Author + information
- Received January 1, 2006
- Revision received March 28, 2006
- Accepted April 24, 2006
- Published online September 5, 2006.
- Cynthia M. Westerhout, MSc⁎,†,⁎ (, )
- Yuling Fu, MD⁎,
- Michael S. Lauer, MD, FACC‡,
- Stefan James, MD, PhD§,
- Paul W. Armstrong, MD, FACC⁎,
- Eyad Al-Hattab, MD‡,
- Robert M. Califf, MD, FACC∥,2,
- Maarten L. Simoons, MD, FACC†,
- Lars Wallentin, MD, PhD, FACC§,1,
- Eric Boersma, PhD†,
- GUSTO-IV ACS Trial Investigators
- ↵⁎Reprint requests and correspondence:
Ms. Cynthia M. Westerhout, 214 Heritage Medical Research Centre, University of Alberta, Edmonton, Alberta, T6G 2S2 Canada.
Objectives The purpose of this study was to develop 30-day and 1-year risk stratification models for non–ST-segment elevation acute coronary syndrome (NSTE-ACS) patients that incorporate quantitative ST-segment depression and novel biomarkers.
Background Several novel biomarkers have changed the risk profile of ACS; thus, the reassessment of traditional indicators such as ST-segment depression in this new context is warranted.
Methods Multivariable logistic regression was used to identify significant predictors of 30-day death and death/myocardial infarction (MI) and 1-year mortality in 7,800 NSTE-ACS patients enrolled in the GUSTO-IV (Global Utilization of Strategies to Open Occluded Arteries-IV ACS) trial between 1998 and 2000.
Results Among all other predictors, the degree of ST-segment depression had the highest prognostic value for 30-day death, 30-day death/MI, and 1-year death. Troponin T (TnT), creatinine clearance, N-terminal pro-brain natriuretic peptide (NT-proBNP), heart rate, and age were also highly influential on adverse outcomes. Unlike TnT and NT-proBNP, C-reactive protein was only predictive of long-term death. In contrast to mortality, the contribution of TnT to predicting 30-day death/MI increased, whereas NT-proBNP’s role was attenuated. The discriminatory power was excellent (c-index [adjusted for over-optimism]: 0.82 [30-day death]; 0.72 [30-day death/MI]; 0.81 [1-year]).
Conclusions In this large contemporary study of NSTE-ACS patients, novel insights into risk stratification were observed—in particular, the utility of quantitative ST-segment depression and multiple biomarkers. Collection of these indicators in future NSTE-ACS populations is recommended to evaluate generalizability and clinical application of these findings.
Demographics, comorbidities, and other classic patient factors have long been the foundation of risk stratification in acute coronary syndromes (ACS) (1–4). More recently, several biomarkers have also been shown to be associated with the risk for subsequent coronary events, including indicators of myocardial necrosis (troponin), inflammation (C-reactive protein), and hemodynamic stress (brain natriuretic peptides).
The electrocardiogram (ECG) has continually played an important role in prognosis of ACS patients. ST-segment depression >0.5 mm at baseline, for instance, is associated with poorer prognosis than absence of ST-segment depression. The extent of ST-segment depression, however, is rarely measured or reported even though this refinement substantially improves risk stratification compared with conventional qualitative ECG data (5–8). The prognostic value of combining quantitative ECG data and the extended biomarker profile with classical patient risk factors is unclear. Thus, the systematic collection of quantitative ST-segment depression, cardiac troponin T (TnT), C-reactive protein (CRP), and N-terminal pro-brain natriuretic peptide (NT-proBNP) in 7,800 non–ST-segment elevation acute coronary syndrome (NSTE-ACS) patients enrolled in the GUSTO-IV (Global Utilization of Strategies to Open Occluded Arteries-IV ACS) trial provided a unique opportunity to investigate the relative roles of factors influencing short- and long-term outcomes (9). Risk scores were also developed to illustrate potential clinical applications.
The details of the GUSTO-IV-ACS trial have been previously described (9,10). Briefly, patients over the age of 21 years were eligible if they presented within 24 h after an episode of ischemic chest pain lasting ≥5 min and had either elevated TnT/TnI above the upper limit of normal (ULN) according to the local quantitative or qualitative assays or transient/persistent ST-segment depression (≥0.5 mm) that was not known to be pre-existing and not attributable to a co-existing disorder or medication on the admission ECG.
