Author + information
- Received September 18, 2007
- Revision received November 29, 2007
- Accepted January 6, 2008
- Published online May 13, 2008.
- Pallav Garg, MBBS, MSc⁎,§,
- David J. Cohen, MD, MSc‡,⁎ (, )
- Thomas Gaziano, MD, MSc† and
- Laura Mauri, MD, MSc⁎,§
- ↵⁎Reprint requests and correspondence:
Dr. David J. Cohen, Saint Luke's Mid America Heart Institute, 4401 Wornall Road, Kansas City, Missouri 64111.
Objectives We sought to define what incremental risk of very late stent thrombosis (VLST) in drug-eluting stents (DES) would outweigh the restenosis benefit.
Background Although there are robust data on the restenosis benefit of DES versus bare-metal stents (BMS), the incremental risk of stent thrombosis, a rare but serious complication of percutaneous coronary intervention (PCI), is not known with certainty.
Methods We developed a decision analytic Markov model comparing DES versus BMS strategies for a contemporary PCI population. Procedure-related morbidity and mortality data from published reports were used to derive the model probabilities. Over a range of incremental risk and duration of risk of VLST, we identified the net benefit of DES versus BMS in terms of quality-adjusted life expectancy (QALE).
Results Under an assumption of equal stent thrombosis rates beyond 1 year, the DES strategy was superior to BMS in terms of QALE (16.262 vs. 16.248 quality-adjusted life years [QALYs], difference = 0.014). Under the alternative assumption of an incremental risk difference of 0.13%/year, the net benefit was substantially reduced (difference = 0.001 QALYs). The threshold excess risk of very late DES thrombosis compared with BMS, above which BMS would be the preferred strategy, was 0.14%/year (over 4 years of follow-up). This threshold increased as the population risk of restenosis increased and decreased as the vulnerable time window lengthened.
Conclusions A small absolute increase in DES thrombosis compared with BMS after 1 year (>0.14%/year) would result in BMS being the preferred strategy for the overall PCI population. Larger clinical trials with longer follow-up are needed to estimate the risk of late stent thrombosis with greater certainty for existing and new DES.
Treatment with percutaneous coronary intervention (PCI) aims to provide sustained relief of coronary ischemia and angina. As an invasive strategy, it is associated with risks during and after the procedure. Periprocedural risks have decreased over time despite treatment of patients with more complex lesions (1). Within the first year of treatment, recurrent angina related to progressive renarrowing of the treated arterial segment is predictable according to stent and patient characteristics and might be prevented by elution of medications from stents that suppress neointimal hyperplasia. Thrombotic occlusion or stent thrombosis, however, is a rare but potentially fatal event that is less well characterized in terms of absolute and relative risks over time.
Despite the dramatic early efficacy of drug-eluting stents (DES) in reducing restenosis compared with bare-metal stents (BMS) (2,3), there is concern that DES might lead to higher rates of stent thrombosis—particularly beyond the first year after implantation (4). Although restenosis might be associated with unstable angina and myocardial infarction (MI), its more common clinical manifestation is pro-gressive angina (5). In contrast, stent thrombosis is usually associated with ST-segment elevation MI and high mortality (6–9). Therefore, the consequences of restenosis and thrombosis are difficult to compare. Furthermore, the incremental risk of late stent thrombosis in DES versus BMS is not known with certainty; randomized trials to date have been limited in their power to detect rare and late events (10), and observational studies have been limited by selection and ascertainment bias (11). Even large meta-analyses of broader randomized trial data have yet to resolve this uncertainty and suggest that DES might be associated with a 2- to 4-fold increased risk of very late thrombotic events (12).
Decision analysis is a tool for combining data from multiple sources that can be useful in guiding complex medical decisions, particularly when there is uncertainty regarding 1 or more key parameters. The precise differences in the risk of stent thrombosis between DES and BMS are not known but have important implications for patient care. Understanding the trade-off that might exist between restenosis prevention and avoidance of stent thrombosis would provide the context for the extended follow-up of existing DES as well as guidance regarding reasonable margins of safety for the development of new stents. Therefore, we sought to define what incremental risk of stent thrombosis in DES would outweigh the benefits of restenosis prevention.
