Is Insulin Sensitivity a Causal Intermediate in the Relationship Between Alcohol Consumption and Carotid Atherosclerosis?: The Insulin Resistance and Atherosclerosis Study

The IRAS was the first large study to make direct measurements of both insulin resistance and subclinical atherosclerosis. A detailed explanation of sampling strategy and examinations can be found elsewhere (24). In summary, the study recruited 1,625 participants from four clinical centers: Los Angeles, CA; Oakland, CA; San Luis Valley, CO; and San Antonio, TX. The sampling strategy in the IRAS was designed to recruit a population with a relatively equal distribution across categories of age, sex, ethnicity, and glucose tolerance (GT) status. Exclusion criteria included recent corticosteroid treatment, insulin treatment within the last 5 years, pregnancy, decompensated congestive heart failure, decompensated emphysema or chronic lung disease, current treatment for cancer (other than skin cancer), renal failure, or other serious illness. Of the 3,416 people who met eligibility criteria, 1,625 agreed to participate in the IRAS (48% response rate).

S_I was assessed by the frequently sampled intravenous GT test (26,27). Specifically, a 50% solution of glucose (0.3 g/kg) and regular human insulin (0.03 units/kg) was injected intravenously at 0 and 20 min, respectively. Twelve blood samples (from a second intravenous line) were taken and assayed for glucose and insulin levels over a 3-h period as follows: −5, 2, 4, 8, 19, 22, 30, 40, 50, 70, 100, and 180 min. Insulin sensitivity was calculated using MINMOD mathematical modeling (28). For a portion of the participants, S_I was calculated by both the Central Laboratory and Coordinating Center (to document reproducibility of the calculation). The above protocol has been shown previously to be a valid and reliable index of S_I compared with the euglycemic clamp method (29).

In the IRAS, alcohol consumption was assessed as part of an extensive 114-item nutritional interview in which alcohol intake was recorded in grams per day. In the present investigation, alcohol consumption was divided into six categories identical to those used by Bell et al. (19). First, “never drinkers” were those participants who responded “no” to the following question, “In your entire life, have you had at least 12 drinks of any kind of alcoholic beverage?” Next, participants were categorized as “ex-drinkers” if they reported that it had been greater than 1 year since they “last drank any kind of alcoholic beverage.” The following final four categories were based on the definition of a single drink being 12 g of alcohol: <0.5, 0.5 to <1, 1 to <3, and ≥3 drinks/day.

In the IRAS, high-resolution B-mode ultrasonography of the internal carotid artery (ICA) and the common carotid artery (CCA) provided an index of atherosclerosis. Evaluations were performed using identical equipment, and sonographers and readers were centrally trained and certified. The scanning and reading protocol was identical to that used in the Cardiovascular Health Study (30). Briefly, three ICA images and one CCA image were obtained bilaterally (for a total of eight images). From these, the maximal ICA and CCA intimal medial thickness (IMT) farthest away from the skin surface (the “far wall”) was obtained from each side and averaged. This approach has been used as an index for atherosclerosis in other population-based investigations (24,30) as well as in clinical trials (31,32).

Data were analyzed using the Statistical Analysis System (SAS) version 6.12 (SAS Institute, Cary, NC). To estimate the relationship between alcohol consumption and IMT, linear regression was used. The dependent variables were the two indexes of atherosclerosis described above (IMT of both the ICA and CCA). Each index was analyzed separately with an identical set of covariates. Alcohol consumption, the main independent variable, was divided into six categories (as outlined above). S_I was logarithmically transformed to produce a more normal distribution. To solve the problem of an S_I equal to 0, 1.0 was added to all S_I values before log_e transformation. This method of transforming S_I has been used in previous investigations (19,24).

