myocardial bloodflow during hyperemia in the presence of astenosis and the maximum achievablemyocardial blood flow in the absence of a

stenosis (6). FFR has a “grey zone” of ischemicthreshold value (between 0.75 and 0.80) (7). Previouscomprehensive trials showed that the use ofFFR-guided revascularization reduced the rate ofcomposite endpoints (death, nonfatal myocardialinfarction, repeat revascularization) and urgentrevascularization (9,10). Computed tomographymyocardial perfusion imaging (CTP) has been pro-posed as an alternative to magnetic resonance imag-ing (MRI) and nuclear cardiology techniques for thediagnosis of hemodynamically significant CAD (11). Inaddition, new noninvasive CT techniques haveemerged, which allow the assessment of potentialhemodynamic changes attributable to CAD withoutmodification of imaging protocols or an additionalradiation dose. Advances in computational fluid dy-namics allow for determination of significantischemia by computation of noninvasive FFR CT(FFRCT) or the transluminal attenuation gradient(TAG) (12–15). The aim of this study was to conduct ameta-analysis to determine the diagnostic perfor-mance of CTA, CTP, FFRCT, and TAG, and the com-bined use of these techniques with CTA versusinvasive FFR as the reference standard for thedetection of hemodynamically significant CAD in pa-tients with suspected or known CAD.

METHODS

This systematic review and meta-analysis was con-ducted in agreement with the PRISMA (PreferredReporting Items for Systematic Reviews and Meta-Analyses) statement (16). The protocol was publishedonline (www.crd.york.ac.uk/PROSPERO, PROSPERO2015 CRD42015017148). PubMed, EMBASE, and Webof Science were systematically searched from

inception to September 7, 2017. The search syntax ispresented in Online Table 1. No search restrictionswere imposed for language or date of publication. Amanual reference check of included articles was per-formed to identify additional articles missed by oursystematic search.

ELIGIBILITY CRITERIA. Two independent re-searchers (C.C. and R.A.P.T.) applied pre-definedcriteria to determine article eligibility. Articles wereincluded based on the following criteria: studydomain–patients with suspected or known CAD; in-dex test–CTA, CTP, FFRCT, and TAG, and the combi-nation of CTA and CTP, CTA, and FFRCT, and CTA andTAG; reference standard–invasive FFR for interme-diate coronary lesions; study result–agreement be-tween the index and reference standard; and studydesign–cross-sectional study. In cases of studies withoverlapping populations, the study with the largerpopulation sample was included. Discordances be-tween reviewers were resolved by consensus. Animalstudies, phantom studies, and case reports (n < 10)were not included.

DATA EXTRACTION AND CRITICAL APPRAISAL.

Data were extracted by one reviewer (C.C.) and veri-fied by a second reviewer (R.A.P.T.). The followingdata categories were extracted from the includedstudies: study population characteristics, index testcharacteristics, and reference test characteristics.True positive, false positive, false negative, and truenegative numbers were obtained. The results weresummarized in 2-by-2 contingency tables. In cases ofcombined modalities, the reviewers included onlythose studies that judged the combined test positiveif both CTA and functional CT were positive for he-modynamically significant CAD. This means that pa-tients with a positive test on CTA but negative onfunctional CT, or a negative CTA and positive test onfunctional CT, were deemed as not having hemody-namically significant CAD. In cases of multiple imag-ing thresholds, the pre-defined primary endpoint wasselected. If the primary endpoint was not reported,the best performing threshold was chosen. In addi-tion, the performance of FFRCT was evaluated atvessel level by digitizing scatterplots or Bland-Altmanplots, in which invasive FFR was used as the refer-ence standard using a previously described methodby Cook et al. (17). Because individual points in ascatterplot can overlap, studies with >10% missingdata points were omitted from the analysis. If indi-vidual FFRCT values were listed in a table, then thosevalues were extracted. A threshold of invasive FFRof #0.80 was used as the cutoff for hemodynamically

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . - , N O . - , 2 0 1 8 Celeng et al.- 2 0 1 8 :- –- Functional CT for Diagnosing CAD

3

significant CAD. Two independent reviewers (C.C.and R.A.P.T.) assessed the relevance and validity ofthe included studies using the QUADAS-2 (QualityAssessment of Diagnostic Accuracy Studies) criteria(18). Discrepancies were resolved by consensus. Themaximum time between noninvasive and invasivestenosis degree assessment was also recorded. If thisperiod was >2 months, the flow and timing of thestudy was scored as high bias.

