Computed tomography coronary angiography accuracy in women and men at low to intermediate risk of coronary artery disease

Objectives To investigate the diagnostic accuracy of CT coronary angiography (CTCA) in women at low to intermediate pre-test probability of coronary artery disease (CAD) compared with men. Methods In this retrospective study we included symptomatic patients with low to intermediate risk who underwent both invasive coronary angiography and CTCA. Exclusion criteria were previous revascularisation or myocardial infarction. The pre-test probability of CAD was estimated using the Duke risk score. Thresholds of less than 30 % and 30–90 % were used for determining low and intermediate risk, respectively. The diagnostic accuracy of CTCA in detecting obstructive CAD (≥50 % lumen diameter narrowing) was calculated on patient level. P < 0.05 was considered significant. Results A total of 570 patients (46 % women [262/570]) were included and stratified as low (women 73 % [80/109]) and intermediate risk (women 39 % [182/461]). Sensitivity, specificity, PPV and NPV were not significantly different in and between women and men at low and intermediate risk. For women vs. men at low risk they were 97 % vs. 100 %, 79 % vs. 90 %, 80 % vs. 80 % and 97 % vs. 100 %, respectively. For intermediate risk they were 99 % vs. 99 %, 72 % vs. 83 %, 88 % vs. 93 % and 98 % vs. 99 %, respectively. Conclusion CTCA has similar diagnostic accuracy in women and men at low and intermediate risk. Key Points • Coronary artery disease (CAD) is increasingly investigated by computed tomography angiography (CTCA). • CAD detection or exclusion by CTCA is not different between sexes. • CTCA diagnostic accuracy was similar between low and intermediate risk sex-specific-groups. • CTCA rarely misses obstructive CAD in low–intermediate risk women and men. • CAD yield by invasive coronary angiography after positive CTCA is similar between sex-risk-specific groups.


Introduction
Cardiovascular disease is the leading cause of death in the western world [1]. For several years awareness has been growing that cardiovascular disease is also the primary cause of death in women not only in the western world but also in economic developing countries [2]. Part of this high mortality is related to under-recognition, underdiagnosis and undertreatment of coronary artery disease (CAD) in women [3][4][5][6][7].
The interpretation of chest pain in women caused by CAD is often difficult because of the "less typical" symptoms in women compared with the classical "typical" symptoms in men [3,4,8]. This may cause a delay or lead to incorrect diagnosis [3,4]. Additionally first-line non-invasive diagnostic tests such as exercise electrocardiography and single photon emission computed tomography (SPECT) imaging in women are less sensitive and specific [5,6].
Paradoxically, because of the diagnostic uncertainty there appears to be an overuse of invasive coronary angiography (ICA) in women, with a rather low diagnostic yield of obstructive CAD [9,10]. This prompted a "red alert call" for the promotion of advanced non-invasive imaging techniques in women including CT coronary angiography (CTCA) [11]. CTCA has evolved as a reliable gatekeeper of ICA because negative CT findings virtually rule out the presence of obstructive CAD [12][13][14]. Whether this also applies to women remains unresolved because women have been under-represented in studies assessing the diagnostic accuracy of CTCA [12][13][14][15][16][17]. The pre-test probability (PTP) of obstructive CAD plays a significant role in the assessment of the clinical utility of CTCA. This PTP of obstructive CAD can be derived from reported clinical prediction rules [18] and may influence the post-CTCA probability of obstructive CAD and thus the need for further testing and management.
The purpose of our study was to assess the diagnostic accuracy and clinical utility of CTCA to detect or exclude obstructive CAD in women with low to intermediate PTP of obstructive CAD in comparison to men.

Study population
Patients referred by their treating physician for ICA between July 2004 and June 2009 on the basis of chest pain presentation (typical, atypical and non-anginal complaints) with or without outcome of stress testing were invited to undergo CTCA within 2 weeks of ICA. The CTCA outcome did not affect referral to ICA. We excluded patients with known iodine allergy, impaired renal function and patients with previous revascularisation. A total of 907 patients gave consent for CT accuracy studies and were entered in our single-centre registry. Out of these 907 patients we excluded patients with previous myocardial infarct (n054) as well as patients with a high PTP (>90 %) of having obstructive CAD (n0283). The final study sample comprised a total of 570 symptomatic patients. The study was approved by the institutional review board of our medical hospital.

