Objective: To evaluate the performance of dual source CT coronary angiography (DSCT-CA) in the detection of in-stent restenosis (⩾50% luminal narrowing) in symptomatic patients referred for conventional angiography (CA).
Design/patients: 100 patients (78 males, age 62 (SD 10)) with chest pain were prospectively evaluated after coronary stenting. DSCT-CA was performed before CA.
Setting: Many patients undergo coronary artery stenting; availability of a non-invasive modality to detect in-stent restenosis would be desirable.
Results: Average heart rate (HR) was 67 (SD 12) (range 46–106) bpm. There were 178 stented lesions. The interval between stenting and inclusion in the study was 35 (SD 41) (range 3–140) months. 39/100 (39%) patients had angiographically proven restenosis. Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of DSCT-CA, calculated in all stents, were 94%, 92%, 77% and 98%, respectively. Diagnostic performance at HR <70 bpm (n = 69; mean 58 bpm) was similar to that at HR ⩾70 bpm (n = 31; mean 78 bpm); diagnostic performance in single stents (n = 95) was similar to that in overlapping stents and bifurcations (n = 83). In stents ⩾3.5 mm (n = 78), sensitivity, specificity, PPV, NPV were 100%; in 3 mm stents (n = 59), sensitivity and NPV were 100%, specificity 97%, PPV 91%; in stents ⩽2.75 mm (n = 41), sensitivity was 84%, specificity 64%, PPV 52%, NPV 90%. Nine stents ⩽2.75 mm were uninterpretable. Specificity of DSCT-CA in stents ⩾3.5 mm was significantly higher than in stents ⩽2.75 mm (OR = 6.14; 99%CI: 1.52 to 9.79).
Conclusion: DSCT-CA performs well in the detection of in-stent restenosis. Although DSCT-CA leads to frequent false positive findings in smaller stents (⩽2.75 mm), it reliably rules out in-stent restenosis irrespective of stent size.
Statistics from Altmetric.com
Computed tomography angiography is a non-invasive diagnostic tool to visualise coronary arteries.1 The evaluation of stents is, however, hampered by the occurrence of high-density artifacts (“blooming effect”) caused by the stent struts. These artifacts cause an apparent enlargement of the stent and preclude appropriate assessment of the in-stent lumen.
In particular, the lumens of small stents, overlapping stents and bifurcation stents are difficult to assess.2 Motion artifacts may further hinder the evaluation of stents.3 Several investigations have evaluated the diagnostic performance of computed tomography in assessing stent patency or the presence of in-stent restenosis;4–16 these studies have included patients with low heart rates and pre-scan preparation with β-blockers.
The introduction of dual source 64-slice computed tomography scanners, with an improved temporal resolution,17 may be helpful to more accurately assess coronary stents.
In this study we evaluated the diagnostic performance of dual source computed tomography coronary angiography (DSCT-CA) for the detection of in-stent restenosis in patients with anginal symptoms after stent implantation.
From April 2006 to January 2007, 133 patients with chest pain and prior stent implantation were considered for inclusion in this prospective study. All patients were scheduled for diagnostic conventional angiography. Serum creatinine >120 μmol/l, irregular heart rhythm and known allergy to iodinated contrast agents were exclusion criteria. The recruitment procedure is described in fig 1. The institutional review board approved the study protocol and all the included patients gave informed consent.
All patients were examined with a dual source CT scanner (Somatom Definition, Siemens, Forchheim, Germany). The scanner design consists of two x-ray tubes and two detector arrays mounted at an angle of 90°. Scan parameters were: 120 kV, 330 ms gantry rotation time, 2×32×0.6 mm collimation with z-flying focal spot for both detectors. Using this scan protocol, spatial resolution was 0.4×0.4×0.4 mm.3 17 Pitch values were automatically adapted to the heart rate after an estimation based on the last 10 heartbeats preceding the scan, and varied between 0.20 and 0.43. Current x-ray tube modulation was used at full current for 25–70% of the R-R interval. Each tube provided a maximum of 412 mAs/rotation. In patients with heart rates <70 bpm, x-ray exposure was (mean (SD)) 15.0 (4.1) mSv in men and 16.7 (5.0) mSv in women; in patients with heart rates ⩾70 bpm, x-ray exposure was 12.1 (2.6) mSv in men and 13.7 (4.7) in women (values calculated using ImPACT®, version 0.99x, St George’s Hospital, Tooting, London, UK).