DATA SYNTHESIS AND ANALYSIS. Analyses wereperformed at both vessel and patient level. Possiblepublication bias per imaging technique was assessedgraphically by generating funnel plots for modalitiesthat included $10 studies. Sensitivity, specificity,positive likelihood ratio (þLR) and negative LR (�LR),and the log diagnostic odds ratio, including the95% confidence interval, were calculated. Pooledresults were used to calculate the summary receiver-operating curves (SROCs). Heterogeneity wasassessed by visual inspection of SROCs (19). Due toanticipated methodological heterogeneity, Bayesianrandom effects analysis was performed, because itprovides more robust results in cases of studies with arelatively small sample size. Furthermore, it alsotakes the potential sources of variation (e.g., impre-cision of sensitivity and specificity) into account (20–23). The logit link function was used. The penalizedcomplexity prior framework was applied, whichresulted in more precise estimates (23). The contrastwas chosen at p ¼ 0.05 (sigma > 3), which corre-sponded to sensitivities or specificities within an in-terval of 0.5 to 0.95, with a 95% probability (22). Post-test probabilities per imaging technique were calcu-lated based on various pre-test probabilities and thepooled LRs. Subgroup analyses were performed toreveal potential performance differences between on-site FFRCT versus off-site FFRCT and dynamic CTPversus static CTP. For the subgroup analysis of indi-vidual FFRCT at vessel level, a graph was generatedthat displayed the sensitivity and specificity atdifferent FFRCT values. Youden’s index was used todetermine the optimal threshold (i.e., by identifica-tion of the threshold that maximizes the sum ofsensitivity plus specificity). In addition, the FFRCT

value was determined, at which sensitivity or speci-ficity was $90%. Agreement between invasive FFRand FFRCT values was considered when both testswere #0.80 or >0.80. All statistical analyses wereperformed with the statistical software R (version3.42, R Foundation for Statistical Computing, Vienna,Austria) and the package “meta4diag” (version 2.0.6,arXiv:1512.06220v2 [stat.AP]) and MedCalc (version18.5, Mariakerke, Belgium).

RESULTS

After removal of duplicates, the search retrieved 1,377potentially relevant articles. Full text review wasperformed in 150 articles that met our pre-definedinclusion criteria. Included studies were publishedbetween 2008 and 2017. From 54 studies (n ¼ 5,330),39 studies reported on CTA, 12 reported on CTP, 10assessed CTA and CTP, 6 reported on TAG, 4 reportedon CTA and TAG, 18 assessed FFRCT, and 3 reportedon CTA and FFRCT (Figure 1). The average weightedradiation exposure was 5.0 mSv for prospectivelygated CTA, 11.4 mSv for retrospectively gated CTA,and 6.9 mSv for stress CTP only (not including restCTA). Funnel plots did not reveal obvious publicationbias (Online Figure 1). Study population characteris-tics, index test characteristics, and reference testcharacteristics were retrieved and are listed in OnlineTables 2 to 4. Quality assessment using QUADAS-2criteria showed that patient selection specificallywas a source of bias (Figure 2).

POOLED SENSITIVITY AND SPECIFICITY AT VESSEL

AND PATIENT LEVEL. At vessel level, pooled sensi-tivity was the best for CTA (0.87), followed by FFRCT

(0.85), CTA and CTP (0.82), CTP (0.81), CTA andFFRCT (0.76), CTA and TAG (0.70), and finally TAG(0.59). The highest pooled specificity was reached byCTA and TAG (0.92), followed by CTA and CTP (0.88),CTP (0.86), CTA and FFRCT (0.80), FFRCT (0.78), TAG(0.77), and CTA (0.61). More detailed diagnostic per-formances are listed in Table 1 and Online Figure 2.The subanalysis of on-site (11 studies) versus off-site(7 studies) FFRCT on vessel level revealed no sub-stantial differences between the sensitivity (0.84 vs.0.85) and specificity (0.80 vs. 0.73, respectively) ofthe 2 techniques (Online Figure 3). The second sub-analysis on dynamic CTP (7 studies) versus static CTP(5 studies) showed that dynamic CTP had highersensitivity (0.85 vs. 0.72) but lower specificity (0.81vs. 0.90) compared with static CTP (Online Figure 4).

At patient level, pooled sensitivity was the best forCTA (0.94), followed by FFRCT (0.89), CTA and CTP(0.89), CTP (0.83), and TAG (0.69). The highestpooled specificity was reached by CTA and CTP (0.81),followed by CTP (0.79), FFRCT (0.76), CTA (0.48), andTAG (0.39). More detailed diagnostic performancesare listed in Table 2 and Online Figure 5.

POOLED DIAGNOSTIC ACCURACY. The SROC atvessel level demonstrated superior diagnostic accu-racy for CTP, FFRCT, and CTA in combination withCTP, compared with CTA. CTA in combination withFFRCT, TAG, and CTA in combination with TAG were

FIGURE 1 Flowchart of Selection of the Included Studies

Potentially relevant articles identified throughPubMed (n = 404), Embase (n = 739) and Web

of Science (n = 976)

Studies reviewed for title and abstract(n = 1377)

Articles reviewed for full text (n = 150)

Studies included in the meta-analysis (n = 54)

CTA(n = 39)

CTP(n = 12)

CTA and CTP(n = 10)

TAG(n = 6)

CTA and TAG(n = 4)

FFRCT(n = 18)

CTA and FFRCT(n = 3)

Duplicates excluded (n = 742)

Studies excluded (n = 96)- No FFR as reference standard (n = 32)- Overlapping patient population (n = 23)- Animal study (n = 1)- Case report (n = 3)- Conference abstract (n = 16)- Review (n = 2)- Subgroup/not in domain (n = 19)

Studies excluded on basis of abstract(n = 1228)

Studies included based on referencecheck (n = 1)

Flowchart of selection of the included studies. CTA ¼ computed tomography angiography; CTP ¼ computed tomography perfusion; FFRCT ¼fractional flow reserve computed tomography; TAG ¼ transluminal attenuation gradient.

