Even among asymptomatic persons at low risk (<10%) according to the Framingham risk score, high coronary artery calcium (CAC) scores signify a greater predicted risk of coronary heart disease events. We sought to determine the noninvasive factors (without radiation exposure) significantly associated with CAC in low-risk, asymptomatic persons. In a cross-sectional analysis, we studied 3,046 participants from the Multi-Ethnic Study of Atherosclerosis at a low 10-year predicted risk (Framingham risk score <10%) of coronary heart disease events. Multivariate logistic regression analysis was used to assess the association of novel markers with the presence of any CAC (CAC >0) and advanced CAC (CAC ≥300). A CAC level of >0 and of ≥300 was present in 30% and 3.5% of participants, respectively. Factor VIIIc, fibrinogen, and soluble intercellular adhesion molecule were each associated with the presence of CAC (p ≤0.02), and C-reactive protein, D-dimer, and the carotid intima-media thickness with advanced CAC (p ≤0.03). The base model combining the traditional risk factors had excellent discrimination for advanced CAC (C-statistic 0.808). The addition of the 2 best-fit models combining the biomarkers with or without carotid intima-media thickness improved the c-statistic to 0.822 and 0.820, respectively. All 3 models calibrated well but were similar in estimating the individual risk probabilities for advanced CAC (prevalence 9.97%, 10.63%, and 10.10% in the greatest quartiles of predicted probabilities vs 0.26%, 0.26%, and 0.26% in the lowest quartiles, respectively). In conclusion, in low-risk persons, the traditional risk factors alone predicted advanced CAC with high discrimination and calibration. The biomarker combinations with and without carotid intima-media thickness were also significantly associated with advanced CAC; however, the improvement in the prediction and estimation of the clinical risk were modest compared to the traditional risk factors alone.
The Framingham risk score (FRS) has been validated as a useful tool in the estimation of coronary heart disease (CHD) risk. CHD events, however, might still occur in those at a low predicted 10-year CHD risk, because the FRS provides only an average probability of event occurrence. One proposed method for improving the identification of those at risk of coronary events has been to implement widespread screening for coronary artery calcium (CAC) in intermediate-risk adults. The presence of CAC has been highly correlated with plaque burden, plaque rupture, and coronary events, and a CAC level of ≥300 can predict coronary events beyond the traditional Framingham risk factors in intermediate- and low-risk persons. The factors associated with CAC presence (CAC >0) and/or advanced CAC (CAC ≥300) in low-risk persons (constituting 75% of the general population ) have not been identified. The present study, therefore, aimed to assess the novel markers significantly associated with the presence of CAC, or advanced CAC, among low-risk, asymptomatic adults, thereby avoiding radiation exposure, the discovery of incidental findings requiring follow-up computed tomography, and the potentially increased costs associated with CAC measurement in low-risk persons.
Methods
The Multi-Ethnic Study of Atherosclerosis (MESA) was a prospective cohort study examining the measures of subclinical atherosclerosis, the progression of subclinical atherosclerosis, and conversion to clinical events. The details of the study design, inclusion and exclusion criteria, and baseline characteristics have been previously described. In brief, at baseline, the cohort included 6,814 participants (3,213 men and 3,601 women), aged 45 to 84 years, from 4 different racial/ethnic groups (38% white, 28% black, 22% Hispanic, and 12% Asian) in 6 United States communities (Baltimore, Maryland; Chicago, Illinois; Forsyth, North Carolina; Los Angeles, California; New York, New York; and St. Paul, Minnesota). The participants were free of clinical cardiovascular disease at first examination (July 2000 to August 2002).
For the present study, we included men and women aged ≤79 years at baseline, categorized as having a low 10-year risk of hard CHD events (10-year predicted risk <10%) according to the FRS. Those >79 years old could not have a calculated FRS and were therefore excluded. Consistent with the Adult Treatment Panel III definitions of patients with coronary risk equivalents, the present analyses also excluded participants with a diagnosis of diabetes, peripheral arterial disease (ankle brachial index <0.9), carotid artery disease (≥50% carotid artery stenosis), a history of abdominal aortic aneurysm, severe chronic renal insufficiency, or end-stage renal disease (glomerular filtration rate <30 ml/min/1.73 m 2 according to the Modification of Diet in Renal Disease equation). We further excluded those participants who were already receiving lipid-lowering therapy, because this could have affected their FRS estimate.
The baseline examination findings, laboratory data, cardiac computed tomographic and carotid ultrasound findings have been previously described. The body mass index was defined as the weight in kilograms divided by the height in meters squared. The participants’ medication use was derived from the medication lists and the clinical staff entry of prescribed medications. Aspirin use was defined as ≥3 days/wk at baseline. The Agatston CAC score >0 was defined as CAC presence and a CAC score ≥300 as advanced CAC. Carotid ultrasonography was performed by obtaining images of the right and left common carotid and internal carotid arteries (including near and far wall images), using high-resolution B-mode ultrasonography.
