Introduction

Carotid intima-media thickness (IMT) measurement by ultrasound is a noninvasive marker of early atherosclerotic disease¹ that has been proven as a useful predictor of cardiovascular events in epidemiological^2,3 and clinical^4,5 studies. Traditional cardiovascular risk factors have been shown to be determinants of IMT in the general population.⁶

See accompanying editorial on page 1910

As knowledge about the pathogenesis and epidemiology of atherosclerosis progresses, there is growing interest for the identification of novel risk factors.⁷ Major depression disorder,⁸ socioeconomic level,⁹ nuclear¹⁰ and mitochondrial¹¹ genetic variances, and some biomarkers, as higher growth differentiation factor-15 levels,¹² higher uric acid,¹³ and lower adiponectin levels,¹⁴ for example, have all been associated with atherosclerotic disease.

Most articles describing empirical evidence about these new risk factors focus on the statistical significance of measurements of association and risk, as odds ratios and relative risks. However, it is important to identify how much of the atherosclerotic process is already explained by traditional risk factors, and, consequently, how much room is left for these novel risk factors. Recently, Rundek et al¹⁵ analyzed data of 1790 stroke-free individuals from the Northern Manhattan Cohort Study (NOMAS) and found that traditional risk factors explained ≈11% of total variance in mean total carotid IMT. The addition of other less traditional risk factors (eg, adiponectin) raised this proportion only to about 16%.

In clinical practice, several different measurements for blood pressure, glucose metabolism, lipid profile, and adiposity are commonly used, but little is known about their specific contributions to the IMT values. In specific, it is unknown how these variables contribute to preclinical atherosclerosis in individuals considered with low cardiovascular risk, in whom blood pressure, glucose, lipid, and body mass index levels are not high enough to establish diagnoses of hypertension, diabetes mellitus, dyslipidemia, and obesity according to the current guidelines. In addition, identifying characteristics that contribute the most to IMT variance may help defining IMT values that are most predictive of cardiovascular disease (CVD) and classifying individuals according to the probability of abnormal IMT values.

In this article, we aim to study to which extent mean common carotid IMT values are determined by traditional risk factors in the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil), a multicenter cohort study in Brazil. In addition, we aim to evaluate which commonly used measurements of blood pressure, glucose metabolism, lipid profile, and adiposity contribute the most to the IMT values.

Materials and Methods

Materials and Methods are available in the online-only Data Supplement.^16–21

Results

Table 1 shows study sample characteristics. Age was similar according to sex. Men had a worse cardiovascular risk profile compared with women, with higher blood pressure measurements, fasting glucose, triglycerides, low-density lipoprotein/high-density lipoprotein (LDL/HDL) ratio, lower HDL-cholesterol, and more current or past smokers.

Models including only age and race (stratified by sex) yielded coefficients of determination (R²) between 0.071 and 0.298. When we added traditional risk factors to the models (hypertension, diabetes mellitus, dyslipidemia diagnoses, smoking status, and family history of premature cardiovascular disease [CVD]), R² values varied between 0.108 and 0.336 (Table II in the online-only Data Supplement). Table 2 shows the results for main multiple linear models. Models’ R² values varied between 0.141 and 0.373. We observed that most part of the explained variance was because of age and race, as subsequent models for both sexes resulted in small R² increments.

The variables for blood pressure, glucose metabolism, lipid profile, and adiposity that most frequently composed main models (selected by highest R²) were pulse pressure, glycohemoglobin, LDL/HDL ratio, and neck circumference (Table 2). Comparably, however, β-coefficients associated with 1 SD increase in glycohemoglobin and LDL/HDL ratio was smaller than β-coefficients for pulse pressure and neck circumference in most models, suggesting that these 2 latter variables are more important contributors to common carotid artery (CCA)-IMT variance. In the same direction of this finding, β-coefficients for diabetes mellitus and dyslipidemia diagnoses were not statistically significant in most models. β-coefficients associated with hypertension diagnosis, however, were significant in all models in which this variable was included, except in the subgroup of men with 10-year Framingham Heart Study (FHS) coronary heart disease risk ≥10%, probably because of the smaller subsample size. In addition, although family history of premature CVD is not part of FHS coronary heart disease risk calculations, it is noteworthy that we could detect significant effects of this risk factor toward higher IMT values, especially in individuals with lower 10-year FHS coronary heart disease risk.