Celeng et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . - , N O . - , 2 0 1 8

Functional CT for Diagnosing CAD - 2 0 1 8 :- –-

4

associated with wide confidence intervals due to thelimited number of studies; hence, the true diagnosticperformance was indeterminate (Figure 3A). In linewith these findings, a large change in post-test

FIGURE 2 Quality of the Included Studies Assessed by QUADAS-2

Patient Selection

Index Test

Reference Standard

Flow and Timing

0% 25%

69%

56%

33% 6%

Low Bias

The chart illustrates that patient selection specifically is a potential sou

Studies.

probability was observed for functional CT over CTAalone, whereas TAG alone demonstrated only amodest change in post-test probability (Figure 4).Specifically, at a pre-test probability of 30%, in case of

50%

61%

13%

28%

89%

19%

17%

11%

75% 100%

Unclear High Bias

rce of bias. QUADAS ¼ Quality Assessment of Diagnostic Accuracy

TABLE 1 Outcome Summary Per Vessel

Imaging ModalityNumber ofVessels

Sensitivity (%)(95% CI)

Specificity (%)(95% CI) þLR (95% CI) �LR (95% CI)

LogDOR(95% CI)

CTA 6,400 0.87 (0.84–0.91) 0.61 (0.54–0.68) 2.27 (1.87–2.76) 0.21 (0.15–0.27) 2.40 (2.00–2.80)

CTP 2,118 0.81 (0.72–0.88) 0.86 (0.76–0.93) 6.28 (3.17–11.91) 0.23 (0.13–0.34) 3.29 (2.43–4.22)

CTA and CTP 1,785 0.82 (0.76–0.87) 0.88 (0.81–0.92) 6.97 (4.29–11.39) 0.21 (0.13–0.30) 3.49 (2.81–4.24)

FFRCT 2,432 0.85 (0.82–0.87) 0.78 (0.72–0.83) 3.86 (2.91–5.22) 0.20 (0.16–0.24) 2.97 (2.56–3.41)

CTA and FFRCT 362 0.76 (0.50–0.93) 0.80 (0.69–0.88) 4.00 (1.85–7.05) 0.31 (0.08–0.69) 2.64 (1.09–4.23)

TAG 726 0.59 (0.46–0.72) 0.77 (0.61–0.89) 2.89 (1.41–5.98) 0.53 (0.33–0.76) 1.64 (0.64–2.70)

CTA and TAG 327 0.70 (0.46–0.87) 0.92 (0.77–0.99) 15.98 (2.63–59.63) 0.33 (0.12–0.63) 3.56 (1.75–5.53)

CI ¼ confidence interval; CTA ¼ computed tomography angiography; CTP ¼ computed tomography perfusion; FFRCT ¼ fractional flow reserve CT; LogDOR ¼ log diagnostic odds ratio; þLR ¼positive likelihood ratio; �LR ¼ negative likelihood ratio; TAG ¼ transluminal attenuation gradient.

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . - , N O . - , 2 0 1 8 Celeng et al.- 2 0 1 8 :- –- Functional CT for Diagnosing CAD

5

a negative test, post-test probabilities were between8% and 11% for CTA, CTP, FFRCT, and the combina-tion of CTA and CTP as well as CTA and FFRCT. In caseof a positive result, CTA yielded a post-test proba-bility of 49%, whereas CTP achieved 73% and FFRCT

achieved 62%. On a patient level, the SROCs showedsimilar good diagnostic performance for CTP, CTA incombination with CTP, and FFRCT compared withCTA (Figure 3B). TAG had a lower diagnostic perfor-mance than CTA, with wide confidence intervals.

DIAGNOSTIC ACCURACY OF INDIVIDUAL FFRCT AT

VESSEL LEVEL. We extracted data on 1,370 vesselsfrom 14 studies (Online Table 4) and found an optimalFFRCT threshold of #0.80 using the Youden index,which resulted in an area under the curve of 0.87,with a sensitivity of 84% (95% CI: 81% to 87%) and aspecificity of 77% (95% CI: 74% to 80%). Using pre-determined thresholds, we observed $90% sensi-tivity at a FFRCT value of >0.82 and $90% specificityat a FFRCT value of #0.74 (Figure 5A). In vessels withFFRCT values between 0.74 and 0.82, diagnosticagreement with invasive FFR was weak (54% in 316vessels), whereas diagnostic agreement for FFRCT

values outside of this range was high (87% in 1,054

TABLE 2 Outcome Summary Per-Patient

Imaging Modality NSensitivity (%)

(95% CI)Specifici

(95%

CTA 3,101 0.94 (0.91–0.97) 0.48 (0.3

CTP 697 0.83 (0.71–0.92) 0.79 (0.6

CTA and CTP 619 0.89 (0.78–0.96) 0.81 (0.7

FFRCT 1,069 0.89 (0.85–0.92) 0.76 (0.6

CTA and FFRCT NA NA NA

TAG 188 0.69 (0.22–0.97) 0.39 (0.0

CTA and TAG NA NA NA

Abbreviations as in Table 1.

vessels). The distribution of the data also presentedas the frequency of invasive FFR at a different FFRCT

value (Figure 5B). The agreement between invasiveFFR and FFRCT is displayed in Figure 5C.

