Author + information
- Received November 21, 2011
- Revision received January 9, 2012
- Accepted January 26, 2012
- Published online July 1, 2012.
- Afshin Farzaneh-Far, MD, PhD⁎,†,
- Harry R. Phillips, MD†,
- Linda K. Shaw, MS‡,
- Aijing Z. Starr, MS‡,
- Mona Fiuzat, PharmD‡,
- Christopher M. O'Connor, MD†,‡,
- Ashwani Sastry, MD†,
- Leslee J. Shaw, PhD§ and
- Salvador Borges-Neto, MD†,‡∥,⁎ ()
- ↵⁎Reprint requests and correspondence:
Dr. Salvador Borges-Neto, Duke University, Medicine and Radiology, Duke University Medical Center, Durham, North Carolina 27710
Objectives The aim of this study was to evaluate the independent prognostic significance of ischemia change in stable coronary artery disease (CAD).
Background Recent randomized trials in stable CAD have suggested that revascularization does not improve outcomes compared with optimal medical therapy (MT). In contrast, the nuclear substudy of the COURAGE (Clinical Outcomes Utilizing Revascularization and Aggressive Drug Evaluation) trial found that revascularization led to greater ischemia reduction and suggested that this may be associated with improved unadjusted outcomes. Thus, the effects of MT versus revascularization on ischemia change and its independent prognostic significance requires further investigation.
Methods From the Duke Cardiovascular Disease and Nuclear Cardiology Databanks, 1,425 consecutive patients with angiographically documented CAD who underwent 2 serial myocardial perfusion single-photon emission computed tomography scans were identified. Ischemia change was calculated for patients undergoing MT alone, percutaneous coronary intervention, or coronary artery bypass grafting. Patients were followed for a median of 5.8 years after the second myocardial perfusion scan. Cox proportional hazards regression modeling was used to identify factors independently associated with the primary outcome of death or myocardial infarction (MI). Formal risk reclassification analyses were conducted to assess whether the addition of ischemia change to traditional predictors resulted in improved risk classification for death or MI.
Results More MT patients (15.6%) developed ≥5% ischemia worsening compared with those undergoing percutaneous coronary intervention (6.2%) or coronary artery bypass grafting (6.7%) (p < 0.001). After adjustment for established predictors, ≥5% ischemia worsening remained a significant independent predictor of death or MI (hazard ratio: 1.634; p = 0.0019) irrespective of treatment arm. Inclusion of ≥5% ischemia worsening in this model resulted in significant improvement in risk classification (net reclassification improvement: 4.6%, p = 0.0056) and model discrimination (integrated discrimination improvement: 0.0062, p = 0.0057).
Conclusions In stable CAD, ischemia worsening is an independent predictor of death or MI, resulting in significantly improved risk reclassification when added to previously known predictors.
Recent randomized trials have shown no improvement in death or myocardial infarction (MI) with an early revascularization strategy compared with initial optimal medical therapy (MT) in patients with stable coronary artery disease (CAD) (1,2). However, the COURAGE (Clinical Outcomes Utilizing Revascularization and Aggressive Drug Evaluation) nuclear substudy found that revascularization more frequently resulted in ≥5% ischemia reduction and suggested in an unadjusted analysis that this ischemia reduction may be associated with improved outcomes (3). Importantly, this association was not maintained after adjustment to treatment arm, and multivariate adjustment to other known predictors was not possible, because this exploratory substudy was not powered to examine clinical outcomes. Thus, in patients with stable CAD, the relative effects of MT and revascularization on ischemia require further investigation, and the independent prognostic significance of ischemia change is unclear.
Despite these uncertainties, in clinical practice, ischemia is 1 of the primary drivers of decisions regarding revascularization in patients with stable CAD (4). Therefore, clarification of these questions has significant implications for both patient management and healthcare resource utilization.
The aims of this study were threefold: 1) to compare the change in ischemia with MT, percutaneous coronary intervention (PCI), or coronary artery bypass grafting (CABG) as measured by serial myocardial perfusion single-photon emission computed tomography (MPS) scans in a large cohort of patients with stable CAD; 2) to evaluate the independent prognostic utility of ischemia change; and 3) to assess whether the addition of ischemia change to traditional predictors results in improved risk classification for death or MI.
