Author + information
- Received August 2, 2012
- Revision received September 21, 2012
- Accepted October 9, 2012
- Published online February 5, 2013.
- Javed Butler, MD, MPH⁎,
- Haris Subacius, MA†,
- Muthiah Vaduganathan, MD, MPH‡,
- Gregg C. Fonarow, MD§,
- Andrew P. Ambrosy, MD∥,
- Marvin A. Konstam, MD¶,
- Aldo Maggioni, MD#,
- Robert J. Mentz, MD⁎⁎,
- Karl Swedberg, MD††,
- Faiez Zannad, MD‡‡,
- Mihai Gheorghiade, MD†,⁎ (, )
- EVEREST Investigators
- ↵⁎Reprint requests and correspondence:
Dr. Mihai Gheorghiade, Center of Cardiovascular Innovation, Northwestern University Feinberg School of Medicine, 645 North Michigan Avenue, Suite 1006, Chicago, Illinois 60611
Objectives The study investigated whether the number of participants enrolled per site in an acute heart failure trial is associated with participant characteristics and outcomes.
Background Whether and how site enrollment volume affects clinical trials is not known.
Methods A total of 4,133 participants enrolled among 359 sites were grouped on the basis of total enrollment into 1 to 10, 11 to 30, and >30 participants per site and were compared for outcomes (cardiovascular mortality or heart failure hospitalization).
Results Per-site enrollment ranged from 0 to 75 (median 6; 77 sites had no enrollment). Regional differences in enrollment were noted between North and South America, and Western and Eastern Europe (p < 0.001). Participants from sites with fewer enrollments were more likely to be older and male, have lower ejection fraction and blood pressure as well as worse comorbidity and laboratory profile, and were less likely to be on angiotensin-converting enzyme inhibitors or aldosterone antagonists. During a median follow-up of 9.9 months, 1,700 (41%) participants had an outcome event. Compared to event rate at sites with >30 participants (32%), those with 1 to 10 (51%, hazard ratio [HR]: 1.77, 95% confidence interval [CI]: 1.56 to 2.02) and 11 to 30 (42%, HR: 1.44, 95% CI: 1.28 to 1.62) participants per site groups had worse outcomes. This relationship was comparable across regions (p = 0.43). After adjustment for risk factors, participants enrolled at sites with fewer enrollees were at higher risk for adverse outcomes (HR: 1.26, 95% CI: 1.08 to 1.46 for 1 to 10; HR: 1.22, 95% CI: 1.07 to 1.38 for 11 to 30 vs. >30 participant sites). Higher proportion of participants from site with >30 participants completed the protocol (45.5% for <10, 61.7% for 11 to 30, and 68.4% for sites enrolling >30 participants; p < 0.001).
Conclusions Baseline characteristics, protocol completion, and outcomes differed significantly among higher versus lower enrolling sites. These data imply that the number of participant enrolled per site may influence trials beyond logistics.
Heart failure (HF) remains a major global health concern (1). The overall prevalence of HF and the number of hospitalizations for acute heart failure (AHF) are high, and outcomes for these patients remain poor (1–3). Although many therapies have been evaluated in the last decade, none of them reduced mortality or readmission rates among AHF patients (4,5). The reasons behind this are complex including issues related to the therapies studied and trial conduct (6). AHF patients constitute a heterogeneous group and patient characteristics affect outcomes. Moreover, important differences between continents, and regions within continents, in HF etiology, severity, and management exist, which affect patient outcomes as well (7–10).
Recently, attention has been focused on the difficulties of enrolling participants in clinical trials in the United States (11). In order to enroll the required number of participants, it is increasingly common for trials to have several hundred sites in operation. The performance and the number of participants enrolled by these sites vary widely. In a clinical trial, slower enrollment rate affects the duration of the trial, however the participants are expected to be clinically homogenous due to the pre-specified strict inclusion and exclusion criteria. Therefore the overall enrollment by sites may not affect outcomes of the trial besides costs and logistics. Conversely, if outcomes of participants from high versus low enrollment sites are different, this may have significant implications for trial design and conduct. To assess whether the site enrollment affects outcomes, we performed a post hoc analysis of the data from the EVEREST (Efficacy of Vasopressin Antagonism in Heart Failure: Outcome Study with Tolvaptan) trial.