DISCUSSION

This meta-analysis demonstrated the additionalvalue of functional CT over anatomic CTA alone forthe assessment of hemodynamically significant CAD.The main results of this investigation can be sum-marized as follows. 1) In line with previous obser-vations, CTA showed excellent sensitivity; however,it had limited specificity both on vessel and patientlevels. 2) Regarding the functional CT imagingtechniques, CTP and FFRCT yielded high to excellentsensitivity. Interestingly, the specificity of boththese imaging techniques, especially CTP, was high.Evidence for TAG was limited and showed moderatesensitivity on both levels, whereas its specificityshowed substantial differences in performance atvessel and patient levels (0.77 vs. 0.39). Also, dis-crepancies in TAG could be dependent on theacquisition technique and scanner hardware, which

ty (%)CI) þLR (95% CI) �LR (95% CI)

LogDOR(95% CI)

7–0.59) 1.85 (1.48–2.36) 0.12 (0.06–0.19) 2.78 (2.13–3.51)

8–0.87) 4.10 (2.46–7.01) 0.22 (0.09–0.40) 2.97 (1.98–4.02)

3–0.90) 4.75 (3.16–6.99) 0.14 (0.04–0.30) 3.59 (2.54–4.78)

9–0.82) 3.76 (2.76–5.21) 0.15 (0.10–0.21) 3.25 (2.70–3.84)

NA NA NA

7–0.82) 1.55 (0.32–4.72) 1.36 (0.07–5.42) 0.45 (�2.70–3.61)

NA NA NA

FIGURE 3 SROC at Vessel and Patient Levels

CTA CTP CTA and CTP FFRCT

CTA and FFRCT TAG CTA and TAG

CTA CTP CTA and CTP FFRCT TAG

A

B

SROC

SROC

1−Specificity

Sens

itivi

ty

0 0.2 0.4 0.6 0.8 1

0

0.2

0.4

0.6

0.8

1

Sens

itivi

ty

0

0.2

0.4

0.6

0.8

1

1−Specificity0 0.2 0.4 0.6 0.8 1

Summary receiver-operating characteristic curve (SROC) at (A) vessel level and (B)

patient level. Each symbol represents an imaging modality and the corresponding circle

around the confidence intervals, except for CTA and FFRCT at vessel level and TAG at

patient level due to limited data points. Abbreviations as in Figure 1.

Celeng et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . - , N O . - , 2 0 1 8

Functional CT for Diagnosing CAD - 2 0 1 8 :- –-

6

might have resulted in differences in coronarycontrast concentrations and thus, could also haveinfluenced TAG results (24). 3) With the combinationof CTA and CTP as well as CTA and FFRCT, vessel-level sensitivity remained high, whereas CTA andTAG showed moderate sensitivity. The specificity ofall the combined techniques was high to excellent. 4)Regarding the diagnostic accuracy on both vessel-and patient-level all CT functional imaging tech-niques except TAG showed better performance thanCTA alone; however, on vessel-level the combinationof CTA and FFRCT as well as CTA and TAG showedwide confidence intervals. Therefore, their truediagnostic accuracy was unclear. 5) Concerning thechanges in the diagnostic probability, at a pre-testprobability of 30%, a negative test resulted in post-test probabilities of 8% to 11% for CTA, CTP, FFRCT,and the combination of CTA and CTP as well as CTAand FFRCT. Hence, a negative test decreased theabsolute probability of disease by w20%, althoughno improvement was observed over CTA alone. Incases of a positive result, CTA yielded a post-testprobability of 49%, whereas CTP achieved 73% andFFRCT achieved 62%; thus, these techniques moreaccurately identified hemodynamically significantCAD. This meta-analysis demonstrated that the useof functional CT imaging techniques might aid theclinical management of patients with suspected orknown CAD.

ADVANTAGES AND LIMITATIONS OF CTP. A previ-ous meta-analysis demonstrated that the perfor-mance of CTP was similar to MRI and positronemission tomography for the detection of hemody-namically significant CAD (25). In the present study,which included twice the number of vessels, weshowed that CTP yielded the same specificity. Despitethese advantages, the integration of CTP into clinicalworkflow is limited. A possible limitation, whichmight hamper the use of CTP, is the complexity of theimaging procedure. Currently, there are 3 acquisitionmethods for the detection and evaluation ofmyocardial perfusion defects: static single energy(myocardial contrast enhancement is obtained at theearly arterial phase); static dual energy (iodine ma-terial distribution map is created by using dual-energy acquisition); and dynamic CTP (myocardialcontrast enhancement is acquired several times dur-ing the first-pass) (26). The subanalysis of dynamicCTP versus static CTP demonstrated that the sensi-tivity of dynamic CTP was higher, but its specificitywas lower than that of static CTP. Both static anddynamic CTP can be performed during rest and/orstress.