Computed tomography imaging protocol and image reconstruction
Patients underwent single-source CT (Somatom Sensation, Siemens Healthcare, Forchheim, Germany) between 2004 and 2006 and subsequently underwent dual-source CT (Somatom Definition, Siemens Healthcare). The imaging settings are described in Table 1. Heart-rate-lowering medication (Metoprolol, Seloken, Astra Zeneca, Zoetermeer, Netherlands) was administered when the heart rate was above 65 beats per minute before the single-source CT in the absence of contraindications to achieve better quality images. To achieve better visualisation of the small coronary arteries the vasodilating agent Nitrolingual (Nitroglycerin Pumpspray, G.Pohl-boskamp, Itohenlockstedt, Germany) was administered before imaging in the absence of contraindications.
All patients initially underwent unenhanced CT to calculate the calcium score using the Agatston method [19]. Subsequently a bolus tracking technique was used to synchronise the start of image acquisition with the arrival of the iodinated contrast agent [Iomeprol, iomeron (400 milligram iodine per millilitre), Bracco, Milan, Italy; Ultravist, iopromide (370 milligram iodine per millilitre), Schering Berlin, Germany] in the coronary arteries followed by a saline chaser.
Between 2004 and 2006 a full-dose ECG-synchronised spiral imaging technique was used. After the introduction of ECG-triggered tube current modulation, the spiral imaging technique with ECG pulsing was used in the subsequent years to reduce the effective radiation dose. Data sets were reconstructed retrospectively according to the heart rate to obtain motion-free images. For heart rates of at least 80 beats per minute the data sets were reconstructed in 31-47 % of the R-R interval (systolic phases). For heart rates of no greater than 65 beats per minute this value was 60-76 % of the R-R interval (diastolic phases). For heart rates between 66 and 79 beats per minute reconstructions were needed in both systolic and diastolic phases, 30-77 % of the R-R interval. Images were analysed using medium-to-smooth convolution kernels for non-calcified lesions and sharp convolution kernels for calcified lesions.

Computed tomography image evaluation
All data sets were transferred for analyses to an offline proprietary workstation (Leonardo, Siemens, Erlangen, Germany); the total calcium score was calculated using dedicated proprietary software (CaScoring). Two experienced observers with more than 2 years of experience in cardiac CT, blinded to the ICA results, independently evaluated all CTCAs for the presence of CAD, using axial source images, multiplanar, curved reformatted reconstructions, and thinslab maximum intensity projections using "Circulation" software on the proprietary workstation. Interobserver disagreements were resolved by a joint consensus reading.
The modified 17-segment American Heart Association model was used to classify each segment [20]. Each segment was visually scored as obstructive in the presence of at least 50 % lumen diameter narrowing and non-obstructive when the lumen diameter narrowing was less than 50 % in comparison with the proximal and distal lumen. All anatomically available coronary segments with a diameter of at least 1.5 mm, irrespective of image quality or calcification, were included and scored with the intention to diagnose. Un-evaluable segments of poor quality due to calcification, stack, motion artefacts, or low contrast enhancement were scored as obstructive. Coronary segments distal to a total occlusion could not be optimally visualised by ICA and thus were excluded from the analyses. The image quality was scored on a segment level as diagnostic and non-diagnostic. The average of this score per patient was compared between women and men.

Invasive coronary angiogram image evaluation
One experienced cardiologist, blinded to the CT results, visually assessed each coronary segment (American Heart Association model) for the presence of luminal narrowing in two orthogonal planes. Stenoses scored as having more than 20 % narrowing on visual assessment were quantified using the validated quantitative coronary angiography (QCA) algorithm (CAASII [Cardiovascular Angiography Analysis System II]; Pie Medical Imaging Maastricht, the Netherlands). The segments were considered obstructive when the quantified lumen diameter narrowing in one of the two planes was at least 50 %. Obstructive lesions on a segment level were used to determine obstructive CAD on a patient level as having one or more obstructive stenoses (≥50 % lumen diameter stenosis) irrespective of the segment in which they were located.