Patients were randomly assigned to abciximab (0.25 mg/kg bolus plus 0.125 μg/kg/min infusion for 24 or 48 h) or placebo. Because abciximab did not reduce the primary or secondary end points, patients from all 3 treatment arms were combined for this analysis. Coronary angiography was not to be performed during or within 12 h after the completion of the study agent infusion. Patients also received aspirin and either unfractionated or low-molecular-weight heparin. Other clinically indicated medications were used at the discretion of the treating physician.
Standard 12-lead baseline ECG data obtained at baseline were centrally evaluated by 2 experienced readers without knowledge of the clinical outcomes in 7,741 patients (99.2%) at the ECG core laboratory (Cleveland Clinic Foundation, Cleveland, Ohio) and were entered and analyzed at the Canadian VIGOUR Centre (University of Alberta, Edmonton, Canada). ST-segment depression was prospectively defined and measured with the aid of a magnifying caliper to the nearest 0.5 mm in all leads except aVR. ST-segment depression was judged to be present if the J point was depressed by 0.5 mm and was followed by a horizontal or down-sloping ST-segment for at least 0.08 s. Patients were classified as having ST-segment depression if the ST-segment was depressed by ≥0.5 mm in 2 of the limb leads or at least 2 contiguous precordial leads. Patients were categorized into 5 mutually exclusive groups: no ST-segment depression, 0.5-mm, 1- to 1.5-mm, ≥2-mm ST-segment depression, and “confounders.” Confounders included left bundle branch block, paced rhythm or ventricular rhythm in 225 patients, or ST-segment elevation on the baseline ECG in 289 patients.
Blood samples were drawn at baseline and analyzed at a core laboratory (Department of Clinical Chemistry, University of Uppsala, Uppsala, Sweden). Details of the assays used have been previously published (11). Baseline levels of TnT, CRP, and NT-proBNP were determined per protocol in all but 685 (8.8%), 692 (8.9%), and 991 (12.7%) patients, respectively. The loss of 500 samples during transport accounts for much of these missing data. Complete data for all 3 biomarkers were available in 6,809 (87.3%) patients.
Baseline serum creatinine concentration (mg/dl) was measured at local laboratories in 7,703 of 7,800 patients. To assess renal function, creatinine clearance was calculated with the Cockcroft-Gault equation, which adjusts for age, gender, and weight (12).
End point definitions
The end points of the current study included 30-day death, 30-day death or MI, and 1-year death, as described previously (9). Double counting of patients with more than 1 event was avoided by classifying each patient according to the event with the greatest severity. Thus, a patient with MI who subsequently died was classified as experiencing death but not MI. At 1 year post-randomization, follow-up data on vital status were obtained and were complete in 7,746 patients (99.3%). Possible cases of MI were adjudicated by a clinical end point committee that was not aware of biomarker levels evaluated in the core laboratory (i.e., TnT, CRP, NT-proBNP) or the extent of ST-segment depression on the baseline ECG (9).
Baseline characteristics were summarized by frequency and percentage for categorical variables and by median and interquartile range for continuous variables. Creatinine clearance, TnT, CRP, and NT-proBNP were categorized into quartiles of their distributions for ease of presentation (11). The choice of TnT over creatine kinase-MB as the preferred marker of myocardial damage was based on established guidelines as well as by an independent analysis of their relative prognostic power (Appendix 1) (13).
Logistic regression evaluated the associations between baseline characteristics and adverse outcomes. Indicators were entered into the full multivariable model if the p value of the univariable association was <0.25. The final multivariable model was constructed by backward stepwise elimination of the least significant factors and the Akaike information criteria. Given that 1-year follow-up was complete in 99.3% of patients, logistic regression was repeated to determine predictors of 1-year mortality. Unadjusted and adjusted odds ratios and their corresponding 95% confidence intervals are reported.
The Hosmer-Lemeshow goodness-of-fit test statistic and calibration of the predicted versus observed event rate according to deciles of predicted risk were calculated to assess model performance. The discriminatory power was estimated by the c-index (i.e., probability of concordance between observed and predicted survival on the basis of pairs of individuals). The models were developed in the entire study population and validated by bootstrapping techniques: 100 bootstrap samples were drawn, with replacement, to estimate the amount of “over-optimism” in the models on the basis of the entire population and quantified by the decrease in the area under the receiver operating characteristic curve (14).
