Diagnostic accuracy of coronary computed tomography angiography for the evaluation of obstructive coronary artery disease in patients referred for transcatheter aortic valve implantation: a systematic review and meta-analysis

Objective To evaluate the diagnostic accuracy of coronary computed tomography angiography (CCTA) for the evaluation of obstructive coronary artery disease (CAD) in patients referred for transcatheter aortic valve implantation (TAVI). Methods EMBASE, PubMed/MEDLINE, and CENTRAL were searched for studies reporting accuracy of CCTA for the evaluation of obstructive CAD compared with invasive coronary angiography (ICA) as the reference standard. QUADAS-2 tool was used to assess the risk of bias. A bivariate random effects model was used to analyze, pool, and plot the diagnostic performance measurements across studies. Pooled sensitivity, specificity, positive ( + LR) and negative (−LR) likelihood ratio, diagnostic odds ratio (DOR), and hierarchical summary ROC curve (HSROC) were evaluated. Prospero registration number: CRD42021252527. Results Fourteen studies (2533 patients) were included. In the intention-to-diagnose patient-level analysis, sensitivity and specificity for CCTA were 97% (95% CI: 94–98%) and 68% (95% CI: 56–68%), respectively, and + LR and −LR were 3.0 (95% CI: 2.1–4.3) and 0.05 (95% CI: 0.03 – 0.09), with DOR equal to 60 (95% CI: 30–121). The area under the HSROC curve was 0.96 (95% CI: 0.94–0.98). No significant difference in sensitivity was found between single-heartbeat and other CT scanners (96% (95% CI: 90 – 99%) vs. 97% (95% CI: 94–98%) respectively; p = 0.37), whereas the specificity of single-heartbeat scanners was higher (82% (95% CI: 66–92%) vs. 60% (95% CI: 46 – 72%) respectively; p < 0.0001). Routine CCTA in the pre-TAVI workup could save 41% (95% CI: 34 – 47%) of ICAs if a disease prevalence of 40% is assumed. Conclusions CCTA proved an excellent diagnostic accuracy for assessing obstructive CAD in patients referred for TAVI; the use of single-heartbeat CT scanners can further improve these findings. Key Points • CCTA proved to have an excellent diagnostic accuracy for assessing obstructive CAD in patients referred for TAVI. • Routine CCTA in the pre-TAVI workup could save more than 40% of ICAs. • Single-heartbeat CT scanners had higher specificity than others in the assessment of obstructive CAD in patients referred for TAVI. Supplementary Information The online version contains supplementary material available at 10.1007/s00330-022-08603-y.


Introduction
Transcatheter aortic valve implantation (TAVI) represents the gold standard for treatment of severe aortic stenosis in patients at high and intermediate surgical risk [1]. Candidates for TAVI are an elderly and frail population with a high prevalence (up to 60%) of coronary artery disease (CAD).
Current guidelines recommend pre-procedural screening for CAD before valvular intervention [1]. Computed tomography angiography (CTA) has become the standard imaging method for pre-procedural TAVI assessment because it can evaluate non-invasively both the aortic arch and the peripheral vessels [1,2]. Coronary computed tomography angiography (CCTA) has been recommended as an initial test in patients with low clinical likelihood of CAD due to its very high negative predictive value in this population [2], but the clinical value of this approach in the TAVI setting is still under evaluation. Indeed, patients with severe aortic stenosis have a high pre-test probability of obstructive CAD and present specific technical challenges to CCTA interpretation, including extensive coronary calcifications, frequent abnormal cardiac rhythm, and contraindication to nitrate administration [1].
To the best of our knowledge, only two meta-analyses [3,4] have evaluated the diagnostic accuracy of CCTA in patients with aortic stenosis referred for aortic valve replacement (surgical or transcatheter). In the last few years, technological advances in CT scanners have resulted in improved image quality, allowing accurate assessment of coronary anatomy even in difficult technical settings [5,6]; this aspect was not assessed by the two abovementioned meta-analyses, both published in 2018. Moreover, indications to TAVI are progressively extending to low-surgical-risk patients with lower pre-test probability of CAD [7,8].
In this framework, the incorporation of coronary artery assessment into pre-TAVI CT evaluation has the potential to reduce the need for unnecessary ICA and total amount of contrast medium applied, making pre-procedural evaluation safer and faster with reduced cost [9,10].
The aim of this systematic review and meta-analysis is to provide an updated overview of the diagnostic accuracy of CCTA for the evaluation of obstructive CAD among patients referred for TAVI.