Other sensitivity analyses (including those for model diagnosis) and simulation techniques to estimate prediction accuracy are provided in the online-only Data Supplement.

Discussion

We found that traditional cardiovascular risk factors explained <40% of the total variance of CCA-IMT in ELSA-Brasil. Indicators of blood pressure, lipid profile, and adiposity that most frequently composed the best models were pulse pressure, LDL/HDL ratio, and neck circumference. Indicators of glucose metabolism were only marginal contributors to the IMT variance.

The association between pulse pressure and CCA-IMT has been studied previously. Consistently to our results, Winston et al²² reported a significant linear association between pulse pressure and CCA-IMT in 6726 individuals from the Multiethnic Study of Atherosclerosis (MESA). Similar findings were reported in individuals with hypertension²³ and in normotensives with known coronary disease.²⁴ A bidirectional causality has been proposed to explain the direct association between pulse pressure and IMT.²⁵ An elevated pulse pressure induces endothelial dysfunction by shear stress and thus may accelerate the atherosclerotic process. However, atherosclerosis also increases arterial stiffness. Because arterial compliance is a major determinant of pulse pressure, individuals with more advanced atherosclerotic disease may present with high pulse pressure. In favor to the hypothesis of pulse pressure as a determinant of atherosclerotic disease, the Young Finns Study analyzed 2146 individuals and concluded that those with higher pulse pressure during childhood and adolescence had higher carotid IMT values in adulthood, even after adjustment for other cardiovascular risk factors.²⁶ Zureik et al²⁷ analyzed 957 older individuals (mean age, 65.2 years) from the Étude du Vieillissement Artériel (EVA study) and found that higher pulse pressure was associated with CCA-IMT progression after a 4-year follow-up.

Neck circumference is an obesity measure related with obstructive sleep apnea and its association with central obesity, metabolic syndrome,²⁸ and cardiovascular risk is increasingly recognized.²⁹ The effect of neck circumference on IMT has been less studied. Rosenquist et al³⁰ studied 3274 participants of the FHS offspring cohort and found a significant association between CCA-IMT and neck circumference, body-mass index and waist circumference, with the strongest association between CCA-IMT and neck circumference. It is possible to speculate, therefore, that neck fat mass may exert a local effect on carotid arteries and this could explain a stronger association between CCA-IMT and neck circumference, compared with more traditional adiposity measures as the waist/hip ratio. If the deposition of fat mass in the neck is associated only locally with higher carotid IMT or if this influences intima-media thickening in arteries located otherwise is still unclear. Recently, Pokharel et al³¹ did not observe a significant association between coronary artery calcium and neck circumference in 845 retired National Football League players. However, it is reasonable to assume that neck circumference was more attributable to the muscle mass in retired high-level professional athletes than to the fat deposition, when compared to the general population. Also, participants were relatively young (median age, 54 years) and significant coronary artery calcium scores are uncommon at this age.

In our study, the proportion of CCA-IMT variance explained by traditional risk factors, blood pressure, glucose metabolism, lipid profile, and adiposity measurements varied between 14.1% and 37.3%. Although these are still relatively low values, our analyses resulted with higher R² coefficients compared with the previous work by Rundek et al.¹⁵ Some population characteristics can explain different results between the 2 studies. Models including only age and race as explanatory variables in ELSA-Brasil yielded higher R² coefficients (≤0.298, Table II in the online-only Data Supplement) than the R² coefficients yielded in NOMAS. Although NOMAS consists of individuals from multiethnic groups, the differences in mean IMT across race–ethnicities were less pronounced in NOMAS than in our study. These results suggest that race (and possibly age) may be acting as a proxy for socioeconomic status and healthcare access (including treatment for risk factors as hypertension) in our population.³² Other methodological procedures may also explain greater R² coefficients in our study. In NOMAS, the authors opted for a stepwise forward approach for inclusion of the variables in the models using a P value cutoff at 0.10. We opted for a modeling technique that was structured on pathophysiological reasoning and prioritized the selection of models that led to higher R² values, even if some of the β-coefficients were not significantly different from zero. We opted for this strategy because models built based on P values are sensitive to sample sizes, and comparisons between models for all individuals and for subsamples would be inadequate. We used P values to determine included variables only in the post hoc model, to avoid excessive colinearity among included variables. Finally, in NOMAS, IMT measurements included segments in the near and the far wall of the left and the right carotid bifurcations, and the internal and the common carotid arteries, whereas in ELSA-Brasil a specific segment of the far wall CCA-IMT was used.