Simplified risk scores were developed to demonstrate the potential for clinical application. The top 5 contributing factors in the model were assigned weights that were equivalent to the logistic regression coefficient multiplied by 10. For patient “i”, a weighted risk score was estimated by summing the weighted risk for each variable (risk score = Σ10βixi, where beta was the regression coefficient associated with the factor, and x was equal to 1 when the factor was present and 0 when absent). The discriminatory power of the simplified risk scores was also adjusted for over-optimism.
Overall, 301 (3.9%) of the 7,800 enrolled patients died within 30 days after randomization, with an additional 348 patients (4.5%) dying between 30 days and 1 year (Table 1).Of the 659 (8.4%) patients who had an MI or died within 30 days, 358 had a non-fatal MI, 74 had an MI and died, and 227 died without an MI. Patients who experienced an end point were more often older and had a higher frequency of comorbidities and prior cardiac history. Baseline biomarkers (i.e., TnT, NT-proBNP, CRP), heart rate, and the extent of ST-segment depression were significantly higher in patients who died within 30 days or 1 year.
Predictors of 30-day death
Thirteen of 19 possible predictors of 30-day mortality remained statistically significant in the final model. (Table 2).Compared with those without ST-segment depression, the likelihood of 30-day death was twice as high in patients with 1- to 1.5-mm and nearly 4-fold higher with ≥2-mm ST-segment depression, after baseline adjustment (p < 0.001). ST-segment depression contributed the most to the model, as indicated by the percentage of the total chi-square (Fig. 1).Creatinine clearance was also strongly associated with 30-day mortality: patients with creatinine clearance ≤58.4 ml/min had a 5-fold increase in the risk of 30-day death relative to >98.6 ml/min (p < 0.001). Patients with TnT >0.47 μg/l were more than 3 times as likely to die within the first 30 days than those with TnT ≤0.01 μg/l (p < 0.001). Compared with NT-proBNP ≤237 ng/l, the odds of 30-day mortality steadily rose with increasing levels of NT-proBNP, particularly in those >1,896 ng/l, who had nearly a 3-fold increase in the risk of death (p < 0.012). Strong prognostic associations were also observed for heart rate and age and, to a lesser extent, histories of MI and heart failure; weight; current smoker; and histories of hyperlipidemic therapy, percutaneous coronary intervention (PCI), and stroke.
Predictors of 30-day death/MI
Among the 8 independent predictors of 30-day death/MI, ST-segment depression and TnT made the greatest contributions. (Table 2, Fig. 1). Compared with patients without ST-segment depression, the odds ratio of 30-day death/MI ranged from 1.2 to 3 as the extent of ST-segment depression increased (p < 0.001). Also, patients with TnT >0.47 μg/l were over 2.5 times more likely to experience the composite within 30 days compared with those ≤0.01 μg/l (p < 0.001). Creatinine clearance, age, and NT-proBNP were important risk factors, similar to the 30-day mortality model. However, the relative contribution of NT-proBNP was attenuated compared with the mortality models (Fig. 1). Unlike the mortality model, heart rate was not a significant independent predictor of 30-day death/MI.
Predictors of 1-year death
ST-segment depression was the strongest contributor to the model predicting 1-year mortality, similar to the 30-day mortality model (Table 2, Fig. 1). A marked gradient in risk relative to ST-segment depression was observed, even after baseline adjustment: patients with ≥2 mm were 2.4 times more likely to die within the first year compared with those without ST-segment depression (p < 0.001). Other important predictors included creatinine clearance, NT-proBNP, age, heart rate, and TnT. For instance, patients who had ≤58 ml/min creatinine clearance were nearly 4 times more likely to die within the first year compared with >98.6 ml/min. In contrast to 30-day mortality, CRP, diabetes, and prior angina were significant prognostic factors, whereas past use of hyperlipidemic therapy and PCI were no longer significant factors to long-term prognosis. There was also a significant interaction between NT-proBNP and age (p = 0.006). In 30-day survivors, the risk profile changed to some extent (Appendix 2, Table A2.1). Whereas TnT was no longer a significant factor, 30-day post-randomization PCI reduced the likelihood of death by 50% (p = 0.001).
