Author + information
- Received January 18, 2011
- Revision received April 14, 2011
- Accepted May 20, 2011
- Published online August 30, 2011.
- Vikas Veeranna, MD⁎,†,
- Sandip K. Zalawadiya, MD⁎,†,
- Ashutosh Niraj, MD, MS⁎,†,
- Jyotiranjan Pradhan, MD⁎,†,
- Brian Ference, MD, MPhil⁎,†,
- Robert C. Burack, MD, MPH†,
- Sony Jacob, MD⁎,† and
- Luis Afonso, MD⁎,†,⁎ ()
- ↵⁎Reprint requests and correspondence:
Dr. Luis Afonso, Internal Medicine, Division of Cardiology, Wayne State University, Detroit Medical Center, 8 Brush, Harper University Hospital, 3990 John R, Detroit, Michigan 48201
Objectives The purpose of this study was to examine whether adding homocysteine (Hcy) to a model based on traditional cardiovascular disease (CVD) risk factors improves risk classification.
Background Data on using Hcy to reclassify individuals in various risk categories beyond traditional approaches have not been adequately scrutinized.
Methods We performed a post hoc analysis of the MESA (Multi-Ethnic Study of Atherosclerosis) and NHANES III (National Health and Nutrition Examination Survey III) datasets. Hcy was used to predict composite CVD and hard coronary heart disease (CHD) events in the MESA study and CVD and CHD mortality in the NHANES III survey using adjusted Cox-proportional hazard analysis. Reclassification of CHD events was performed using a net reclassification improvement (NRI) index with a Framingham risk score (FRS) model with and without Hcy.
Results Hcy level (>15 μmol/l) significantly predicted CVD (adjusted hazard ratio [aHR]: 1.79, 95% confidence intervals [CI]: 1.19 to 1.95; p = 0.006) and CHD events (aHR: 2.22, 95% CI: 1.20 to 4.09; p = 0.01) in the MESA trial and CVD (aHR: 2.72, 95% CI: 2.01 to 3.68; p < 0.001) and CHD mortality (aHR: 2.61, 95% CI: 1.83 to 3.73; p < 0.001) in the NHANES III, after adjustments for traditional risk factors and C-reactive protein. The level of Hcy, when added to FRS, significantly reclassified 12.9% and 18.3% of the overall and 21.2% and 19.2% of the intermediate-risk population from the MESA and NHANES cohorts, respectively. The categoryless NRI also showed significant reclassification in both MESA (NRI: 0.35, 95% CI: 0.17 to 0.53; p < 0.001) and NHANES III (NRI: 0.57, 95% CI: 0.43 to 0.71; p < 0.001) datasets.
Conclusions From these 2 disparate population cohorts, we found that addition of Hcy level to FRS significantly improved risk prediction, especially in individuals at intermediate risk for CHD events.
Cardiovascular disease (CVD) is the leading cause of mortality in the United States, accounting for more than one-third of all deaths (1). Although traditional risk factors account for most of the CVD risk, prediction models such as the Framingham risk score (FRS) are inherently limited in their ability to discriminate among individuals who will or will not experience adverse CVD events (2,3). Evidently, significant residual risk for CVD exists in the population independent of these risk factors (4,5).
To better comprehend and eliminate this residual risk, hosts of novel risk factors and biomarkers have been studied to date. Some of these emerging markers, such as C-reactive protein (CRP), an inflammatory biomarker (6,7); concentration of coronary artery calcium (CAC) (8); and level of homocysteine (Hcy), an amino acid that can be easily and reliably measured in human plasma (5,9), have shown promise (5,10,11). Indeed, several follow-up and case-control studies, as well as meta-analyses, have attempted to profile the risk conferred by Hcy level with inconsistent results. Of note, the large majority of these studies stemmed from predominantly Caucasian cohorts or individuals with preexisting cardiac disease (11–14).
To date, no study has systematically explored the predictive value of Hcy beyond existing risk predicted by the FRS (particularly in individuals without overt CVD) or assessed whether Hcy level contributes to reclassification of individuals in the intermediate-risk FRS category. These deficiencies were highlighted in the recent U.S. Preventive Services Task Force (USPSTF) recommendation statement for nontraditional risk factors in coronary heart disease (CHD) risk assessment (5).