All patients received sublingual nitroglycerin just before scanning. Contrast agent (Iomeron® 400 mg/ml, Bracco, Italy) was injected into the antecubital vein at a flow rate of 5.0 ml/s, followed by a saline chaser (40 ml). We calculated the contrast volume with the following equation: estimated scan time + scan delay (7 s). The contrast volume varied between 60 and 100 ml depending on the scan time, which in turn varied between 5 and 13 seconds. We used a bolus-tracking technique to synchronise the start of the scan with the arrival of contrast agent in the coronary arteries. A circular region-of-interest (ROI) was positioned in the ascending aorta and the scan was automatically started when a threshold of +100 Hounsfield Units was reached inside the ROI.
Given the scanner geometry, a monosegmental algorithm using data from a single heartbeat obtained during a quarter gantry rotation was used for reconstruction. This translated into a temporal resolution equal to one-fourth of the gantry rotation time, that is, 330/4 = 83 ms.
First, 0.75 mm-thick images were retrospectively reconstructed during the mid-diastolic to end-diastolic phase. The position of the reconstruction window within the R-R interval varied according to the heart rate (from −400 ms before the R wave for low heart rates to −175 ms for high heart rates). Additional data sets were reconstructed during the end-systolic phase (from +400 ms after the R wave for low heart rates to +200 ms for high heart rates). The reconstruction increment was 0.4 mm. The data set with the fewest motion artifacts was chosen for analysis and reconstructed using a dedicated sharp convolution kernel (B46f, “Heart View”).
Quantitative coronary angiography (QCA)
A single observer unaware of the CT results examined the angiograms before contrast injection to identify the sites of stent implantation. The stents were then located within the coronary tree following a 17-segment modified AHA model,17 as previously described.18 19 Stents and stent edges, the latter defined as 5 mm-long coronary segments proximal and distal to the stents, were evaluated on multiple projections; luminal narrowing ⩾50% was classified as significant (restenosis). A validated quantitative coronary angiography software (CAAS II®, Pie Medical, Maastricht, The Netherlands) was used to determine the minimal lumen diameter and derive the per cent diameter stenosis by means of the user-independent method of the interpolated reference diameter, using the angiographic catheter diameter as reference for calibration.20
Two experienced readers evaluated the DSCT-CA studies independently; the readers were unaware of the findings of conventional angiography. In the event of diverging opinions, a consensus was reached and used in the final analysis.
Using axial images, multiplanar reconstructions (MPR) and curved MPR, the stents were visually screened for the presence of in-stent restenosis (⩾50% lumen diameter reduction). When multiple stents were implanted contiguously to treat one lesion, they were considered as one single lesion. The assessment of restenosis was based exclusively on the visualisation of the in-stent lumen. The presence of distal run-off was not considered as an indicator of patency because retrograde or collateral filling in an occluded stent can also give distal contrast enhancement (fig 2). All stent edges, defined as 5 mm-long coronary segments proximal and distal to the stents, were also included in the evaluation. In the event of bifurcation stenting, each of the three branches was evaluated. When the stent lumen was uninterpretable (eg, obscured by high-density artifacts), and in-stent restenosis could not be excluded by DSCT-CA, stents were considered to have restenosis (worst case scenario)6 7 for the purpose of the analysis.
The statistical analysis was performed with SPSS, version 12.1 (SPSS Inc., Chicago, Illinois, USA). Results are reported in accordance with the STARD criteria.21 Continuous variables are expressed as mean (standard deviation). Categorical variables are presented as counts and percentages.
Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of DSCT-CA for the detection of ⩾50% in-stent restenosis, as determined by QCA as reference standard, were computed in two subgroups defined as low heart rate and high heart rate; the definition of the two groups was based on a cut-off heart rate of 70 bpm as previously described by Leber et al22 and Mollet et al.19 The low heart rate subgroup comprised patients with heart rates <70 bpm, the high heart rate subgroup comprised patients with heart rates ⩾70 bpm.
Sensitivity, specificity, PPV and NPV were then computed in two subgroups defined as simple configuration and complex configuration: the simple configuration subgroup comprised lesions treated with the deployment of one single stent; the complex configuration subgroup comprised overlapping stents and bifurcation stenting.
We also computed sensitivity, specificity, PPV and NPV in three subgroups defined by stent diameters, that is, ⩾3.5 mm, 3 mm and ⩽2.75 mm. When multiple stents were employed to treat one lesion, the diameter of the proximal stent determined the diameter subgroup. This criterion for subclassifying lesions with multiple stents was based on the expectation that small stents were more difficult to evaluate than large stents; we preferred to underestimate the diagnostic performance in the larger stent subgroups rather than overestimating the diagnostic performance in the smaller stent subgroup.
Sensitivity, specificity, PPV and NPV were also computed separately for the right coronary artery (RCA), left main (LM) stem, left anterior descending (LAD) and left circumflex (LCx) arteries. Because the accuracy of DSCT-CA to detect occluded stents might be greater than the accuracy to detect restenosis, we performed separate analyses after excluding totally occluded segments. Diagnostic test results are presented with corresponding 95% confidence intervals based on binomial probabilities. The χ2 test was used to compare the frequency of occurrence of restenosis in the different subgroups. The Mann–Whitney U test was used to compare the mean stent diameters in the heart rate subgroups and in the configuration subgroups.
We determined the effect of heart rate, stent configuration and stent diameter on sensitivity, specificity, PPV, and NPV using logistic regression analysis including patient identification to correct for possible correlation within the individuals who had multiple stents.23 To compensate for multiple testing, we used a significance level of 0.01 and computed 99% confidence intervals (CI).
Interobserver and intraobserver agreement for the detection of restenosis were determined by κ-statistics.
Baseline characteristics and angiographic findings
Of the 133 patients screened for inclusion in our study, 33 were excluded because of serum creatinine level >120 μmol/l (n = 4), known contrast allergy (n = 3), irregular heart rate (n = 6) and refusal to undergo DSCT-CA (n = 20), so that 100 patients underwent DSCT-CA (fig 1).
DSCT-CA and conventional angiography were performed 35 (SD 41) months after stenting (range 3–140 months). Seventy patients were on treatment with β-blockers; none of the patients received additional β-blockers before the scan.
The average heart rate during the scan was 67 (SD 12) (range 42–106) bpm. Sixty-nine patients had heart rates <70 bpm and were included in the low heart rate subgroup (the average heart rate in this subgroup was 58 (SD 6) (range 42–69) bpm); 31 patients had heart rates ⩾70 bpm and were included in the high heart rate subgroup (the average heart rate in this subgroup was 78 (SD 9) (range 70–106) bpm).
We examined 178 stented lesions (247 stents used, 1.4 (SD 0.8) stents per lesion); 95 lesions consisted of single stents (simple configuration subgroup); the remaining 83 lesions consisted of overlapping stents (n = 62) and bifurcations (n = 21) (complex configuration subgroup). All but one complex lesion consisted of stents of the same type. One complex lesion was a stent-in-stent implantation consisting of two partially overlapped BX Velocity stents plus three Taxus stents implanted 1 year later (fig 2); this lesion was classified as a bare metal stent (BMS) lesion. BMS accounted for 37% (65/178) of the stented lesions, drug-eluting stents (DES) accounted for 63% (113/178) of the stented lesions. Restenosis was diagnosed angiographically in 39/178 (22%) stented lesions in 39/100 (39%) patients. Restenosis was found in 27/65 (42%) BMS and in 12/113 (11%) DES. Table 1 summarises patient baseline characteristics, main angiographic findings and stent types.