The Hosmer-Lemeshow goodness-of-fit test statistic indicated adequate fit of all models (30-day mortality: chi-square = 4.72 [p = 0.787]; 30-day death/MI: chi-square = 11.84 [p = 0.158]; 1-year mortality: chi-square = 5.72 [p = 0.679]), and calibration was excellent (Fig. 2).The c-index was 0.83 for the 30-day mortality model, which reflected excellent discriminatory power. The level of over-optimism determined by bootstrapping was 0.01, which reduced the c-indexadjustedto 0.82. The model predicting 1-year mortality had comparable discriminatory power of c-indexadjusted= 0.81 (over-optimism = 0.01), whereas it was lower for 30-day death/MI (0.72 [over-optimism = 0.01]).
Incremental value of novel risk factors
The incremental value of quantitative ST-segment depression and these biomarkers over “traditional” patient characteristics (e.g., age, comorbidities, history of cardiovascular disease, heart rate) was assessed with 2 approaches. First, the relative contribution of each independent predictor in the model was estimated (Fig. 1). Second, as each of these novel risk factors was added to the model of traditional baseline characteristics, a noticeable increase in the c-indexadjustedwas observed (Fig. 3).
Simplified risk scores
As shown in Figure 4,the 5-factor risk score ranged from 0 in patients with no risk factors to 51 in those with all (and the most severe degree) of the risk factors for 30-day mortality, from 0 to 39 for 30-day death/MI, and from 0 to 48 for 1-year mortality. The accompanying nomogram translates the risk score into the likelihood of the outcome. The discriminatory power of the risk scores was slightly attenuated but remained reasonably strong (c-indexadjusted= 0.78, 0.67, and 0.77, respectively).
This large study of contemporary NSTE-ACS patients highlights the striking prognostic value of quantitative ST-segment depression even in the context of an expanded biomarker profile including creatinine clearance, NT-proBNP, TnT, and CRP as well as other more traditional baseline risk factors such as creatinine clearance and age.
Electrocardiography has long been an essential clinical tool in the evaluation of ACS as it is inexpensive, readily available, and non-invasive. Numerous studies on the basis of clinical trial and registry samples have also demonstrated that ST-segment depression on the admission ECG is highly influential on short- and long-term outcomes (6,15,16). Although the binary indicator (i.e., presence vs. absence of ST-segment depression) is often used in risk stratification, an increasing body of evidence, including this study, has demonstrated that quantitative ST-segment depression is a substantial refinement (7,8). The PARAGON-A (Platelet IIb/IIIa Antagonism for the Reduction of Acute Global Organization Network) investigators (7) noted a steady increase in 30-day and 1-year mortality as the extent of ST-segment depression rose (30-day mortality: 0.7% [no ST-segment depression], 2.8% [1-mm], and 6.3% [≥2-mm ST-segment depression]; 1-year mortality: 2.0%, 7.8%, and 13.4%, respectively). After baseline adjustment, ST-segment depression made the highest relative contribution to the prediction of 1-year mortality, which was similarly observed in this study.
The quantitative collection of indicators of renal function, myocardial necrosis, hemodynamic stress, and inflammation offered a unique opportunity to explore the pathophysiology of ACS. Abnormal renal function is a well-documented predictor of morbidity and mortality in a wide variety of patients (17). In concert with other ACS studies, lower levels of creatinine clearance were significantly associated with poorer short- and long-term prognosis, even though patients with advanced renal dysfunction were excluded from this trial (9). Impaired renal function might also alter the prognostic value of other serum biomarkers, because their clearance might be inhibited; however, previously published findings dispel this concern (11,18).
In the past, creatine kinase-MB was the reference standard in MI diagnosis, but troponins have replaced creatine kinase-MB with improved sensitivity and specificity (13). In the current study, TnT was ranked second and third among the predictors of 30-day death/MI and 30-day death, respectively, which reflects its role as a highly sensitive marker of myocardial necrosis in the acute phase. Although TnT remained a significant predictor of 1-year mortality (in all-comers), its influence declined over time after the index coronary event.
Elevated levels of BNP have been linked with left ventricular dysfunction and poorer prognosis in patients with heart failure and other associated disorders. Recently, these observations have been extended to ACS patients in whom elevated BNP levels also might be the result of transient ischemia (19–21). As demonstrated previously, increasing quartiles of NT-proBNP were strongly related to an increase in the odds of 30-day and 1-year mortality but not of 30-day MI, after multivariable adjustment (21). Although this relationship remained robust in the current study, it was no longer the top contributor of prognostic information after introducing the extent of ST-segment depression.