We conducted an observational analysis of patients from the Duke Cardiovascular Disease and Nuclear Cardiology Databanks. The Duke Cardiovascular Disease Databank is a prospective longitudinal record of the details and clinical courses of >53,000 cardiovascular patients in the Duke University Health System with angiographically significant CAD (>75% stenosis of at least 1 major epicardial coronary artery). All patients are followed at 6 months, 1 year, and annually thereafter, with recording of major clinical events. The Duke Nuclear Cardiology Databank contains the clinical details of >50,000 MPS scans performed at Duke over the past 15 years.
We identified 1,425 consecutive patients with angiographically documented CAD who underwent 2 serial MPS scans between September 1993 and June 2009 within a 36-month time frame (Fig. 1). All patients had CAD documented by coronary angiography within 180 days of their MPS studies. Patients with incomplete angiographic or nuclear data were excluded.
Demographic and clinical characteristics were recorded prospectively at the time of coronary angiography and MPS. Prospectively collected variables included age, sex, race, left ventricular (LV) ejection fraction (LVEF), hypertension, diabetes, smoking history, hyperlipidemia, New York Heart Association functional classification, cardiac medications, CAD severity as reflected by the number of diseased vessels, prior MI, history of revascularization, and a modified Charlson comorbidity index (5).
Stress testing and single-photon emission computed tomography imaging
Patients capable of exercise underwent treadmill stress testing with the Bruce protocol, unless an alternative protocol was requested by the ordering physician. Patients unable to exercise underwent pharmacological stress testing. MPS scans were performed according to previously described Duke University nuclear laboratory protocols (6,7). In brief, single-photon emission computed tomography images were obtained with multiple-head detectors, with 30 s/projection at rest and 20 s/projection during stress. The studies were clinically interpreted by 3 nuclear cardiologists in our laboratory, without attenuation correction. Commercially available software (Cedars Sinai QGS/QPS [Cedars-Sinai Medical Center, Los Angeles, California] or Emory Toolbox [Emory University, Atlanta, Georgia]) was used to determine the gated single-photon emission computed tomography LVEF.
The MPS studies were evaluated semiquantitatively for the severity and extent of abnormalities, with relative perfusion recorded in each myocardial segment (0 = no defect, 1 = mild defect, 2 = moderate defect, and 3 = severe defect) at rest and during stress. The summed difference score, which is the sum of the differences between the stress and rest perfusion scores (reversible defects), was determined for each patient. The Duke Nuclear Cardiology Databank has used a 4-point severity scale since its inception and was initiated before the current American Society of Nuclear Cardiology–recommended 5-point scale. We have kept the same system of scoring to maintain consistency throughout our database over time. At the time these data were collected, we used a 12-segment model. We used a previously reported algorithm for the conversion of 12-segment perfusion scores to 17-segment scores, which is highly correlated with expert reading of the same studies using the 17-segment model (8). Thus, we have a robust method for converting 12-segment data into 17-segment data. In addition, we have previously shown that the prognostic value of the converted data is nearly identical to the prognostic information derived from our 12-segment model data (8). The percentage LV ischemia for each scan was calculated as: 100 × summed difference score/51 (8). For each patient, the change in LV ischemia between the 2 MPS scans was calculated.
Follow-up and outcomes
Treatment group was assigned on an intention-to-treat basis. Patients were included in the PCI (n = 419) or CABG (n = 135) group on the basis of the first revascularization procedure occurring within 60 days of the first MPS scan (Fig. 1). If no revascularization occurred within this period, they were assigned to the MT group (n = 871). Follow-up time was initiated at the second MPS scan. The primary endpoint was a composite of death or MI. An independent clinical events committee reviewed and classified major clinical events without knowledge of the clinical data or MPS results. Follow-up was 96% complete.
Differences in baseline characteristics were compared using t tests or chi-square statistics. When appropriate, paired t tests were used to compare paired data. Continuous variables that were not distributed normally were compared using Wilcoxon rank sum tests for independent groups and Wilcoxon signed rank tests for paired data. On the basis of the nuclear substudy of COURAGE, a threshold change in ischemia of ≥5% of the LV was used. The COURAGE investigators chose this value because it represented a change that exceeded test repeatability (9).