The design of the EVEREST trial has been described previously (3,12,13). Briefly, EVEREST was a prospective, international, randomized, double blind, placebo-controlled program that examined the short- and long-term efficacy and safety of tolvaptan added to optimal medical therapy in participants hospitalized for worsening HF. The EVEREST program included 2 identical short-term clinical status trials during hospitalization (trials A and B) embedded within the long-term post-discharge outcome study that combined all participants. Adults ≥18 years of age with left ventricular ejection fraction ≤40% who were hospitalized primarily for worsening HF and with 2 or more signs or symptoms of fluid overload (i.e., dyspnea, pitting edema, or jugular venous distension) were randomized within 48 h of admission to oral tolvaptan (30 mg/day) or matching placebo in addition to conventional therapy. Exclusion criteria included cardiac surgery within 60 days of enrollment, cardiac mechanical support, biventricular pacemaker placement within 60 days, expected survival of <6 months, acute myocardial infarction at the time of hospitalization, hemodynamically significant uncorrected valvular disease, end-stage HF, dialysis, systolic blood pressure <90 mm Hg, serum creatinine >3.5 mg/dl, serum potassium >5.5 mEq/l, and hemoglobin <9 g/dl. Background therapy was at the discretion of the treating physician, but recommendations for guideline-based therapy were included in the protocol.
Clinical trial sites
In the EVEREST trial, overall 4,133 participants were randomized from 359 sites in 20 countries across North America, South America, and Europe between October 7, 2003, and February 3, 2006. Countries were grouped as follows: North America (United States and Canada), South America (Argentina and Brazil), Western Europe (Italy, Belgium, Norway, Netherlands, Germany, Spain, France, United Kingdom, Sweden, and Switzerland), and Eastern Europe (Poland, Romania, Czech Republic, Russia, Bulgaria, and Lithuania).
The participants were divided into the following groups on the basis of overall participants enrolled by an individual site into those sites that enrolled 1 to 10, 11 to 30, and >30 participants. This grouping was virtually identical to sample tertiles on the basis of total numbers of participants enrolled within these sites (1 to 11, 12 to 30, and ≥30) and was selected for ease of communication. Outcomes for patients within these 3 individual groups were homogeneous and further subdivision did not improve overall model fit to the data. Hence these larger groups were used for all further analysis and reporting of the data.
For this analysis the primary outcome was defined as a composite of cardiovascular death or HF hospitalization, 1 of the 2 coprimary endpoints in EVEREST. All-cause mortality, the other coprimary endpoint in EVEREST, was also compared among the 3 groups. Multiple other clinical outcomes were also assessed in secondary analysis including modes of death, other cardiovascular outcomes, and hospitalizations. An independent event committee adjudicated the mode of death and the cause of hospitalizations for all participants. An independent and blinded adjudication committee determined the cause of all hospitalizations and deaths during follow-up. Rehospitalization was defined as a nonelective hospital admission for medical therapy with a duration that extended over a change in calendar date. HF hospitalization was defined as hospitalization that included substantive worsening of HF symptoms and/or signs resulting in augmentation of oral medications or new administration of intravenous HF therapies including ultrafiltration. Mode of death was adjudicated as cardiovascular, noncardiovascular, or unknown. Cardiovascular deaths were further classified as sudden cardiac death, HF death, acute myocardial infarction, stroke, or other. Noncardiovascular death was defined as a death due to a specific noncardiovascular event, while unknown death was defined as a death for which no information surrounding the event was available.
Demographics; physical and laboratory findings; medical history; and medical, surgical, and device therapies were compared among the 3 groups using the analysis of variance for parametric, and Kruskal-Wallis test for nonparametric variables. The Pearson chi-square test was used for categorical variables. Continuous variables are reported as either mean ± SD or median (interquartile range [i.e., 25th percentile to 75th percentile]) in cases where the distribution was not normal. Categorical variables are reported as count and percentage. For analysis of QRS duration, this interval was not reported in 142 participants, and 1,029 participants were excluded due to the presence of a pacemaker and/or implantable cardioverter-defibrillator.