FIGURE 4 Absolute Change in Post-Test Probability for the Various CT Techniques

Pre-Test Probability

Post

-Tes

t Pro

babi

lity

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

CTACTPCTA and CTP

TAG

Reference lineCTA and TAG

FFRCT

CTA and FFRCT

Absolute change in post-test probability of hemodynamically significant coronary artery disease by various pre-test probabilities. The

reference line shows no effect of the test on post-test probability. A large change in post-test probability was observed for functional CT over

CTA alone, whereas TAG alone demonstrated only a modest change in post-test probability. Abbreviations as in Figure 1.

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . - , N O . - , 2 0 1 8 Celeng et al.- 2 0 1 8 :- –- Functional CT for Diagnosing CAD

7

Based on our findings, most institutions use astress perfusion protocol with dynamic acquisition(53%) to improve the detection of perfusion defects(27,28). In terms of radiation exposure, static restCTP—in line with coronary CTA—requires a single-snapshot for the detection of contrast attenuationand is therefore associated with a lower radiationdose. We found that the weighted radiation exposureof stress CTP plus rest CTA leads to substantiallyhigher radiation (11.9 mSv) compared with that ofprospectively electrocardiographically triggered CTAalone (5.0 mSv). With use of low-kilovolt protocols,dynamic CTP is feasible at a lower radiation dose(29,30). Nevertheless, radiation exposure should belimited whenever feasible, and this might entail apreference for FFRCT.

ADDED VALUE OF FFRCT. Our meta-analysis showedthat the diagnostic performance of FFRCT was similarto that of CTP; moreover, it did not require additionalradiation nor the administration of adenosine.Furthermore, it allows to estimate functional changesin coronary artery flow after stent implantation (31).

Recently, Cook et al. (17) performed a systematic re-view, in which they included 5 FFRCT studies anddigitized plots to acquire individual correlations ofFFRCT and invasive FFR values. They concluded thatFFRCT around the cutpoint of 0.80 showed less cer-tainty, and that FFRCT values of >0.90 and #0.60provided almost complete certainty. In our meta-analysis, we performed a similar subanalysis on alarger number of vessels (1,370 vs. 908) and observedgood agreement with invasive FFR for FFRCT values>0.82 and #0.74. Of note, when evaluating valuesclose to a threshold (i.e., intermediate values), loweragreement will be observed, which is also true forrepeated measures of invasive FFR itself (32). Somediscrepancy between modeling and directlymeasured FFR is expected to be present, althoughwith advanced FFRCT algorithms, the use of machinelearning, and improvements in CT hardware, thisdiscrepancy is expected to decrease (33,34). Heart-Flow (Redwood City, California) FFRCT analysis wasthe first commercially available software that wasable to compute a CT-derived FFR value. In recentyears, various vendors developed other FFRCT

FIGURE 5 Individual FFRCT Values

0

50

FFRCT

≤0.30

0.30-035

0.35-0.40

0.40-0.45

0.45-0.50

0.50-0.55

0.55-0.60

0.60-0.65

0.65-0.70

0.70-0

.75

0.75-0

.80

0.80-0.85

0.85-0.90

0.90-0.95

0.95-1.00

100

150

200

250

Freq

uenc

y

B

0

10

20

30

40

50

60

70

80

90

100

Compared to Invasive FFR

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0FFRCT

(%)

Specificity

Sensitivity

A

40

50

60

70

80

90

100

0.3 0.4 0.5 0.6 0.7 0.8 1.00Average Between Invasive FFR and FFRCT

Agre

emen

t (%

)

0.9

C

(A) Sensitivity and specificity (with 95% confidence intervals) at different FFRCT values.

(B) The frequency of invasive FFR #0.80 (orange) and invasive FFR >0.80 (green) at

different FFRCT values. The dotted line represents the moving average trendline. (C) Per-

range agreement by plotting agreement against the average of invasive FFR and FFRCT

value. Abbreviations as in Figure 1.

Celeng et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . - , N O . - , 2 0 1 8

Functional CT for Diagnosing CAD - 2 0 1 8 :- –-

8

platforms. When using HeartFlow, CTA data aretransmitted to an off-site location and are analyzedby an independent laboratory. In case of on-siteassessment, FFRCT data are analyzed at the local

institution. Within the frame of this meta-analysis,we performed a subanalysis to determine thediagnostic accuracy of off-site FFRCT versus on-siteFFRCT. Our results indicated that the performanceof on-site FFRCT algorithms was as high as thatof off-site FFRCT. Of note, on-site FFRCT post-processing might be desirable because it allowsthe physician to access functional data in a moretimely manner. Nevertheless, most FFRCT algorithmsare proprietary, and the exact mechanism remainsunclear.

COSTS ATTRIBUTABLE TO FUNCTIONAL CT IMAGING.

