An alternative 2-level classification of atherosclerosis severity, based on the presence or absence of atherosclerotic plaques/shadowing, was evaluated in statistical models with and without the first severity variables (pre-existing disease and high risk status) with which there was overlap. These analyses were limited to persons with information on plaques or shadowing at any carotid site (n = 7,381).

Statistical analysis
The analyses were performed using SAS v9.1.2 (SAS Institute Inc., Cary, N.C.). Baseline characteristics between disease severity groups were compared using one-way analysis of variance (ANOVA) and chi-square tests. Multiple linear regression analyses were used to determine both the parameter estimates for age and for atherosclerosis severity variables after adjustment for height, race, and gender (basic adjustment) and after adding statistically significant (p ≤ 0.05) atherosclerotic risk or preventive factors from the following list: body mass index (BMI), hypertension, diabetes, systolic and diastolic blood pressures (SBP and DBP), smoking status, drinking status, usual grams of alcohol consumption per week, LDL and HDL cholesterols, cholesterol lowering medication use, white blood count, and fibrinogen. To test for effect modification, variables representing interaction terms between age, gender, and race with variables for atherosclerosis severity were tested and retained at an alpha of 0.05. Adjustments were to the sample covariate means unless specified otherwise. Age centering was used in conjunction with interaction terms to estimate the effects of atherosclerosis severity at specific ages.

The gender-, height-, and age-specific means were estimated with ordinary least squared regression models adjusting for race (to the proportion in the low risk subset). The 5th and 95th percentiles around the means were estimated using quantile (percentile) regression available in experimental SAS procedure QUANTREG.

Results
Of the 9,109 participants with right CCA diameter B-mode ultrasound measurements, 8,528 participants had adequate information for classification of disease severity and 8,163 participants had data for risk adjustment. Among persons with data from all 6 carotid sites (right or left, internal, bifurcation, and CCA), 36% were found to have plaque or shadowing (evidence of calcification) of at least 1 site. The low risk, high risk, and pre-existing disease groups were comprised of 1,397, 6175 and 591 participants respectively.

Characteristics of study population and subsets
Characteristics used to define the pre-existing disease, high risk, and low risk subsets varied as expected (Table 1): smoking status, lipid profiles, and BP were most favorable in the low risk group, and heart disease and stroke were present only in the pre-existing group. However, other characteristics varied as well. The mean age, proportion of men, proportion of blacks, white blood count, and fibrinogen values were lowest in the low risk subset and were highest in the pre-existing disease subset except for black race. The low risk subset also had the largest proportion of current drinkers and the lowest ethanol consumption reported by drinkers. The group with pre-existing heart disease or stroke contained a greater proportion of former smokers, former drinkers, diabetics, hypertensive persons, users of cholesterol-lowering medication, and persons with plaques/shadowing, but a smaller proportion of current smokers, current drinkers, never smokers, and never drinkers than the high risk group. The pre-existing disease group also had a less favorable lipid profile (higher LDL and lower HDL), a lower DBP, and higher cigarette years smoked for current smokers than the high risk subset.

Table 1. Characteristics of the cohort sample and the cohort subsets*: pre-existing disease, high risk, and low risk subsets, the Atherosclerosis Risk in Communities Cohort Limited Access Data, 1987–89.

Age-diameter relationships: adjustment to study population means. When disease severity dummy variables (pre-existing disease, high risk status) were used as indicators of atherosclerosis severity, the risk-adjusted models retained age, race, height, BMI, current smoker status and cigarette years of smoking, current drinker status and usual ethanol intake, SBP, and DBP, and fibrinogen. When interactions were tested, the age-diameter relationships were not the same among the pre-existing disease, high risk and low risk subsets (p < 0.05 for deviance tests comparing models with and without interaction terms). Therefore, interaction terms (age-high risk status and age-pre-existing disease status) were retained and independent variable betas are shown in Table 2.