Unadjusted Kaplan-Meier estimates of the rates of cardiovascular death and HF hospitalization were calculated. Hazard ratios (HR) and corresponding 95% confidence intervals (CI) for the primary endpoint and all-cause mortality were calculated relative to sites that enrolled >30 participants over the entire follow-up period using Cox proportional hazards model with and without adjustment for other baseline covariates. Confounders were adjusted on the basis of clinical relevance and included region, study arm assignment, age, gender, ejection fraction, QRS duration, systolic blood pressure, heart rate, New York Heart Association functional class, revascularization history, atrial fibrillation or flutter on baseline electrocardiogram, diabetes, mellitus, chronic renal disease, implanted defibrillator, serum sodium and blood urea nitrogen concentrations, natriuretic peptide levels, and randomization use of ACE inhibitors or angiotensin receptor blockers, beta-blockers, aldosterone antagonists, and inotropes. Proportionality of hazards through time was assessed using empirical score process. The sponsor performed database management according to a pre-specified plan and the analysis was performed at the Northwestern University using SAS software, version 9.3 (SAS Institute Inc., Cary, North Carolina). The authors had full access to the data and take responsibility for its integrity, and had complete control and authority over manuscript preparation and the decision to publish.
Site and participant distribution
Overall, 436 sites participated in the EVEREST trial. Of these, 77 sites did not enroll any patients. The remaining 359 sites enrolled 4,133 participants. The enrollment per site ranged from 0 to 75 (median 6) participants. Of the sites that enrolled participants, 224 (62%) sites enrolled ≤10, 105 (29%) sites enrolled 11 to 30, and 30 (9%) sites enrolled >30 participants. These represented 1,052 (26%), 1,792 (43%), and 1,289 (31%) participants enrolled among sites with ≤10, 11 to 30, and >30 participants enrolled by sites.
Significant differences in site enrollment were observed among the 4 regions studied (Fig. 1). The highest proportion of sites that did not enroll any participants was observed in Western Europe (22 of 100, 22%) and North America (46 of 220, 20.9%), whereas fewer nonenrolling sites were from South America (5 of 40, 12.5%) and Eastern Europe (3 of 77, 3.9%). Of the 359 sites that did enroll, 173 (48%) sites were in North America and enrolled 1,251 participants (median, 7 participants per site); 35 (10%) sites were in South America and enrolled 699 participants (20 participants per site); 77 (21%) sites were in Western Europe and enrolled 564 participants (7 participants per site); and 74 (21%) sites were in Eastern Europe and enrolled 1,619 participants (22 participants per site). Significant regional differences were noted in the percent of sites within each enrollment number grouping (North America 75.1%, 23.1%, 1.8%; South America 25.7%, 54.3%, 20.0%; Western Europe 77.9%, 20.8%, 1.3%; and Eastern Europe 23.0%, 50.0%, 27% for ≤10, 11 to 30, and >30 participants enrolled per site, respectively; p < 0.001).
Table 1 illustrates the baseline participant characteristics and Table 2 shows medication at randomization and discharge, stratified into groups on the basis of enrollment. Participants at sites with high enrollment had an overall lower risk profile for multiple clinical characteristics. Sites that enrolled >30 participants tended to enroll younger participants who were more likely to be female and white, had higher left ventricular ejection fraction and systolic blood pressure, and shorter QRS duration. They had less diabetes mellitus, chronic kidney disease, chronic lung disease, and ischemic heart disease, and underwent less coronary revascularization procedures as compared with participants enrolled in sites with ≤10 enrollees. Similarly, participants enrolled at higher enrolling sites had lower blood urea nitrogen, white blood cell count, and natriuretic peptide levels at baseline. Participants at sites with >30 participants had more dyspnea, rales, edema, and New York Heart Association functional class IV symptoms but less jugular venous distension and lower natriuretic peptide levels.
Similar differences were noted in therapy at randomization with participants enrolled from sites with >30 enrollees more likely to be on ACE inhibitors or angiotensin receptor blockers and aldosterone antagonists therapy and less likely to be on inotropes or had implanted defibrillators or pacemakers; however, participants enrolled at centers with ≤10 participants were more likely to be on beta-blockers.
Of the 4,133 participants randomized, 2,467 (59.7%) completed the protocol. Of the rest, 761 (18.4%) and 252 (6.1%) were early withdrawals from the protocol due to death or adverse events. The other 653 (15.8%) of the participants did not complete the protocol for reasons other than death or adverse event. The reasons for these early withdrawals are listed in Table 3; they were all numerically lower among sites with higher enrollment. Both adverse events and other reasons for early treatment termination were less common, and completion of protocol more common, among sites with >30 enrolled participants (all p < 0.001).
