Prediction of coronary artery disease by a systemic atherosclerosis score index derived from whole-body MR angiography

The study population consisted of 50 patients scheduled to undergo elective cardiac catheterization for suspected coronary artery disease at the Department of Cardiology, University Hospital Heidelberg, Germany. Patients were recruited from the cardiology outpatient clinic during February and March 2007. A total of 74 consecutive patients with the indication for elective cardiac catheterization for suspected CAD were screened. Pre-defined exclusion criteria for the study were [numbers of patients excluded based on each criteria]: known extra-cardiac atherosclerosis [n = 3], age < 18 years [n = 0], inability or unwillingness to give informed consent [n = 7], pregnancy [n = 0], renal failure with a glomerular filtration rate < 60 ml/min/1.73 m² [n = 5] and general contraindications to MRI such as presence of pacemakers [n = 0], implantable cardioverter defibrillators [n = 0], intracranial clips [n = 1], other foreign metallic objects [n = 3] or known severe claustrophobia [n = 5]. WB-MRA was performed several days before the date of the scheduled cardiac catheterization and ankle-brachial blood pressure index (ABI) was measured at the time of the MRI scan. PROCAM [11] and Framingham [12] scores were calculated for each patient as measures of the likelihood to suffer from a coronary event within the next 10 years. The study was performed in accordance to the Declaration of Helsinki. The ethics committee of the University of Heidelberg approved the study protocol and all patients gave their written informed consent.

Whole-body MRA was performed on a 1.5T clinical scanner (Intera Achieva^®, Philips Medical Systems, Best, The Netherlands) equipped with a table top extender using MobiTrak^® software (Philips Medical Systems). The system integrated quadrature body coil was used for signal transmission and reception. Patients were placed in a supine position feet-first. A survey scan with a time-of-flight sequence was performed for subsequent planning. MRA was acquired with a 3-dimensional T1 RF-spoiled gradient echo sequence in 4 slightly overlapping coronal data sets ("stations"). The first station included the adominal aorta and its branches, the second station started at the external iliac arteries and continued to the popliteal arteries, the third station contained the arteries of the lower leg. The fourth station consisted of the thoracic aorta and its supraaortic branches. Typical scan parameters were: TR 5 ms, TE 1.5 ms, flip angle 40°, FOV 451 mm, acquired voxel sizes 1.66 × 1.74 × 2 mm³ (station 1), 1.41 × 1.71 × 1.8 mm³ (station 2), 1.23 × 1.23 × 0.75 mm³ (station 3) and 1.66 × 1.74 × 2 mm³ (station 4). Reconstructed voxel size for all stations was 0.88 × 0.88 × 1 mm³. A breath-hold was used for station 1 only [5, 6]. After acquisition of unenhanced images, 26 ml of gadopentetate dimeglumine (Magnevist^®, Bayer Schering Pharma, Germany) were power injected (flow rate of 1.2 ml/sec for the first 13 ml, 0.4 ml/sec for the second 13 ml, followed by 25 ml saline with a flow rate of 0.4 ml/sec). Scanning of the first three stations was timed using real-time bolus monitoring (BolusTrak^®, Philips Medical Systems), the image acquisition was started when the distal aorta and aortic bifurcation were clearly visible with contrast. After a wash-out period of 7 minutes, images of the fourth station were acquired in a similar fashion with the injection of 14 ml contrast media followed by 25 ml saline (flow rate for both 2 ml/sec). The scan was started when the ascending aorta was filled with contrast. For all stations, unenhanced images were subtracted from enhanced images. Maximum-intensity-projections were obtained from subtracted images in standard frontal and lateral projections. Multiplanar reconstructions were generated for image interpretation.

Interpretation of WB-MRA data was performed on a commercially available work station (ViewForum^®, Philips Medical Systems). For image analysis the arterial tree was divided into 40 segments ranging from the internal carotid artery to the arteries of the lower leg (Figure 1A). Using source images and multiplanar reconstructions all segments were graded with regard to their diagnostic quality (0 = non-diagnostic, 1 = poor image quality, diagnosis suspected, 2 = moderate quality, some blurring/artefacts but definite diagnosis possible 3 = good quality and 4 = excellent image quality) and their most severe stenosis (≤25%, 26- 50%, 51-75%, 76-99%, occlusion) by consensus reading of two experienced observers (S.L. and H.S.) blinded to the results of coronary angiography and other clinical data. Reasons for non-diagnostic quality were noted and included venous overlap, poor signal-to-noise ratio (SNR), motion artefacts and others (e.g. anatomic variants, orthopedic implants). All evaluable vessel segments received a score according to the severity of stenosis, ranging from 1 for a normal vessel to 6 for a total occlusion. The atherosclerosis score index (ASI) for each patient was calculated by dividing the sum of vessel scores by the number of analysed segments (Figure 1B).