A total of three models were created for each atherosclerosis index. Model 1 adjusted for demographic variables (age, sex, and an ethnicity/clinic variable that accounted for IRAS’s unique sampling strategy) and smoking status (“never,” “former,” or “current”). These covariates represented known confounders, not potential causal intermediates, of the alcohol-atherosclerosis relationship. Model 2 represented model 1 with the additional inclusion of the log_e-transformed S_I variable. Model 3 represented model 2 with the following potential causal intermediates and confounders: LDL-to-HDL ratio, hypertension status (hypertensives were identified by systolic blood pressure >140, diastolic blood pressure >90, or current use of antihypertensive medication), BMI (kg/m²), waist-to-hip ratio (measured at the umbilicus), total energy expenditure (kcal · kg^–1 · year^–1), and GT status (based on the World Health Organization criteria, [33]). Least squares mean (and 95% CIs) for IMT was calculated for each category of alcohol consumption within every model. We also examined the potential interactions between alcohol consumption and the major risk factors for cardiovascular disease in model 3.

To quantify the intermediary effect of S_I, the least squares mean that was calculated for the never drinkers was compared with the following: 1) the alcohol consumption category with the minimum least squares mean IMT (i.e., the nadir of the expected J-shaped relationship) and 2) the maximum alcohol intake category (anticipated to be the maximal IMT category). The difference in IMT between never drinkers and the alcohol intake category with the minimum IMT represents the protective effect of alcohol (in the form of reduced IMT). The difference in IMT between never drinkers and the ≥3 drinks/day category represents the adverse effect of alcohol (in the form of increased IMT). We anticipated that adjustment for S_I would attenuate both the protective and adverse effects of alcohol consumption if S_I was indeed a significant intermediate responsible for the J-shaped association between alcohol and atherosclerosis.

Of the 1,625 IRAS enrollees, 1,482 had baseline values recorded for S_I; therefore, this group represented the study population. On preliminary analysis, there was no statistically significant association between level of alcohol consumption and CCA IMT. However, a statistically significant association was demonstrated between alcohol consumption and ICA IMT. Therefore, all analyses presented hereafter focus on ICA IMT as the index of atherosclerosis. Table 1 shows the mean ICA IMT associated with various population characteristics. As expected, male sex as well as increasing age, waist-to-hip ratio, LDL cholesterol, and decreasing S_I were associated with increased ICA IMT. Furthermore, light to moderate drinkers demonstrated the lowest ICA IMT, followed by nondrinkers, and then heavy drinkers. Other well-known risk factors for atherosclerosis, such as tobacco use, hypertension, and diabetes, were also associated with increased ICA IMT in our study.

In Fig. 1, the least squares mean ICA IMT was plotted for each category of alcohol consumption (for all three models). A J-shaped relationship was observed between level of alcohol consumption and ICA IMT. The ICA IMT for ex-drinkers fell between the average IMTs for the two highest alcohol consumption categories. Alcohol consumption remained statistically significant in its association with ICA IMT in all three models (P = 0.0071, P = 0.0130, and P = 0.0238, respectively).

In Fig. 1, from model 1 to model 3, the adjusted mean IMT for “never-drinkers” decreases, whereas the adjusted mean IMT for the 0.5 to <1 drink/day category (the lowest mean IMT category for all models) increases. This intra-model mean IMT difference (the “protective effect” of moderate alcohol consumption) was 62 μm of IMT in model 1. As Fig. 1 demonstrates, there is an attenuation of this protective effect of moderate alcohol consumption after adjusting for S_I (in model 2) and other covariates listed in Table 1 (model 3). Specifically, this attenuation was on the order of 25% for S_I (from 62 to 46 μm) and 60% after adjusting for all covariates (from 62 to 25 μm). The intra-model mean IMT difference between never-drinkers and ≥3 drinks/day drinkers (the “adverse effect” of excessive alcohol consumption) was 68 μm in model 1. Interestingly, instead of attenuating the adverse effect of excessive alcohol consumption, there is a slight enhancement in model 2 (from 68 to 69 μm), with a more marked enhancement in model 3 (from 68 to 111 μm).

When tested within the framework of model 3, the interaction term for alcohol consumption and GT status was associated with a P value of 0.08. Given this evidence for potential effect modification, we calculated least squares mean ICA IMT for each level of alcohol consumption within each of the three categories of GT status (normal, impaired, and diabetes). These results are presented in Fig. 2.

