Author + information
- Received May 18, 2011
- Revision received August 2, 2011
- Accepted August 12, 2011
- Published online November 8, 2011.
- Andrew P. DeFilippis, MD, MSc⁎,†,
- Michael J. Blaha, MD, MPH⁎,
- Chiadi E. Ndumele, MD⁎,
- Matthew J. Budoff, MD‡,
- Donald M. Lloyd-Jones, MD, ScM§,
- Robyn L. McClelland, PhD∥,
- Susan G. Lakoski, MD, MS¶,
- Mary Cushman, MD#,
- Nathan D. Wong, PhD⁎⁎,
- Roger S. Blumenthal, MD⁎,
- Joao Lima, MD, MBA†† and
- Khurram Nasir, MD, MPH‡‡,⁎ ()
- ↵⁎Reprints requests and correspondence:
Dr. Khurram Nasir, Blalock 524C–Ciccarone Center, Johns Hopkins Hospital, 600 North Wolfe Street, Baltimore, Maryland 21287
Objectives The purpose of this study was to compare the association of the Framingham risk score (FRS) and Reynolds risk score (RRS) with subclinical atherosclerosis, assessed by incidence and progression of coronary artery calcium (CAC).
Background The comparative effectiveness of competing risk algorithms for identifying subclinical atherosclerosis is unknown.
Methods MESA (Multi-Ethnic Study of Atherosclerosis) is a prospective cohort study of 6,814 participants free of baseline cardiovascular disease. All participants underwent risk factor assessment, as well as baseline and follow-up CAC testing. We assessed the performance of the FRS and RRS to predict CAC incidence and progression using relative risk and robust linear regression.
Results The study population included 5,140 individuals (mean age 61 ± 10 years, 47% males, mean follow-up: 3.1 ± 1.3 years). Among 53% of subjects (n = 2,729) with no baseline CAC, 18% (n = 510) developed incident CAC. Both the FRS and RRS were significantly predictive of incident CAC (relative risk: 1.40 [95% confidence interval (CI): 1.29 to 1.52] and 1.41 [95% CI: 1.30 to 1.54] per 5% increase in risk, respectively) and CAC progression (mean CAC score change: 6.92 [95% CI: 5.31 to 8.54] and 6.82 [95% CI: 5.51 to 8.14] per 5% increase). Discordance in risk category classification (<10% or >10% per 10-year coronary heart disease risk) occurred in 13.7%, with only the RRS consistently adding predictive value for incidence and progression of CAC. These subclinical atherosclerosis findings are supported by a coronary heart disease events analysis over a mean follow-up of 5.6 ± 0.7 years.
Conclusions Both the RRS and FRS predict onset and progression of subclinical atherosclerosis. However, the RRS may provide additional predictive information when discordance between the scoring systems exists.
- calcium progression
- coronary artery
- Framingham risk score
- Reynolds risk score
- risk prediction
- subclinical atherosclerosis
Coronary heart disease (CHD) is the leading cause of death for men and women in the United States (1). The National Cholesterol Education Program (NCEP) Adult Treatment Panel III (ATP-III) guidelines recommend that all adults undergo an office-based assessment to evaluate their risk of CHD events, based on the Framingham risk score (FRS) (1). This widely used prediction algorithm incorporates age, sex, levels of total cholesterol, high-density lipoprotein cholesterol, smoking status, systolic blood pressure, and antihypertensive therapy to estimate a 10-year risk of a myocardial infarction (MI) or death occurring as a result of CHD. The ATP-III guidelines set thresholds for lipid treatment based on the 10-year CHD risk estimated by the FRS.
In an effort to improve cardiovascular disease (CVD) risk prediction, the Reynolds risk score (RRS) was derived in a cohort of 25,000 healthy U.S. women (2). The RRS includes traditional risk factors used in the FRS and adds parental family history of premature CHD and high-sensitivity C-reactive protein. This, and an RRS calibrated for men, provided superior prediction of CVD events compared with the FRS model in 2 studies (2,3). Notably, the RRS was developed and tested in predominantly non-Hispanic white populations.
The underlying pathophysiology of most CHD events, atherosclerosis, now can be measured noninvasively, including via quantification of coronary artery calcification (CAC) as assessed by computed topography. Identification of individuals with subclinical atherosclerosis, via CAC, has been shown to predict future cardiovascular events in multiple asymptomatic populations, including the young (<40 years of age), the middle aged (50 to 70 years of age), the elderly (>70 years of age), men, and women (4–12). Serial CAC measurements may provide information on disease progression; importantly, CAC progression has been shown to be a powerful predictor of future CVD events (4,13–15).