Protocol and registration
This systematic review and meta-analysis was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) extension for Diagnostic Test Accuracy (DTA) Studies [11]. The protocol was prospectively registered in the PROSPERO International register of systematic reviews with the ID number CRD42021252527.

Eligibility criteria
The primary aim of this study was to evaluate the diagnostic accuracy of CCTA for obstructive coronary stenosis among patients referred for TAVI. Studies reporting data on CCTA for the evaluation of obstructive CAD were deemed eligible if all the following inclusion criteria were respected: (1) CCTA performed with at least a 64-slice CT scanner; (2) ICA performed in all patients and used as the reference standard; (3) sensitivity and specificity were reported or assessed by the published data. Obstructive CAD was defined in the metaanalysis as a narrowing of the coronary lumen by more than 50% on CCTA and a lumen diameter reduction of more than 50% on ICA.

Study endpoints
The primary endpoint was the patient-level accuracy of CCTA to identify obstructive CAD. For the purpose of this analysis, non-evaluable segments were considered positive based on an intention to diagnose approach [12]. Secondary analyses included the evaluation of the accuracy of CCTA for obstructive CAD at the patient level, excluding patients with nonevaluable segments, at the vessel and at the segment level, evaluating also coronary artery bypass grafting (CABG) and stented coronary artery segments. Sensitivity was analyzed according to the risk of bias and applicability. Subgroup analysis was performed based on CT scanner characteristics. We identified three main CT scanner subgroups: (1) whole-heart coverage CT scanner-scanners with extensive detector coverage on the z-axis (i.e., the 160-mm scanners); (2) high temporal resolution CT scanners (i.e., dual-source scanners); (3) single-heartbeat CT scanners-scanner capable of acquiring the entire heart volume in a single beat, including both wholeheart CT scanners and high temporal resolution scanners with a large number of detectors (e.g., Somatom Force, Siemens Healthineers). In particular, we conducted three subgroup analyses to determine the effect of these technical CT parameters on diagnostic accuracy: (1) whole-heart coverage CT scanner vs. other CT scanners; (2)  Data collection, data extraction, and risk of bias and applicability Two researchers (F.B. and A.S.) independently searched for studies fulfilling the inclusion criteria in a two-stage process: first by using title and abstract of the papers and then the full text. The reasons for excluding studies in this second phase were recorded. The results from both searches were compared and the discrepancies were discussed. In some cases of disagreement, the decision was reached by consultation with a third researcher (G.G.). All selected articles were automatically downloaded, imported, and de-duplicated in Microsoft Excel (Microsoft).
General characteristics included total number of patients, age, sex, body mass index, cardiovascular risk factor (i.e., diabetes, hypercholesterolemia, smoking, history, hypertension), known CAD, previous percutaneous coronary intervention or CABG and atrial fibrillation. Moreover, main CT scanner characteristics were recorded: number of detector rows, dual-energy techniques, tube voltage tube current, contrast media concentration, contrast media volume, heart rate during acquisition, and mean dose-to-length product.
The quality assessment of diagnostic accuracy studies 2 (QUADAS-2) tool [13] was used to assess the risk of bias of included studies (reported in Supplementary Material).

Statistical analysis
Two-by-two contingency tables were extracted from each study and used to calculate sensitivity, specificity, positive (+LR) and negative (−LR) likelihood ratio, and diagnostic odds ratio (DOR) with 95% confidence interval (CI) of CCTA for the detection of significant coronary artery stenosis (in general, + LR > 10 and −LR < 0.1 demonstrates a satisfactory diagnostic performance [14]). A bivariate random effects model was used to analyze, pool, and plot the diagnostic performance measurements across studies. Derived logit estimates of sensitivity, specificity, and respective variances were used to construct a hierarchical summary ROC curve (HSROC). Heterogeneity between studies was evaluated utilizing Cochran's Q and Higgins I 2 statistics. Deeks' funnel plot was used to assess publication bias. The patient-level clinical accuracy of CCTA was evaluated using the likelihood ratios to calculate post-test probability based on Bayes' theorem with the use of Fagan's nomograms, Likelihood ratio scattergram, and probability modifying plot.
The analyses were performed with STATA (version 16.1, Stata Corp LP) using the MIDAS module [15] and MetaDTA (Diagnostic Test Accuracy Meta-Analysis v2.01) [16]. A p value of less than 0.05 was considered statistically significant.