Kaplan-Meier methods were used to evaluate time to the primary outcome of death or MI. Cox proportional hazards regression modeling was used to identify factors that were independently associated with death or MI. After examining the results of a flexible Cox model-fitting approach involving cubic polynomial spline functions (10), the linearity of the unadjusted relationship between each continuous variable and death or MI was assessed. The proportionality assumption was verified using Schoenfeld residuals. For the multivariate model, covariates were chosen on the basis of known clinical risk factors as well as by stepwise selection (and backward elimination) at p < 0.05 from the list of baseline characteristics. To explore potential modifying effects of baseline ischemia, the interaction between ≥5% ischemia improvement or worsening and ischemia on the first scan was tested in the adjusted Cox model.
To assess the added prognostic value of ≥5% ischemia worsening, the final model was compared with a model in which ≥5% ischemia worsening was not included. The global chi-square statistic for both models was calculated, and a likelihood ratio test was performed. Model discrimination was compared by calculating the integrated discrimination improvement, which measures the improvement in the average sensitivity and specificity of the model with the new predictor (11). In addition, the C-statistic for each model was derived. Formal risk reclassification analyses were conducted by examining the net reclassification improvement (NRI) statistics (11). Ten-year risk for death or MI was categorized as 0% to 20%, >20% to 40%, >40% to 60%, and >60%. Risk categories determined on the basis of the models with and without ≥5% ischemia worsening were cross-tabulated to describe the number and percent of patients who were reclassified appropriately (i.e., to a lower risk group for nonevents or to a higher risk group for events) and inappropriately (i.e., to a lower risk group for events or to a higher risk group for nonevents). The NRI was estimated as: ([number of events reclassified higher − number of events reclassified lower]/number of events) + ([number of nonevents reclassified lower − number of nonevents reclassified higher]/number of nonevents). Assessment of overall model fit was performed using the survival-adapted Hosmer-Lemeshow chi-square statistic (12).
All tests were 2-tailed, and statistical significance was declared at alpha <0.05. Statistical analyses was performed using SAS version 8.2 (SAS Institute, Inc., Cary, North Carolina). The study protocol was reviewed and approved by the Duke University Medical Center institutional review board.
Baseline clinical characteristics
The median patient age was 62 years, and 68% were men, with 61% having undergone prior revascularization. There were histories of diabetes in 32% and of MI in 39%. The median LVEF was 58%. The median time between MPS scans was 12 months. The distribution of clinical characteristics in the 3 treatment groups is shown in Table 1.
Indications for performance of second scan
Symptoms of chest pain or shortness of breath or fatigue were the main clinical indications for performance of the second scan (Online Table 1). A few had arrhythmias (most commonly atrial fibrillation) as their indication for scanning. The remainder had a variety of miscellaneous indications, including pre-operative testing, abnormal electrocardiographic findings, and dizziness.
Pre-treatment and post-treatment ischemia
There was significantly greater baseline ischemia in the CABG and PCI groups compared with the MT group (p < 0.001) (Table 1). However, there was no significant difference in ischemia among the 3 groups after treatment (p = 0.634).
There was an overall significant reduction in ischemia in all 3 groups (Online Fig. 1). Reduction in ischemia was significantly greater in the CABG and PCI groups compared with the MT group (Online Fig. 2). In addition, significantly greater ischemia reduction occurred in the CABG group compared with the PCI group (p = 0.0218). Change in ischemia in the 3 treatment groups was not significantly different when comparing subjects in the study from before the year 2000 with those afterward.
Patients with improvement in ischemia
In the MT group, 175 patients (20.1%) had improvements in ischemia of ≥5%, compared with 219 (52.3%) in the PCI group and 84 (62.2%) in the CABG group. The proportion of patients with ≥5% ischemia improvement was significantly greater in the CABG and PCI groups compared with the MT group (p < 0.0001) (Online Fig. 3). Significantly more of the CABG population had ≥5% improvement in ischemia compared with the PCI group (p = 0.0470). The proportion of patients with ≥5% ischemia improvement was not significantly different when comparing subjects in the study from before the year 2000 with those afterward.
Patients with worsening ischemia
In the MT group, 136 patients (15.6%) had worsening ischemia of ≥5%, compared with 26 (6.2%) in the PCI group and 9 (6.7%) in the CABG group. The proportion of patients with ≥5% ischemia worsening was significantly greater in the MT group compared with the PCI and CABG groups (p < 0.0053) (Fig. 2). Similar proportions of patients in the PCI and CABG groups had ≥5% ischemia worsening (p = 0.8400). The proportion of patients with ≥5% ischemia worsening was not significantly different when comparing subjects in the study from before the year 2000 with those afterward.