The role of inflammation in the pathophysiology of ACS is increasingly important, with CRP as one of the best recognized markers. Although once thought to have a passive role in vascular inflammation, CRP might be an active contributor to atherogenesis, as suggested by recent evidence (22). Similar to the current study, others observed that the prognostic impact of elevated CRP was not evident early after the coronary event but emerged later as an index of the ongoing evolution of coronary artery disease (23–25).
Classical risk factors such as age and heart rate played a significant role in risk prediction; however, compared with previously published risk models, these factors lost considerable prognostic value in the current study with the inclusion of quantitative ST-segment depression and biomarkers (1,2,26). Heart rate was not an independent predictor of 30-day death/MI, a finding similar to those of other studies, (2).
Whether risk stratification that defines the optimal management strategy should be driven by noninvasive determinants of ischemia or knowledge of the coronary anatomy remains controversial. Although there is general consensus that patients at highest baseline risk tend to benefit most from early invasive therapy, others have argued that despite risk, invasive angiography and/or revascularization should become the standard of care in virtually all such patients. Higher rates of early invasive treatment, however, do not necessarily translate into improved outcomes in all patients, especially in those without ST-segment changes (27–29). International practice patterns revealed that the use of angiography and angioplasty in NSTE-ACS patients was inversely related to the extent of ST-segment depression (30). Thus, improved risk assessment as proposed in the current study would better identify patients at increased risk, in whom coronary intervention should be considered, and its implementation might result in the increased efficiency of care. It seems clear that application of this approach depends, to some extent, on what risk one aims to predict and when. As evident in Figure 1, the shorter term risk of death is strongly influenced by ST-segment depression, renal function, cardiac troponin, and NT-proBNP in addition to traditional risk factors. The relative contribution of ST-segment depression and troponin is actually enhanced as it relates to 30-day death/MI. By 1 year some repositioning occurs such that age, diabetes, prior angina, and CRP appear as factors associated with mortality: whereas the influence of ST-segment depression remains strong, NT-proBNP is strengthened, and troponin is attenuated.
Identifying appropriate treatment strategies might be facilitated through the use of simplified risk models in clinical practice. Several user-friendly risk scores have been developed from clinical trial and registry populations, such as the Thrombolysis In Myocardial Infarction (TIMI) risk score in NSTE-ACS and the recently published risk score from the GRACE (Global Registry of Acute Coronary Events) registry (3,4). However, there are notable limitations. Although the TIMI risk score contains few factors, its discriminatory power is modest (c-index = 0.65), which limits its clinical application. And although the GRACE registry risk score extends beyond the TIMI risk score to include serum creatinine, heart rate, and systolic blood pressure, it is lacking quantitative ST-segment depression, NT-proBNP, and CRP, which were shown to have significant prognostic value in the current study and other studies.
Some limitations of our study should be addressed. First, baseline systolic blood pressure was not available and could have influenced our findings. However, this seems unlikely. An ad hoc analysis of the PURSUIT (Platelet Glycoprotein IIb/IIIa in Unstable Angina: Receptor Suppression Using Integrilin Therapy) 30-day death model revealed that if systolic blood pressure was not included, the top 5 predictors (as measured by percent Sigma chi-square) to the model—heart rate(interaction with AMI/UA), age, rales, ST-segment depression (yes/no), and region of enrollment—essentially remained unchanged from the model that included systolic blood pressure (2). Notably, if one of the top indicators, such as heart rate, was not included in the model, the structure of the overall model changed significantly. Second, the extrapolation of these findings to the general ACS population might be in question. According to the protocol, coronary angiography was restricted at least 12 h after the completion of the study drug infusion. Although this might not be routine procedure in many tertiary facilities, it likely is representative of the sizeable proportion of the overall ACS population who do not present directly to these facilities. And finally, the external validation of these risk models was not provided. To our knowledge, no other clinical trial or population-based cohort possesses the requisite ECG and/or biomarker data required for this task. Validation of clinically meaningful cut points will also be important. Future investigations should incorporate these indicators into their design to achieve this.