Literature search and study characteristics
The PRISMA 2020 flow diagram [17] for systematic reviews is reported in Figure 1. Fourteen studies with a total of 2533 patients were included in the analysis. Tables 1 and 2 report the baseline characteristic of the included patients and of the CT scanner used.

Assessment of study quality
The QUADAS-2 Domain assessment is reported in Figure 2. The supplementary material contains the details of this analysis.

Diagnostic accuracy: primary endpoint
A total of 2228 patients were included in the analysis. For the purpose of this analysis, performed at the patient level, non-evaluable segments were considered positive based on an intention-to-diagnose approach. The pooled sensitivity and specificity for CCTA were 97% (94-98%) and 68% (56-68%), respectively, and the + LR and −LR were 3.0 (2.1-4.3) and 0.05 (0.03-0.09), with a DOR of 60 . The HSROC had an AUC = 0.96 (0.94-0.98). Table 3 shows sensitivity and specificity with % (95% CI) derived from each study included in the analysis. The summary forest plot and HSROC plot are reported in Figure 3.
The per-patient analysis revealed a + LR of 3.03 (2.12-4.33) and a −LR of 0.05 (0.03-0.09) (i.e., with an estimated pre-test probability of CAD of 40%, a positive CCTA could increase the post-test probability to 67% and a negative CCTA can decrease the post-test probability to 3%, whereas in a hypothetical population with pre-test probability of 15%, the post-test probability can reduce to less than 1%).
Fagan's nomograms, with estimated pretest probability of 40% and 15%, Likelihood ratio scattergram and probability modifying plot are reported in Figure 4. In summary, estimating a disease prevalence of 40% in a population of 1000 patients, the study of coronary arteries with CCTA before the TAVI procedure would correctly avoid 409 (95% CI 335-470) ICAs ( Figure 5). Table 4 provides a summary of the CCTA diagnostic performance for the evaluation of obstructive CAD among patients referred for TAVI at a patient, vessel, and segment level. The secondary analysis is reported in the Supplementary Material.

Sub-analysis: sensitivity and subgroup analysis
We found a high value for Cochran's Q and I 2 , which indicates the presence of heterogeneity in the studies. As a result, we visually assessed the forest plot and HSROC and a significant heterogeneity in specificity was found, particularly in the forest plot (Figure 4a), where some studies fell outside the combined 95% CI.
For the purpose of the sub-analysis, performed at the patient level, non-evaluable segments were considered positive based on an intention-to-diagnose approach.
The results of the subgroup analysis based the various CT scanner features are summarized in Table 5.
In summary, the use of a whole-heart coverage CT increased specificity (p < 0.001) but did not affect sensitivity (p = 0.26); the use of high temporal resolution scanners increased sensitivity (p = 0.02) but decreased specificity (p < 0.001); and the use of single-heartbeat scanners increased specificity (p < 0.001) with no effect on sensitivity (p = 0.37).
To translate our findings into clinical practice, we estimated a disease prevalence of 40% in a 1000-patient population and evaluated coronary arteries with different CT scanners: a whole-heart coverage CT scanner could correctly avoid 477 (95% CI 340-552) ICAs, a high temporal resolution CT scanner could correctly avoid 357 (95% CI 259-444) ICAs, and a single-heartbeat CT scanner could correctly avoid 494 (95% CI 398-550) ICAs ( Figure 6).

