Alcohol Consumption and the Prevalence of the Metabolic Syndrome in the U.S.: A cross-sectional analysis of data from the Third National Health and Nutrition Examination Survey

We analyzed data on 8,125 individuals from the Third National Health and Nutrition Examination Survey (NHANES III) (1988–1994); the details of the study design and sampling and examination techniques have been described elsewhere (16,17). Individuals participating in NHANES III were considered eligible for the present study if they were age 20 years or older, not pregnant, had completed the adult household interview and questionnaire, had a physical examination and laboratory evaluation performed at one of the examination centers, and had fasted for at least 8 hours before phlebotomy. Furthermore, they were required to have data on 1) abdominal waist circumference, 2) fasting serum triglyceride, 3) blood pressure, 4) fasting serum HDL cholesterol, 5) fasting serum glucose, 6) fasting serum insulin, and 7) each of the covariates detailed below.

We defined the metabolic syndrome using the National Cholesterol Education Program definition (6). This definition was satisfied if a subject possessed three or more of the following five criteria: an abdominal waist circumference >102 cm in men or >88 cm in women; serum triglycerides ≥150 mg/dl (≥1.69 mmol/l); serum HDL cholesterol <40 mg/dl (<1.04mmol/l) in men or <50 mg/dl (<1.29 mmol/l) in women; average blood pressure ≥130/85 mmHg or currently taking antihypertensive medication; or fasting serum glucose ≥110 mg/dl (≥6.1 mmol/l) or on oral anti-diabetic medication. We averaged the second and third systolic and diastolic blood pressure readings for each subject. Subjects were also classified as hyperinsulinemic if the fasting insulin level was ≥90th percentile of its distribution among subjects in the sample (this value corresponded to a fasting insulin ≥166 pmol/l). This percentile was selected based on work by Meigs et al. (18).

In NHANES III, alcohol consumption was assessed by asking each participant to quantify the number of drinks consumed in the past month before the survey of three different types of alcohol: 1) beer or light beer; 2) wine, wine coolers, sangria, and champagne; and 3) hard liquor such as tequila, gin, vodka, scotch, rum, whiskey, and liqueurs, either alone or mixed. We defined alcohol consumption as 1) currently drinking (≥1 drink of any type per month) or not currently drinking (<1 per month); 2) the number per month of any type of alcoholic drinks (<1, 1–19, or ≥20); and 3) the number of beverage-specific drinks per month (<1, 1–19, or ≥20). Further categorization of the alcohol variable was not possible due to inadequate numbers. The not currently drinking category (<1 drink per month) included subjects consuming at least 12 drinks of any type over a subject’s lifetime but not currently drinking alcohol (past drinkers) and subjects consuming <12 drinks in their entire lifetime (lifetime abstainers). Types of alcohol consumed were divided into three categories: beer, wine, and liquor.

The definitions of the covariates used in this study were based on metabolic syndrome risk factors described by Park et al. (11). Race/ethnicity was based on the screener and subject self-report. We used three categories: non-Hispanic white (white), non-Hispanic black (black), and Mexican American. Other racial/ethnic groups were not included in our analysis because of inadequate numbers.

Level of education was categorized into <8th grade, between 8th and 12th grade, and >12th grade education. Annual household income from the previous year was classified as <$20,000 or ≥$20,000. Subjects were considered physically inactive if they expended <3.5 metabolic equivalents/month participating in walking, jogging, running, bicycle riding, swimming, aerobics, aerobic dancing, other types of dancing, calisthenics, exercises, or garden work. Tobacco use was divided into current user, past user (≥100 cigarettes in their lifetime but currently not smoking), or never smoker. The percentage of daily total energy intake from carbohydrates (kilocalories) was calculated and divided into <40, 40–60, or >60%. Adequate data on hormone replacement therapy were not available; we did, however, use postmenopausal status as a covariate because it has been shown to be a risk factor for the metabolic syndrome (11). Women were considered postmenopausal if they had not had a menstrual period in the past 12 months. Age was used as continuous variable as well as categorized into categories of 20–29, 30–39, 40–49, 50–59, 60–69, and ≥70 years of age.

For each sex and race/ethnicity, we calculated the prevalences of the metabolic syndrome, of each component of the syndrome, and of all covariates. Multiple logistic regression analyses were used to estimate the ORs for the prevalence of the metabolic syndrome and each of its components using all three definitions of alcohol consumption as independent variables (separate analyses for each) while adjusting for covariates. For race/ethnicity-specific models, interaction was assessed using an interaction term between race/ethnicity and the number of alcoholic drinks per month for men and women. Because BMI was highly correlated with waist circumference (r = 0.87, P < 0.0001), which was one of the constituent criteria for our dependent variable, we felt BMI served more as a proxy than a predictor for waist circumference. We did not, therefore, include BMI as a covariate in our primary analyses. We did perform secondary analyses with BMI as a covariate, and the inclusion of BMI did not alter the relationship between alcohol and the metabolic syndrome (data not shown).