Clinical CHD risk assessment scores correlate with the presence and burden of atherosclerosis identified by these noninvasive methods (14). However, recent studies have suggested that the FRS may misclassify both subclinical and clinical CHD risk in some individuals (16,17). These inaccuracies may be particularly pronounced among women, who rarely have a risk estimate high enough to meet ATP-III treatment thresholds. As a result, some individuals who do not qualify for primary prevention pharmacotherapies in fact may have advanced subclinical atherosclerosis (18,19).
We evaluated the associations of the FRS and RRS with incident and progression of subclinical atherosclerosis, as estimated by CAC, in nondiabetic MESA (Multi-Ethnic Study of Atherosclerosis) participants with emphasis on instances where the scoring systems provided discordant results. The MESA cohort is ideal for studying these relationships because of its large size, sex balance, multiethnic composition, serial measures of subclinical atherosclerosis, and rigorous assessment of CHD events.
The MESA is a prospective epidemiologic study of the prevalence, risk factors, and progression of subclinical cardiovascular disease in a sex-balanced, multiethnic cohort. The study design and methods have been published previously (20). Briefly, 6,814 participants 45 to 84 years of age who identified themselves as white, African American, Hispanic, or Chinese were recruited from 6 U.S. communities from 2000 through 2002. Participants were free of clinical cardiovascular disease (MI, angina, stroke, transient ischemic attack, heart failure, atrial fibrillation, revascularization, valve replacement, pacemaker or defibrillator implantation, or taking nitroglycerin) at enrollment. All participants gave informed consent, and the study protocol was approved by the institutional review board at each site.
Medical history, anthropometric measurements, and laboratory data for the present analysis were assessed as described previously (20). Parental history of CHD was considered premature according to NCEP ATP-III guidelines (age <55 years in men, >65 years in women).
FRS was calculated in accordance with NCEP ATP-III guidelines (1). Because the NCEP ATP-III FRS calculation does not include adults with diabetes (1), participants with diabetes were not included in this study. Additionally, the FRS and RRS algorithms are applicable only for those younger than 80 and 81 years, respectfully; therefore, participants 80 (or 81) to 85 years were assigned a risk as though they were 79 (or 80) years of age. Otherwise, the sex-specific RRS was calculated as described by Ridker et al. (2,3).
Measurement of coronary artery calcium
Coronary artery calcium was measured using either electron beam computed tomography (3 sites) or multidetector computed tomography (3 sites). Participants were scanned twice consecutively, and each scan was read by a single trained physician–reader (M.J.B.) independently at a centralized reading center (Harbor-UCLA Medical Center/Los Angeles Biomedical Research Institute, Torrance, California). The methodology for acquisition and interpretation of the scans, as well as reproducibility of the readings, has been reported previously (21). The results from the 2 scans were averaged to provide a more accurate point estimate of the amount of calcium present.
Calcium scores were adjusted using a standard calcium phantom that was scanned along with the participant. The phantom contained 4 bars of known calcium density and was used to calibrate the x-ray attenuation level between measurements conducted on different machines (22). The presence of CAC was defined as a volume score of more than 0, and a minimum focus of calcification was based on at least 4 contiguous voxels, which resulted in identification of calcium of 1.15 mm3 for the multidetector computed tomography scanners and 1.38 mm3 for the electron beam computed tomography scanners. The nominal section thickness was 3.0 mm for electron beam computed tomography scanners and 2.5 mm for multidetector computed tomography.
To quantify CAC progression, a second CAC measurement was performed on approximately one-half of the cohort (randomly selected) at a second examination (September 2002 through February 2004) and on most of the remaining subjects at a third examination (March 2004 through October 2005). A small portion of subjects underwent their second CAC measurement at examination 4 (October 2005 through February 2008), whereas approximately one-quarter of participants (randomly selected) underwent a third CAC measurement at examination 4. CAC progression was calculated between the baseline and latest scan date, with an average time between scans of 3.1 ± 1.3 years. CAC progression is known to be highly dependent on time between scans. Given the difference in follow-up duration between subjects, we adjusted for time between scans in our regression model to factor out the importance of this variable.