We developed a decision analytic Markov model comparing 2 common strategies in PCI: stenting with DES versus stenting with BMS. A Markov model simulates transitions between distinct health states that would occur over a lifetime in a cohort of patients undergoing a selected treatment strategy (13). Procedure-related morbidity and mortality data derived from published reports were used to derive the model probabilities, and the outcome from each strategy was quantified in terms of quality-adjusted life years (QALYs). On the basis of this model, we sought to identify the threshold risk of very late stent thrombosis (VLST) above which the strategy of DES use was no longer superior.
Our model was designed to be applied to a typical U.S. patient requiring and undergoing PCI amenable to either BMS or DES. Where possible, the characteristics of the patient population were chosen to match those of contemporary population-based PCI registries.
Decision model structure
A decision analytical Markov model was constructed covering the possible outcomes for a patient over a lifetime after the index procedure. The first year of the model was divided into 3 distinct periods: 0 to 30 days after the index procedure, 30 days to 6 months, and 6 months to 1 year. After the first year, subsequent health states in the Markov model were based on a cycle length of 1 year. These periods were chosen to correspond to the known biology of restenosis and thrombosis after stent implantation. Stent thrombosis was considered “early” during the first 30 days after implantation, “late” between 1 month and 1 year, and “very late” beyond 1 year, for consistency with recently developed terminology by the Academic Research Consortium (ARC) (14).
Figure 1A depicts the initial treatment strategy (DES vs. BMS) and the immediate and 30-day outcomes of the chosen procedure in the form of a decision tree. During the initial 30-day follow-up period, patients could experience noncardiac death or cardiac death unrelated to stent thrombosis. In the absence of these events, patients were at risk for stent thrombosis. All patients with stent thrombosis were assumed to either die or suffer a nonfatal MI (6). Survivors of stent thrombosis were assumed to undergo target vessel revascularization (TVR) with a variety of possible strategies, including conventional balloon angioplasty (percutaneous transluminal coronary angioplasty [PTCA]), stenting with DES or BMS, or coronary artery bypass graft surgery (CABG). Patients were not considered at risk for restenosis within the first month, consistent with known biology.
During the ensuing 11 months, survivors of the first 30 days after the index procedure were at risk of clinically significant restenosis and stent thrombosis in addition to background risks of noncardiac and cardiac mortality. The basic structures of the sub-trees for the periods from 1 to 6 months and from 6 to 12 months were identical (Fig. 1B), but with differing probability estimates. Patients who experienced late stent thrombosis could either die or suffer a nonfatal MI. For simplicity, all patients with clinical restenosis were assumed to undergo TVR and were at risk for nonfatal MI (both at the time of presentation and as complication of the procedure) as well as procedure-related mortality. The options for TVR for clinical restenosis included PTCA and repeat stenting with either DES or BMS in both arms of the model with subsequent risk of procedure-related mortality and CABG. Because brachytherapy is no longer generally available in the U.S., it was not included as a treatment strategy in our model. Nonfatal MI in the absence of restenosis and stent thrombosis was also modeled. Patients who survived months 1 to 6 (with or without restenosis) were at risk for the same events during months 6 to 12 (with different probabilities). We assumed that patients could experience up to a maximum of 3 episodes of stent thrombosis (1/distinct model period) and 2 episodes of clinical restenosis within the first year after the index procedure.
The Markov model that describes potential health states and transitions beyond the first year after PCI is depicted in Figure 1C. The specific health states that we considered included: 1) survivors without TVR; 2) survivors after TVR; 3) post-CABG survivors; 4) survivors beyond the interval of risk for VLST; and 5) death. Each of the first 4 health states was further stratified according to the presence or absence of prior MI (not shown in the figure). Because most studies indicate that the absolute difference in TVR between DES and BMS observed at 1 year is maintained over the next 3 to 4 years (15), we assumed that clinically significant restenosis and TVR would occur only during the first year after the index procedure. During each cycle, patients in the first 2 health states were at risk for VLST. Patients in each of the health states were at risk for additional cardiac events, including cardiac death (unrelated to stent thrombosis) as well as noncardiac death.
Data sources for decision model
Probabilities of procedural success and complications after stenting were derived from a review of the published medical reports as well as unpublished data presented at scientific meetings. Details of these data sources are provided in the Online Appendix and are summarized in Tables 1 to 3⇓⇓ (16–38). To accurately reflect the contemporary real-world clinical experience with coronary stenting, whenever possible, absolute event rates were derived from registry data that describe the “real-world” outcomes of nonemergent PCI in the U.S. population, whereas randomized controlled trial data were used preferentially to obtain unbiased estimates of the relative risks of specific outcomes between the DES and BMS strategies and also as a source of absolute event rates where registry data were lacking. Estimates of risk for the 2 approved DES (Cypher, Cordis Corp., Miami Lakes, Florida, and TAXUS, Boston Scientific, Natick, Massachusetts) and all BMS were pooled. Pooled estimates across publications were calculated with inverse variance weighting.