Accordingly, we sought to define: 1) the predictive value of Hcy levels in predicting CV outcomes in a community-based, multiethnic, healthy U.S. adult population and a population cohort representative of U.S. adults; and 2) the incremental value, if any, of adding Hcy level to the FRS for CHD risk prediction.
The MESA Study
MESA (Multiethnic Study of Atherosclerosis) was a population-based study (n = 6,814) of individuals belonging to various ethnicities, ranging between 45 and 84 years of age at study enrollment, without a prior history of clinical CVD. After obtaining institutional review board approval, we undertook a post hoc analysis of the MESA Limited Access dataset, obtained from the National Heart, Lung, and Blood Institute. A detailed description of the study design, methods, and objectives has been published previously (15). In brief, the MESA study was designed to identify the characteristics of subclinical CVD and risk factors that predict progression to clinically overt CVD or progression of the subclinical disease. After excluding patients with missing data, we identified 6,450 adults.
Plasma Hcy levels were measured using a fluorescence polarization immunoassay (IMx Hcy assay, Axis Biochemicals ASA, Oslo, Norway) with the IMx analyzer (Abbott Diagnostics, Abbott Park, Illinois). High-sensitivity C-reactive protein (hs-CRP) was measured using a BNII nephelometer (Dade Behring, Deerfield, Illinois). Assays were performed exclusively at the Biochemical Genetics Clinical Laboratory, Fairview-University Medical Center (Minneapolis, Minnesota). Details of blood sample collection and analyses are provided elsewhere (16).
Demographic information was obtained using standard questionnaires. The endpoints were defined per the MESA steering committee protocol. All CVD events and hard CHD events were considered the primary outcomes in separate survival analyses. All CVD events was a composite outcome consisting of myocardial infarction, resuscitated cardiac arrest, definite angina, probable angina (if followed by revascularization), stroke, stroke death, CHD death, other atherosclerotic death, and other CVD death. A hard CHD event was defined as myocardial infarction, resuscitated cardiac arrest, or CHD death.
The NHANES III Study
NHANES III (National Health and Nutrition Examination Survey III) was a longitudinal study of noninstitutionalized nationally representative individuals surveyed between 1988 and 1994. The protocols for the NHANES III study were approved by the National Center for Health Statistics ethics/institutional review board, and all participants who underwent standardized interviews, physical examinations, and various laboratory measurements signed an informed consent (17).
Hcy levels were measured using the high-performance liquid chromatography method as described by Araki and Sako (18). CRP was measured using latex-enhanced nephelometry in the NHANES III population (not hs-CRP). Detailed descriptions of the data collection and laboratory methods have been described elsewhere (19). It is worth noting that different biochemical assays were used to measure serum Hcy levels in the MESA and NHANES populations; however, earlier research has shown that no significant differences exist in their analytic performance (20).
Mortality data were ascertained using the National Death Index and are available through the National Center for Health Statistics record linkage program. The primary outcome of interest was mortality due to CVD and CHD. CVD mortality was defined as death due to ischemic heart disease, cerebrovascular disease, and atherosclerotic heart disease, whereas CHD mortality included death due to ischemic heart disease only (21).
Distribution of baseline characteristics across Hcy categories (category 1: <10 μmol/l; 2: 10 to 14.9 μmol/l; and 3: ≥15 μmol/l) was compared using the chi-square test for categorical variables (%) and 1-way analysis of variance for continuous variables (mean ± SD). All covariates were tested for normality by visual inspection using frequency distribution curves. Log transformations were performed to normalize Hcy and CRP levels. To assess the predictive role of Hcy, adjusted multivariable Cox proportional hazard models were generated as follows: model 1: Hcy + age, sex, and race; model 2: model 1 + body mass index (BMI), systolic blood pressure, current smoking, high-density lipoprotein cholesterol, total cholesterol, lipid-lowering therapy, diabetes, and use of antihypertensive therapy; model 3 included variables in model 2 and log(CRP); and model 4 included variables in model 2 and creatinine. Assumption of proportionality for all of the models was tested using Schoenfeld residuals, which was found to be statistically nonsignificant. Furthermore, we also performed formal statistical analyses for the interaction between sex and Hcy and ethnicity and Hcy to evaluate for any effect modification. We found the interaction to be statistically nonsignificant, suggesting no sex or ethnic variation in the predictive value of Hcy on CV outcomes. Hazard ratios were also estimated for CRP using similar analyses.