CHD events consisted of MI, death from CHD, definite angina, probable angina followed by coronary revascularization, or resuscitated cardiac arrest. Events were recorded over a median follow-up of 5.8 years. At intervals of 9 to 12 months, an interviewer contacted each participant or a family member by telephone to inquire about interim hospital admissions, outpatient diagnoses of CHD, and deaths. To verify self-reported diagnoses, copies were requested of all death certificates and medical records for all hospitalizations and outpatient cardiovascular diagnoses. Next-of-kin interviews for out-of-hospital cardiovascular deaths were obtained. The MESA was successful in obtaining medical records for approximately 98% of hospitalized CHD events and information on 95% of outpatient cardiovascular diagnostic encounters. Follow-up telephone interviews were completed for 92% of living participants.
Trained personnel abstracted data from medical records reporting possible cardiovascular events. Two physician members of the MESA mortality and morbidity review committee independently classified each event, and if there was disagreement, the full committee made the final classification. The diagnosis of myocardial infarction was based on a combination of symptoms, electrocardiography findings, and levels of cardiac biomarkers. The adjudicators graded angina as definite, probable, or absent based on clinical judgment. Angina required documentation of symptoms distinct from the diagnosis of MI. A classification of definite angina additionally required objective evidence of reversible myocardial ischemia or obstructive coronary artery disease. The reviewers classified deaths resulting from CHD as present or absent based on hospital records and interviews with families. A more detailed description of the MESA follow-up methods is available online (23).
We used chi-square tests for categorical variables and the t test or analysis of variance for continuous variables to assess for baseline differences in demographics and cardiovascular risk factors between participants across FRS and RRS categories (<10% and ≥10%).
Changes in CAC were defined in 2 ways, as previously described by Kronmal et al. (24): incident CAC, defined as detectable CAC at the follow-up examination (examination 2, 3, or 4) in a participant free of CAC at baseline; and change in CAC Agatston score in participants who had detectable CAC at examination 1. The 2 CAC endpoints were modeled separately. The probability of incident CAC was modeled as a function of covariates using a generalized linear model with log link and binomial error distribution (relative risk regression). Relative risk regression was used rather than logistic regression because the incidence of new calcification was >10%, so the odds ratio would overestimate the relative risk. To estimate the progression of CAC among those with detectable CAC at baseline, we used multivariate-adjusted robust linear regression. We performed a robust regression using iteratively reweighted least squares, that is, we assigned a weight to each observation, with higher weights given to better behaved observations. Models were adjusted for race, MESA site, and follow-up duration.
The practical clinical application of cardiovascular risk score calculation is to categorize individuals into risk categories. Numerous national consensus guideline recommendations are based on risk categories, and therefore individuals who have discordant classification based on the risk prediction model used would be impacted most by which prediction model was used to make guideline-driven treatment recommendations. Therefore, the groups of individuals who were classified into a different risk category, <10% or >10% risk, depending on which risk prediction model was applied (discordant risk score prediction), were evaluated for a difference in the study outcome variables (incident CAC and CAC progression). If the discordant risk prediction group predicted more or less risk than the risk category for which it was being compared (the concordant higher or lower risk group) and revealed more or less incident CAC, CAC progression, or both than predicted by the risk prediction model being evaluated, then we concluded that the model being evaluated added additional predictive information.
Hazard ratios for incident CHD events were calculated using Cox proportional hazards regression per 5% increase in risk prediction score for both the RRS and FRS. In a similar fashion, hazard ratios were calculated for incident CHD events among those with a risk prediction score of <10% versus >10% using both the RRS and the FRS. Hazard ratios for incident CHD among those with discoordinate risk score classification were calculated using RRS <10% and FRS <10% as the reference group. The overall diagnostic performance of each risk prediction model to predict CHD as a continuous variable or dichotomous variable was evaluated by comparing c-indexes for each model.
All statistical analyses were completed using Stata software version 9 (StataCorp, College Station, Texas).
Of the 6,814 MESA participants, 767 were excluded because of missing follow-up CAC testing, 684 because of baseline diabetes, and 223 because of a missing variable of interest. The final study population included 5,140 individuals with characteristics shown in Table 1 (mean age 61 ± 10 years, 47% males, mean follow-up 3.1 ± 1.3 years).