Stent thrombosis estimates
We assumed the identical probability of early (<30 days) and late stent thrombosis (30 days to 1 year) for the 2 treatments, because there is no evidence of difference in stent thrombosis rates at 30 days or at 1 year in the randomized trials (10,15), and examined a range of risks of VLST (>1 year) from the published reports. For our primary analysis, we used values derived from a pooled analysis of the long-term (4 year) results of the pivotal randomized controlled trials comparing DES with BMS with the ARC definitions (10). These studies did not demonstrate a statistically significant excess of definite or probable VLST in DES over BMS, but the absolute difference over 2 to 4 years was 0.39% (DES 0.90% vs. BMS 0.51%). Therefore, we evaluated 2 possible baseline sets of assumptions: 1) that the incidence of VLST was identical for the 2 stent types (0.22% annually; Scenario A); or 2) that the incremental risk of VLST for DES versus BMS was 0.13%/year for DES (i.e., absolute risk of 0.30%/year for DES vs. 0.17% for BMS; Scenario B). In either scenario, the risk of VLST was assumed to persist through 4 years after initial treatment. These assumptions were varied extensively in sensitivity analyses. Further details and explanation of the data sources are provided in the Online Appendix.
The outcomes of each treatment strategy were quantified in terms of quality-adjusted life years (QALYs) over a patient's lifetime, as noted previously. In this context, 1 year of life without angina, revascularization, or hospital stay was assumed to be a year of perfect health and was assigned a value of 1.0 QALY. Data on utilities and QALYs for patients with coronary artery disease undergoing revascularization were obtained from a previous study (Stent-PAMI [Stent-Primary Angioplasty in Myocardial Infarction]), in which the Euro-QOL health status instrument was used to assess utilities for 771 PCI patients during the year after initial treatment (35,36,39). In that study, the mean quality-adjusted life expectancies (QALEs) for patients with and without repeat revascularization during the first year of follow-up were 0.79 and 0.85, respectively (p < 0.001). The difference between these values was assumed to represent the mean disutility associated with revascularization, which was applied only during the year in which the revascularization actually occurred. Studies have previously demonstrated that, by 12 months, there is virtually no quality-of-life difference between patients with and without repeat revascularization (35).
We assumed that MI (regardless of the underlying cause) would decrease both long-term survival and quality of life beyond the first year of follow-up but that revascularization would not, because previous studies have not shown a consistent association between coronary restenosis and long-term mortality (40). We also adjusted quality-of-life measurements to account for the short-term morbidity of nonfatal MI and CABG by basing these adjustments on the estimated duration of hospital stay and recuperation from such an event (37,38). Details of each of these assumptions and their data sources are provided in the Online Appendix.
Analytic method and sensitivity analysis
For each of the 2 strategies we calculated QALE and considered the strategy associated with a higher value to be preferred. Because our model was based on a number of assumptions, we performed 1-way and multi-way deterministic sensitivity analyses to examine whether and how plausible variations in these assumptions and risks (in particular, the probability of VLST) would alter our findings. We also performed probabilistic sensitivity analysis in which we allowed each of the key variables of the model to vary simultaneously according to its underlying distribution; in general, these distributions were based on the log-normal or beta distribution and were derived by iterative fitting on the basis of published data (where available) or plausible ranges where not published (see Online Appendix). The Markov model was designed and all analyses were performed with TreeAge Pro Suite 2007 software package (TreeAge Software, Inc., Williamstown, Massachusetts).
QALE of DES and BMS strategies
The model-predicted QALE for a typical 62-year-old patient undergoing PCI was 16.262 years after DES treatment and 16.248 years after BMS treatment under the assumption of no difference in VLST risk (Scenario A). Thus, for a prototypical PCI patient, stenting with DES would be preferred over BMS with a small net gain in QALE. However, the QALE for the 2 treatments were virtually identical under the assumption of 0.13% difference in VLST risk (Scenario B; DES 16.254 vs. BMS 16.253 QALYs). The predicted numbers of events/10,000 patients according to treatment strategy are shown in Table 4. Under the assumption of 0.13%/year difference in VLST rates between DES and BMS (Scenario B), there were 37 additional episodes of VLST resulting in 7 deaths in DES compared with 890 additional episodes of TVR in BMS. At 4 years, there were 6 additional deaths in DES, largely due to stent thrombosis.