Discussion
In this systematic review and meta-analysis, we investigated the diagnostic accuracy of CCTA for the assessment of obstructive CAD among patients referred for TAVI. Overall, CCTA prior to TAVI procedure provides high sensitivity (97%) with a good −LR (0.05). This result highlights the effectiveness of CCTA to rule out significant CAD and to reduce unnecessary ICA procedures by 40.9%, aspiring its potential role as a gatekeeper test in this subgroup of patients. Furthermore, the use of new CT scanners, particularly single-heartbeat CT, has the potential to save the number of ICAs by up to 49.4%.
A recent meta-analysis on the same topic by van den Boogert et al [3] with included seven studies on a total of 1275 patients (all of these studies/patients were also incorporated into our meta-analysis) reported sensitivity, specificity, positive predictive value, and negative predictive value of 95.3% (93.3-96.9%), 65.3% (61.6-68.9%), 70.8% (68.6-72.9%), and 94.0% (91.6-95.8%) respectively. These results are very similar to our findings, despite the fact that the authors did not use a bivariate random effects model to summarize sensitivity, specificity, and their 95%, but rather a fixed effects model. In addition, they did not estimate +LR, −LR, and DOR.
Another meta-analysis on the performance of CCTA in patients with aortic stenosis undergoing surgery or transcatheter intervention by Chaikriangkrai et al [4] was performed. It included thirteen studies with a total of 1498 patients (6 of these studies, for a total of 1135 patients, were also incorporated into our meta-analysis); the results showed sensitivity = 95% (93-97%), specificity= 79% (68-86%), +LR = 4.48 (2.96-6.78), −LR = 0.06 (0.04-0.09) and AUC= 0.96 (0.95-0.98). The obtained specificity is slightly higher than our finding. This may be partially due to the population included in their study: in fact, in the subgroup analysis comparing patients who underwent surgery with those who underwent percutaneous intervention, the latter group showed a lower specificity (albeit not significant) (83% (77-87%) vs. 74% (51-88%)). This difference would be further increased by removing the study of Andreini et al [30] from their meta-analysis, thus increasing the specificity in the percutaneous intervention group even if it excluded non-evaluable patients. Another explanation could be the difference in disease prevalence between the two groups (48% in transcatheter vs. 29% in surgical): subjects who underwent transcatheter replacement were typically more fragile and at higher risk than those who underwent surgery, and this may increase the number of nonevaluable segments, thus reducing specificity.
In accordance with the European Society of Cardiology (ESC) guidelines [1], CCTA with its high negative predictive value is considered useful to exclude CAD in patients undergoing TAVI who are at low risk for atherosclerosis. This is in perfect agreement with our findings: a patient with a pre-test probability of 15% has a post-test probability of less than 1%, if the CCTA is negative.
However, our data seem to support a slightly broader usefulness and feasibility of CCTA even for patients with intermediate risk of CAD. In fact, on the basis of our findings, a patient with a pre-test probability of 40% (i.e., the prevalence of the disease in our population = 40%), in case of CCTA negativity, has a post-test probability of about 3%. This last finding is supported by a recent meta-analysis [31] of prospective studies comparing CCTA with coronary angiography as the reference standard. It highlighted the role of CCTA in patients with a low-to-intermediate pretest probability of CAD, emphasizing how this population could benefit the most from CCTA to rule in and rule out significant CAD.
The European Society of Cardiovascular Radiology (ESCR) consensus document [32] recently reported that Fig. 2 Quality assessment of diagnostic accuracy studies 2 (QUADAS-2) tool for risk of bias and applicability concern. Green represents low; yellow, high; and blue, unclear risk. On the top panel, QUADAS-2 was reported for each study and summarized in a bar graph on the bottom panel by stacked bars for each item CCTA should not be used routinely for pre-procedural assessment of CAD; however, they also added that, as technology evolves, CCTA can be used on a "case-by-case" basis, according to local expertise and available equipment and primarily to exclude significant coronary stenosis. This last statement was one of the starting points for our subgroup analyses. Technological advances in CT scanners were notable in recent years: the ideal CT scanner for cardiac imaging has high spatial and temporal resolution, covers the entire cardiac volume in a single rotation, and is ALARP (as low as reasonably practicable) compliant [33].
The abovementioned meta-analyses on a similar population [3,4], published in 2018, did not analyze this aspect: (a) Regarding the high temporal resolution CT scanners, only three studies [12,19,27] analyzing 100, 475, and 145 patients, respectively, were included. (b) Concerning whole heart and single-heartbeat CT scanner, only one study [25] containing 66 patients was included.