Overall, 53% (n = 2,729) had no detectable CAC at baseline, of which 18% (n = 510) had incident CAC at follow-up examinations. Of those with no detectable CAC at baseline, 2,244 (82%) had FRS <10% and RRS <10%, 109 (4%) had FRS <10% and RRS ≥10% 158 (6%) had FRS ≥10% and RRS <10%, and 218 (8%) had FRS ≥10% and RRS ≥10%.
The FRS and the RRS were associated with incidence and progression of CAC of similar magnitude and statistical significance (Tables 2 and 3)⇓⇓ when assessed as both continuous (per 5% increase in predicted event rate) and dichotomized (>10% or <10% risk) variables. Stratification of these results by race reveled relative risk ratios for incident CAC that were similar to those of the total population; additionally, the interaction term for race was not significant in multiple analyses (Table 2). When stratified by sex, the relative risk for developing CAC for those classified as intermediate to high risk (>10%) compared with those classified as low risk (<10%) by the FRS was 2.41 (95% confidence interval [CI]: 1.57 to 3.72) for women and 1.62 (95% CI: 1.16 to 2.27) for men (p = 0.07 for interaction). The respective relative risks by the RRS were 2.50 (95% CI: 1.63 to 3.41) for women and 2.40 (95% CI: 1.68 to 3.41) for men (p = 0.30 for interaction) (Table 2). Similar sex-specific findings were observed for both risk prediction scores when assessed as a continuous variable (Table 2).
Among those with baseline CAC >0, the mean change in CAC score consistently was greater among white persons as compared with Chinese, African Americans, and Hispanics when evaluated as a continuous or dichotomized variable (more or less than a risk prediction of 10% by RRS or FRS) (Table 3). When stratified by sex, the mean incremental change in CAC score for those classified as intermediate to high risk (>10%) compared with those classified as low risk (<10%) by the FRS was 8.69 Agatston units (AU) (95% CI: 0.40 to 16.89) for women and 18.79 AU (95% CI: 10.85 to 26.66) for men (p = 0.39 for interaction). The respective mean incremental change in CAC score by the RRS was 14.33 AU (95% CI: 6.92 to 21.73) for women and 23.38 AU (95% CI: 15.43 to 31.31) for men (p = 0.72 for interaction) (Table 3). Differences in the prediction of CAC progression by sex were not observed when applying the FRS or FRS as a continuous variable (Table 3).
Discordance in risk category classification (>10% or <10% CHD risk) between the FRS and the RRS occurred in 13.7% of participants. Of those classified with predicted risk of <10% by the FRS, 369 (7.2%) were classified as intermediate to high risk (>10%) by the RRS; these participants had a relative risk of 2.41 (95% CI: 1.57 to 3.72) of having incident CAC compared with those classified as low risk (<10%) by both scoring systems (Table 4). These subjects also had a CAC progression score of 19 AU higher (95% CI: 12 to 26) as compared with those who were classified as low risk by both scoring systems (Table 4). Of those classified with predicted risk >10% by the FRS, 336 (6.5%) were classified low risk (<10%) by the RRS; these participants had a relative risk of 0.54 (95% CI: 0.33 to 0.88) of having incident CAC compared with those classified as intermediate to high risk (>10%) by both scoring systems (Table 4). These subjects also had a CAC progression score of 16 AU lower (95% CI: −30 to −1) as compared with those who were classified as intermediate to high risk by both scoring systems (Table 4).
Of those classified with predicted risk <10% by RRS, 336 (6.5%) were classified as intermediate to high risk (>10%) by the FRS; these participants had a relative risk of 1.47 that did not meet statistical significance (95% CI: 0.98 to 2.21) for the development of incident CAC compared with those classified as low risk by both scoring systems (Table 4). However, these subjects had a CAC progression score of 11 AU higher (95% CI: 3 to 19) compared with those who were classified as low risk by both scoring systems (Table 4). Of those classified with predicted risk of more than 10% by the RRS, 369 (7.2%) were classified low risk (<10%) by the FRS; these participants had no statistical difference in progression of or development of incident CAC when compared with those classified as low risk by both scoring systems (−6 [95% CI: −18 to 7] progression; 0.86 [95% CI: 0.52 to 1.47] incident) (Table 4).
Participants with concordant risk prediction classification of intermediate to high risk (>10%) category by both scores had a higher odds ratio for incident CAC and mean change in CAC score than those classified as intermediate to high risk (>10%) by one risk prediction model, but low risk (<10%) by the other risk prediction model (Table 5).