VLST risk threshold for drug-eluting stenting strategy
Figure 2 demonstrates the effect of varying risks of VLST on the gain in QALE with DES versus BMS. As the risk of VLST increased, the gain in QALE with DES implantation decreased. When the absolute excess risk of VLST in DES exceeded 0.14%/year (over 4 years of follow-up), BMS implantation became the preferred treatment strategy. Given that the duration of the risk of stent thrombosis is also currently uncertain, we explored this threshold risk as a function of time (Fig. 3). The annual excess risk for VLST in DES that would be acceptable decreased as the duration of the risk period was increased, as expected. For example, if the excess risk of VLST were to persist for 5 years after PCI, the maximum excess risk above which BMS would be the preferred strategy was 0.11%/year; if the risk were to persist for 10 years, the threshold would fall to 0.05%/year.
Figure 4 shows the impact of other key model parameters on the estimated threshold of excess late stent thrombosis risk above which DES would no longer be preferred. The most influential parameters were the stent thrombosis case fatality rate, the magnitude of differences in clinical restenosis and VLST rates, and the net disutility associated with clinical restenosis. The threshold risk of late stent thrombosis was insensitive to plausible variations in most other model parameters.
In 2-way sensitivity analyses, we explored the effect of varying the baseline BMS target vessel revascularization rate on the threshold excess risk of DES thrombosis. As the baseline risk of TVR increased, the annual probability of VLST above which DES would no longer be the preferred revascularization strategy increased (Fig. 5). For example, if the BMS TVR rate was 20% (as compared with the average population rate of 14%), then the maximum tolerable risk of VLST in DES increased to 0.20%/year over 4 years of follow-up.
Because DES might vary in the degree of suppression of neointimal hyperplasia and restenosis, we performed a sensitivity analysis to explore the effect of varying the relative risk reduction in TVR rate with DES as compared with BMS. Not surprisingly, we found that as the relative efficacy of a DES decreased, the threshold incremental VLST risk decreased (Fig. 6). For example, if the relative risk reduction in TVR with DES decreased to 50% (compared with our baseline estimate of 65%), the maximum tolerable rate of excess VLST decreased to 0.11%/year over 4 years of follow-up.
Because individual patients might differ in their willingness to tolerate recurrent anginal symptoms as well as the inconvenience and discomfort associated with additional revascularization procedures, we performed a 2-way sensitivity analysis to explore the impact of variations in the disutility associated with restenosis on the optimal treatment selection (Fig. 7). As the disutility associated with restenosis decreased, the maximum tolerable DES thrombosis risk decreased. However, even under the extreme situation of a patient who assigned no value to avoidance of restenosis (i.e., disutility = 0), a small excess risk of very late thrombosis with DES would be acceptable (i.e., <0.08%/year), owing to the non-negligible mortality risk associated with restenosis and its subsequent treatment.
Finally, we performed probabilistic sensitivity analyses by simultaneously varying all of the parameters listed in Figure 4. Under Scenario A (equal VLST risk for DES and BMS), we found that the DES strategy was the preferred strategy over BMS in 91.1% of trial iterations with a mean life-expectancy of 16.30 QALYs in DES (95% confidence interval 14.74 to 17.72) and 16.29 QALY in BMS (95% confidence interval 14.73 to 17.71). Under Scenario B (0.13%/year excess VLST with DES), however, the DES strategy was preferred in only 56% of iterations.
Decisions regarding percutaneous treatment of obstructive coronary disease have become increasingly challenging for patients and physicians since the observation of delayed stent thrombosis in DES. The main risk attributable to bare-metal stenting was restenosis requiring repeat revascularization—a risk that largely ended within 1 year after stenting (41,42). Beyond this period, events attributable to the stent were rare (21). In particular, in-stent thrombotic complications occurred in <1% of patients, almost exclusively within the first month after BMS implantation. By limiting neointimal hyperplasia within the stent, the current generation of DES have reduced the occurrence of clinical restenosis by 50% to 70% (2,3). However, there is concern that DES might be associated with increased risks of delayed stent thrombosis. In this study, we sought to quantify the degree to which current uncertainty in the rate of very late thrombotic complications after DES implantation would affect the choice of an optimal PCI strategy. Because both the absolute risk and duration of risk of stent thrombosis after DES implantation are uncertain, we used the techniques of decision analysis to define what threshold of incremental risk with DES would outweigh the benefits of reduced restenosis in clinical practice.