Because health care costs of noninvasive versusinvasive imaging are substantially different, severalstudies aimed to assess the cost-effectiveness ofnovel noninvasive imaging techniques for identifi-cation of patients with significant CAD. The clinicaland economic outcomes of FFRCT versus usual carewere investigated within the framework of 3comprehensive trials (DISCOVER-FLOW [Diagnosis ofISChemia-Causing Stenoses Obtained Via Noninva-sivE FRactional FLOW Reserve], HeartFlowNXT[HeartFlow analysis of coronary blood flow using CTangiography: NeXt sTeps] and PLATFORM [Prospec-tive LongitudinAl Trial of FFRct: Outcome andResource IMpacts]) (35–37). Beyond improved orequivalent clinical outcome (12% to 19% fewerevents at 1 year), the use of FFRCT resulted inreduced costs (30% to 33% lower) by more accuratelyidentifying patients for revascularization. Based onthese promising results, HeartFlow Inc. receivedFood and Drug Administration clearance inNovember 2014. In the United Kingdom, a studydemonstrated that the integration of HeartFlowFFRCT as a diagnostic strategy for stable chest painresulted in cost savings of £200 per patient (38),which led the National Institute for Health and CareExcellence to recommend the adoption of HeartFlowFFRCT into current practice (39). The National Insti-tute for Health and Care Excellence consideredHeartFlow FFRCT cost-effective at a list price of £700(excluding VAT) per analysis compared with thecurrent treatment pathway of all functional imagingtests, which includes single-photon emissioncomputed tomography, MRI, and echocardiography.In case of TAG, no studies were reported on theadditional cost of TAG software, because the analysiscan be performed on-site using a dedicated work-station, thus the costs are expected to be limited.The cost-effectiveness of CTP was evaluated only in1 study, which reported lower costs and an addi-tional gain in quality-adjusted life years when using

PERSPECTIVES

COMPETENCY IN MEDICAL KNOWLEDGE 1: Current clinical

guidelines do not encompass the application of functional CT

imaging techniques for the detection of hemodynamically rele-

vant ischemia in patients with stable CAD.

COMPETENCY IN MEDICAL KNOWLEDGE 2: CTP and FFRCT

provide improved diagnostic accuracy compared with coronary

CTA alone, which has high sensitivity but moderate specificity.

COMPETENCY IN PATIENT CARE AND PROCEDURAL

SKILLS: Both functional CT imaging techniques might refine the

critical pathway of patients to invasive catheterization.

COMPETENCY IN INTERPERSONAL AND COMMUNICA-

TION SKILLS: It is relevant to assemble robust evidence to

clarify the role of functional CT techniques and to discuss the

available options with patients who will undergo coronary CTA.

TRANSLATIONAL OUTLOOK: Further clinical trials using

sophisticated imaging protocols and machine-learning algo-

rithms are needed to validate the novel functional CT tests in the

clinical setting and to further refine the probability of potential

outcomes of patients with CAD.

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . - , N O . - , 2 0 1 8 Celeng et al.- 2 0 1 8 :- –- Functional CT for Diagnosing CAD

9

dual-energy CT compared with single-photon emis-sion computed tomography (40).

STUDY LIMITATIONS. The results of our meta-analysis need to be considered in the context of theincluded studies, which were mainly performed atexpert centers and limited in number for some mo-dalities. CT radiation exposure was not reported in allincluded studies. Verification bias was an importantlimitation, which occurred when studies only per-formed the reference standard in those with a posi-tive initial test. This might result in a high pre-testprobability, which could increase the specificity of agiven imaging modality. Another limitation was theheterogeneity among studies (including differencesin diagnostic criteria), which might have influencedthe reported summary values of the different imagingmodalities. Finally, we were not able to obtain indi-vidual FFRCT values from all studies, including theNXT trial (41), because only 312 individual pointscould be identified in the Bland-Altman plot of the484 vessels included in the study. For the DeFACTO(Determination of Fractional Flow Reserve byAnatomic Computed Tomographic AngiOgraphy)data, a substudy (42) was used in which 150 vessels(407 in total in the original study population) of in-termediate stenosis on CTA were evaluated.

CONCLUSIONS

This meta-analysis indicated that despite recentdevelopments in CT hardware, the specificity ofCTA remains moderate for the detection of hemo-dynamically significant stenosis. The specificity offunctional CT imaging tests and their combined usewith CTA was high to excellent, except for TAG,which showed discrepancies in specificity on vesseland patient levels. CTP, FFRCT and its combineduse with CTA yielded higher diagnostic performancethan CTA alone. The diagnostic performance ofindividual FFRCT at vessel level was good, with

a relative narrow area of uncertainty aroundthe threshold of 0.80. CTP and FFRCT both havethe potential to improve identification of patientswith hemodynamically significant CAD at reducedhealth costs, therefore their integration intothe routine clinical workflow is recommended. Dueto limited evidence the exact role of TAG isuncertain.

ADDRESS FOR CORRESPONDENCE: Dr. Csilla Celeng,Department of Radiology, University Medical CenterUtrecht, Heidelberglaan 100, 3584 CX Utrecht, theNetherlands. E-mail: c.celeng@umcutrecht.nl.

