Author + information
- Received July 24, 2013
- Revision received October 1, 2013
- Accepted October 4, 2013
- Published online February 1, 2014.
- Hanna K. Gaggin, MD, MPH,
- Jackie Szymonifka, MA,
- Anju Bhardwaj, MD,
- Arianna Belcher, BA,
- Benedetta De Berardinis, MD,
- Shweta Motiwala, MD,
- Thomas J. Wang, MD and
- James L. Januzzi Jr., MD∗ ()
- ↵∗Reprint requests and correspondence:
Dr. James L. Januzzi, Jr., Cardiology Division, Massachusetts General Hospital, Yawkey 5984, 55 Fruit Street, Boston, Massachusetts 02114.
Objectives This analysis aimed to perform a head-to-head comparison of 3 of the promising biomarkers of cardiovascular (CV) outcomes in heart failure (HF)—soluble ST2 (sST2), growth differentiation factor (GDF)-15, and highly-sensitive troponin T (hsTnT)—and to evaluate the role of serial measurement of these biomarkers in patients with chronic HF.
Background sST2, GDF-15, and hsTnT are strongly associated with CV outcomes in HF.
Methods This post-hoc analysis used data from a study in which 151 patients with chronic HF due to left ventricular systolic dysfunction were followed up over 10 months. At each visit, N-terminal pro–B-type natriuretic peptide (NT-proBNP), sST2, GDF-15, and hsTnT were measured and any major CV events were recorded.
Results Baseline values of all 3 novel biomarkers independently predicted total CV events even after adjusting for clinical and biochemical characteristics, including NT-proBNP, with the best model including all 3 biomarkers (p < 0.001). Adding serial measurement to the base model appeared to improve the model's predictive ability (with sST2 showing the most promise), but it is not clear whether this addition is a unique contribution. However, when time-dependent factors were included, only sST2 serial measurement independently added to the risk model (odds ratio: 3.64; 95% confidence interval: 1.37 to 9.67; p = 0.009) and predicted reverse myocardial remodeling (odds ratio: 1.22; 95% confidence interval: 1.04 to 1.43; p = 0.01).
Conclusions In patients with chronic HF, baseline measurement of novel biomarkers added independent prognostic information to clinical variables and NT-proBNP. Only serial measurement of sST2 appeared to add prognostic information to baseline concentrations and predicted change in left ventricular function. (Use of NT-proBNP Testing to Guide Heart Failure Therapy in the Outpatient Setting (PROTECT)]; NCT00351390).
The introduction of B-type natriuretic peptide and N-terminal pro–B-type natriuretic peptide (NT-proBNP) as biomarkers of heart failure (HF) has dramatically altered the standard of care for HF patients. Inclusion of these biomarkers in determining the diagnosis and prognosis in HF is now a frequent element of standard HF care. Additionally, a decrease in natriuretic peptide levels with proven HF therapy and parallel improvement in prognosis has led to the concept of biomarker-“guided” HF management, with promising results (1).
Fueled by this success of natriuretic peptides, together with an accumulation of data regarding the pathophysiology of HF development and progression, there has been a surge of interest in novel HF biomarkers. Promising novel biomarkers for HF evaluation include soluble ST2 (sST2), growth differentiation factor (GDF)-15, and highly-sensitive troponin T (hsTnT). Each has a growing body of data supporting its use, and sST2 and troponin measurements were both recently included in the American College of Cardiology/American Heart Association guidelines for the evaluation of HF (2). Elevated circulating concentrations of all 3 markers have been closely linked with adverse clinical outcomes, with an ability to predict prognosis often surpassing that of the natriuretic peptides; change in the concentration of each also appears to predict prognosis, suggesting that their serial measurement could potentially be of use for HF evaluation and management (3–5).
Despite the growing number of studies that have explored sST2, GDF-15, and hsTnT in chronic HF, almost nothing is known regarding the value of their measurement at more than 2 time points, and data regarding a direct comparison between all 3 novel biomarkers in a multimarker analysis are lacking. Further, despite biological links to myocardial remodeling, it is unclear whether any of these biomarkers can predict changes in left ventricular (LV) structure and function. Lastly, it is not known whether medications commonly used for HF affect concentrations of the biomarkers. It is in this context that we aimed to characterize serial measurements of sST2, GDF-15, and hsTnT at multiple time points in the cohort from the PROTECT (ProBNP Outpatient Tailored Chronic Heart Failure study (1,6).