We then assessed whether the addition of Hcy level to FRS (calculated for each individual using adult treatment panel III guidelines ) resulted in better predictive accuracy for CHD events. To assess discrimination, which provides information about whether the new marker has an ability to differentiate among individuals who do and do not experience an event, we generated area under the receiver operating characteristics (AU-ROC) curves for models with and without Hcy. We also calculated the Harrell c-statistic, which allows censored data and permits appraisal of the predictive accuracy for the model with and without Hcy. Similar analysis was performed for CRP. Considering the drawbacks of ROC curves, which need a large odds ratio for even a small improvement, we further assessed measures that may have clinical impact by reclassification, such as the net reclassification index (NRI) and integrated discrimination index (IDI) (23,24).
The estimated risk for 6 years of follow-up in the MESA cohort was calculated using multivariate logistic analysis. Because the FRS provides estimates for 10-year risk, we divided the risk estimates by 1.67 to obtain 6-year estimated risk in the MESA cohort. This approximation might result in a slightly higher incidence rate for a given age and risk factor level at entry, but it allowed us to compare the primary outcome variable during a defined period of observation instead of extending it beyond the actual follow-up duration (25). Recalibrated risk categories for 6-year estimated Framingham risk were as follows: very low: <3%, low: 3% to <6%, intermediate: 6% to 12%, and high: >12%, based on the third report of the National Cholesterol Education Program Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults Framingham point scoring (22).
The estimated risk for 10 years of follow-up, adjusted from 15 years of follow-up, in the NHANES cohort was also calculated using multivariate logistic analysis. The following risk categories were analyzed: very low: <5%, low: 5% to <10%, intermediate: 10% to <20%, and high: >20%, as described earlier (22). The NRI for CHD events, as described by Pencina et al. (23), was calculated by comparing estimated risk probabilities derived from multivariate logistic models incorporating FRS with and without Hcy level.
Individuals were then compared based on their classifications by these 2 models. The NRI was calculated for the entire cohort and the intermediate-risk category. Statistical significance for all components of NRI and overall NRI was tested by calculating the Z score for discordance in a manner analogous to the McNemar test. Two-tailed p values <0.05 were considered statistically significant. We also calculated the IDI, which provides information about the probability differences in the discrimination slopes of events and nonevents by a new marker (23). It can be expressed as (EY1 − EX1) + (EX0 − EY0), in which EY1 and EY0 represent the mean expected probabilities of events and nonevents, respectively, for the model including the new marker, and EX1 and EX0 are the mean expected probabilities of events and nonevents, respectively, for the model without the new marker. Relative IDI, an index calculated when the incidence of events is relatively small, is defined as ([EY1 − EY0]/[EX1 − EX0]) − 1. We further analyzed both of the cohorts for improvement in “categoryless” NRI, which is a recently proposed analytic measure applicable to time to event data (24). As the name suggests, categoryless NRI does not depend on pre-defined risk categories and measures upward and downward reclassification as any change in predicted probabilities derived from the model without and with the new marker. Reclassification >0 for events (+1) and <0 for nonevents (+1) is classified as improvement, with the maximum possible value of 1 + 1 = 2 and the opposite as nonimprovement.
To assess calibration, which provides information about how closely the predicted probabilities of estimated risk by the novel biomarker reflect the actual observed risk, we performed a Hosmer-Lemeshow chi-square test for models with and without Hcy. A predictive model with calibration estimates of less than 20 is considered to be a model with adequate calibration. We also performed a −2 log likelihood ratio test and the Bayesian information criterion, which provides information about the probability that a given independent variable is a part of the true model, to assess the global fit of the models. All statistical analysis was performed using STATA version 10 (StataCorp, College Station, Texas).
Baseline characteristics for the MESA cohort
Table 1 summarizes the baseline characteristics of the 2 study cohorts. In the MESA cohort, there was a graded increase in mean age and systolic blood pressure and a decrease in total cholesterol and high-density lipoprotein cholesterol with increasing Hcy categories. Proportions of patients with hypertension or diabetes, smokers, and those using lipid-lowering therapy increased with increasing Hcy categories. In the NHANES cohort, similar results were found for age, systolic blood pressure, and patients with hypertension or diabetes, smokers, and those using lipid-lowering therapy. We found that mean total cholesterol and the proportion of men increased with increasing Hcy categories in the NHANES cohort.
