ABC | Volume 112, Nº3, March 2019

Original Article Eickemberg et al Abdominal adiposity and C-IMT in the ELSA-Brasil Arq Bras Cardiol. 2019; 112(3):220-227 Figure 1 – Study sample selection flowchart. Note: Percentage of exclusion (sample with valid images and final sample): 23%. Baseline initial sample ELSA-Brazil n = 15105 Sample with valid images of the carotid arteries n = 10943 Exclusion criteria Final sample n = 8449 569 cardiovascular disease 36 serum triglycerides > 800 mg/dL 1974 use of lipid-lowering medication 144 body mass index > 40 Kg/m 2 120 bariatric surgery 32 body dystrophies and abdominal hernias 297 Asian ethnicity/skin color 136 Brazilian Native ethnicity/skin color 150 no information on ethnicity/skin color 15 no information on abdominal adiposity Statistical analysis A data descriptive analysis was carried out to evaluate the distribution of participants according to the characteristics of interest. Due to the asymmetric distribution of some variables it was decided to show the continuous variables as median and interquartile range. Categorical variables were expressed as absolute and relative frequencies. The frequency of high CIMT (≥75 th percentile) and abdominal adiposity throughWC, WHR, C index, LAP and VAI indicators were estimated. Regression coefficients and odds ratios (OR), with their respective 95% confidence intervals, were calculated using linear regression and multivariate logistic analyses, respectively. Regression analyses were used to identify the magnitude of the effect of the abdominal adiposity presence, measured by the indicators in a categorical scale, on the mean of the CIMT in the linear model and on the diagnosis of high CIMT in the logistic analysis. Due to the asymmetric distribution, CIMT values were transformed into natural logarithm for linear regression. For the logistic regression, the dichotomized CIMT was used in the 75 th percentile of the distribution. The main independent variables (abdominal adiposity indicators) were introduced separately in five models for each regression analysis (linear and logistic) by gender. All models were adjusted for age, ethnicity/ skin color, level of schooling, smoking status, HDL-cholesterol, LDL-cholesterol, and arterial hypertension, chosen for their proximity to the atherosclerosis condition. 21 An effect modification analysis was performed to test the variables gender and ethnicity/skin color in all proposed models using the maximum likelihood ratio test. No effect modification was detected; however, we maintained the analyses stratified by gender based on theoretical references. 5,22 A diagnostic evaluation of the multiple linear regression models was carried out through graphic analysis of residues, evaluation of influential points and multicollinearity. The Hosmer-Lemeshow test, goodness-of-fit test using the Pearson’s residuals and Deviance residues, McFadden's Adjusted R 2 and ROC curve, were used to diagnose logistic model adjustment. A significance level of 5% was established and the Stata 12 software (Stata Corporation, College Station, Texas, USA) was used for the analyses. Results The sample characteristics are shown in Table 1. Men and women with high CIMT had an older median age (47 and 48 years versus 57 years) and a higher frequency of abdominal adiposity (men 71.9% to 78.4%; and women 66% to 73.1%). The values of abdominal adiposity indicators were higher in men and in men and women with CIMT > 75 th percentile. The men had a median CIMT of 0.59 mm (0.52-0.69), and women of 0.56 mm (0.50-0.64) (data not shown). In both genders, the adiposity measured by the five indicators was associated with the mean log of CIMT. The C index showed the smallest effect (Table 2). According to the multivariate logistic regression analysis (Table 3), there was an association between the diagnosis of adiposity by WC, WHR, LAP and VAI with CIMT in both genders. The adiposity diagnosed by WC showed a greater effect on CIMT in both genders. According to the diagnostic analyses of the models, there were no assumption violations, indicating the models’ adequacy. Discussion Using data from the ELSA-Brasil study, associations were observed between abdominal adiposity measurements and CIMT, a noninvasive marker of subclinical atherosclerosis capable of predicting cardiovascular disease. 23 It has been documented, in a study carried out in Southeast Brazil, the definition of CIMT as the thickening of the intima-media complex starting from 1.0mm. 24 Considering this value, in our study, the presence of abdominal adiposity diagnosed by WC, WHR, LAP and VAI showed an important effect, with a variation of 0.02 mm to 0.04 mm in the log of CIMT in both genders. Polack et al., 23 using data from the Framingham offspring cohort study, found that an annual change in CIMT > 0.02 mm was associated with a more than two-fold risk of cerebrovascular accident. 23 222