Study design and patient population
The PROTECT study was a prospective, randomized, controlled, single-center trial of 151 patients with New York Heart Association (NYHA) functional class II to IV symptoms and left ventricular ejection fraction (LVEF) ≤40%; the study was designed to evaluate NT-proBNP–guided HF management versus standard HF care over the course of 10 months (1). These patients were recruited in the outpatient clinic if they had a history of recent HF decompensation. The primary endpoint of the PROTECT study and this post-hoc analysis was total cardiovascular (CV) events—a composite outcome defined as worsening HF (new or worsening symptoms/signs of HF requiring unplanned intensification of decongestive therapy), hospitalization for acutely decompensated HF, clinically significant ventricular arrhythmia, acute coronary syndromes, cerebral ischemia, and cardiac death. For secondary analysis, time to first CV event was used as a secondary outcome. There were a total of 160 endpoints in the PROTECT study, and 15 patients had a single event, 18 patients had 2 events, 9 patients had 3 events, 7 patients had 4 events, 8 patients had 5 events, 1 patient had 6 events, and another patient had 8 events. Total number of events for the PROTECT study (1) was updated to reflect a correction to a coding issue. The Partners Healthcare Institutional Review Board approved all study procedures, and all patients gave informed consent.
Study subjects were seen every 3 months at a minimum and more frequently as needed to achieve an aggressive guideline-compliant medication regimen. At each visit, a detailed medication list and a blood sample for routine laboratory tests and biomarker measurements were obtained. An echocardiogram was performed at study enrollment and at the final follow-up visit when possible; LVEF, LV end-systolic volume index, and LV end-diastolic volume index were measured.
Plasma was sampled at each visit and stored at –80°C with a single freeze-thaw cycle. A total of 145 patients had at least 2 plasma samples. (Online Table 1 reports the number of blood samples available at 0, 3, 6, and 9 months.) Biomarkers measured included NT-proBNP (Elecsys proBNP, Roche Diagnostics, Indianapolis, Indiana), sST2 (Critical Diagnostics, San Diego, California; coefficient of variation ≤1.4%), GDF-15 (Roche Diagnostics, Rotkreuz, Switzerland; coefficient of variation ≤2.3%), and a 5th-generation hsTnT (Roche Diagnostics; coefficient of variation ≤6.2% at the 99th percentile). The lot of hsTnT reagents used for our analysis were affected by calibration issues, as reported (7); to address this, a new standard curve was utilized to recalibrate concentrations.
For our initial analysis, consistent with results found in the prior literature, concentrations of sST2, GDF-15, and hsTnT were expressed relative to previously defined thresholds for each; the cutoff points were 35 ng/ml for sST2, 2,000 ng/l for GDF-15, and 14 pg/ml for hsTnT. Patient response was defined as achievement of a concentration below each cutoff point subsequent to baseline; thus, responders had concentrations below the cutoff, whereas nonresponders had concentrations above.
In a more comprehensive analysis, to evaluate whether there may be other relevant prognostic thresholds, we examined each novel biomarker as a continuous variable as well as a categorical variable. The changes in concentrations over time were also treated as continuous variables (absolute change and percent change from baseline) as well as categorical variables (study-determined optimal cutoff points as determined by receiver-operating characteristic curve analysis and the optimal area under the curve, and change greater than literature-defined biological variability of sST2 >30% increase from baseline, GDF-15 >7% increase from baseline and hsTnT >85% increase from baseline [8–11]).
Because we had the benefit of multiple measurements across an extended period of time for each subject, a percent time in response for each biomarker was derived as the proportion of time spent below the prognostic threshold relative to the total time enrolled in the study.
