Details about the study design of the Multi-Ethnic Study of Atherosclerosis (MESA) have been previously published (10
). Briefly, 6,814 participants, aged 45–84 years, who identified themselves as white, African American, Hispanic, or Asian (predominantly of Chinese ancestry), were recruited from six U.S. communities between July 2000 and August 2002. All participants were free of clinically apparent CVD at enrollment. Each study site recruited an approximately equal number of men and women, according to prespecified age and race/ethnicity proportions. All participants gave informed consent, and the study protocol was approved by the institutional review board of each center. A1C was measured at exam 2 in MESA, which was conducted ~16 months after the baseline study exam. Of 6,233 participants attending exam 2, 1,112 were excluded because of missing A1C (n
= 96) or fasting glucose (n
= 4) data, diabetes based on fasting glucose ≥7.0 mmol/l (11
) or reported use of oral hypoglycemic medication or insulin (n
= 962), or prevalent CVD at exam 2 (n
= 50), leaving 5,121 participants for this cross-sectional analysis.
All measurements were obtained at exam 2, unless otherwise specified. Medication use and smoking status were ascertained by questionnaire. Smoking status was coded as current, quit <1 year ago, quit ≥1 year ago, and never smoked. Physical activity was measured by questionnaire. Participants reported the average time spent doing intentional exercise activities per week during the past month, and the total estimated metabolic energy expenditure (in MET units per minute) for each activity was summed (12
). The summary scores were categorized as no intentional exercise, ≤735 MET min/week, 736–1,785 MET min/week, and >1,785 MET min/week. Fasting blood samples were obtained for measurements of lipids, glucose, A1C, and highly sensitive CRP. CRP was not measured at exam 2, so baseline measurements were used in the analysis. A morning urine sample was obtained to measure urine albumin-to-creatinine ratio, which was categorized as normal (<30 mg/g), microalbuminuric (30–299 mg/g), or macroalbuminuric (≥300 mg/g).
Common and internal CIMTs were measured by ultrasound on all participants at baseline, as previously described (13
). CIMT was measured on the right and left side, and the maximum value was used for analysis. CIMT is expected to change only ~0.008–0.012 mm per year (14
), hence we assumed that baseline CIMT measurements were a reasonable approximation of CIMT at exam 2.
CAC was measured using either electron-beam computed tomography at three field centers or multidetector computed tomography at three field centers. Each participant was scanned twice consecutively, and these scans were read independently at a centralized reading center, as previously reported (15
). All participants had CAC measured at baseline. Follow-up CAC measurements were performed on about half the cohort (2,438 randomly selected participants) at exam 2 (~16 months after baseline) and the other half of the cohort (n
= 2,246) at exam 3 (~39 months [range 23–59] after baseline). For 2,246 participants who did not have CAC measured concurrently with A1C at exam 2, we used a linear interpolation of their CAC scores from baseline and exam 3 as an estimate of their exam 2 CAC. Either measured or interpolated CAC data for exam 2 were available for 4,684 participants, and 437 participants had missing CAC data.
Participants who had CAC measured at exam 2 were similar to those with interpolated CAC with regard to age, sex, smoking status, BMI, A1C, fasting glucose, systolic blood pressure, LDL cholesterol, HDL cholesterol, common CIMT, and internal CIMT (all P > 0.08). Compared with participants with CAC measured at exam 2, participants with interpolated CAC had higher prevalence of detectable CAC (57 vs. 50%, χ2 test P < 0.001), lower median CAC score if CAC >0 (66.6 vs. 85.1 Agatston units, Wilcoxon rank sum test P < 0.001), and lower mean natural logarithm (ln) CAC score if CAC was more than zero (3.93 vs. 4.36 ln Agatston units, Student's t test P < 0.001). Differences in CAC measurements by interpolated status were not accounted for by adjustment for age, sex, race, smoking status, BMI, systolic blood pressure, LDL cholesterol, HDL cholesterol, antihypertensive medication use, or lipid-lowering medication use (all P < 0.001).
For CIMT, associations with A1C were analyzed using linear regression. In the MESA, about half of the participants had a CAC score of zero, with the positive scores being highly skewed. We modeled this using a two-stage approach. The association between presence or absence of detectable CAC and A1C was analyzed using relative risk regression (16
). Specifically, we used generalized linear models with log link, Gaussian error structure and used a Huber-White sandwich (robust) estimator of variance to calculate the prevalence ratio (PR). Logistic regression was not used because the presence of CAC is not a rare outcome, and the odds ratio would overestimate the PR. Among those with positive CAC scores, we modeled ln of the value as a function of A1C using linear regression. Interactions between A1C and sex or A1C and race/ethnicity were considered. To confirm that results were not dependent on A1C outliers, we repeated all analyses excluding values >6.1% (95th percentile); results were similar, so the findings presented in this report are based on all available A1C data. We also modeled results using all A1C quartiles entered as dummy variables with quartile 1 as the reference group. These models were used to calculate adjusted means or prevalences of CVD measurements by A1C quartile using the mean value for each covariate. To confirm that the linear interpolation of CAC values did not alter the findings, we repeated all CAC analyses using only the participants with a CAC measurement at exam 2. Because multiple comparisons were made, P
values near 0.05 should be interpreted with caution. All statistics were calculated using Stata/SE software (version 8.0 for Windows; Stata, College Station, TX).