As Fig. 2A demonstrates, when considering only participants without diabetes but with normal GT, there was no longer a J-shaped association between alcohol consumption and ICA IMT. Rather, never drinkers had the highest index of atherosclerosis. Those who consumed any alcohol (irrespective of the amount) had lower ICA IMT than never drinkers. There was some attenuation of the ICA IMT difference between drinkers and nondrinkers with successive models (although there was minimal attenuation of this relationship when adjusting for S_I only in model 2). The observed associations between alcohol consumption and ICA IMT were not statistically significant in this subgroup (model 1 P = 0.20, model 2 P = 0.25, and model 3 P = 0.34).

Figure 2B demonstrates the relationship between alcohol consumption and ICA IMT in people who had impaired GT but who did not meet the World Health Organization criteria for diabetes (35). This relationship more closely takes the form of the J-shaped association seen in Fig. 1. Specifically, the lowest index of atherosclerosis is in the light drinkers (<0.5 drinks/day), whereas heavy drinkers have the highest index of atherosclerosis. This J-shaped relationship held for all three models and in fact changed little after adjusting for S_I and the additional risk factors for atherosclerosis listed in Table 1. The observed associations between alcohol consumption and ICA IMT were associated with P values of 0.071, 0.081, and 0.058 for models 1, 2, and 3, respectively.

Figure 2C demonstrates the relationship between alcohol consumption and ICA IMT in people with diabetes. A J-shaped association was not observed in this analysis. Rather, in people with diabetes, never drinkers had the lowest index of atherosclerosis, with all levels of alcohol consumption being associated with a higher index of atherosclerosis. The observed associations between alcohol consumption and ICA IMT were associated with P values of 0.054, 0.055, and 0.151 for models 1, 2, and 3, respectively.

As expected, a J-shaped association was demonstrated between alcohol consumption and atherosclerosis in the IRAS sample. No attempt was made to integrate the ICA IMT for ex-drinkers into the curves of ICA IMT by level of alcohol consumption. This approach was chosen because the reason for discontinuing alcohol intake (which was not recorded in the IRAS interview) almost certainly confounds the relationship between ex-drinker status and ICA IMT.

As mentioned above, only ICA IMT (and not CCA IMT) demonstrated such an association. There does not appear to be a clear consensus in the literature regarding which measure is a better index of atherosclerosis. Several investigators have found asymmetry between ICA and CCA IMT, depending on the atherosclerosis risk factor profile (34–36). However, focal atherosclerotic lesions have been shown to be more prevalent in the ICA compared with the CCA (30). Furthermore, ICA sonography has been shown to be more sensitive in detecting carotid atherosclerotic lesions (compared with gold-standard angiography) than CCA sonography (37). Therefore, we feel confident in presenting our results based on ICA IMT as an index of atherosclerosis.

As outlined above, our results support the contention that S_I may be a causal intermediary of the protective aspect of the alcohol-atherosclerosis relationship observed at mild to moderate levels of alcohol consumption. However, the interpretation of these findings is limited by the possible interaction between alcohol consumption and GT status. In our analysis of the association between alcohol consumption and atherosclerosis by category of GT status, the apparent intervening effect of S_I was weaker than that represented for the overall sample. Therefore, it appears that S_I may play only a minimal role in mediating the alcohol-atherosclerosis relationship when analyzed by GT status. However, there is a significant loss of power and precision resulting from the separation of the study sample into three subpopulations based on their GT status.

Despite a decreased ability to answer our primary question regarding the intermediary role of S_I, the transformation of the association between alcohol consumption and ICA IMT by category of GT status was an interesting finding. First, our results suggest that alcohol consumption may be cardioprotective in people with normal GT status at all levels of consumption observed in IRAS. On the contrary, even moderate consumption of alcohol may be harmful in persons with diabetes. The specific physiological mechanisms that underlie this apparent interaction are uncertain and deserve additional investigation if the interaction is confirmed by future studies.