We found, on the basis of the best data currently available, that the DES strategy was preferred for a prototypical PCI patient under the assumption of no difference in the rates of VLST (Scenario A). Although the benefit was small in absolute terms (0.014 QALYs), this gain is plausible given the time-limited nature of the restenosis process and the absence of long-term mortality benefit associated with restenosis avoidance in most studies. This finding was confirmed to be robust with probabilistic sensitivity analysis, which allows one to consider a range of plausible probabilities rather than relying on fixed estimates alone. Nonetheless, we found that our results were highly sensitive to the absolute risk and duration of risk for VLST, leading to uncertainty in the optimal decision over a range of risk that is plausible on the basis of current data. In particular, for a prototypical patient, we found that even a small excess risk of VLST (>0.14%/year over 4 years) would be sufficient to negate any advantage of DES over BMS in terms of QALE. Furthermore, if the at-risk period extended beyond 4 years, the incremental annual risk that could be tolerated was even smaller.
Whether the true risk of very late thrombosis with existing DES exceeds this risk is uncertain at present. Restenosis risks can be predicted with reasonable precision in overall populations and according to well-understood patient and lesion-based factors, because restenosis was a frequent occurrence over the past decade of practice. In contrast, stent thrombosis risks have only recently been studied with similar rigor. Although pooled analyses of randomized trials of the 2 approved DES platforms to 4 years of follow-up have not shown significant differences in risk of thrombosis between drug-eluting and bare-metal stents, the confidence intervals of these estimates are wide, suggesting that up to a 1.4% absolute risk difference at 4 years cannot be excluded (10). A larger network meta-analysis suggested that up to 4-fold increase in late hazard for DES cannot be excluded (12). Currently, the best data regarding VLST rates are from the pooled analysis of the pivotal randomized trials of DES versus BMS (10,12,15). In these studies, the individual point estimates range of excess risk range from 0.13% to 0.18%/year, depending on the study and definition used (Fig. 8). Thus, the available data do not provide a definitive answer regarding the preferred stent strategy at the present time.
One should consider our results in the context of our study design. Ideally, one would base the choice between DES and BMS on large, long-term, randomized trials conducted in a population that directly reflects the general population. As in any decision analysis, we were limited by the available data. To derive results applicable to the “real world” we chose to populate our decision analytic model with absolute risks derived from observational studies and relative risks derived from randomized trials. This approach allowed us to derive unbiased estimates of relative risk while applying these to the absolute risks seen in the general population undergoing PCI rather than the more selected patients who would be eligible for randomized trials. Furthermore, we addressed the inherent uncertainty in the reported data by conducting sensitivity analyses (including probabilistic sensitivity analysis) to determine the robustness of our results.
In addition, we assumed a constant risk of stent thrombosis beyond the first year (11). If the stent thrombosis risk was a declining function rather than constant, then the VLST risk threshold would be somewhat higher in the initial years. We did not explicitly model the effects of clopidogrel in this study, because of the conflicting data on the efficacy of clopidogrel in decreasing the risk of VLST. It is likely that the impact of extended dual antiplatelet therapy, as seen in large trials of patients with acute coronary syndromes, is predominantly mediated by prevention of MI outside the stent territory, because the absolute risk of progression of other atherosclerotic disease after treatment is greater than the risk of stent thrombosis. Lastly, we specifically chose not to include cost in our model, because the primary question we intended to answer was a purely clinical one: what excess risk of late stent thrombosis would be acceptable for a typical patient undergoing PCI with stents to suppress restenosis?
Implications for clinical trials and new DES evaluation
Our observations suggest that it is clinically relevant to determine relatively small differences in risk of stent thrombosis. To date, the available randomized trials, even when pooled, have been underpowered to detect such differences. For example, for a randomized noninferiority comparison to detect an absolute difference in VLST rates of 0.5% or more, a sample size of >10,000 patients would be required. Furthermore, the duration of risk is an important determinant of net clinical benefit. Our analysis provides the context of what upper limit of thrombosis rate can be tolerated once follow-up data to 5 years and beyond become available.