During a median follow-up of 9.9 months, 1,700 (41%) participants had an event. The overall cardiovascular death or HF hospitalization event rate was lowest among participants at highest enrolling sites (event rate 51%, 42%, and 32% for ≤10, 11 to 30, and >30 participants per site enrolled, respectively; p < 0.001) (Fig. 2). Compared to participants enrolled at sites with >30 enrollees, ≤10 (HR: 1.77, 95% CI: 1.56 to 2.02) and 11 to 30 (HR: 1.44, 95% CI: 1.28 to 1.62) groups had worse outcomes. This risk was not proportional over time for sites enrolling ≤10 participants (HR: 2.15, 95% CI: 1.83 to 2.52 during the first 100 days; HR: 1.42, 95% CI: 1.19 to 1.69 after 100 days compared to sites with >30 enrollees). The association between enrollment and outcome was comparable across regions (p = 0.43). Similarly, no interaction was observed between the effect of the study drug (tolvaptan) and enrollment in this analysis (p = 0.13). After adjustment for major risk factors, participants enrolled at sites with fewer enrollees were at a higher risk for outcome events (HR: 1.26, 95% CI: 1.08 to 1.46 for ≤10; HR: 1.22, 95% CI: 1.07 to 1.38 for 11 to 30 vs. >30 participant enrolling sites).
There was a higher all-cause mortality rate among participants enrolled at lower enrollment sites (33%, 26%, and 21% for ≤10, 11 to 30, and >30 per site enrollment, respectively, p < 0.01; HR: 1.50, 95% CI: 1.27 to 1.76 for ≤10; HR: 1.23, 95% CI: 1.06 to 1.43 for 11 to 30 versus >30 participant enrolling sites). These differences, however, were reduced and no longer reached the threshold of statistical significance after controlling for baseline participants differences (HR: 1.12, 95% CI: 0.92 to 1.35, p = 0.26 for 1 to 10; HR: 1.07, 95% CI: 0.91 to 1.26, p = 0.39 for 11 to 30 vs. ≥30 participant enrolling sites) (Fig. 3).
Table 4 shows the distribution of other outcomes in the 3 groups. Besides all-cause mortality and cardiovascular mortality, worsening HF, and HF hospitalizations, all other causes of hospitalizations except for those related to acute myocardial infarction were also more likely among participants enrolled in sites with overall lower enrollment.
In this analysis we demonstrate that, despite strict inclusion and exclusion criteria for eligibility to enroll in a clinical trial, there are significant differences in baseline clinical characteristics among study sites according to enrollment volume. Such variations have been previously reported related to regional differences among participants. To our knowledge, this is the first report that shows significant differences in participants' baseline clinical characteristics on the basis of the number of participants enrolled by any individual clinical trial site. These differences spanned demographic and clinical characteristics, comorbidity burden, and laboratory parameters. Participants from the trial sites that enrolled fewer individuals had worse health at baseline. The less prevalent dyspnea and rales coupled with lower blood pressure and worse renal function, and higher use of inotropes all suggest that participants enrolled at lower enrolling sites likely represent a proportion of participants in low cardiac output state as well. The higher ejection fraction and lower proportion of participants with implanted defibrillators on the contrary further suggest that participants enrolled in higher enrolling sites had less severe HF. Not surprisingly, these differences correlated with clinical outcomes and protocol completion rates, with participants enrolled at sites with fewer participants being at higher risk for adverse outcomes and less likely to complete the protocol. Interestingly, such differences in the outcome of cardiovascular death or HF hospitalization persisted even after controlling for baseline clinical differences and the number of participants enrolled by a site remained as an independent predictor of this outcome. Therefore, factors related to trial sites beyond the participants' clinical characteristics that were available and adjusted for, might also play a role in patient outcomes. Finally, these characteristics were not only related to clinical outcomes but also to completion of the entire trial protocol. Early withdrawal related to both adverse events and other reasons were more common among sites that enrolled fewer participants.