Hcy and CV risk
Table 2 shows Hcy concentration as a predictor of CV events in the MESA cohort. As shown, when Hcy was evaluated as a continuous variable, each 0.1 log-unit increase in Hcy level was associated with an adjusted hazard ratio (aHR) of 1.87 (95% confidence interval [CI]: 1.29 to 2.69; p = 0.001) for all CVD events and 2.90 (95% CI: 1.69 to 4.95; p < 0.001) for hard CHD events. Higher Hcy level was also significantly associated with a higher incidence of all CVD (aHR: 1.79, 95% CI: 1.19 to 1.95; p = 0.006 in category III vs. I) and hard CHD events (category III vs. I: model 3 aHR: 2.22, 95% CI: 1.20 to 4.09; p = 0.011) after adjustments when evaluated as a categorical variable.
Table 3 shows Hcy as a predictor of CVD deaths in the NHANES cohort. As shown, when Hcy was evaluated as a continuous variable, each 0.1 log-unit increase in Hcy was associated with an aHR of 1.97 (95% CI: 1.52 to 2.55; p < 0.001) for CVD deaths and 1.95 [95% CI: 1.43 to 2.64; p < 0.001) for CHD deaths. Higher Hcy level was significantly associated with a higher incidence of CVD mortality (aHR: 2.72; 95% CI: 2.01 to 3.68; p < 0.001 in category III vs. I) and CHD mortality (category III vs. I: model 3 aHR: 2.61, 95% CI: 1.83 to 3.73; p < 0.001) when evaluated as a categorical variable.
Further, no attenuation of HRs was observed following adjustments for serum creatinine in either cohort. Upon similar analyses, CRP was not found to be a significant predictor of CVD/CHD events (data not shown).
Table 4 details the calibration properties of the model with a new marker (Hcy); as shown, the addition of Hcy level to FRS significantly improved Bayesian information criterion, along with a statistically significant likelihood ratio test.
Risk stratification and restratification
In the ROC analysis for the MESA cohort, the c-statistic for the model with FRS alone was 0.745, which increased significantly with the addition of Hcy level (c-statistic of 0.760; p for improvement <0.001). A similar magnitude of effect was observed in the NHANES cohort (c-statistic increased from 0.844 to 0.868 after Hcy level was added to FRS; p < 0.001). The AU-ROC curve also improved significantly after addition of Hcy level to FRS compared with the FRS alone for both MESA and NHANES cohorts (Table 4). There was no significant increment in either AU-ROC or c-statistic when CRP level was added to FRS in either cohort, and hence, newer reclassification indexes were not attempted (data not shown).
Upon examining the reclassification properties of Hcy, we observed an overall improvement in the net risk stratification of up to 12.9% (number of participants reclassified of overall sample was 832) in the MESA cohort and 18.3% (number of participants reclassified of overall sample was 1,243) in the NHANES cohort (p < 0.001 for both). For individuals at intermediate risk of events, the net reclassification was 21.2% in the MESA cohort and 19.1% in the NHANES cohort (p < 0.001 for both) (Tables 5 and 6).⇓⇓ Overall, the absolute IDI for the MESA cohort was 0.006 (p for improvement <0.001) and 0.015 (p for improvement <0.001) for the NHANES cohort, with relative IDIs of 46% and 33% in the MESA and NHANES cohorts, respectively.
The categoryless NRI also improved significantly for both the MESA (0.35, 95% CI: 0.17 to 0.53; p < 0.001) and NHANES (0.57, 95% CI: 0.43 to 0.71; p < 0.001) cohorts; the improvement was largely driven by downward reclassification of nonevents (83% in the MESA and 67% in the NHANES cohorts).
To our knowledge, these observations represent the only analyses of Hcy and CV risk in 2 large, low-risk, multiethnic, representative U.S. populations to date. Our results lend support to previously published data exploring the association between Hcy and CVD and CHD events. Of note, unlike any previous study, our analyses were adjusted for CRP, an independent predictor of future CVD events (11,13,26,27).