Cardiac catheterization was performed via the femoral or radial artery and coronary angiograms were obtained from all patients. Signficant coronary artery disease was defined as presence of vessel segments with > 50% luminal narrowing and was classified into 1- 2- and 3 vessel disease according to number of affected vessels. Coronary revascularization (percutaneous intervention or bypass grafting) was recommended at the discretion of the attending physician and either performed directly or at a later time point.

The ABI was measured and calculated according to the recommendations of the American Heart Association [13]. A blood pressure cuff and a doppler ultrasonic sensor were used to determine systolic blood pressure in the brachial arteries and the posterior tibial and dorsalis pedis arteries. The ABI for each leg equaled the ratio of the higher of the two systolic pressures (posterior tibial or dorsalis pedis) and the average of the brachial pressures. In cases where there was a difference of > 10 mmHg between brachial artery pressures, the higher value was used. An ABI of < 0.9 was considered a pathological finding.

Data were expressed as mean ± SEM or as median with interquartile range (IQR; 25-75th percentile) if not normally distributed. Differences between two groups were compared by Student's t-test or Mann-Whitney-U-Test for non-parametrical continuous variables. Categorical variables between groups were compared by chi-square test. The associations between ASI and Framingham and PROCAM scores were assessed using Pearson's correlation coefficient. ROC analysis was performed to determine the optimal ASI cut-off for diagnosis of significant CAD. Logistic regression analysis was carried out to provide the optimal model for prediction of the odds for significant CAD. For all analyses, a p-value of < 0.05 was regarded statistically significant. All statistical analyses were carried out using the Statistical Package for Social Science software version 16.0 (SPSS Inc., Chicago, USA).

In 50 patients a total number of 1793/2000 (90%) vessel segments were evaluable. Excellent and good image qualities were found in 249 (14%) and 1235 (69%) segments respectively, while 270 (15%) segments showed moderate and only 39 (2%) poor diagnostic quality. Reasons for the non-diagnostic quality included motion artefacts (22/207), poor SNR (151/207), venous overlay (12/207) and others (22/207). Of the non-evaluable segments 52% were located in the abdominal area, while 18% were found in the supraaortic branches and 30% in the lower extremities.

A total of 27 stenoses > 50% were detected with a distribution among the different vessel territories as follows: 1. thoracic aorta/supraaortic branches n = 5 (19%); 2. abdominal aorta with visceral and pelvical branches n = 10 (37%); 3. lower extremities starting with the common femoral artery n = 12 (44%). A total of 18 patients (36%) were found to exhibit at least one luminal narrowing > 50%, among these, 6 patients had two ore more > 50% stenoses.

The mean ASI among the study cohort was 1.43 ± 0.05, ranging from 1 to 2.56. A higher ASI was not associated with older age (> 70 years), gender or traditional cardiovascular risk factors. However, ASI was markedly higher in patients with an ABI < 0.9 (Table 2). Significant correlations were found between the ASI and PROCAM as well as Framingham risk scores (Figure 2).

Patients with significant coronary artery disease (n = 27) exhibited a higher ASI compared to patients without significant coronary disease (n = 22; 1.56 vs. 1.28, p = 0.004). Furthermore, ASI was markedly elevated in patients undergoing coronary revascularization (1.6 vs. 1.34, p = 0.01). ROC analysis revealed an ASI cut-off of >1.54 as the optimal discriminator for the presence of significant CAD with a sensitivity of 59%, a specificity of 86% and a positive predictive value of 84%. Likewise, a cut-off for severe CAD necessitating coronary revascularization of >1.6 could be determined with a sensitivity of 59%, a specificity of 90% and a positive predictive value of 63% (Figure 3). Significant CAD was associated with the presence of higher grade extra-cardiac stenoses - while 15/27 (55%) of patients with CAD exhibited at least one extra-cardiac stenosis > 50%, only 3/22 (14%) of those patients without CAD did (p = 0.003, Figure 4). The incidence of a significant extra-cardiac stenosis when CAD was present differed between vessel territories. While 4/27 (15%) patients with CAD suffered from carotid artery stenoses, higher grade stenoses in the abdominal/pelvical regions were present in 7/27 (26%) patients and a total of 12/27 (44%) patients had concomitant stenoses of the arteries in the lower extremities.

The presence of a stenosis > 50% was slightly less powerful compared to the ASI discriminator in predicting the presence of CAD with a sensitivity of 56% and a specificity of 86% (positive predictive value 83%). Figure 5 displays an example of a patient with a high ASI and 3-vessel coronary disease.

Together with other variables associated with CAD in univariate analysis, the ASI cut-off of 1.54 was used to build logistic regression models predicting the odds of significant CAD. Table 3 displays the best-fit model. Among the variables entered (age > 70 yrs, male gender, ABI < 0.9, PROCAM score), the odds ratio was statistically different from 1 for ASI > 1.54 only. In the specific model, an ASI > 1.54 was associated with a 11-fold increase in likelihood to suffer from significant CAD. When entered into logistic regression models as a continuous variable ASI did not retain its independent predictive value.