In the period between those meta-analyses and our study, seven more papers using these technologies were published [18,[20][21][22][23][24]29]. Therefore, we planned a sub-group analysis to explore the benefits of high intrinsic TR obtained by dual-  source CT scanner, whole-heart coverage CT scanner, and single-heartbeat CT scanner. According to our findings, the use of CT scanners with high temporal resolution seems to improve sensitivity with lower specificity, whereas the use of whole-heart coverage and single-heartbeat CT scanners resulted in a higher specificity. Single-heartbeat CT scanners in the evaluation of obstructive CAD prior to TAVI could correctly avoid up to 49.4% of ICAs. This data may be at least partially related to the increased number of segments, which allows accurate assessment of coronary arteries even in challenging population. The nonassessable segments were considered positive, leading to higher sensitivity and lower specificity in case of a high number of non-assessable segments. Our results confirm in a larger population the findings of Meier et al [18], who reported that the use of a 64-row scanner (compared to a 256-row scanner) for evaluating CAD in a pre-TAVI population was the only parameter in a multivariate analysis to be associated with a high risk of unanalyzable images. Our findings are consistent also with the meta-analysis of Haase et al [31], who reported that using a CT scanner with more than 64 detector rows led to higher sensitivity (93.4% vs 86.5%, p = 0.002) and specificity (84.4% vs 72.6%, p = 0.001) in ruling out or confirming CAD patients with a pretest probability of CAD ranging from 7 to 67%.
From a clinical point of view, it has been reported that patients with a large area of myocardium at ischemic risk Fig. 4 a The conditional probability modifying plot is a graphical sensitivity analysis of predictive value along a prevalence continuum designating low-risk to high-risk populations. It shows distinct curves for positive and negative testing. The user draws a vertical line from the chosen pre-test probability to the appropriate likelihood ratio line, then reads the post-test probability from the vertical scale. b The likelihood ratio scattergram represents the summary point of likelihood ratios calculated as functions of mean sensitivity and specificity. The summary point is located in the left lower quadrant: the CCTA has a likelihood ratio positive < 10 and a likelihood ratio negative < 0.1. Based on these considerations, the CCTA in patients referred for TAVI is useful for exclusion of CAD (when negative) rather than confirmation (when positive). c, d Fagan's nomograms, with estimated pretest probability of 40% and 15%, respectively. A Fagan plot has a vertical axis on the left with the prior log-odds, a vertical axis in the middle with the log-likelihood ratio, and a vertical axis on the right with the posterior log-odds. The lines are then traced from the prior probability on the left to the likelihood ratios in the center, and then to the posterior probabilities on the right. Both plots highlight the strength of the CCTA in excluding the presence of CAD, with residual post-test probabilities of 3% and 1%, respectively The numbers in brackets represent 95% confidence intervals may benefit the most from revascularization intervention prior to TAVI, and the proximal segment stenosis is prognostically more important than distal [34]. Unfortunately, the only study included in this meta-analysis that allows this type of analysis is that of Andreini et al [30], where the exclusion of nonassessable segments gives a boost to the specificity of CCTA, which appears superior even in the study of distal vs. proximal segments [35,36].
Further studies are needed to evaluate the real impact of CCTA in clinical practice; a randomized controlled trial could be proposed to evaluate the prognostic impact of patients undergoing CCTA pre-TAVI vs. a group not undergoing CCTA; also, the actual potential of CCTA in the study of proximal segments in such a selected population remains to be explored.
This meta-analysis has some limitations. First of all, a relatively low number of studies met the selection criteria. Unfortunately, only a fraction of the studies reported the analyses at patient level, either by considering nonanalyzable segments as positive or by excluding them,  6 Impact of high temporal resolution CT scanners (i.e., dual-source CTs) (a), whole-heart coverage CT scanner (b), and single-heartbeat CT scanner (c) in 1000 patients referred for TAVI. The numbers in brackets represent 95% confidence intervals and therefore it was not possible to create 3 × 2 tables [37]. Moreover, despite relevance from a clinical point of view, we were not able to compare performance among high vs. low calcium and arrhythmic and vs. nonarrhythmic patients. Almost all studies included are retrospective cohorts; only one is prospective and none a randomized controlled study. In addition, many of the included studies are small in size. The studies' overall quality was however adequate for analysis according to the QUADAS-2 evaluation, despite the presence of some unclear or high-risk items.
In conclusion, CCTA proved to have excellent diagnostic accuracy for assessing obstructive CAD in patients referred for TAVI. Routine CCTA assessment of coronary arteries as part of the pre-TAVI workup could save more than 40% of ICAs. The use of single-heartbeat CT scanners, which provide higher specificity, can further improve these findings.
Funding The authors state that this work has not received any funding.

Declarations
Guarantor The scientific guarantor of this publication is Prof. Paolo Fonio, chief of Diagnostic and Interventional Radiology at the University of Turin.

Conflict of interest
The authors of this manuscript declare no relationships with any companies whose products or services may be related to the subject matter of the article.
Statistics and biometry Dott. Marco Gatti, one of the authors, kindly provided statistical advice for this manuscript.
Informed consent The protocol was prospectively registered in the PROSPERO International register of systematic reviews with the ID number CRD42021252527.
Ethical approval Institutional Review Board approval was obtained.

Methodology • Meta-analysis
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.