The frequency of in-stent restenosis was 23% (29/124) in the low heart rate subgroup and 19% (10/54) in the high heart rate subgroup. In the simple and complex configuration subgroups, frequencies were 18% (17/95) and 26% (22/83), respectively. In the ⩾3.5 mm, 3 mm and ⩽2.75 mm subgroups, frequencies of restenosis were 19% (15/78), 20% (12/59) and 31% (13/41), respectively. All p values were not significant (p>0.05).
The mean stent diameter in the entire sample was 3.15 (SD 0.58). The mean stent diameters in the simple and complex configuration subgroups were 3.16 (SD 0.60) mm and 3.15 (SD 0.61) mm, respectively. The mean stent diameters in the low and high heart rate subgroups were 3.17 (SD 0.61) mm and 3.15 (SD 0.66) mm, respectively. All p values were not significant (p>0.05).
Diagnostic performance of DSCT-CA
All 178 stented lesions were detected by DSCT-CA. The stent lumen was judged interpretable in 169/178 (95%) stents. In the remaining nine stents (5%), all of which were ⩽2.75 mm in diameter, the lumen was uninterpretable due to high-density artifacts obscuring the in-stent lumen; these stents were scored as having in-stent restenosis. However, these nine small stents were angiographically normal (no in-stent restenosis, DSCT-CA false positives).
Sensitivity, specificity, PPV and NPV in detecting ⩾50% restenosis, calculated on all stents, were 94%, 92%, 77% and 98%, respectively. The diagnostic performance of DSCT-CA at heart rates <70 bpm did not differ significantly from that obtained at heart rates ⩾70 bpm, as witnessed by widely overlapping confidence intervals. The diagnostic performances obtained in simple stents and overlapping stents/bifurcations were also similar. Likewise, no significant differences were seen between the four major coronary vessels. Table 2 gives sensitivity, specificity, PPV and NPV obtained in the heart rate subgroups, in simple and complex configuration subgroups and in the four major coronary vessels.
In the diameter-based subanalysis, we found that NPV was in the range 90–100% in all subgroups. Sensitivity was 100% in ⩾3.5 mm stents and in 3 mm stents, but it dropped to 84% in ⩽2.75 mm stents. Specificity and PPV were both 100% in the ⩾3.5 mm subgroup; the values for specificity and PPV were 97% and 91%, respectively, in the 3 mm subgroup, and 64% and 52%, respectively, in the ⩽2.75 mm subgroup. Table 2 gives the diagnostic performance of DSCT-CA in detecting ⩾50% luminal narrowing in the different diameter subgroups.
In the logistic regression analysis predicting specificity, a stent diameter of ⩾3.5 mm had an odds ratio of 6.14 (99%CI 1.52 to 9.79). In predicting the PPV, a stent diameter of ⩾3.5 mm had an odds ratio of 3.70 (99%CI 0.98 to 8.90). All other variables were not significant.
The interobserver agreement in detecting restenosis was good (κ-value = 0.78). The intraobserver agreement was very good (κ-value = 0.87).
Cardiac catheterisation is the technique of choice for the detection of in-stent restenosis. However, it may involve life-threatening complications and is relatively expensive. The diagnostic accuracy of non-invasive techniques such as exercise testing is known to be suboptimal;24 therefore an alternative non-invasive “gatekeeper” to invasive coronary angiography would be desirable. The reliability of such a technique would have to be demonstrated in various clinical situations (ie, various stent sizes and configurations, higher heart rates) before it could be routinely used in patients with prior coronary artery stenting.
Feasibility of DSCT-CA at high heart rates
This study showed that DSCT-CA was accurate in detecting in-stent restenosis without the use of pre-scan β-blockers. Although the rate of false positives was slightly higher at high heart rates (PPV = 69%) than at low heart rates (PPV = 80%), the number of false negatives was low in both subgroups (NPV in the range 97–98%). In addition, the capability to perform DSCT-CA at higher heart rates without the use of β-blockers may be advantageous in terms of reducing radiation exposure. Table speed increases with increasing heart rates in DSCT-CA, which in turn translates into shorter scan times and reduced radiation exposure.