Hcy and CV risk assessment
The FRS has been characterized in relation to Hcy level with variable results. In a study by Kullo et al. (28), Hcy was significantly associated with CAC in individuals with intermediate risk (6% to 20%) and was shown to be highly predictive of CV mortality in the very elderly, independent of conventional risk factors (29). On the other hand, a more recent study by Wilson et al. (6) showed that use of Hcy level in CVD risk assessment did not improve predictive accuracy for discrimination of events in the Framingham Offspring Study population. In contrast, Hcy level significantly improved risk prediction beyond the FRS in our study involving 2 disparate, large, multiethnic U.S. populations.
Data on the impact of Hcy level when added to the FRS to reclassify individuals at intermediate risk are essentially lacking (5,30). Numerous recent reports employing novel risk metrics such as brachial flow-mediated dilation, CAC, and CRP have attempted to reclassify CVD risk beyond the FRS, with mixed results. Compared with the FRS alone, the addition of flow-mediated dilation (25) and CAC (31) to the FRS has resulted in NRI values of 29% and 28% and 25% and 55% in overall study samples and subjects at intermediate risk, respectively. In this context, our results are comparable in magnitude and are encouraging, with an overall NRI of 12.9% and 18.3% and for intermediate-risk category NRI values of 21.2% and 19.1% for MESA and NHANES cohorts, respectively. Furthermore, upon analyzing Hcy level for improvement in the categoryless NRI, a measure that is applicable to time to events data and is independent of pre-defined risk categories, we observed significant reclassification in both cohorts (Table 4). This is particularly important when the relative ease of obtaining an Hcy level is juxtaposed against radiation exposure and the expense involved in obtaining an individual's CAC score (31). Of note, our analysis in the MESA and NHANES cohorts showed that hs-CRP and CRP, respectively, were not independent predictors of CHD events, with no significant increment in AU-ROC or c-statistic. These findings are in sharp contrast to that of a previous study using hs-CRP, which reported an NRI of 5.6% for CVD and 11.8% for hard CHD events (6).
Homocysteine: biomarker or risk factor?
Current guidelines do not support the use of Hcy level for CVD risk stratification. Indeed, a debate as to whether Hcy should be viewed as a risk factor or a biomarker still exists. Observational studies have revealed an association between plasma Hcy and CVD events, whereas several randomized controlled trials and a recent meta-analysis have shown that Hcy-lowering interventions did not reduce the occurrence of CVD events, especially CHD (32–37). However, it should be noted that although the duration of observational studies exceeded a decade, most of the interventional trials have had a limited follow-up of <5 years. Such truncated follow-up periods may not permit definitive comments on risk reduction, particularly when one considers the natural history of atherosclerosis, a process that could take decades to progress from plaque formation to a clinical event (32). Further, the majority of these interventional trials recruited patients with pre-existing CVD or a risk equivalent (34). This bias is reflected in the recent USPSTF statement on screening for intermediate risk factors for CHD: “all trials to date have been of tertiary prevention and conducted among individuals with prevalent CHD, cerebrovascular disease or diabetes. Whether treatment of elevated Hcy levels before an individual develops vascular disease will be beneficial is not resolved by these trials of tertiary prevention” (30). The USPSTF guidelines also noted that “no studies calculated a Framingham risk score, assessed predictive value beyond Framingham risk scoring, or assessed whether homocysteine levels contribute to reclassification from intermediate to another risk category” (5). The inference from the current guidelines is that significant uncertainties in the applicability of Hcy concentration for risk prediction of individuals at intermediate risk exist; in this context, our results provide compelling and substantive evidence filling this void in knowledge.
The USPSTF also pointed out that current guidelines fail to account for differences in risk prediction among racial and ethnic groups (5). For example, African Americans and Hispanics may need more aggressive interventions for a particular risk score because of their high absolute risk of CVD events relative to whites, who accounted for most of the original Framingham cohort (38). In contradistinction, FRS has been shown to overestimate the risk of CHD among Chinese populations (39). Based on the available evidence, it appeared imperative to investigate the effect modification by ethnicity on the association between Hcy and CV outcomes. We found that the interaction term between ethnicity and Hcy level was statistically nonsignificant for all CV outcomes in both the NHANES and MESA cohorts, a noteworthy finding suggesting that ethnicity might not play a significant role when an individual's risk for future CV events is evaluated based on serum Hcy levels.
