Combined Quantitative Assessment of Myocardial Perfusion and Coronary Artery Calcium Score by Hybrid ⁸²Rb PET/CT Improves Detection of Coronary Artery Disease

Patients

From consecutive patients undergoing ⁸²Rb PET/CT MPI at Cedars Sinai Medical Center between January 2008 and September 2012 (n = 3,515), we excluded those with known CAD (myocardial infarction, percutaneous coronary intervention, or coronary bypass surgery [n = 1,416]), those with invasive coronary angiography greater than 6 mo of indexed PET MPI (n = 1,884), and those who had no CAC scan at the time of PET/CT imaging (n = 63). All selected patients (n = 152) had no cardiac events in the interval between MPI and invasive coronary angiography. Clinical and demographic data were collected during patient interviews before the scanning. All patients provided written informed consent for the use of their clinical and imaging data for research purposes. Pretest likelihood for having CAD was calculated as previously described (13).

Imaging and Analysis Protocols

A detailed description of PET MPI/CAC imaging protocols and perfusion deficit quantitation is provided in the supplemental materials (available at http://jnm.snmjournals.org). Briefly, the patients underwent same-day rest–stress gated ⁸²Rb PET MPI and CAC imaging on a hybrid Biograph 64 PET/CT scanner (Siemens) (14). A 6-min rest electrocardiogram-gated acquisition was started immediately before the injection of 925–1,850 MBq (25–50 mCi) of ⁸²Rb. Pharmacologic stress was performed using intravenous adenosine infusion (15) or regadenoson bolus (16) in 99 and 53 patients, respectively. No adjunctive exercise was performed. A 6-min stress electrocardiogram-gated acquisition was started immediately with the ⁸²Rb injection of 925–1,850 MBq (25–50 mCi). CT attenuation-correction scans were acquired for both stress and rest studies. After completion of PET MPI, without moving patients from the scan table, CAC scanning was performed. CAC scanning consisted of 30–40 slices encompassing the heart from the carina to the apex, with a 30- to 35-cm field of view sufficient to include the entire heart as well as the ascending and descending thoracic aorta. Radiation doses from ⁸²Rb PET and CAC ranged between 1.8 and 3.8 mSv and 0.7 and 1.2 mSv, respectively.

Left-ventricular segmentation for PET images was performed as recently described (17). For regional analysis, the myocardial wall division into vascular territories was performed according to the 17-segment model of the American Heart Association (18). An abnormality threshold of 3.0 mean absolute deviations (∼2.5 SDs) below the normalized counts of subjects in the normal reference database was applied to estimate the local extent of hypoperfusion in polar map coordinates. The perfusion defect extent was calculated as the percentage of the total surface area of the left ventricle for which test data were below the abnormality threshold. The quantitative perfusion variable used was total perfusion deficit (TPD), which reflects a combination of both defect severity and the extent of the defect as recently validated for ⁸²Rb PET perfusion analysis (17). Per-vessel and global (per-patient) ITPD was defined as stress TPD − rest TPD, computed separately for each vascular territory—left anterior descending (LAD), left circumflex (LCX), and right coronary artery (RCA)—and for the entire left ventricle. Standard per-vessel Agatston CAC scores and global (per-patient) CAC scores were calculated (19). The CAC score for LAD was calculated as the sum of CAC score of the left main (LM) and LAD. The CAC score for LCX was the sum of CAC scores of LM and LCX. The same CAC categories of 0, 1–99, 100–399, and 400 or greater were used for per-patient and per-vessel measures of the severity of coronary calcifications.

Invasive Coronary Angiography

Clinically indicated invasive coronary angiography was performed according to standard clinical protocols (mean time difference, 14 ± 27 days; range, 0–170 days) after the PET/CT MPI. Coronary angiograms were visually analyzed and reported by 2 experienced interventional cardiologists. A stenosis of 50% or greater of luminal diameter of the LM or 70% or greater in the LAD, LCX, and RCA was considered to signify obstructive CAD. Diagonal lesions were considered as LAD, obtuse marginal and ramus arteries were considered as LCX, and posterior descending and posterolateral branches were considered as RCA.

Statistical Analysis

All continuous variables were described as mean ± SD. A student 2-sample t test and χ² test were used to compare the differences in continuous and categoric variables, respectively. All tests of hypothesis were 2-sided, with a critical significance level of 5%. The natural logarithm of CAC score (logCAC) was used because of a wide range of the CAC scores. Multivariable logistic regression was used to develop an integrated (ITPD/logCAC) quantitative score predicting obstructive CAD. The first combined model used per-vessel ITPD/per-vessel logCAC, and the second model used per-vessel ITPD/per-patient logCAC. The prediction of obstructive CAD by the models was compared with the prediction by ITPD alone, as previously reported (17). A 10-fold cross-validation (CV) technique (20) that ensured that none of the image data creating the logistic regression model were used in the subsequent evaluation of the diagnostic performance of the same model was applied. To further clarify, in 10-fold CV each of the 10 models was built from approximately 90% of the 456 vessels—that is, from 410 vessels—and was then tested in the remaining 46 vessels. This procedure was repeated 10 times, each time leaving out a different 10% of the vessels for testing. Thus, for each procedure the specific model was tested only in unseen vessels, but there were 10 separate models. In our implementation of the 10-fold CV, the patients’ arteries were always clustered at the patient level; thus, in each-fold of the CV, there were no patient data for which one artery belonged to the training set and the other artery of the same patient to the validation set. The procedure was performed in this manner to ensure the real-life test of the application, in which the method is tested always in new patients and no single patient is used in the training and validation at the same time. 10-fold CV has been definitively shown to have the smaller bias for discriminant analysis than other methods, such as split-sample, as traditionally used (20), in which just 1 arbitrary division into test and validation groups is performed. Receiver-operating-characteristic area-under-the-curve (AUC) analysis (21) and integrated discrimination improvement (IDI) (22) were applied to evaluate the discrimination ability of relevant predictors. IDI was defined as the difference between the integral of test sensitivity (IS) over all possible cutoff values from the (0, 1) interval and the corresponding integral of 1 minus specificity (IP). Thus, the IDI is defined as follows:The subscript new in the above expression refers to the model with the new marker, and subscript old refers to the model without it. Because the integrals of sensitivity and 1 minus specificity over the (0, 1) interval can be seen as average sensitivity and 1 minus specificity, the IDI can be viewed as a difference between improvement in average sensitivity and any increase in average 1 minus specificity. Statistical analyses were performed with STATA software (version 12; Stata Corp LP) and SAS software (version 9.2; SAS Institute Inc.).