MESA is a longitudinal cohort study that has been described previously (8
). The MESA cohort includes 6,814 men and women aged 45–84 years at baseline recruited at 6 field centers: Baltimore, Maryland; Chicago, Illinois; Forsyth County, North Carolina; Los Angeles, California; New York, New York; and St. Paul, Minnesota. Only persons free of clinical CVD at baseline were eligible. The baseline visit for the cohort (on which these analyses are based) took place between July 2000 and September 2002. The study was approved by institutional review boards at each site, and all participants gave written, informed consent.
Coronary artery calcification (CAC) was assessed by chest computed tomography that used either a cardiac-gated electron-beam computed tomography scanner (9
) or a multidetector computed tomography system (9
). Scans were read blindly with respect to scan pairs and to other participant data by using a computer interactive scoring system similar to that described by Yaghoubi et al. (11
). The average Agatston score (12
) for the 2 scans was used in all analyses. Kappa statistics for intra- and interreader reproducibility of CAC prevalence were both 0.92. Intraclass correlation coefficients for intra- and interreader reproducibility of CAC scores exceeded 0.99.
Trained technicians in each field center performed B-mode ultrasonography of the right and left near and far walls of the internal carotid and common carotid arteries (13
). An ultrasound-reading center measured maximal carotid intimal media thickness (cIMT) of the internal and common carotid sites as the mean of the maximum cIMT of the near and far walls of the right and left sides. Intraclass correlation coefficients for intrareader reproducibility of common and internal cIMT both exceeded 0.98 and, for interreader reproducibility, were 0.87 and 0.94, respectively.
Questionnaires administered as part of the baseline visit in English, Spanish, or Chinese were used to obtain information on sociodemographic indicators. Race and ethnicity were characterized on the basis of participants' responses to the ethnicity and race questions modeled on the year 2000 US Census. All participants who reported their ethnicity as Hispanic were classified as Hispanic in these analyses. All others were classified into 3 groups (non-Hispanic whites, African Americans, and Chinese Americans) by their responses to the race question. Centrally trained clinical teams collected information on CVD risk factors during the baseline examination. Hypertension was defined as a systolic blood pressure of ≥140 mm Hg, diastolic blood pressure of ≥90 mm Hg, or use of medication prescribed for hypertension. A central laboratory measured total and high density lipoprotein cholesterol and glucose levels from blood samples obtained after a 12-hour fast. Low density lipoprotein cholesterol was calculated with the Friedewald (14
) equation. Diabetes was defined as a fasting glucose level of ≥126 mg/dL or use of hypoglycemic medication, and impaired fasting glucose was defined as a fasting glucose level of ≥110 mg/dL. Body mass index was calculated as weight (kg)/height (m)2
Usual dietary intake over the preceding year was quantified by a 120-item food frequency questionnaire at baseline (15
). Dietary patterns were empirically derived by using principal components analysis with orthogonal (varimax) rotation, creating uncorrelated dietary patterns (mean = 0, standard deviation = 1) (15
). Four dietary patterns were retained and named according to the food groups loading highest on the respective dietary patterns in . A higher score indicated greater conformity with the pattern being calculated. Physical activity was measured by an activity questionnaire adapted from the Cross-Cultural Activity Participation Study (16
). The physical activity measure investigated was intentional activity measured in metabolic equivalent hours/day.
Characteristics of the Study Population, Stratified by Race/Ethnicity and Sex, Multi-Ethnic Study of Atherosclerosis, 2000–2002
Psychosocial factors were assessed by using standardized questionnaires written in English, Spanish, or Chinese. Depression was assessed by the Center for Epidemiology Studies Depression (CES-D) Scale (17
), anger and anxiety were assessed by use of the Spielberger trait anger and the Spielberger trait anxiety scales (18
), and chronic psychologic stress was assessed by using the chronic burden scale (19
). Scores were analyzed as continuous variables.
As part of the baseline examination, participants were asked to select their total gross family income in the past 12 months from 13 categories. Income was collapsed into 4 categories (<$25,000, $25,000–$39,999, $40,000–$74,999, or ≥$75,000) for these analyses. Participants also reported the highest educational level completed. Education was categorized into 4 categories (completed high school or less, some college but no degree/technical school certificate, associate or bachelor's degree, or graduate/professional degree) for these analyses. The 4 wealth variables were as follows: 1) whether the participant, or his/her family, had investments such as stocks, bonds, mutual funds, retirement investments, or other investments (yes/no); 2) whether the participant owned his/her home (yes/no); 3) whether the participant owned a car (yes/no); and 4) whether the participant owned land or another property that was not his/her primary residence (yes/no). A summary adult socioeconomic position score was created by summing scores for income (0–3, from lowest to highest category) and education (0–3 from lowest to highest) and adding 1 point for each wealth indicator present (20
). Adult socioeconomic position was characterized as being low, medium, or high.
CAC scores were classified as a dichotomous variable (0 = absence; 1 = presence if the calcium score > 0). The absolute risk or probability of having any CAC was modeled by using the GENMOD procedure in SAS software (SAS Institute, Inc., Cary, North Carolina); these generalized linear models use maximum likelihood estimation for β, where β is interpreted as the risk difference for each unit change in the predictor variable (21
). Models were fit in several stages, in order to determine whether any observed heterogeneity in sex differences by race/ethnicity was accounted for by different sets of factors. The first set of models adjusted for age (continuous). The second set of models also adjusted for traditional CVD risk factors, including diabetes (no diabetes, impaired fasting glucose, diabetes), cigarette smoking (never, former, current), presence of hypertension, low density lipoprotein cholesterol levels (continuous), high density lipoprotein cholesterol levels (continuous), and triglycerides (continuous). The third set of models also adjusted for behavioral factors, specifically, dietary patterns (all 4 patterns, each score modeled as a continuous variable) and physical activity level (continuous). The fourth set of models also adjusted for psychosocial factors (continuous), and the fifth set of models adjusted for all types of risk factors including adult socioeconomic position.
In addition, we also examined associations of sex with amount of calcium among persons with calcium by modeling the ln(Agatston score) as a function of covariates using linear regression. Exponentiated coefficients derived from these models can be interpreted as relative differences in the amount of calcium in women compared with men. For example, a relative difference of 1.5 represents a 50% difference in the Agatston score. A similar modeling approach was used for common and internal cIMT. As with CAC Agatston scores, we log transformed cIMT values. Models were fit in stages as described for CAC.
To determine if there were significant differences in associations of sex with subclinical measures by race/ethnicity, we tested interactions by including a race/ethnicity × sex interaction term (in addition to the main effects of race and sex) in models pooling across racial/ethnic groups. We also examined interaction terms between each race/ethnicity dummy and sex in order to compare associations of sex with the outcome in each racial/ethnic group to the sex association observed in whites as the reference category. To determine if associations were different among women likely to be postmenopausal, we examined the subset of participants aged greater than 50 years but found similar patterns (results not shown). All statistical analyses were calculated with SAS, version 9.1, software (SAS Institute, Inc.). Probability values correspond to 2-tailed tests.