Differences in categorical variables between 2 groups (≤ cutoff point and > cutoff point) were assessed using the chi-square test, whereas for continuous variables, the Student t test, Mann-Whitney U test, or Kruskal-Wallis test was employed, as appropriate. Continuous variables were expressed as mean ± SD or median (interquartile range), with the latter reported in the context of non-normality. For correlation analysis, biomarker concentrations were natural logarithm transformed, and Pearson correlation analysis was performed. In initial exploratory analysis, all novel biomarkers (sST2, GDF-15, and hsTnT) were examined simultaneously in a single model that included traditional clinical and biochemical characteristics—age, sex, current smoking status, diabetes, prior CV events (i.e., at least 1 myocardial infarction, atrial fibrillation or flutter, hypertension, ventricular tachycardia, or coronary artery disease), NYHA functional class III or IV, and baseline NT-proBNP concentration. Mainly, novel biomarkers were treated as categorical variables relative to previous literature-defined cutoff point for each in determining total CV events in a linear regression model, then in predicting time-to-first CV event in a Cox proportional hazards model. Patients who were lost to follow-up or who did not experience any CV events were censored at the earlier of 1 year or the date last known to be event-free. Additionally, novel biomarkers were treated as categorical variables relative to study-determined optimal cutoff points in predicting total CV events in a linear regression model.
In a more comprehensive analysis, novel biomarkers were treated as continuous variables and the incremental role of each biomarker to a base model adjusting for traditional clinical and biochemical characteristics evaluated using a negative binomial regression model. Similar analyses were performed using Cox proportional hazard methods. Additional analyses adjusting for LVEF and study arm were performed.
For each of the novel biomarkers, the role of serial biomarker measurement was assessed by adding a change in biomarker status from response to nonresponse during the study to a base model containing a baseline biomarker status, and traditional clinical and biochemical characteristics were assessed by performing multivariable linear regression analysis to predict total CV events. Similar analyses were performed with a Cox regression model. Additionally, landmarking approaches with previous literature-determined cutoff values for each biomarker were used to determine the value of additional measurement at specified time points from baseline (3 and 6 months). Comparisons were made using the log-rank test.
Changes in biomarker levels over time were defined in various methods, and their roles in predicting clinical outcomes were assessed. The best definition of a change in biomarker concentrations was then used to examine the role of a change in each biomarker concentration in predicting CV events at 3 and 6 months.
Finally, logistic regression was used to assess the role of time in response in predicting the occurrence of CV events and major remodeling markers (LVEF, LV end-systolic volume index, and LV end-diastolic volume index).
The relationship between changes in specific HF medications and changes in logarithm of each biomarker concentration was evaluated with the use of a generalized estimation equation using a Gaussian family model with an identity canonical link function, without and with adjustment for age, NYHA functional class, and study arm.
In all statistical analyses, a 2-tailed p value <0.05 was considered to indicate statistical significance. All analyses were performed with SAS version 9.2 (SAS Institute Inc., Cary, North Carolina) or PASW versions 17 and 18 (IBM SPSS Statistics, IBM Corporation, Armonk, New York).
Table 1 details baseline patient characteristics for each of the biomarkers by the primary cutoff-point category.
Change in biomarkers over time
There were significant correlations between biomarkers examined, with stronger correlation between NT-proBNP and GDF-15 or hsTnT, and GDF-15 and hsTnT (Online Table 2). Next, after correlating within-patient paired baseline and final measurements of each biomarker (Fig. 1), there was a strong correlation between paired logarithm baseline and final biomarker concentrations of GDF-15 (Pearson r: 0.87; p < 0.001) and hsTnT (Pearson r: 0.86; p < 0.001). In contrast, sST2 showed a lesser correlation (Pearson r: 0.67; p < 0.001), similar to NT-proBNP (Pearson r: 0.67; p < 0.001). Indeed, sST2 showed the greatest change over study procedures, with 40% of the participants changing their response status. This was followed by NT-proBNP (25%), GDF-15 (21%), and hsTnT (15%) (Table 2).
Prognostic value of baseline biomarker measurements
First, when all 3 novel baseline biomarker concentrations (expressed relative to previously-defined cutoff points for each) were included in a single model with traditional clinical and biochemical characteristics, NT-proBNP, sST2, and GDF-15 were independently predictive of total CV events (R2: 0.216; F[10,138]: 3.80; p =< 0.001; sST2 beta: 0.20, p = 0.02; GDF-15 beta: 0.31, p = 0.002; hsTnT beta: 0.07, p = 0.45). All 3 novel biomarkers independently predicted time to first CV event (sST2, p = 0.01; GDF-15, p = 0.01; hsTnT, p = 0.01). When novel biomarkers were expressed relative to study-determined optimal cutoff points (42 ng/ml for sST2, 3,270 ng/l for GDF-15, and 24 pg/ml for hsTnT), all 3 novel biomarkers predicted total CV events (R2: 0.344; F[10,138]: 7.23; p < 0.001; sST2 beta: 0.22, p = 0.004; GDF-15 beta: 0.37, p < 0.001; hsTnT beta: 0.28, p < 0.001).