In conclusion, the degree of ST-segment depression was the highest contributor to the prediction of 30-day and 1-year mortality and 30-day death/MI in a contemporary sample of 7,800 NSTE-ACS patients. Creatinine clearance, TnT, and NT-proBNP also were significant independent prognostic indicators of adverse outcomes, whereas CRP was only significant in the long-term prediction. The current study provides novel contemporary insights into the risk stratification of NSTE-ACS patients, which might be of particular value in identifying strategies for risk reduction and the planning of future studies.
The authors acknowledge Dr. Wilson W. H. Tang and Dr. Hitinder S. Gurm of the Cleveland Clinic Foundation (Cleveland, Ohio) for their contributions to this manuscript and the collection of heart rate and electrocardiographic data. Dr. Jan-Paul Ottervanger (Hospital De Weezenlanden, Zwolle, the Netherlands) and Dr. Timo Lenderink (Erasmus Medical Center, Rotterdam, the Netherlands) are also thanked for their critical review of earlier drafts of this manuscript. Additional statistical expertise was provided by Dr. Wei-Ching Chang of the Canadian VIGOUR Centre (Edmonton, Alberta, Canada), Dr. Ewout Steyerberg of Erasmus Medical Center (Rotterdam, the Netherlands) and Ms. Karen Pieper of Duke Clinical Research Institute (Durham, North Carolina).
For an independent analysis of the relative prognostic power of TnT over creatine kinase-MB as the preferred marker of myocardial damage and adjusted odds ratios and 95% confidence intervals of baseline characteristics in predicting 1-year mortality in 30-day survivors, please see the online version of this article.
↵1 Dr. Wallentin has received research grants from Centocor.
↵2 Dr. Califf has received grants, speaking fees, and/or consulting fees from Centocor, Merck, Millenium, and Schering Plough.
The GUSTO-IV ACS trial was sponsored by Centocor and Eli Lilly. The sponsors did not participate in the analysis or interpretation of the data, manuscript preparation, review or approval, or the decision to publish. Funding support for this study was also not provided by the sponsors.
- Abbreviations and Acronyms
- C-reactive protein
- non–ST-segment elevation acute coronary syndrome
- N-terminal pro-brain natriuretic peptide
- percutaneous coronary intervention
- cardiac troponin T
- Received January 1, 2006.
- Revision received March 28, 2006.
- Accepted April 24, 2006.
- American College of Cardiology Foundation
- Armstrong P.W.,
- Fu Y.,
- Chang W.C.,
- et al.,
- The GUSTO-IIb Investigators
- Boersma E.,
- Pieper K.S.,
- Steyerberg E.W.,
- et al.,
- The PURSUIT Investigators
- Cannon C.P.,
- McCabe C.H.,
- Stone P.H.,
- et al.
- Holmvang L.,
- Luscher M.S.,
- Clemmensen P.,
- Thygesen K.,
- Grande P.,
- The TRIM Study Group
- Kaul P.,
- Fu Y.,
- Chang W.C.,
- et al.,
- PARAGON-A and GUSTO IIb Investigators
- Kaul P.,
- Newby L.K.,
- Fu Y.,
- et al.
- Ottervanger J.P.,
- Armstrong P.,
- Barnathan E.S.,
- et al.
- James S.K.,
- Lindahl B.,
- Siegbahn A.,
- et al.
- Braunwald E.,
- Antman E.M.,
- Beasley J.W.,
- et al.
- Holmvang L.,
- Clemmensen P.,
- Wagner G.,
- Grande P.
- Ruilope L.M.,
- van Veldhuisen D.J.,
- Ritz E.,
- Luscher T.F.
- Omland T.,
- Persson A.,
- Ng L.,
- et al.
- Lindahl B.,
- Toss H.,
- Siegbahn A.,
- Venge P.,
- Wallentin L.,
- FRISC Study Group
- Heeschen C.,
- Hamm C.W.,
- Bruemmer J.,
- Simoons M.L.,
- CAPTURE Investigators
- Lenderink T.,
- Boersma E.,
- Heeschen C.,
- et al.
- Malmberg K.,
- Yusuf S.,
- Gerstein H.C.,
- et al.
- Lagerqvist B.,
- Safstrom K.,
- Stahle E.,
- Wallentin L.,
- Swahn E.,
- FRISC II Study Group,
- FRISC II Study Group Investigators
- Kaul P.,
- Newby L.K.,
- Fu Y.,
- et al.