RE F E RENCE S

1. Miller JM, Rochitte CE, Dewey M, et al. Diag-nostic performance of coronary angiography by64-row CT. N Engl J Med 2008;359:2324–36.

2. Vanhoenacker PK, Heijenbrok-Kal MH, VanHeste R, et al. Diagnostic performance of multi-detector CT angiography for assessment of coro-nary artery disease: meta-analysis. Radiology2007;244:419–28.

3. Budoff MJ, Dowe D, Jollis JG, et al.Diagnostic performance of 64-multidetector rowcoronary computed tomographic angiography for

evaluation of coronary artery stenosis in in-dividuals without known coronary artery disease:results from the prospective multicenter ACCU-RACY (Assessment by Coronary Computed Tomo-graphic Angiography of Individuals UndergoingInvasive Coronary Angiography) trial. J Am CollCardiol 2008;52:1724–32.

4. Budoff MJ, Nakazato R, Mancini GB, et al. CTangiography for the prediction of hemodynamicsignificance in intermediate and severe lesions:head-to-head comparison with quantitative coro-nary angiography using fractional flow reserve as

the reference standard. J Am Coll Cardiol Img2016;9:559–64.

5. Rossi A, Papadopoulou SL, Pugliese F, et al.Quantitative computed tomographic coronaryangiography: does it predict functionally signifi-cant coronary stenoses? Circ Cardiovasc Imaging2014;7:43–51.

6. Pijls NH, De Bruyne B, Peels K, et al. Measure-ment of fractional flow reserve to assess thefunctional severity of coronary-artery stenoses.N Engl J Med 1996;334:1703–8.

Celeng et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . - , N O . - , 2 0 1 8

Functional CT for Diagnosing CAD - 2 0 1 8 :- –-

10

7. De Bruyne B, Sarma J. Fractional flow reserve: areview: invasive imaging. Heart 2008;94:949–59.

8. Jeremias A, Kirtane AJ, Stone GW. A test incontext: fractional flow reserve: accuracy, prog-nostic implications, and limitations. J Am CollCardiol 2017;69:2748–58.

9. Tonino PA, De Bruyne B, Pijls NH, et al. Frac-tional flow reserve versus angiography for guidingpercutaneous coronary intervention. N Engl J Med2009;360:213–24.

10. De Bruyne B, Pijls NH, Kalesan B, et al. Frac-tional flow reserve-guided PCI versus medicaltherapy in stable coronary disease. N Engl J Med2012;367:991–1001.

11. Lubbers M, Coenen A, Kofflard M, et al.Comprehensive cardiac CT with myocardial perfu-sion imaging versus functional testing in sus-pected coronary artery disease: the multicenter,randomized CRESCENT-II trial. J Am Coll CardiolImg 2017 [epub ahead of print].

12. Grunau GL, Min JK, Leipsic J. Modeling offractional flow reserve based on coronary CTangiography. Curr Cardiol Rep 2013;15:336.

13. Choi JH, Min JK, Labounty TM, et al. Intra-coronary transluminal attenuation gradient incoronary CT angiography for determining coronaryartery stenosis. J Am Coll Cardiol Img 2011;4:1149–57.

14. Choi JH, Koo BK, Yoon YE, et al. Diagnosticperformance of intracoronary gradient-basedmethods by coronary computed tomographyangiography for the evaluation of physiologicallysignificant coronary artery stenoses: a validationstudy with fractional flow reserve. Eur Heart JCardiovasc Imaging 2012;13:1001–7.

15. Lu MT, Ferencik M, Roberts RS, et al. Nonin-vasive FFR derived from coronary CT angiography:management and outcomes in the PROMISE trial.J Am Coll Cardiol Img 2017;10:1350–8.

16. Liberati A, Altman DG, Tetzlaff J, et al. ThePRISMA statement for reporting systematic re-views and meta-analyses of studies that evaluatehealthcare interventions: explanation and elabo-ration. BMJ 2009;339:b2700.

17. Cook CM, Petraco R, Shun-Shin MJ, et al.Diagnostic accuracy of computed tomography-derived fractional flow reserve: a systematic re-view. JAMA Cardiol 2017;2:803–10.

18. Whiting PF, Rutjes AW, Westwood ME, et al.QUADAS-2: a revised tool for the quality assess-ment of diagnostic accuracy studies. Ann InternMedicine 2011;155:529–36.

19. Leeflang MM. Systematic reviews and meta-analyses of diagnostic test accuracy. Clin Micro-biol Infect 2014;20:105–13.

20. Ma X, Chen Y, Cole SR, Chu H. A hybridBayesian hierarchical model combining cohort andcase-control studies for meta-analysis of

diagnostic tests: accounting for partial verificationbias. Stat Methods Med Res 2016;25:3015–37.

21. Chu H, Guo H, Zhou Y. Bivariate random effectsmeta-analysis of diagnostic studies using gener-alized linear mixed models. Med Decision Making2010;30:499–508.