Although the present data are not sufficient to draw definite conclusions about the main reasons for this high unexplained variance of IMT, we think that the influence of novel risk factors in the development of atherosclerotic disease may have an important role. The association between IMT and these novel conditions may not be solely mediated by traditional risk factors, and their inclusion in regression models may reduce the unexplained variance. Some of the unexplained variance may be also because of our option of using linear models. It is reasonable to think that nonlinear models could potentially yield better prediction properties. However, other studies^15,33 also adopted linear models, and therefore our results can be readily compared across different studies. In addition, nonlinear models are, in general, more difficult to interpret. One could argue that a higher intraobserver and interobserver variability could interfere with model precision and lead to lower R². The adopted methodology for IMT measurement in ELSA-Brasil, using a standardized, computer-aided protocol, minimized the influence of this factor. In addition, strong associations between some of the variables in model 2 and model 3 (eg, hypertension and blood pressure measurements) may explain little R² increments between these models. In summary, we strongly think that, although some explanation because of methodological features remain, our findings of relatively low R² coefficients in the full models emphasize the importance of investigating novel and nontraditional risk factors for atherosclerotic disease.

Although the focus of this article is CCA-IMT, it is important to note that total plaque area and total plaque volume correlate with different cardiovascular risk factors compared with CCA-IMT.³⁴ Other studies analyzed regression models using traditional risk factors as independent variables and total plaque area, instead of CCA-IMT, as the dependent variable, reporting higher R² values.^35,36 This may be because of some reasons. First, plaque burden may be more representative of atherosclerosis than CCA-IMT.² Plaque burden seems to be more closely correlated with future cardiovascular events^37,38 and CCA-IMT, especially in the absence of plaque, may reflect other phenotypes besides atherosclerosis, as medial muscular hypertrophy.^39,40 In addition, assuming CCA-IMT as a marker of atherosclerosis, it represents an initial and sometimes reversible⁴¹ stage of the atherosclerotic process, which reduces its ability to act as a predictor for future events. However, given the relatively low mean age of ELSA-Brasil participants, we could expect a high number of individuals without carotid plaque at baseline. In 3441 participants of the MESA, with a mean age of 60.3 years at baseline (8.8 years above the mean age in our sample), Tattersall et al⁴² reported that most individuals (52.9%) had no evidence of plaques in common, bifurcation, or internal carotid arteries. It is expected that in our sample the proportion of individuals without plaque would be even higher, impairing the analysis of the influence of traditional risk factors on atherosclerosis, especially in younger subjects.

Our study has several strengths. This is the largest study to date to analyze the contributions of traditional cardiovascular risk factors to IMT variance. The adoption of a computer-aided protocol for IMT measurements provided reliable and accurate IMT data. We opted for a model selection strategy that maximizes the R² coefficients, which strengthens the finding that >60% of CCA-IMT variance is unexplained. Our study also has some limitations. Because of its cross-sectional design, it is not possible to infer causality. This is particularly important for pulse pressure, because of the existence of a plausible hypothesis for reverse causality. Although waist, hip, and neck circumferences correlate well with body fat mass and distribution, there may be some misclassification in individuals with an unusual high proportion of muscle mass in these sites. We did not use ambulatory blood pressure monitoring. Therefore, we may have missed fluctuations in mean systolic or diastolic blood pressure values during the day. However, we think that this random error was minimized because of the large sample in ELSA-Brasil. Finally, we focused on CCA-IMT and not plaque burden, and total plaque area may be more representative of atherosclerosis than CCA-IMT.^2,38

In conclusion, >60% of CCA-IMT could not be explained by traditional cardiovascular risk factors in the large ELSA-Brasil cohort. This observation represents strong evidence to encourage future studies focusing on new determinants of IMT. Pulse pressure, LDL/HDL ratio, and neck circumference were the most consistent and important contributors to the IMT variance.

Acknowledgments

We thank the ELSA-Brasil participants who agreed to collaborate in this study and the research team of the ELSA-Brasil study for their contribution.

Footnotes