PTP of obstructive CAD
We used the Duke risk score [18] to estimate the PTP of obstructive CAD due to the presence of multiple risk factors such as age, gender, symptoms, history of myocardial infarction, ECG, smoking, hypercholesterolaemia and diabetes mellitus. We used less than 30 %, 30-90 %, and greater than 90 % as thresholds for the low, intermediate and high risk group, respectively. As patients with a high risk are directly referred for ICA we did not include these patients in our analyses [21,22].
The observed PTP of obstructive CAD is the prevalence of obstructive CAD defined as having at least one lesion with at least 50 % lumen diameter stenosis per patient detected by ICA.

Statistical analyses
The statistical analysis was performed using a dedicated statistical software program (SPSS, version 16.0, IMB, Chicago, IL, USA). Categorical variables were expressed as percentages and continuous variables were expressed as means ± standard deviation. Continuous variables with a skewed distribution were expressed as median with interquartile range. The estimated PTP was compared in women, men and in the different risk groups with the observed PTP (observed prevalence) using the paired t test.
The patients' characteristics were compared between women and men and between different risk groups using the independent t test; otherwise the Mann-Whitney U test was used (skewed distribution of continuous data). The interobserver variability was tested between the two CT readers on segment and patient level using the κ statistic.
The CTCA results were compared with the reference standard ICA on a patient level to calculate the sensitivity (SN), specificity (SP), positive predictive value (PPV) and negative predictive value (NPV) [23]. The Wilson score [24] was used to calculate the confidence intervals for small groups. The diagnostic accuracy and utility were compared across the different risk groups, between women and men using the chi-squared test or the Fisher's exact test in the presence of less than five observations in a cell of the 2×2 table. A P value of less than 0.05 was considered statistically significant.
The receiver-operating characteristic (ROC) curve was used for visual analyses of the trade-offs between the sensitivity and the specificity of CTCA. The area under the ROC curve (AUC) was calculated to account for the influence of referral bias and prevalence of disease on the diagnostic accuracy of CTCA in our population [25,26].
The clinical utility of CTCA was a measure of diagnostic certainty or remaining uncertainty of absence or presence of CAD that is derived from the post-test probability of CAD. In case of diagnostic certainty no further testing is required, whereas remaining uncertainty requires further diagnostic testing. It was assumed that a post-test probability of obstructive CAD of less than 5% indicated high certainty with no further requirement for diagnostic testing; probabilities between 5 and 90 % represented uncertainty and indicated a requirement for further diagnostic testing; and probabilities of greater than 90 % indicated high certainty with direct referral to ICA, as proposed by previous studies [27].
This study was performed according to the criteria set forth in the Standard for Reporting of Diagnostic Accuracy Initiative [28].

Results
In total, 570 patients at low to intermediate risk were included of which 46 % were women (n0262). The patients' characteristics are provided according to sex and risk group in Table 2. Women overall had a significantly higher heart rate (HR) than men (P00.03). The HR was similar in the different risk groups ( Table 2). The prevalence of CAD (at least one obstructive lesion) in women was 61 % and in men this was 64 % (P00.42). The PTP of obstructive CAD determined by the Duke risk score (estimated PTP) was significantly lower than the observed prevalence of obstructive CAD determined by ICA in women in the overall group (45 % vs. 61 %, P<0.01) as well as in the low ( In this population the prevalence of obstructive CAD and extent of CAD in terms of multi-vessel disease and calcium score were similar between women and men ( Table 2).
The interobserver variability for detection of obstructive stenosis showed a good agreement on segment (κ00.  (Table 3).