22. Guo J, Riebler A, Rue H. Bayesian bivariatemeta-analysis of diagnostic test studies withinterpretable priors. Stat Med 2017;36:3039–58.

23. Menten J, Boelaert M, Lesaffre E. Bayesianmeta-analysis of diagnostic tests allowing forimperfect reference standards. Stat Med 2013;32:5398–413.

24. Stuijfzand WJ, Danad I, Raijmakers PG, et al.Additional value of transluminal attenuationgradient in CT angiography to predict hemody-namic significance of coronary artery stenosis.J Am Coll Cardiol Img 2014;7:374–86.

25. Takx RA, Blomberg BA, El Aidi H, et al. Diag-nostic accuracy of stress myocardial perfusionimaging compared to invasive coronary angiog-raphy with fractional flow reserve meta-analysis.Circ Cardiovasc Imaging 2015;8.

26. Varga-Szemes A, Meinel FG, De Cecco CN,Fuller SR, Bayer RR 2nd, Schoepf UJ. CT myocar-dial perfusion imaging. AJR Am J Roentgenol2015;204:487–97.

27. Gould KL, Lipscomb K. Effects of coronarystenoses on coronary flow reserve and resistance.Am J Cardiol 1974;34:48–55.

28. Ko SM, Park JH, Hwang HK, Song MG. Directcomparison of stress- and rest-dual-energycomputed tomography for detection of myocar-dial perfusion defect. Int J Cardiovasc Imaging2014;30 Suppl 1:41–53.

29. Fujita M, Kitagawa K, Ito T, et al. Dosereduction in dynamic CT stress myocardial perfu-sion imaging: comparison of 80-kV/370-mAs and100-kV/300-mAs protocols. Eur Radiol 2014;24:748–55.

30. Kim SM, Kim YN, Choe YH. Adenosine-stressdynamic myocardial perfusion imaging using 128-slice dual-source CT: optimization of the CT pro-tocol to reduce the radiation dose. Int J CardiovascImag 2013;29:875–84.

31. Kim KH, Doh JH, Koo BK, et al. A novelnoninvasive technology for treatment planningusing virtual coronary stenting and computedtomography-derived computed fractional flowreserve. J Am Coll Cardiol Intv 2014;7:72–8.

32. Petraco R, Sen S, Nijjer S, et al. Fractionalflow reserve-guided revascularization: practicalimplications of a diagnostic gray zone and mea-surement variability on clinical decisions. J Am CollCardiol Intv 2013;6:222–5.

33. Itu L, Rapaka S, Passerini T, et al. A machine-learning approach for computation of fractional

flow reserve from coronary computed tomogra-phy. J Appl Physiol (1985) 2016;121:42–52.

34. Zreik M, Lessmann N, van Hamersvelt RW,et al. Deep learning analysis of the myocardium incoronary CT angiography for identification of pa-tients with functionally significant coronary arterystenosis. Med Image Anal 2017;44:72–85.

35. Hlatky MA, Saxena A, Koo BK, Erglis A,Zarins CK, Min JK. Projected costs and conse-quences of computed tomography-determinedfractional flow reserve. Clin Cardiol 2013;36:743–8.

36. Kimura T, Shiomi H, Kuribayashi S, et al. Costanalysis of non-invasive fractional flow reservederived from coronary computed tomographicangiography in Japan. Cardiovasc Interv Ther2015;30:38–44.

37. Douglas PS, De Bruyne B, Pontone G, et al. 1-Year Outcomes of FFRCT-guided care in patientswith suspected coronary disease: the PLATFORMstudy. J Am Coll Cardiol 2016;68:435–45.

38. Rajani R, Webb J, Marciniak A, Preston R.Comparative efficacy testing - fractional flowreserve by coronary computed tomography for theevaluation of patients with stable chest pain. Int JCardiol 2015;183:173–7.

39. National Institute for Health and Care Excel-lence. HeartFlow FFRCT for estimating fractionalflow reserve from coronary CT angiography.Available at: nice.org.uk/guidance/mtg32. AccessedSeptember 2, 2017.

40. Meyer M, Nance JW Jr., Schoepf UJ, et al.Cost-effectiveness of substituting dual-energy CTfor SPECT in the assessment of myocardialperfusion for the workup of coronary artery dis-ease. Eur J Radiol 2012;81:3719–25.

41. Norgaard BL, Leipsic J, Gaur S, et al. Diagnosticperformance of noninvasive fractional flowreserve derived from coronary computed tomog-raphy angiography in suspected coronary arterydisease: the NXT trial (Analysis of Coronary BloodFlow Using CT Angiography: Next Steps). J AmColl Cardiol 2014;63:1145–55.

42. Nakazato R, Park HB, Berman DS, et al.Noninvasive fractional flow reserve derived fromcomputed tomography angiography for coronarylesions of intermediate stenosis severity: resultsfrom the DeFACTO study. Circ Cardiovasc Imaging2013;6:881–9.

KEY WORDS coronary artery disease,computed tomography, fractional flowreserve, meta-analysis, myocardial perfusionimaging

APPENDIX For supplemental tables andfigures, please see the online version of thispaper.