In a secondary analysis, the novel biomarkers were treated as continuous variables, and the incremental role of each of the novel biomarkers was comprehensively evaluated (Table 3). Adding each of the novel biomarkers to the baseline model containing traditional clinical and biochemical characteristics including NT-proBNP added independent information in predicting total CV events (sST2, beta: 0.23, p < 0.001; GDF-15, beta: 0.17, p = 0.004; hsTnT, beta: 0.13, p = 0.04). As reflected by decreasing Akaike information criterion (AIC) values, adding any 2 of the 3 novel biomarkers to the baseline model further improved our ability to predict clinical outcomes, but the best model was when all 3 novel biomarkers were added to the base model. Similar results were seen in models predicting time to first CV event (Online Table 3). In a model further adjusting for LVEF and study arm, continuous concentrations of all 3 markers remained independently predictive (sST2, hazard ratio [HR]: 1.12, p = 0.02; GDF-15, HR: 1.17, p = 0.001; hsTnT, HR: 1.09, p = 0.02).
Prognostic value of serial biomarker measurements
In a multiple regression model that adjusted for traditional risk factors including NT-proBNP and baseline sST2 status (according to the previously defined cutoff point of 35 ng/ml), adding a change in sST2 status from ≤35 to >35 pg/ml (from response to nonresponse) during the study improved the model (R2 from 0.145 to 0.158), with both models being significant (p = 0.004 and 0.005, respectively). However, sST2 status change did not appear to add any unique contribution to the model (beta: 0.13; p = 0.15). Adding GDF-15 status change to the base model containing traditional risk factors and baseline GDF-15 status improved the model for predicting total CV events (R2 from 0.130 to 0.177; p < 0.001 and p = 0.001, respectively) but did not appear to be uniquely significant (beta: 0.004; p = 0.96). Adding a change in hsTnT status to the baseline model with baseline hsTnT status did not improve the model (R2 from 0.110 to 0.111; p = 0.03 and 0.06, respectively), which, again, was not a unique contribution (beta: –0.02; p = 0.80). Of note, when NT-proBNP status change was included in the base model that adjusted for traditional risk factors including NT-proBNP, this model no longer predicted total CV events (R2: 0.090; p = 0.12).
Next, time-dependent data were incorporated in a Cox regression model. When traditional risk factors as well as baseline sST2 status and a change in the sST2 responder status during the study were included in the model, a baseline sST2 <35 ng/ml was associated with longer time to first CV event (HR: 0.30; 95% confidence interval [CI]: 0.14 to 0.63; p = 0.002), whereas a change in the sST2 responder status from ≤35 ng/ml to >35 ng/ml (from response to nonresponse) during the study was associated with significantly shorter time to first CV event (HR: 3.64; 95% CI: 1.37 to 9.67; p = 0.009). In a similar analysis of GDF-15, only baseline values ≤2,000 pg/l were predictive of longer time to first CV event (HR: 0.32; 95% CI: 0.14 to 0.73; p = 0.007), whereas serial measurements did not add clear value; the results for hsTnT were similar, with baseline values adding overwhelming prognostic information (HR: 0.29; 95% CI: 0.13 to 0.61; p = 0.001) relative to serial measurement.
Next, using landmarking approaches with dichotomous cutoffs, of the 3 novel biomarkers evaluated, sST2 was the only test with information added from measurement at the 3-month (p = 0.03) and 6-month (p = 0.02) landmarked time points beyond baseline (p = 0.005). Only baseline values of GDF-15 (p = 0.001) and hsTnT (p = 0.01) were significant. In regression modeling for time to first event, 3-month sST2 biomarker concentrations added incremental prognostic information to baseline (p = 0.01); such findings were not seen with GDF-15 (p = 0.19) or hsTnT (p = 0.91), not surprisingly.