All analyses used sampling weights to report estimates that would be representative of the U.S. population. We used SUDAAN statistical software (19) to obtain proper variance estimates to calculate weighted means, percentages, standard errors, and ORs given the NHANES III complex sampling design. A two-sided P value of <0.05 was considered statistically significant. The institutional review board at Boston University Medical Center approved this study.

The characteristics of our sample are presented in Table 1. Over one-half (57.9%) of the participants were current drinkers with higher percentages for men (66.0%) than women (50.0%). The prevalence of the metabolic syndrome was slightly higher in women (22.7%) than in men (21.9%). Mexican-American women had the highest proportion of participants with the metabolic syndrome (26.9%) and the smallest proportion of current drinkers (36.0%).

Current drinkers had a lower adjusted prevalence of the metabolic syndrome than subjects not currently drinking (OR 0.57 [95% CI 0.45–0.72]). Compared with lifetime abstainers, the adjusted odds for the prevalence of the metabolic syndrome among past drinkers was 0.96 (0.75–1.22). The reduced odds for the prevalence of the metabolic syndrome in current drinkers were statistically significant in all age-groups except the 50–60 age-group (0.76 [0.46–1.26]) and the ≥70 age-group (0.72 [0.52–1.00]). The inverse relation of alcohol consumption to prevalence of the metabolic syndrome was consistent in men and in women (0.58 [0.44–0.77] and 0.57 [0.44–0.74], respectively).

Overall, subjects consuming higher quantities of alcoholic beverages had a lower OR for the prevalence of the metabolic syndrome as compared with current nondrinkers (P < 0.0001 for trend) (Table 2). Beer, wine, and liquor had different associations with the prevalence of the metabolic syndrome, with statistically significant reductions in the OR for the prevalence of the metabolic syndrome only for beer and wine.

The relation of alcohol intake to prevalence of the metabolic syndrome was stronger in whites as compared with blacks or Mexican Americans (Table 3). Furthermore, unlike among whites, in black and Mexican-American subjects, consumption of ≥20 drinks per month resulted only in a modest reduction in the OR for the metabolic syndrome, as compared with subjects who drank 1–19 drinks per month. This difference, however, was not accompanied by a statistically significant interaction term between race/ethnicity and the number of alcoholic drinks per month for men (P = 0.37) or women (P = 0.13).

Although increasing quantities of alcohol consumption were associated with a lower OR for the prevalence of the metabolic syndrome in the overall population, this association was not consistent for each of the components. Alcohol consumption was significantly associated with a lower OR for the prevalence of three components of the metabolic syndrome: low serum HDL cholesterol, elevated serum triglycerides, and high waist circumference, as well as hyperinsulinemia (Table 4). When type of alcohol was considered, beer and wine drinkers had favorable associations with low HDL cholesterol, elevated serum triglycerides, waist circumference, hyperinsulinemia, and elevated serum glucose. Among beer and wine drinkers, the odds for the prevalence of elevated serum glucose when drinking ≥20 drinks per month were 0.68 (95% CI 0.44–1.05) and 0.51 (0.28–0.90), respectively. In contrast, for liquor drinkers, the only favorable association was with low HDL cholesterol. Furthermore, among liquor drinkers, the odds for the prevalence of elevated serum glucose when drinking ≥20 drinks per month was 1.90 (1.25–2.88).

Overall, our results demonstrate that people who drink alcoholic beverages have a lower prevalence of the metabolic syndrome as compared with current nondrinkers after accounting for confounders. The inverse relation of alcohol consumption and the metabolic syndrome was especially noticeable in people who consume ≥20 alcoholic beverages per month and was stronger when the beverage was beer or wine. Although the association between alcohol consumption and the metabolic syndrome was present for all race/ethnicities, it was particularly strong in white individuals.

Inconsistent results have been reported between alcohol and the metabolic syndrome (11,13,20). Yoon et al. (13) reported that 1–15 g of alcohol per day was associated with decreased prevalence of the metabolic syndrome, whereas Park et al. (11) reported a decrease only in women. This difference may relate to the fact that Park et al. (11) included in the “heavy” drinking category participants who had a history of “ever consuming five or more drinks almost daily” even if they were not currently drinking. Furthermore, participants were included if they fasted for ≥6 h, whereas the present study and that of Yoon et al. (13) required ≥8 and 10–12 h, respectively.