The present study has several limitations. First, given the cross-sectional nature of this analysis, there exists the possibility of temporal ambiguity in the relationship between exposure (alcohol consumption) and outcome (ICA IMT). However, given the wealth of prospective evidence demonstrating the same J-shaped relationship found in the present study, the likelihood that temporal ambiguity significantly altered these results is minimal. Another potential limitation is the self-reported nature of alcohol consumption. If underreporting of alcohol use occurred, a systematic bias may have been introduced, resulting in an underestimate of the protective range of alcohol intake. The possibility of underreporting was minimized by the use of a standardized nutrient intake interview of proven validity (38). It is also important to note that there were few heavy drinkers among IRAS participants. It is therefore more difficult to draw conclusions in this group. Finally, although the protocol for estimation of IMT in the IRAS has established validity as an index of atherosclerosis (30), there exists the possibility that random error and/or systematic bias resulted from error in measurement of IMT.

The IRAS was not a strictly population-based study because selection criteria were designed to over-sample glucose-intolerant subjects (for more favorable numbers of participants across the spectrum of GT status). However, such a sampling strategy is necessary to understand how S_I relates to risk factors for cardiovascular disease in a population with relatively small numbers of glucose-intolerant individuals. Furthermore, IRAS participants were drawn from two existing population-based studies and a health maintenance organization. Therefore, the participants are fairly representative of the general population. In addition, the goal of the present study was to understand the relationship among alcohol intake, S_I, and ICA IMT measured in the study population and does not attempt to address the distribution of risk in the general population.

We feel the present study contributes to the existing work on the relationship among alcohol consumption, S_I, and atherosclerosis. The IRAS was the first population-based investigation to use a direct measure of S_I, to include a sizable proportion of insulin-resistant participants, and to rigorously assess risk factors for cardiovascular disease. The results presented herein support our initial hypothesis that S_I may represent a causal intermediate in the alcohol-atherosclerosis relationship. Furthermore, the demonstration that GT status may be an effect modifier of the alcohol consumption-atherosclerosis relationship was an interesting finding that, to our knowledge, has not yet been described in the literature. The few studies that have examined the association between alcohol consumption and atherosclerosis in people with and without diabetes have concluded that the protective effect of alcohol does not differ significantly between these two groups (39–41). The American Diabetes Association has advocated that people with diabetes, in most cases, can safely consume one to two alcoholic beverages per day (42). This level of moderate alcohol consumption is derived from recommendations made by the American Heart Association for the general population (43). As indicated above, our results appear to suggest otherwise.

Previous studies of alcohol consumption and cardiovascular disease have pooled individuals with impaired GT with those demonstrating normal GT. In the IRAS, when comparing those with normal GT to those with impaired GT (while excluding participants with diabetes), there was a significant interaction between alcohol consumption and GT status. Therefore, in the IRAS, the effect of alcohol consumption on atherosclerosis did differ among people without diabetes, depending on whether they had normal or impaired GT. No previous reports have presented results from analogous analyses.

It may be prudent to consider GT status when making recommendations regarding alcohol consumption. Our results suggest that we should go further than separating patients into those with and without diabetes when risk stratifying drinkers in terms of their risk for cardiovascular disease. Furthermore, if people with diabetes who drink moderate levels of alcohol are potentially at increased risk for atherosclerosis, then perhaps physicians should consider advocating abstinence from alcohol in this population. This is obviously a departure from current recommendations, which suggest that GT status does not significantly alter the protective effect of moderate levels of alcohol consumption. Clearly, additional studies in the field of diabetes and cardiovascular disease are required to shed more light on the effect of alcohol consumption across varying degrees of glucose intolerance.

This study was supported by the National Heart, Lung and Blood Institute (NHLBI) of the National Institutes of Health (NHLBI Grant Nos. HL47887, HL47889, HL47890, HL47892, and HL47902) and by the General Clinical Research Centers Program of the National Center for Research Resources (M01 RR431 and M01 RR01346).