Clearly, the ideal stent would avoid both restenosis and stent thrombosis. Our model suggests that, despite a small increase in the risk of stent thrombosis, DES might still provide a net clinical benefit to patients if the expected risk of restenosis with BMS is nontrivial and the restenosis reduction is profound. However, new stent platforms that differ in mechanism, drug, or polymer from current designs might vary in the degree of suppression of neointimal hyperplasia and separately in the risk of stent thrombosis. Our analysis suggests that, as the relative risk reduction of restenosis decreases, the tolerable excess stent thrombosis risk must also diminish to preserve the benefit of a DES strategy. That is, a DES that is less effective at preventing restenosis must be required to be very similar to or better than BMS in terms of avoidance of thrombosis for a net advantage to exist.
On the basis of a decision analytic model incorporating the best data currently available, we found that even a small (<1%) incremental risk of thrombosis in DES might be sufficient to outweigh the benefit of restenosis prevention and favor BMS use for the overall PCI population. To identify whether risks of restenosis can be safely reduced, evaluation of existing and new DES must be adequately powered and have sufficient length of follow-up in order to determine both the relative and absolute risks of stent thrombosis with greater certainty.
The authors thank Michael Garshick and Manu Varma for their assistance in the preparation of tables and figures for the manuscript.
For a supplementary table and detailed Methods section, please see the online version of this article.
Balancing the Risks of Restenosis and Stent Thrombosis in Bare Metal Versus Drug-Eluting Stents: Results of a Decision Analytic Model
Dr. Cohen has received research grant support from Cordis and Boston Scientific.
- Abbreviations and Acronyms
- bare-metal stent(s)
- coronary artery bypass graft surgery
- drug-eluting stent(s)
- percutaneous coronary intervention
- percutaneous transluminal coronary angioplasty
- quality-adjusted life expectancy
- quality-adjusted life year
- target vessel revascularization
- very late stent thrombosis
- Received September 18, 2007.
- Revision received November 29, 2007.
- Accepted January 6, 2008.
- American College of Cardiology Foundation
- Singh M.,
- Rihal C.S.,
- Gersh B.J.,
- et al.
- Pfisterer M.,
- Brunner-La Rocca H.P.,
- Buser P.T.,
- et al.
- Cutlip D.E.,
- Baim D.S.,
- Ho K.K.,
- et al.
- Kuchulakanti P.K.,
- Chu W.W.,
- Torguson R.,
- et al.
- Ong A.T.,
- Hoye A.,
- Aoki J.,
- et al.
- Beck J.R.,
- Pauker S.G.
- Cutlip D.E.,
- Windecker S.,
- Mehran R.,
- et al.
- Lemos P.A.,
- Serruys P.W.,
- van Domburg R.T.,
- et al.
- Urban P.,
- Gershlick A.H.,
- Guagliumi G.,
- et al.
- Williams D.O.,
- Abbott J.D.,
- Kip K.E.
- Cutlip D.E.,
- Chhabra A.G.,
- Baim D.S.,
- et al.
- Ong A.T.,
- McFadden E.P.,
- Regar E.,
- de Jaegere P.P.,
- van Domburg R.T.,
- Serruys P.W.
- Stone G.W.,
- Ellis S.G.,
- Cox D.A.,
- et al.
- Saia F.,
- Piovaccari G.,
- Manari A.,
- et al.
- Hwang S-J, Jang Y, Yoon J, et al. Incidence, predictors, and outcome of stent thrombosis after successful drug-eluting stent implantation: a results of a multicenter study. Presented at: TCT 2006.
- Massachusetts Data Analysis Center
- Moreno R.,
- Fernandez C.,
- Hernandez R.,
- et al.
- Alfonso F.,
- Auge J.M.,
- Zueco J.,
- et al.
- Alfonso F.,
- Perez-Vizcayno M.J.,
- Hernandez R.,
- et al.
- Cole J.H.,
- Miller J.I. 3rd.,
- Sperling L.S.,
- Weintraub W.S.
- Vaccaro O.,
- Eberly L.E.,
- Neaton J.D.,
- Yang L.,
- Riccardi G.,
- Stamler J.
- Cohen D.J.,
- Taira D.A.,
- Berezin R.,
- et al.
- Bakhai A.,
- Stone G.W.,
- Mahoney E.,
- et al.
- Tsevat J.,
- Goldman L.,
- Soukup J.R.,
- et al.
- Cohen D.J.,
- Breall J.A.,
- Ho K.K.,
- et al.
- Weintraub W.S.,
- Ghazzal Z.M.,
- Douglas J.S. Jr..,
- et al.