Difficulties of successful clinical trial conduct are significant. In particular, recruitment and retention can be a significant barrier to successful trial conduct. The challenges of recruitment are complex. Clinical trials conducted within the United States have significantly decreased over the past decade whereas a parallel increase has been noted in the developing countries (11). Since 2002 the number of active Food and Drug Administration–regulated investigators based outside the United States has grown by 15% annually, whereas those based in the United States has declined by 5.5% (14). One-third of the Phase 3 industry-sponsored trials were being conducted solely outside the United States and a majority of study sites were outside the United States as well according to 1 estimate (15). Among 300 clinical trial reports between 1995 and 2005, the number of countries serving as trial sites outside the United States more than doubled in 10 years (11,16). The underlying reasons for this are complex and include cost savings and shortening timeline. The cost to develop a new drug averaged $802 million in 2000, with time costs accounting for half of the amount (17). This is driven by slow recruitment in trials and higher costs of keeping a large number of sites open.
Imbalances in enrollment across regions however, can pose problems with participant case mix and in turn the generalizability of results. Our data add a new dimension to this debate that even within a region, the total number of participant enrolled in a trial may be associated with participant characteristics and outcomes and, in turn, the trial's results. Our data demonstrate that, irrespective of region, the sites with fewer enrollees tended to enroll participants at higher risk for worse outcomes. The correlation between the higher risk on the basis of clinical characteristics and the actual event rate is consistent, further strengthening our results. Thus site enrollment rate in itself may affect the generalizability of results and outcomes of the drug being tested in the trial. These data are of significance, if replicated, in terms of future trial design and conduct.
Though the results are straightforward, the interpretation of our study is more complex. First, are these data reliable? Considering the large number of participants and event rate in the EVEREST trial, the consistency of findings across the various domains of baseline clinical characteristics, and the consistency between the expected relationship on the basis of differences at randomization and the actual outcomes; it is unlikely that these data represent chance findings. These data need to be replicated, however, in other cohorts, including assessing whether such a paradigm is specific to AHF or exists with chronic HF or clinical trial beyond HF.
A second issue is how to explain these differences in participant characteristics between high versus low enrolling centers. It is possible that some high enrolling centers were lax in following the inclusion and exclusion criteria and enrolled participants who were borderline candidates in terms of eligibility, were less sick, and this led to higher enrollment. On the contrary, or in addition, it is possible that low enrolling sites included participants who were too sick, representing a desire by participants and providers to pursue any options available. Such participants may have other disadvantaged social factors that predisposed them to worse outcomes as well. Whether these reasons underlie our results or if there are other explanations cannot be clearly elucidated by the current data.
Understanding the dynamics that underlie these results, however, is important in terms of future trial design. Assuming that these results are replicated, should we recommend a shift in the direction of sites that enroll more or fewer participants? Are participants from low enrollment centers too sick to benefit regardless of the therapy, or are participant from high enrollment centers not likely to benefit as a certain proportion of them have low baseline risk implying a narrow therapeutic window for benefit. These are important issues to contend with because they may determine the fate of the trial beyond the intrinsic capacities of the drug or device under investigation.
Another important facet of our data is how to explain the independent influence of the sites that enrolled higher versus lower participants on outcomes even after the baseline participant characteristics were accounted for. Again, many hypotheses can be proposed but cannot be proven by the current data. Did centers that enrolled fewer participants more frequently enroll socioeconomically disadvantaged populations? Did they have fewer staff or less sophisticated infrastructures for research and, by extension, for clinical care? Were less frequent appropriate therapies in general for participants in lower enrollment centers a function of quality of care or related to participant severity of illness? Another important issue is the financial incentive structure to the investigators at sites and regions that had more robust enrollment than those that did not. It is possible that direct or indirect financial incentives to the investigators versus to the institutions may play a role in the overall enrollment and the kind of participants that are enrolled in a trial. Furthermore, when a study is slow in enrolling, the current standard default is for new sites to be added, instead of providing more finances and nonfinancial support to the existing sites. These are important issues to understand in order to explain our findings, but nevertheless underscore the importance of careful site selection for clinical trials.
We did not have the data on site initiation dates. Similarly for sites that did not participate in the trial until its closure, we do not have the date for the site termination. In the absence of these dates, we analyzed the data on the basis of overall enrollment per site as opposed to the rate of enrollment over a given unit period of time. Also, although our data show variation in outcomes by site enrollment status, we do not have granular data for the participants' nonmedical determinant of health outcomes, or for the site structure and processes, to further explain our findings. Despite covariate adjustment, other measured and unmeasured factors may have influenced these findings.