A total of 135 participants (2.63%) had CHD events during a mean follow-up of 5.6 ± 0.7 years. Adjusting for race, the hazard ratios for any incident CHD event with every 5% increase in FRS and RRS were 1.42 (95% CI: 1.33 to 1.52) and 1.32 (95% CI: 1.27 to 1.41), respectively. The relative risk of CHD developing for those with intermediate to high risk (>10%) compared with those considered low risk (<10%) by the FRS was 3.62 (95% CI: 2.57 to 5.08). The respective relative risk for the RRS was 4.58 (95% CI: 3.24 to 6.48). The c-indexes for the FRS and RRS for predicting incident CHD events were not significantly different when assessed as either as a continuous variable (0.71 vs. 0.75, p = 0.27) or dichotomous variable (0.71 vs. 0.72, p = 0.21). Analysis of the 6.5% of subjects classified as low risk by the RRS but as high risk (>10%) by the FRS and the 7.2% of subjects classified as low risk by the FRS but as high risk (>10%) by the RRS was limited by the low absolute number of CHD events, but was consistent with the differences found between these 2 risk scores in the prediction of subclinical atherosclerosis, as assessed by incident and progression of CAC (Table 6). Excluding participants older than age 79 years or who did not report a parental history of CVD at examination 1, but who reported a positive family history at examination 2, did not change the results of the above analyses (data not shown).
In this large multiethnic cohort, both the RRS and the FRS are predictive of incidence and progression of subclinical atherosclerosis, as measured by CAC. Importantly, these results were supported by similar findings for CHD events, underscoring the principle that subclinical atherosclerosis is a precursor for CHD events and that the magnitude of this disease process can be assessed by CAC progression. When evaluated as a continuous variable, the RRS and FRS performed equally well in the prediction of incident CAC and CAC progression. However, considerable reclassification (13.7%) occurred among the risk models when classified into conventional risk categories, as is done in clinical practice. When classified into low- and high-risk to intermediate-risk groups, the RRS provided additional predictive power beyond the FRS, both for predicting incident CAC and progression of CAC. The converse was not true: the FRS did not consistently add predictive power beyond the RRS for predicting incident CAC, CAC progression, or incident CHD events. An analysis of discoordinate classification between the RRS and FRS for CHD events revealed similar results as those seen with incident and progression of CAC; however, this analysis was limited by the low number of CHD events currently present in this cohort.
Both the FRS and RRS include sex as a weighted variable. In our cohort, sex-specific analyses did not reveal a consistent difference in either risk score's capacity to predict progression or incident CAC based on sex. Both risk prediction scores performed equally across 4 race groups for predicting incident CAC, but predicted significantly greater CAC progression in whites as compared to Chinese, African Americans, or Hispanics—a finding the warrants validation in independent cohorts.
The reclassification observed in our cohort was considerably less than the reclassification found in the cohort used to develop and test the RRS; nondiabetic women classified as intermediate risk (10% to 20% 10-year risk of a major cardiovascular disease event) by the FRS would be reclassified with the RRS as follows: 21% of women would be reclassified as high risk (>20% risk) and 24% of women would be reclassified as low risk (<10% risk) (2). Importantly, both the analysis of the cohort used to develop the RRS and this analysis are limited by the fact that these 2 prediction models differ in that the RRS predicts the composite outcome of MI, ischemic stroke, coronary revascularization, and cardiovascular death, whereas the FRS was designed to predict MI and CHD death only. Additional limitations include an upper age limit in the FRS and RRS algorithms of 79 and 80 years, respectively; therefore, for participants 80 (or 81) to 85 years of age, we assigned them a risk as though they were 79 (or 80) years of age. Detailed family history was evaluated in the MESA at examination 2, whereas baseline CAC was evaluated at examination 1. However, exclusion of subjects older than 79 years or who did not report a CHD event in any first-degree relative at examination 1, but who had a positive family history at examination 2, did not change the results of the analysis. Finally, our definition of parental history of CHD is consistent with that of the NCEP ATP-III guidelines (age <55 years in men, <65 years in women), but differs slightly from that used in the RRS (age <60 years in men or women for CHD or stroke). We believe this difference in definitions of family history of CVD is minimal and unlikely to detract from our results.