Change in ischemia and clinical outcomes
Patients were followed for a median of 5.8 years. There was a significant difference in Kaplan-Meier event rates (p < 0.0001) between patients with ≥5% ischemia worsening and those with <5% ischemia worsening (Fig. 3). After 5 years, the composite endpoint of death or MI occurred in 43% of patients with ≥5% ischemia worsening compared with 26% of those with <5% ischemia worsening. There was also a significant difference in Kaplan-Meier event rates (p = 0.0199) between patients with ≥5% ischemia improvement and those with <5% ischemia improvement (Fig. 4). Of the 871 patients initially assigned to the MT group (on the basis of treatment within 60 days of the first MPS scan), 345 (39.6%) subsequently underwent revascularization during follow-up.
After adjustment for clinical and established risk predictors, as well as treatment arm, the association of ≥5% ischemia worsening with increased risk for death or MI remained significant (hazard ratio: 1.634; p = 0.0019) (Table 2). Age, diabetes, noncardiac Charlson index, history of MI, hypertension, white race, smoking, New York Heart Association functional class, LVEF, and number of diseased vessels also remained statistically significant. Residual ischemia on the final scan was not a significant independent predictor when ischemia worsening was included in this multivariate model. In addition, there was no significant interaction between baseline ischemia (at scan 1) and ≥5% ischemia worsening in our final model (p = 0.56). The model remained unchanged when the clinical indications for scan 2 were included as additional variables. In particular, the clinical indications for scan 2 were not significant predictors of events.
When comparing this full model with a model with ≥5% ischemia worsening removed, the full model was a better fit as assessed by the likelihood ratio test (p = 0.002). Assessment of model discrimination also showed significant improvement after the addition of ≥5% ischemia worsening, with an integrated discrimination improvement of 0.0062 (p = 0.0057), despite a very small increase in the C-statistic (from 0.7109 to 0.7149). Assessment of model calibration showed that the survival-adapted Hosmer-Lemeshow chi-square statistic was 2.85 (p = 0.42) for both models, suggesting that neither model had a significant lack of fit.
Although ≥5% ischemia improvement had an association with death or MI (hazard ratio: 0.825; p = 0.0444) in the unadjusted setting, this was no longer significant after adjustment to clinical variables (age, diabetes, history of MI, hypertension, race, smoking, New York Heart Association functional class, and noncardiac Charlson index) (hazard ratio: 0.896; p = 0.2584). In addition, there was no significant interaction between baseline ischemia (at scan 1) and ≥5% ischemia improvement in our final model (p = 0.21), indicating that the effect of ≥5% ischemia improvement on outcome was not significantly different in those with greater baseline ischemia.
Table 3 shows the cross-tabulations of the 10-year estimated risk using the models with and without ≥5% ischemia worsening. The addition of ≥5% ischemia worsening to the model resulted in reclassification of 9.3% of the sample. Of the patients with events, 5.2% correctly moved up in risk category and 2.3% incorrectly moved down in risk category when adding ≥5% ischemia worsening to the model, resulting in a relative improvement in the event group of 2.9%. For the subjects with no events, 6.0% correctly moved down in risk category and 4.3% incorrectly moved up in risk category when adding ≥5% ischemia worsening to the model, resulting in a relative improvement in the nonevent group of 1.7%. Thus, the NRI for the entire study cohort was 4.6% (p = 0.0056).
Overall, 129 patients in the cohort were reclassified: 64 to a higher risk category, with a 10-year event rate of 39.1%, and 65 to a lower risk category, with a 10-year event rate of 16.9%.
To our knowledge, this is the largest reported series assessing changes in ischemia on serial MPS. We found that ≥5% worsening ischemia was a strong, independent predictor of death or MI after adjustment for established predictors and irrespective of treatment arm. Finally, the addition of ≥5% worsening ischemia to traditional predictors resulted in significant improvement in the classification of risk for prediction of death or MI.
Comparison with the nuclear substudy of COURAGE
The nuclear substudy of COURAGE examined 314 patients within the overall trial who underwent second MPS scans (3). Similar to our findings, they showed that mean ischemia reduction with PCI (2.7%) was greater than with MT alone (0.5%). They also found that a greater proportion of patients undergoing PCI had significant ischemia reduction. The proportion of MT patients achieving ≥5% ischemia reduction in COURAGE (19%) was very similar to that in our study (20%), although ours was an observational experience. The overall crossover rate from MT to revascularization was 39.6% in our study, similar to that in the COURAGE (33%) and BARI-2D (Bypass Angioplasty Revascularization Investigation 2 Diabetes) (42%) trials (1,2).