To the best of our knowledge, the population evaluated in the present study is the largest available in the literature on coronary CT angiography including overlapping and bifurcation stenting (n = 83). Although the differences were not statistically significant, we found that specificity and PPV of DSCT-CA in bifurcation stenting and overlapping stents were slightly lower than those obtained in single stents, probably due to the large amount of metal at the ostium of side branches and at overlapping sites. This excess of metal is a major source of high-density artifacts in CT and may lead to lesion overestimation (fig 3).
It may be argued that restenosis occurs more frequently in patients with complex lesion characteristics than in patients with simple lesions.25 This might have led to an overestimation of sensitivity in our complex configuration subgroup. However, the difference between the frequency of restenosis in our simple and complex configuration subgroups was not statistically significant.
We found that stent diameter was the most important feature influencing the diagnostic performance of coronary CT angiography. This is in keeping with the findings of Gilard et al8 and of Rixe et al,13 who performed 16-slice CT coronary angiography and 64-slice CT coronary angiography, respectively, in patients with low heart rates. In those studies, however, up to 50% of the stents were judged unevaluable. With DSCT-CA, we judged unevaluable only 5% (n = 9) of the stents, and found a low rate of false negatives irrespective of stent diameter (NPV in all stents = 98%, NPV in stents ⩽2.75 mm = 90%). When stent diameter was ⩽2.75 mm, DSCT-CA had a high rate of false positives (specificity = 64%; PPV = 52%) and could not predict with certainty the presence of restenosis; in particular, the specificity of DSCT-CA in this subgroup was significantly lower than that obtained in stents ⩾3.5 mm (OR = 6.14; 99%CI: 1.52 to 9.79). Therefore our study confirms that DSCT-CA is not a suitable gatekeeper to conventional angiography in symptomatic patients with ⩽2.75 mm stents.
In a recent study, Cademartiri et al4 reported the relationship between diagnostic accuracy of 64-slice CT and stent size in terms of the rate of false diagnosis (false positives and false negatives). In contrast to our findings, the rate of false diagnosis reported in that study did not decrease with increasing stent diameter. In stents <3 mm, the rate of false diagnosis was 6.1%; in 3 mm stents, the rate of false diagnosis was 1.6%; and, surprisingly, in stents >3 mm, the rate of false diagnosis was 12%. An explanation of these findings compared with our study is that only patients with stents larger than 2.5 mm were included. Secondly, complex stent configurations such as overlapping stents and bifurcation stenting were not included in that study, which may have led to an overestimation of the diagnostic performance in smaller stents. Lastly, a 64-slice CT scanner was used, rather than the dual source CT scanner in this study.
It is conceivable that future developments in CT technology might further increase the diagnostic performance in patients with stents by improvements in spatial (eg, flat panel technology) and temporal resolution.
Totally occluded stents
Totally occluded stents might be easier to recognise on DSCT-CA. In our sample, 17 stents were totally occluded. Overall, the analysis after exclusion of totally occluded segments yielded similar sensitivity, specificity and NPV but lower PPV (table 2).
We examined a selected symptomatic patient cohort; applicability to a wider population may therefore yield different results. This limitation was inevitable in order to compare DSCT-CA with the reference technique, that is, conventional coronary angiography.
X ray radiation exposure is a general limitation of multi-slice CT coronary angiography.26 27 However, using x-ray tube modulation, full dose radiation may be given for a shorter duration of the cardiac cycle than used in this study. Radiation dose is further reduced by automated pitch adaptation with higher heart rates. Other widely accepted imaging techniques, such as technetium sestamibi scans, may deliver radiation doses as high as 20 mSv.26 27
This study shows that, in patients with recurrent chest pain after stent implantation, DSCT-CA performs well in the detection of in-stent restenosis. Stent diameter is an important predictor of DSCT-CA diagnostic performance; when stent diameter is ⩽2.75 mm, the technique is associated with frequent false positives. However, due to the high NPV, DSCT-CA reliably rules out in-stent restenosis irrespective of stent size.
Competing interests: None.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.