Although robust data exist on the prognostic value of CAC severity in predicting cardiovascular outcomes, fewer data exist as to whether CAC progression is informative regarding the risk of future events. Multiple studies have demonstrated an association between CAC progression and several traditional and emerging CVD risk factors in this cohort and others (14,24,25). Emerging data now suggest additive value of CAC progression beyond CAC alone in prediction of CVD events (4,13–15).
Our study findings raise 2 important questions. First, do cardiovascular risk prediction models accurately predict the progression of subclinical atherosclerosis? In our study, both the RRS and FRS were associated significantly with progression of subclinical atherosclerosis as measured by incident CAC and CAC progression. Of particular interest is that this association occurred in an ethnically diverse and sex-diverse population that was unlike the cohort used to develop either risk score. Our incident CHD event data were consistent with the large body of data that demonstrates the association of the FRS and RRS with future cardiovascular events, but this study is unique in that it demonstrates that these risk scores also predict progression of subclinical atherosclerosis.
Second, are there differences in the ability of RRS and FRS to predict accurately progression of subclinical atherosclerosis? Reclassification (13.7%) did occur among the risk models, and the RRS provided additional predictive power beyond the FRS for predicting incident CAC and progression of CAC. This finding of RRS superiority in the prediction of subclinical atherosclerosis was substantiated further with our finding that the RRS had additive value in predicting CHD events in individuals who had discordance between the 2 risk scores; the same was not true of the FRS.
The RRS and FRS may be useful in predicting the development and progression of subclinical atherosclerosis as assessed by incidence and progression of CAC. When risk classification differences occurred between these 2 risk scoring systems, the RRS was able to provide additional predictive information more consistently. The importance of these findings was supported by an analysis of the risk models' ability to predict CHD events in this cohort. Further studies are needed to delineate more clearly the role of risk prediction models in identifying subclinical atherosclerosis and to delineate the role of CAC progression in risk prediction and the evaluation of treatment efficacy.
The authors thank the other investigators, the staff, and the participants of the MESA study for their valuable contributions. A full list of participating MESA investigators and institutions can be found at http://www.mesa-nhlbi.org.
This research was supported by grant R01 HL071739 and contracts N01-HC-95159 through N01-HC-95169 from the National Heart, Lung and Blood Institute. Dr. DeFilippis is supported by a National Research Service Award Training Grant (T32-HL-07227). Dr. Budoff is on the Speaker's Bureau for General Electric (<$10,000/year). All other authors have reported that they have no relationships relevant to the contents of this paper to disclose. John J. P. Kastelein, MD, PhD, served as Guest Editor for this paper.
- Abbreviations and Acronyms
- Adult Treatment Panel III
- Agatston units
- coronary artery calcification
- coronary heart disease
- confidence interval
- cardiovascular disease
- Framingham risk score
- myocardial infarction
- National Cholesterol Education Program
- Reynolds risk score
- Received May 18, 2011.
- Revision received August 2, 2011.
- Accepted August 12, 2011.
- American College of Cardiology Foundation
- ↵(2002) Third Report of the National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III) final report. Circulation 106:3143–3421.
- Ridker P.M.,
- Paynter N.P.,
- Rifai N.,
- Gaziano J.M.,
- Cook N.R.
- Arad Y.,
- Goodman K.J.,
- Roth M.,
- Newstein D.,
- Guerci A.D.
- Kondos G.T.,
- Hoff J.A.,
- Sevrukov A.,
- et al.
- LaMonte M.J.,
- FitzGerald S.J.,
- Church T.S.,
- et al.
- Taylor A.J.,
- Bindeman J.,
- Feuerstein I.,
- Cao F.,
- Brazaitis M.,
- O'Malley P.G.
- Vliegenthart R.,
- Oudkerk M.,
- Hofman A.,
- et al.
- Budoff M.J.,
- Hokanson J.E.,
- Nasir K.,
- et al.
- McEvoy J.W.,
- Blaha M.J.,
- Defilippis A.P.,
- et al.
- Raggi P.,
- Callister T.Q.,
- Shaw L.J.
- Mosca L.,
- Benjamin E.J.,
- Berra K.,
- et al.
- Bild D.E.,
- Bluemke D.A.,
- Burke G.L.,
- et al.
- Carr J.J.,
- Nelson J.C.,
- Wong N.D.,
- et al.
- ↵MESA Web. http://www.mesa-nhlbi.org. Accessed September 29, 2011.
- Kronmal R.A.,
- McClelland R.L.,
- Detrano R.,
- et al.