In both this study and the COURAGE substudy, ≥5% ischemia improvement was associated with decreased risk for death or MI in unadjusted analysis. In COURAGE, this association was not maintained after adjustment to treatment arm, and multivariate adjustment to other known predictors was not possible because of insufficient numbers. Similarly, in our study, ≥5% ischemia improvement was no longer a significant predictor after adjustment to clinical variables. In contrast, we have shown that ≥5% ischemia worsening is a strong independent predictor, resulting in significant risk reclassification improvement when added to known predictors. In the COURAGE substudy, residual ischemia was an unadjusted predictor of events, but again, this association was not significant when adjusted for treatment arm, and they were unable to perform multivariate adjustment to other known predictors because of insufficient numbers. Consistent with this, we found that the association of residual ischemia to events was not significant when adjusted in a multivariate model that included known predictors and amount of ischemia worsening. These results highlight the importance of ischemia worsening as a novel independent predictor of events, perhaps by identifying patients with more rapidly progressive atherosclerotic disease.
Other studies assessing ischemia change during treatment of stable CAD
Several other studies have assessed ischemia change during treatment of stable CAD, but none examined its independent prognostic significance. In the INSPIRE (Adenosine Sestamibi SPECT Post-Infarction Evaluation) trial, 205 stable post-MI patients with ≥10% baseline ischemia were randomized to MT or revascularization (13,14). Ischemia reduction was similar with MT and revascularization (15% vs. 16%) and occurred in a similar percent of patients (80% vs. 81%). The prognostic significance of ischemia reduction was examined in a subgroup of 44 patients in this study and was a univariate predictor of major adverse cardiac events (death, MI, and unstable angina). However, the number of patients was insufficient to adjust for other predictors (15).
Berman et al. (16) reported the change in ischemia with MT and revascularization in 421 patients using serial MPS scans. In those with moderate baseline ischemia, they noted similar reductions in ischemia with MT and revascularization (1.8% vs. 2.7%). However, in those with severe baseline ischemia, there was significantly greater reduction in ischemia with revascularization than with MT (13.0% vs. 0.9%). In patients with small baseline defects, there were no significant changes in ischemia seen with either treatment.
In the ACME (Angioplasty Compared to Medicine) trial, 270 patients with documented ≤2-vessel disease underwent serial MPS scans before and after treatment with MT or percutaneous transluminal coronary angioplasty (17). Although presence of residual ischemia was found to be a predictor of survival, neither change in ischemia nor its prognostic significance was reported.
Implications for imaging in risk assessment of stable CAD
In current clinical practice, risk assessment of patients with stable CAD is determined on the basis of a combination of clinical factors, LVEF, number of diseased vessels (if known), and amount of ischemia on objective stress testing. However, it has previously been shown that the amount of ischemia is actually a relatively weak predictor of death and a much better predictor of ischemic events such as nonfatal MI, unstable angina, and revascularization (18,19). Nevertheless, amount of ischemia is 1 of the primary measures driving decisions regarding revascularization. This is based primarily on observational data from 10,000 patients referred for MPS studies, suggesting that the benefits of revascularization were confined to patients with >10% ischemia (20,21).
We have shown that worsening ischemia is an independent predictor of death and MI regardless of treatment and other clinical factors, including LVEF, number of diseased vessels, and residual ischemia. There was no significant interaction between baseline ischemia (at scan 1) and worsening ischemia, indicating that the effect of worsening ischemia on outcome was not significantly different in those with greater baseline ischemia. In addition, the absolute amount of residual ischemia was not a significant predictor when adjusted to the effects of ischemia worsening and other clinical factors. This is not surprising, given that the majority of events in our population were deaths rather than MIs.
We have shown that there is a significant amount of risk reclassification when ≥5% ischemia worsening is added to risk assessment of patients with stable CAD. When looking at patients who were reclassified to higher risk, 39% had events (Table 3). However, if they were reclassified to lower risk, only 17% had events. This is a 22% relative risk change. This is a modest change, but that is to be expected for any individual risk marker, especially in the presence of all the other established prognostic variables. To put this into perspective, the magnitude of the NRI for ≥5% ischemia worsening was 4.6% in this study, which compares favorably with those of other well-established risk predictors, such as cholesterol (3.2%), high-density lipoprotein (4.0%), and parental history of early MI (3.2%) in the Women's Health Study (22). However, whether ischemia worsening predicts which patients may accrue a survival benefit with revascularization versus MT is unclear from this study. This is important to know before asserting that ischemia worsening should drive decisions regarding revascularization.