All analyses were performed using Statistical Analysis Systems software, version 9.2 (SAS Institute, Cary, North Carolina). A p value of <0.05 was considered significant. The baseline characteristics ( Table 1 ) were compared according to the CAC classification using general linear models for continuous variables and cross-tabulations for categorical variables. For multivariate analyses, the associations of individual biomarker levels with the presence of CAC or an advanced CAC burden were examined (separately) using logistic regression models, and the standardized estimates (multivariate-adjusted odds ratios and their 95% confidence intervals) were assessed. The Framingham 10-year risk scores for all subjects were calculated according to the National Cholesterol Education Program guidelines.
| Characteristic | CAC | Advanced CAC | ||||
|---|---|---|---|---|---|---|
| 0 (n = 2,114) | >0 (n = 932) | p Value | <300 (n = 2,938) | ≥300 (n = 108) | p Value | |
| Age (years) | 55.4 ± 8.0 | 60.8 ± 9.1 | <0.01 | 56.8 ± 8.6 | 64.7 ± 8.7 | <0.01 |
| Women | 75.3% | 70.2% | <0.01 | 74.0% | 66.7% | 0.09 |
| Race | ||||||
| White | 37.3% | 46.6% | <0.01 | 39.3% | 61.0% | <0.01 |
| Black | 27.4% | 21.6% | 13.0% | 9.3% | ||
| Hispanic | 23.2% | 17.3% | 26.0% | 16.7% | ||
| Asian | 12.1% | 14.6% | 21.7% | 13.0% | ||
| Current smoking | 10.2% | 9.9% | 0.91 | 10.0% | 12.0% | 0.5 |
| Systolic blood pressure (mm Hg) | 118 ± 19 | 123 ± 19 | <0.01 | 120 ± 19 | 125 ± 17 | <0.01 |
| Diastolic blood pressure (mm Hg) | 70 ± 10 | 71 ± 10 | 0.02 | 70 ± 10 | 72 ± 10 | 0.13 |
| Antihypertensive medication use | 18.1% | 25.1% | <0.01 | 19.8% | 31.5% | <0.01 |
| Body mass index (kg/m 2 ) | 28.0 ± 5.8 | 28.0 ± 5.6 | 0.92 | 30.0 ± 5.7 | 27.7 ± 5.6 | 0.64 |
| Family history of premature coronary heart disease | 36.6% | 50.2% | <0.01 | 40.0% | 62.4% | <0.01 |
| Physical activity (MET-min/week) | 896 ± 2,607 | 1,031 ± 2,940 | 0.14 | 930 ± 2,724 | 1,146 ± 2,400 | 0.42 |
| Total cholesterol (mg/dl) | 195 ± 35 | 200 ± 35 | <0.01 | 197 ± 35 | 206 ± 37 | <0.01 |
| High-density lipoprotein cholesterol (mg/dl) | 55 ± 15 | 55 ± 16 | 0.66 | 55 ± 15 | 58 ± 14 | 0.03 |
| Low-density lipoprotein cholesterol (mg/dl) | 116 ± 30 | 122 ± 32 | <0.01 | 118 ± 30 | 125 ± 34 | 0.03 |
| Triglycerides (mg/dl) | 117 ± 69 | 119 ± 62 | 0.43 | 118 ± 67 | 119 ± 71 | 0.88 |
| Mean Framingham risk score (%) | 2.8 ± 2.4 | 4.3 ± 2.5 | <0.01 | 3.2 ± 2.5 | 5.1 ± 2.6 | <0.01 |
| Medication use | ||||||
| Estrogen use (in women) | 29.4% | 30.6% | 0.68 | 29.4% | 38.9% | 0.08 |
| Aspirin | 10.3% | 14.5% | <0.01 | 11.3% | 19.6% | 0.01 |
| Angiotensin-converting enzyme inhibitors/angiotensin receptor blocks | 4.3% | 7.2% | <0.01 | 5.8% | 9.3% | 0.1 |
| β Blockers | 4.7% | 5.8% | 0.15 | 5.3% | 6.5% | 0.61 |
| Nitrates | 0.1% | 0.1% | 0.92 | 0.07% | 0.0% | 0.79 |
| Calcium blockers | 6.6% | 8.4% | 0.06 | 6.9% | 13.9% | <0.01 |
Several different approaches were used in developing the models for assessing the association of the novel marker combinations with CAC presence or advanced CAC. These included data-driven methods (a combination of novel markers independently associated with CAC in the MESA) clinical/mechanistic approaches, and backward stepwise selection techniques. The clinical/mechanistic approaches included a combination of novel markers from each major group, including lipoprotein, inflammatory, and hemostatic factors, and measures of subclinical atherosclerosis, endothelial dysfunction, and chronic kidney disease or according to clinical perception.