In a secondary analysis, changes in biomarker levels were defined in various ways: absolute change in concentration from baseline, percent change, and change greater than literature-defined biological variability for each biomarker. Each of these definitions of changes from baseline to 3 months and baseline to 6 months was used in a model to predict CV events. Of these, the best definition appeared to be a change greater than biological variability for each biomarker. This definition of change in biomarker was used in a regression model to assess whether there were any improvements in prediction of CV events. When a change in sST2 from baseline to 3 months, as defined earlier, was added to a base model that adjusted for traditional risk factors including NT-proBNP, the predictive power of the model improved, with AIC decreasing from 403.87 to 318.80, but the contribution of the change in sST2 to the model was not significant (beta: –0.19; p = 0.68). Similar findings were seen with GDF-15 (AIC decreased from 411.14 to 323.88; beta: 0.31; p = 0.38) and hsTnT (AIC decreased from 415.53 to 324.88; beta: 0.50; p = 0.65). Lastly, in a logistic regression analysis harnessing time-integrated prognostic information across all blood draws (Table 4), increasing percent time spent below prognostic thresholds (time in response, scaled by a factor of 10%) predicted lower CV events for all of the novel biomarkers. The relationship between categories of percent time spent in response and CV events is shown in Figure 2.
Biomarker concentrations and LV remodeling
More time spent in sST2 response predicted decreasing left ventricular end-diastolic index (odds ratio: 1.22; 95% CI: 1.04 to 1.43; p = 0.01) after adjusting for relevant baseline characteristics. Other biomarkers did not show any significant relationship with major remodeling markers evaluated.
Medication effects on biomarker levels
Online Table 4 summarizes significant medication effects on each biomarker in adjusted analyses. The greatest magnitude of interaction was a significant inverse association between β-blocker dose changes and sST2 concentrations.
We rigorously examined a combination of emerging risk markers in chronic HF. In doing so, we modeled each marker individually and collectively, and examined their results in various ways, including using continuous concentrations, previously established cutoffs, as well as integrating their results over time in a manner mimicking their use in clinical practice.
Individually, single measurements of sST2, GDF-15, and hsTnT concentrations have been reported to be predictive of adverse HF outcomes, but no study has evaluated them all together in patients with chronic HF, or with extensive serial measurement. In a general-population cohort from the Framingham Heart Study, we found that these 3 markers were additively predictive of risk (12). In the present study in chronic HF patients, the 3 markers examined were relatively loosely correlated, and each revealed prognostic information independent and additive of each other when adjusted for traditional clinical and biochemical characteristics including NT-proBNP. This reflects the intricate pathophysiology of HF; the best approach to determine prognosis (and perhaps select therapies) may thus be a multimarker profile using several complementary biomarkers, a concept first reported by Ky et al. (13). Beyond the baseline measurement, only sST2 appeared to provide incremental prognostic information and reflect changes in myocardial remodeling over time. A novel biomarker’s dynamic ability to reflect the underlying HF biology makes it an ideal candidate for potentially monitoring and guiding HF management.
Much in the way that natriuretic peptides are induced when cardiomyocytes are stretched, concentrations of sST2 are thought to represent a cellular response to cellular stress; ST2 biology appears to play a pivotal role in LV remodeling and fibrosis (14). Concentrations of sST2 are increasingly accepted to reflect important prognostic information not already revealed by natriuretic peptides (15). Although data from a pilot study of repeated sST2 testing in chronic HF were promising (5), very little is known about the merits of serial sST2 measurement at multiple time points in chronic HF. In the present analysis, sST2 appeared to add prognostic information above the natriuretic peptides across multiple time points of measurement, and to indicate significant dynamic change of the biomarker in parallel with risk for adverse events and myocardial remodeling.
GDF-15 is a member of the transforming growth factor-β cytokine superfamily. Expression of GDF-15 is strongly induced in cardiomyocytes in response to metabolic stress, and appears to be involved in the regulation of cell differentiation and tissue repair. GDF-15 is thereby closely linked with tissue remodeling and is prognostic of adverse outcomes in HF (3). Anand et al. (3) reported that baseline values of GDF-15 were strongly prognostic in chronic HF but that adding a follow-up value did not inform substantial extra prognostic information; in our analysis of sampling at multiple time points, we now report similar results.
In the past decade, highly-sensitive troponin assays have been developed that provide ability to detect even minute degrees of cardiac injury. Elevation of highly-sensitive troponin above the 99th percentile of a normal population has been shown to be common in chronic HF patients and of prognostic importance (16). Much like with GDF-15, we found a single measurement of hsTnT provided much of the ability to predict adverse events, but serial measurement did not add significant incremental prognostic data. Our results with multiple measurements agree with those reported by Masson et al. (4) using paired measurements.