We do not believe that our study can be used as justification for performing serial MPS scans. Randomized prospective trials are required before any such recommendations can be proposed. However, our study provides preliminary data suggesting that if a patient has had 2 consecutive MPS studies performed for appropriate clinical reasons, the information regarding ischemia change may be used to improve prognostication of these individuals. It is possible that better risk stratification with serial imaging may lead to more optimal decisions regarding need for revascularization and subsequent cost savings in some cases, but this remains unproven.
Because this was an observational study, there were significant baseline differences between the treatment groups. Thus, it is difficult to make definitive conclusions regarding the relative effectiveness of these treatments. Prospective, randomized studies with similar amounts of baseline ischemia in all 3 groups are required to address this issue. However, the differing baseline variables in our study do not affect our main findings regarding the predictive power of ≥5% ischemia worsening, because this was independent of baseline ischemia, as well as other clinical variables in our multivariate models.
Patients' treatment was at the discretion of their cardiologists, and therefore, there may have been management differences among patients. Our study encompassed a span of about 16 years, during which there were changes in MT and treatment goals as well as revascularization technologies, particularly PCI. However, neither change in ischemia nor the proportion of patients with worsening or improving ischemia was significantly different when comparing subjects in the study from before the year 2000 with those afterward.
Clearly, there was some selection bias in the patients included in this study, because they were all seen at a single academic medical center and all underwent cardiac catheterization. In addition, patients did not receive second MPS scans routinely but rather at the discretion of their cardiologists, thus selecting out a subpopulation that underwent 2 serial MPS scans. However, to some extent, the patients in this study are more representative of the population seen in daily clinical practice. They include those with the full range of CAD severity, as well as low ejection fractions, prior revascularizations, and heart failure, which are often excluded from clinical trials.
In this observational study of patients with stable CAD undergoing serial MPS scans, ≥5% LV ischemia worsening was a significant independent predictor of death and MI. The addition of ≥5% LV worsening ischemia to traditional predictors resulted in significant improvement in the classification of risk.
Our study provides preliminary data suggesting that if patients have had 2 consecutive MPS studies performed for appropriate clinical reasons, the information regarding ischemia change may be used to improve prognostication in these patients.
For supplementary figures and tables and their legends, please see the online version of this article.
Dr. Phillips has served on the Speakers' Bureau for Gilead and Daiichi Sankyo; has been a consultant to Frederick Medical; and his wife holds stock in Abbott. Dr. O'Connor has consulted for and received research funding from GE Healthcare and Medtronic. Dr. Leslee J. Shaw has received grant support from Astellas Healthcare and Bracco Diagnostics.
All other authors have reported that they have no relationships relevant to the contents of this paper to disclose. H. William Strauss, MD, served as Guest Editor for this paper.
- Abbreviations and Acronyms
- coronary artery bypass grafting
- coronary artery disease
- left ventricular
- left ventricular ejection fraction
- myocardial infarction
- myocardial perfusion single-photon emission computed tomography
- medical therapy
- net reclassification improvement
- percutaneous coronary intervention
- Received November 21, 2011.
- Revision received January 9, 2012.
- Accepted January 26, 2012.
- American College of Cardiology Foundation
- Shaw L.J.,
- Berman D.S.,
- Maron D.J.,
- et al.
- Farzaneh-Far A.,
- Borges-Neto S.
- Shaw L.J.,
- Narula J.
- Stone C.J.,
- Koo C.Y.
- Mahmarian J.J.,
- Dakik H.A.,
- Filipchuk N.G.,
- et al.
- Mahmarian J.J.,
- Shaw L.J.,
- Filipchuk N.G.,
- et al.
- Dakik H.A.,
- Kleiman N.S.,
- Farmer J.A.,
- et al.
- Parisi A.F.,
- Hartigan P.M.,
- Folland E.D.
- Hachamovitch R.
- Hachamovitch R.,
- Hayes S.W.,
- Friedman J.D.,
- Cohen I.,
- Berman D.S.