Discussion
In our clinical evaluation of CTCA in low to intermediate risk women and men with suspected CAD who were  In our study the overall predicted probability of having obstructive CAD using the Duke risk score was underestimated in women (45 % vs. 61 %, P<0.01) compared with the actual presence of CAD assessed by ICA. This finding probably reflects the difficulties of evaluating women with symptoms suspected of CAD and has been described in previous studies [3,8]. Physicians may misclassify the  symptoms as non-anginal or atypical angina in women, which consequently affects the Duke risk score where symptom presentation plays a significant role in the calculation of the PTP. Underestimation of the Duke risk score may also be related to the referral bias of our study population, in particular in women who were referred by their treating physicians for ICA. It is noteworthy that the diagnostic accuracy of CTCA in detecting obstructive CAD was similar in women and men (  [29]. This is in contrast to the reported technical limitations and diminished accuracy in women of other noninvasive ischaemia tests [6,30] and underscores the gender neutrality (equality) of the diagnostic accuracy of CTCA.
The clinical utility of CTCA, i.e. reflecting the remaining diagnostic certainty of the presence or absence of significant CAD, is an important criterion in the overall clinical assessment of CTCA. The clinical utility is derived from the posttest probability of CAD and even with the same diagnostic accuracy of a test (sensitivity/specificity) it is different in patients with low or intermediate PTP of CAD. We arbitrarily assumed that in patients with a post-test probability of less than 5 % or greater than 90 % sufficient diagnostic certainty was achieved and no further diagnostic testing was required. In patients with a post-test probability between 5 % and 90 % diagnostic certainty was not sufficient and in these patients further diagnostic testing was deemed necessary. In our population we found that the clinical utility of a positive CTCA was high (>90 %) in men at intermediate PTP of CAD and further diagnostic testing was not necessary. These patients may then be referred to ICA. The clinical utility of a positive CTCA was moderate in women at low to intermediate PTP and in men of low PTP of CAD. In these patients diagnostic certainty was insufficient and further diagnostic testing may be needed. Higher certainty may be then achieved by functional testing for ischaemia e.g. by SPECT or by stress echocardiography [31,32].
The clinical utility of a negative CTCA was high in women and men at low or intermediate PTP of CAD providing high certainty (<5 %) of absence of obstructive CAD, and no further testing would be necessary. It has been shown in numerous studies that the absence of CAD detected by CTCA is associated with an excellent prognosis [33,34], which further lends support that these patients may be safely discharged. It should be noted that in patients with persistent chest complaints due to microvascular dysfunction, normal CT findings with no obstructive epicardial CAD may be seen [35]. In these patients, who are often women, further diagnostic testing for ischaemia to detect microvascular dysfunction will be needed to adjust clinical management and survival [36].
Recent studies reported that the yield of elective ICA to demonstrate the presence of obstructive CAD in women and men is 27-49 % and 47-67 %, respectively, in patients referred for ICA [9,10], whereas by performing CTCA before diagnostic ICA the yield may improve to 86 % for women and 92 % for men (Fig. 1) and may decrease the number of unnecessary ICA in women and men by 30 % and 31 %, respectively, as suggested by our study.
Our study has limitations. Referral bias was present, because the patients in our population were referred by their treating physicians to undergo invasive coronary imaging on the basis of their chest pain presentation with or without outcome of stress testing. This may explain the relatively high prevalence of obstructive CAD, in particular in women. To account for the influence of referral bias on the diagnostic accuracy [25,26] of CTCA we also reported the AUC which was not different between women and men (0.87 vs. 0.92, P0 0.056). Our study population only consisted of stable patients and our results may not apply to the wider spectrum of patients with suspected CAD who did not undergo ICA or who have unstable symptoms. Of concern was the rather high radiation exposure of CTCA (12 mSv) due to the use of a retrospective imaging protocol, which was standard in first-generation CT [14,37]. Currently with the newer-generation systems and optimal imaging protocols much lower levels of radiation exposure (<3 mSv) can be achieved in patients with low heart rates (<65 beats per minute) [38].
In conclusion, computed tomography coronary angiography has similar diagnostic accuracy in women and men with low and intermediate risk, and may function as an efficient gatekeeper for ICA in women as well as in men.