To date, no specific medication changes have been shown to be associated with change in the concentration of any of the biomarkers studied. In this hypothesis-generating analysis, we found potential associations between therapy changes and biomarker concentrations. Notably, we found a significant relationship between β-blocker changes and sST2. As β-blockers have been shown to reverse myocardial remodeling (17), we are currently examining the potential link between this class of agents and sST2 in more depth. As stated earlier, the link between β-blocker dose changes and sST2 changes may be leveraged in identifying patients who may particularly benefit from aggressive titration of β-blockers.
With the growing number of unique biomarkers that may be available in HF, we have previously argued that a rigorous assessment process is necessary to best understand which markers would be of greatest use and how to best deploy them (18). Our results show value for each marker measured at baseline, but only sST2 appeared to provide incremental prognostic data beyond initial measurement. Further data in this regard are needed.
This was a post-hoc analysis of a small, single-center study. Small numbers of subjects might limit the ability to detect subtle changes in biomarker values over time; however, we had numerous measures from each subject over time, and were able to extensively characterize our cohort, following each medication change, clinical events, and biomarker changes over time. Another issue is that we did not have uniform clinical follow-up time intervals. This resulted in smaller numbers of patients available for repeated-measures analyses. Nonetheless, leveraging the unique strength of the volume of biomarker measures available, our use of time in response is unique, and allows for prognostic comparisons of various markers drawn at the same time points using a time-integrated approach. This time-in-response approach (widely used in studies of response to anticoagulant therapy ) is likely to be more widely employed as more studies of HF-biomarker testing across multiple time points become available. Lastly, although the use of discrete cutoffs facilitates analysis of prognostic ramification of changes from “normal” to “elevated,” the starting level of the biomarker as well as the degree of change should both be considered when interpreting a change in concentration.
In patients with chronic HF, baseline measurements of novel biomarkers added independent prognostic information to clinical variables and NT-proBNP. Only serial measurement of sST2 appeared to add prognostic information to baseline concentrations and predicted change in LV function.
Dr. Gaggin is supported in part by the Ruth and James Clark Fund for Cardiac Research Innovation. Drs. Bhardwaj and Motiwala were supported by the Dennis and Marilyn Barry Cardiology Fellowship. Dr. Wang has received research or assay support from DiaSorin, Brahms, Critical Diagnostics, LabCorp, and Siemens Diagnostics; has received honoraria from Roche, DiaSorin, and Quest Diagnostics; has served on the medical advisory board of Singulex; and is named as coinventor on patent applications relating to the use of metabolomic or neurohormonal biomarkers in risk prediction. Dr. Januzzi is supported in part by the Roman W. DeSanctis Clinical Scholar Endowment and has received grants from Roche Diagnostics, Siemens, Critical Diagnostics, Singulex, and Thermo Fisher. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose. John R. Teerlink, MD, served as Guest Editor for this paper.
- Abbreviations and Acronyms
- Akaike information criterion
- growth differentiation factor
- heart failure
- highly-sensitive troponin T
- left ventricular
- left ventricular ejection fraction
- N-terminal pro–B-type natriuretic peptide
- New York Heart Association
- soluble ST2
- Received July 24, 2013.
- Revision received October 1, 2013.
- Accepted October 4, 2013.
- American College of Cardiology Foundation
- Januzzi J.L. Jr..,
- Rehman S.U.,
- Mohammed A.A.,
- et al.
- Yancy C.W.,
- Lopatin M.,
- Stevenson L.W.,
- De Marco T.,
- Fonarow G.C.
- Anand I.S.,
- Kempf T.,
- Rector T.S.,
- et al.
- Masson S.,
- Anand I.,
- Favero C.,
- et al.
- Apple F.S.,
- Jaffe A.S.
- Vasile V.C.,
- Saenger A.K.,
- Kroning J.M.,
- Jaffe A.S.
- Wang T.J.,
- Wollert K.C.,
- Larson M.G.,
- et al.
- Ky B.,
- French B.,
- Levy W.C.,
- et al.
- Ky B.,
- French B.,
- McCloskey K.,
- et al.
- Latini R.,
- Masson S.,
- Anand I.S.,
- et al.
- van Kimmenade R.R.,
- Januzzi J.L. Jr..
- Jones M.,
- McEwan P.,
- Morgan C.L.,
- Peters J.R.,
- Goodfellow J.,
- Currie C.J.