Numerous studies show serum HDL cholesterol concentrations increase in a dose-dependent response to alcohol consumption (3,21). Serum triglycerides are higher in heavy drinkers (21) but can be lower in light to moderate drinkers (22). The relation with obesity, on the other hand, has not been reported in a consistent fashion (23,24). Nevertheless, some data demonstrate that moderate amounts of alcohol, especially wine, are associated with smaller waist circumference as compared with no alcohol consumption (25,26). Although we did not observe an association for all beverages between alcohol consumption and hyperglycemia, several prospective studies have reported a reduced incidence of diabetes with alcohol consumption (27–29). Furthermore, several studies, including the present study, have demonstrated lower insulin concentrations and increased insulin sensitivity among drinkers (19,30). In contrast, only a few studies have shown a beneficial effect of alcohol consumption on blood pressure (31,32), whereas the majority of studies support our finding that alcohol consumption results in elevated blood pressure (33). The strength of the association between alcohol consumption and the metabolic syndrome may depend heavily upon the prevalence of these individual criteria in a given population.

We found that the reduced prevalence of metabolic syndrome was stronger among drinkers of wine and beer than of liquor. This could result if liquor consumers drank larger amounts or had a different pattern of consumption (e.g., binge drinking) (34,35). Indeed, our data showed that among individuals consuming ≥50 total drinks per month, the mean number of drinks of liquor consumed (18.7) was more than twice that of wine (8.1; P = 0.001), whereas there was no difference in the light-drinking categories. However, we did not have adequate data to describe drinking patterns in detail.

Unlike whites, blacks had only a modest reduction in the odds for the prevalence of the metabolic syndrome when currently drinking, and in black women there was no further reduction in the odds with increasing quantities of alcohol. Although testing of interaction was statistically not significant, we had limited power to evaluate for effect modification by race at the highest quantities of alcohol consumption. Previous studies have reported a relation between increased alcohol consumption and hypertension (33) and a higher prevalence of hypertension in blacks (36), and our data show that blacks were heavier drinkers. Of the current drinkers in our sample, 9.3% of black men and 3.4% of black women currently consumed at least 50 drinks per month, whereas this was lower in white men (5.8%) and women (2.0%).

Differences by race/ethnicity may also relate to insulin sensitivity or hyperinsulinemia (37). Whereas hyperinsulinemia is not a criterion for the metabolic syndrome, studies suggest that it is an important element of the syndrome (7). Using factor analysis, Meigs et al. (18) concluded that a unified metabolic syndrome included three independent factors: hyperinsulinemia, obesity, and dyslipidemia. Among black nondiabetic subjects, however, insulin sensitivity is decreased as compared with white subjects (38). Similarly, fasting insulin concentrations are higher among black women as compared with white women at similar levels of adiposity (39). In the present study, we found that the odds for the prevalence of hyperinsulinemia, after adjusting for our covariates including BMI, was higher among blacks as compared with whites (1.47 [1.10–2.00]).

Although the strengths of our investigations include the use of a large sample representative of the adult U.S. population (thereby enhancing our generalizability), the evaluation of both the type and the quantity of alcohol consumed, and the assessment of effect modification by race/ethnicity, there are several limitations that merit comment. First, given the cross-sectional design, we cannot draw any causal inferences regarding the association of alcohol consumption with the metabolic syndrome. Second, the data on alcohol consumption are based on self-report with the possibility of misclassification of exposure (e.g., under reporting). However, this bias, if nondifferential, would be expected only to increase the amounts of alcohol associated with a reduced prevalence of the metabolic syndrome. Third, categories with the highest quantities of alcohol consumption had the fewest number of individuals, thus limiting our ability to comment on the relation of alcohol consumption to the prevalence of the metabolic syndrome in heavy drinkers. Finally, we cannot comment on the effects of daily patterns of alcohol consumption, as these data were not available.

In summary, mild to moderate alcohol consumption, especially of beer and wine, is associated with a lower prevalence of the metabolic syndrome and with a favorable influence on serum lipids, waist circumference, and fasting serum insulin. This observed association between alcohol consumption and the metabolic syndrome was particularly strong in white individuals.

This work was supported in part through NRSA Training Grant 2T 32 HP 100028–06 (to M.S.F.), a research career award K24 HL 04334 from the NHLBI/NIH (to R.S.V.), and a National Institutes of Health Grant AA13304-01 (to R.C.E.).