The novel markers were added individually to the base model to assess their independent associations with CAC presence. Using this approach, various models were fitted to estimate the associations of combinations of novel markers with the presence of CAC. The base model, which included traditional risk factor covariates only (see Table 2 footnotes), and the backward stepwise selection model used p <0.10 as the criterion for retention (beginning with all available covariates). Similarly, for associations with advanced CAC, several models with combination novel markers were fitted. The models used for our analysis included the base model with covariates only, in addition to high-sensitivity C-reactive protein (CRP), D-dimer, and carotid intima-media thickness (CIMT) (model 1); CRP, D-dimer, low-density lipoprotein particle number, cystatin C, and soluble intercellular adhesion molecule (model 2); and the unbiased approach using a backward stepwise selection model, with p <0.10 as the criterion for retention.
| Model | OR (95% CI) per SD |
|---|---|
| Fibrinogen | 1.11 (1.01–1.22) |
| Soluble intercellular adhesion molecule-1 | 1.16 (1.04–1.298) |
| Factor VIIIc | 1.12 (1.03–1.23) |
| Carotid intima media thickness | 1.08 (0.99–1.18) |
| Age | 1.09 (1.07–1.10) |
| Gender | 3.09 (2.47–3.86) |
| Race | 0.80 (0.74–0.86) |
| Systolic blood pressure | 1.01 (1.00–1.01) |
| Total cholesterol | 1.01 (1.00–1.01) |
| High-density lipoprotein cholesterol | 0.99 (0.98–1.00) |
| Current smoking | 1.40 (1.04–1.89) |
| Hypertension treatment | 1.36 (1.10–1.68) |
To avoid potential bias and power limitations owing to missing cases in complete case analyses, the missing data for soluble intercellular adhesion molecule were replaced by multiple imputations, with all covariates and novel markers taken as predictors. Likelihood ratio tests were used to obtain p values to determine the level of significance of each model relative to the base model. The Akaike information criteria were used to assess the level of informativeness of each model, with lower Akaike information criteria depicting greater informativeness. The C-statistic was used to measure the discrimination of each model, with a greater c-statistic reflecting greater discrimination. The receiver operator characteristic curves were then plotted for the base model and the combination models exhibiting the greatest levels of discrimination for advanced CAC (best-fit models) with and without CIMT.
Results
From a total of 6,814 MESA participants, the study sample included 3,046 persons (44.7%) aged ≤79 years, predicted to have a low risk of CHD events according to the FRS, and who had also undergone computed tomography for CAC measurement. The mean age of the sample was 57 ± 9 years, and 74% were women (mean age 59 ± 9 years for women and 52 ± 6 years for men). Of the 3,046 participants, 932 had CAC (29% of women and 35% of men). The baseline characteristics stratified by the presence of CAC or advanced CAC are listed in Table 1 . CAC presence was associated with adverse levels of most of the established cardiovascular risk factors. Smoking and lipid-lowering medications were not associated with CAC presence in this lower risk subsample of the MESA cohort.
Only 108 of 3,046 participants predicted to be at low risk using the FRS had advanced CAC (3.0% of women and 4.5% of men). Many traditional risk factors were found to be significantly associated with advanced CAC. Race/ethnicity, antihypertensive medications use, and aspirin use were related to both CAC presence and advanced CAC.
On univariate analysis (data not shown), most of the novel markers, with the exception of CRP and D-dimer, were significantly associated with CAC presence (all p <0.01). Cystatin-C, CRP, D-dimer, factor VIIIc, homocysteine, and CIMT were significantly associated with advanced CAC (all p <0.01). No significant interactions between the gender and any of the novel markers were found. The multivariate analysis showed that factor VIIIc, fibrinogen, and soluble intercellular adhesion molecule (imputed and unimputed) were significantly associated with CAC presence, and CRP, D-dimer, and CIMT were significantly related to advanced CAC.
Several biomarker/measure combination models were examined using the data-driven methods (from multivariate analysis), clinical/mechanistic approaches, and backward stepwise selection techniques. In the examination of the association of biomarker combinations with CAC presence, the backward stepwise selection process ( Table 2 ) provided the greatest informativeness (i.e., Akaike information criteria) and discrimination for CAC presence, with a C-statistic of 0.731 (base model C-statistic 0.725). The soluble intercellular adhesion molecule-1, factor VIIIc, fibrinogen, and CIMT, in addition to many of the traditional risk factors, were selected into the model.
Table 3 lists the biomarker combination model selection for the association of biomarker combinations with advanced CAC. All the models provided excellent discrimination for advanced CAC, with a C-statistic >0.80. When compared to the base model (C-statistic 0.808), model 1 (base model plus CRP, D-dimer, and CIMT) showed modest improvement in the discrimination for advanced CAC (C-statistic 0.822; p <0.01). With the exclusion of CIMT, model 2, which combined the base model covariates with CRP, D-dimer, low-density lipoprotein particle number, cystatin C, and soluble intercellular adhesion molecule, exhibited the greatest level of discrimination for advanced CAC, again with a slight improvement compared to the base model (c-statistic 0.820, p = 0.04). Accordingly, the receiver operator characteristic curves for these 3 models substantially overlapped ( Figure 1 ). Many established risk factors (in addition to the same biomarker/measure combinations from model 1) were selected by the backward selection model, which again exhibited the greatest informativeness, with the lowest Akaike information criteria. No age, gender, or race interactions were found in the association between the biomarker combinations and advanced CAC.