In these analyses, we prospectively validated and showed the incremental value of Hcy level in predicting adverse CVD events beyond the FRS, thereby fulfilling the criteria set for a “novel” marker (40). Although specific interventions aimed at lowering Hcy levels in individuals with pre-existing CVD do not appear to be beneficial, similar interventions pursued as part of a primary prevention strategy are yet to be evaluated at this time. Because aggressive risk reduction in individuals classified as high risk has been shown to be beneficial (5), approaching individuals as high risk, based on the reclassification using Hcy, may also seem reasonable. Along these lines, it would seem imprudent to discount the utility of Hcy in CVD risk prediction solely because interventions to lower plasma Hcy levels do not appear to favorably modulate the risk of incident CVD. Although prospective research in this arena is clearly warranted, elevated Hcy level should be construed in the interim more as a biomarker portending elevated CVD risk rather than a causally related “modifiable risk factor” (32).
A single sample of Hcy was measured at baseline in both the MESA and NHANES III cohorts. Hcy levels are subject to variation based on food intake, diurnal changes, and position during blood draw; these variations might have led to nondifferential misclassification and attenuation in effect sizes (9). Low cobalamin or folate levels, which were not accounted for, could lead to elevated Hcy (9). The possibility of unmeasured confounders and residual confounding affecting study results cannot be ruled out—such issues are inherent limitations of observational cohort studies.
Our observations are novel and demonstrated that elevated Hcy concentration predicts future CVD and CHD events in disparate, representative populations of U.S. adults across ethnic subgroups. Our findings revealed that plasma Hcy levels enhance risk prediction when added to the FRS, enabling the reclassification of a significant number of individuals at “intermediate risk” of CHD events.
The MESA trial was conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with the MESA Study Investigators. This paper was prepared using the limited access dataset obtained from the NHLBI and does not necessarily reflect the opinions or views of the MESA trial or the NHLBI. The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
A part of this analysis was presented (oral presentation) at the Annual Scientific Sessions of the American Heart Association, November 13 to 17, 2010, Chicago, Illinois.
- Abbreviations and Acronyms
- coronary artery calcium
- coronary heart disease
- C-reactive protein
- cardiovascular disease
- Framingham risk score
- high-sensitivity C-reactive protein
- integrated discrimination index
- net reclassification index
- U.S. Preventive Services Task Force
- Received January 18, 2011.
- Revision received April 14, 2011.
- Accepted May 20, 2011.
- American College of Cardiology Foundation
- Lloyd-Jones D.,
- Adams R.,
- Carnethon M.,
- et al.
- Wilson P.W.,
- D'Agostino R.B.,
- Levy D.,
- Belanger A.M.,
- Silbershatz H.,
- Kannel W.B.
- Smith S.C. Jr..
- Wilson P.W.,
- Pencina M.,
- Jacques P.,
- Selhub J.,
- D'Agostino R.,
- O'Donnell C.J.
- Refsum H.,
- Smith A.D.,
- Ueland P.M.,
- et al.
- Sacco R.L.,
- Anand K.,
- Lee H.S.,
- et al.
- Tanne D.,
- Haim M.,
- Goldbourt U.,
- et al.
- Bild D.E.,
- Bluemke D.A.,
- Burke G.L.,
- et al.
- MESA Coordinating Center
- ↵Centers for Disease Control and Prevention (CDC): National Health and Nutrition Examination Survey: NCHS research ethics review board (ERB) approval. http://www.cdc.gov/nchs/nhanes/irba98.htm. Accessed September 15, 2010.
- ↵Centers for Disease Control and Prevention (CDC): National Health and Nutrition Examination Survey: NHANES III. http://www.cdc.gov/nchs/nhanes/nh3data.htm. Accessed July 5, 2011.
- ↵Centers for Disease Control and Prevention (CDC): Data Access: NHANES III linked mortality file. http://www.cdc.gov/nchs/data_access/data_linkage/mortality/nhanes3_linkage.htm. Accessed July 5, 2011.
- National Cholesterol Education Program (NCEP) Expert Panel on Detection Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III)
- Yeboah J.,
- Folsom A.R.,
- Burke G.L.,
- et al.
- de Ruijter W.,
- Westendorp R.G.,
- Assendelft W.J.,
- et al.
- Helfand M.,
- Buckley D.,
- Fleming C.,
- et al.
- Marti-Carvajal A.J.,
- Sola I.,
- Lathyris D.,
- Salanti G.
- Hennekens C.H.,
- D'Agostino R.B.
- Hlatky M.A.,
- Greenland P.,
- Arnett D.K.,
- et al.