One of the most important limitations is that most of the data regarding site characteristics such as practice setting (academic vs. nonacademic, teaching vs. nonteaching, availability of advanced services); location (rural vs. urban); hospital and practice size; physician, nursing, and research personnel volume; and total patient volume in the practice were not available. Thus further explaining these results on the basis of site-specific characteristics is not possible. It is rare that these data are routinely presented in results or taken into consideration when interpreting clinical trials data. This raises another important issue that the absence of these data might indicate that site selection and trials conduct are performed without detailed knowledge of site-specific characteristics. Our results underline the importance of uniformly collecting more data regarding site characteristics when conducting future clinical trials.
Recent problems with recruitment in clinical trial in the United States and Western Europe are well known (11). Variation in participant characteristics and outcomes among the various regions of the world has been documented (7–10,18). Regional differences in outcomes may be explained on the basis of differences in quality of care and the social and genetic makeup of study populations (19–23). We add a new dimension to the complexities of conducting clinical trials, namely that overall enrollment per site is associated with participant characteristics, protocol completion, and outcomes. This raises the hypothesis that for novel drug or device therapy, outcomes can be affected by site performance and characteristics, and that these factors should be taken under consideration when designing AHF trials.
Dr. Butler has a relationship with Takeda, Gambro, Ono Pharma, Trevena, CardioMEMS, Alere, Bayer, AMGEN, Corthera, and Medtronic. Dr. Fonarow receives research support from the Agency for Healthcare Research and Quality and serves as a consultant for Medtronic and Novartis. Dr. Konstam has received research support and/or served as a consultant for Merck, Otsuka, Johnson & Johnson, Amgen, Novartis, and Cardiokine. Dr. Maggioni receives honoraria from Otsuka. Dr. Mentz has received grants from Medtronic and research funding from Gilead Sciences. Dr. Swedberg receives research grants from AstraZeneca, Servier, and Amgen; honoraria from AstraZeneca, Otsuka, Servier, and Amgen; and is a consultant for Cytokinetics, Servier, and Novartis. Dr. Zannad has received honoraria from and served on advisory boards for Pfizer Inc. Dr. Gheorghiade has a relationship with Abbott Laboratories, Astellas, AstraZeneca, Bayer Schering Pharma AG, Cardiorentis Ltd, Corthera, Cytokinetics, CytoPherx, Inc., DebioPharm S.A., Errekappa Terapeutici, GlaxoSmithKline, Ikaria, Intersection Medical, Inc., Johnson & Johnson, Medtronic, Merck, Novartis Pharma AG, Ono Pharmaceuticals USA, Otsuka Pharmaceuticals, Palatin Technologies, PeriCor Therapeutics, Protein Design Laboratories, Sanofi-Aventis, Sigma Tau, Solvay Pharmaceuticals, Sticares InterACT, Takeda Pharmaceuticals North America, Inc., and Trevena Therapeutics; and has received support from Bayer Schering Pharma AG, DebioPharm S.A., Medtronic, Novartis Pharma AG, Otsuka Pharmaceuticals, Sigma Tau, Solvay Pharmaceuticals, Sticares InterACT, and Takeda Pharmaceuticals North America, Inc. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- angiotensin-converting enzyme
- acute heart failure
- confidence interval
- heart failure
- hazard ratio
- Received August 2, 2012.
- Revision received September 21, 2012.
- Accepted October 9, 2012.
- American College of Cardiology Foundation
- Lloyd-Jones D.,
- Adams R.J.,
- Brown T.M.,
- et al.
- Gheorghiade M.,
- Zannad F.,
- Sopko G.,
- et al.
- Blair J.E.,
- Zannad F.,
- Konstam M.A.,
- et al.
- Gheorghiade M.,
- Orlandi C.,
- Burnett J.C.,
- et al.
- Getz K.
- Truelove C.
- O'Connor C.M.,
- Fiuzat M.,
- Swedberg K.,
- et al.
- Bristow M.R.,
- Murphy G.A.,
- Krause-Steinrauf H.,
- et al.
- Larson H.N.,
- Zhou J.,
- Chen Z.,
- Stamler J.S.,
- Weiner H.,
- Hurley T.D.