Atherosclerosis Risk in Communities Study

Transcription

1 American Journal of Epidemiology Copyright 2001 by the Johns Hopkins University Bloomberg School of Public Health All rights reserved Vol. 154, No. 8 Printed in U.S.A. Alcohol Consumption and the Risk of Diabetes Mellitus Kao et al. Alcohol Consumption and the Risk of Type 2 Diabetes Mellitus Atherosclerosis Risk in Communities Study W. H. Linda Kao, 1 Ian B. Puddey, 2 Lori L. Boland, 3 Robert L. Watson, 4 and Frederick L. Brancati 1,5 Evidence regarding the association between alcohol consumption and type 2 diabetes risk remains inconsistent, particularly with regard to male-female differences. The authors conducted a prospective study of type 2 diabetes risk associated with alcohol consumption in a cohort of 12,261 middle-aged participants of the Atherosclerosis Risk in Communities Study ( ), who were followed between 3 and 6 years. Alcohol consumption at baseline was characterized into lifetime abstainers, former drinkers, and current drinkers of various levels. Incident diabetes was determined by blood glucose measurements and self-report. After adjustment for potential confounders, an increased risk of diabetes was found in men who drank >21 drinks/week when compared with men who drank 1 drink/week (odds ratio = 1.50, 95% confidence interval: 1.02, 2.20) while no significant association was found in women. This increased diabetes risk among men who drank >21 drinks/week was predominantly related to spirits rather than to beer or wine consumption. The relative odds of incident diabetes in a comparison of men who drank >14 drinks of spirits per week with men who were current drinkers but reported no regular use of spirits, beer, or wine were 1.82 (95% confidence interval: 1.14, 2.92). Results of this study support the hypothesis that high alcohol intake increases diabetes risk among middle-aged men. However, more moderate levels of alcohol consumption do not increase risk of type 2 diabetes in either middle-aged men or women. Am J Epidemiol 2001;154: alcohol drinking; cohort studies; diabetes mellitus, non-insulin-dependent; prospective studies Received for publication March 21, 2000, and accepted for publication May 10, Abbreviation: ARIC, Atherosclerosis Risk in Communities. 1 Department of Epidemiology, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD. 2 Department of Medicine, University of Western Australia, Perth, Australia. 3 Division of Epidemiology, School of Public Health, University of Minnesota, Minneapolis, MN. 4 Division of Epidemiology, University of Mississippi Medical Center, Jackson, MS. 5 Department of Medicine, School of Medicine, Johns Hopkins University, Baltimore, MD. Reprint requests to Dr. Frederick L. Brancati, Welch Center for Prevention, Epidemiology, and Clinical Research, the Johns Hopkins Medical Institutions, 2024 East Monument St., Suite 2-600, Baltimore, MD ( fbrancat@welch.jhu.edu). This study was presented in part at the 31st Annual Meeting of the Society for Epidemiologic Research, Chicago, IL, June 24 26, Type 2 diabetes mellitus affects over 15 million Americans and leads to excess risk of cardiovascular diseases and other morbidities (1 3). Aside from obesity and physical inactivity, there are few other well-established modifiable risk factors for type 2 diabetes. Alcohol consumption represents a potentially important, modifiable risk factor of type 2 diabetes, especially given that more than half of the adults in the United States are current drinkers (4). The hypothesized diabetogenic effects of alcohol include its contribution to excess caloric intake and obesity, induction of pancreatitis, disturbance of carbohydrate and glucose metabolism, and impairment of liver function (5 7). However, the association between alcohol and diabetes or other diabetes-related physiologic endpoints has been inconsistent in previous small clinical studies (8 16), animal studies (17, 18), and larger cross-sectional epidemiologic studies (5, 19 24). Prospective epidemiologic studies have yielded similarly equivocal results; furthermore, results of these studies suggest a potential difference in the association of alcohol and incident type 2 diabetes between women and men (25 32). The Atherosclerosis Risk in Communities (ARIC) Study offers a unique opportunity to examine the association between alcohol consumption and the risk of type 2 diabetes mellitus in a large community-based cohort of middle-aged women and men. MATERIALS AND METHODS Study population The ARIC Study is an ongoing prospective study that examines clinical and subclinical atherosclerotic diseases in a cohort of 15,792 persons, aged years at baseline examination, selected by probability sampling from four US communities. The sampling procedure and methods used in the ARIC Study have been described in detail elsewhere (33). This analysis was based on information obtained over 6 years of follow-up, which included two clinic visits sched- 748

2 Alcohol Consumption and the Risk of Diabetes Mellitus 749 uled at 3 (visit 2) and 6 (visit 3) years after baseline. We excluded participants who reported ethnicity other than Black or White (n 48), had diabetes at visit 1 (n 1,867), had missing exposure information (n 89), had missing diabetes information at visit 1 or were lost to follow-up before visit 2 of the study took place (n 1,022), and had missing information on the potential confounders (n 315), as well as persons in whom total caloric intake was within the top 1 percent or bottom 1 percent of the entire cohort due to probable reporting error (n 190). After these exclusions, 12,261 participants (6,838 women and 5,423 men) remained. Exposure assessment Alcohol consumption was assessed using five questions. The following two questions were used to determine the current drinking status of the persons: Do you presently drink alcoholic beverages? and Have you ever consumed alcoholic beverages? Persons were classified as lifetime abstainers if they answered no to both questions. Persons who answered no to the first question and yes to the second question were classified as former drinkers. Persons who answered yes to both questions were considered current drinkers. Among current drinkers, the following three questions were used to determine the amount and type of alcoholic beverage consumed: How many glasses of wine do you usually have per week (4-ounce glasses)? ; How many bottles or cans of beer do you usually have per week (12-ounce bottles or cans)? ; and How many drinks of hard liquor do you usually have per week (1.5-ounce shots)? (One ounce ml.) For the primary analyses, ethanol, rather than a specific beverage, was the main independent variable. Weekly ethanol consumption was derived from the responses to the three beverage questions using the following conversion factors: 4 ounces of wine 10.8 g of ethanol, 12 ounces of beer 13.2 g of ethanol, and 1.5 ounces of spirits 15.1 g of ethanol. To reclassify current drinkers by the number of drinks (nonspecific to type of alcohol) consumed per week, we assumed one generic drink to be equal to 12 g of ethanol. Seven consumption groups were created: lifetime abstainers, former drinkers, current drinkers who consume <1 drink/week (reference group), current drinkers who consume drinks/week, current drinkers who consume drinks/week, current drinkers who consume drinks/week, and current drinkers who consume more than 21 drinks/week. For alcohol-specific analyses, derivation of weekly ethanol consumption from the three beverages was not used (method described in the previous paragraph). Instead, the original number of glasses, bottles, and drinks was used to categorize wine, beer, and spirit consumption, respectively. For each beverage type, persons who identified themselves as current drinkers but reported an average weekly consumption of zero servings of that beverage were used as the reference group. Information on age, gender, race, family history of diabetes, and education was obtained from home and clinic interviews conducted at the baseline visit. A positive family history of diabetes was defined as having either biologic parent with diabetes. Body mass index (weight (kg)/height (m) 2 ) and waist/hip ratio were determined by the anthropometric measurements taken at the baseline clinic visit. Measurements were made with the participants wearing light-weight, nonconstricting underwear and no shoes. Waist and hip measurements were taken at the level of the umbilicus and the level of maximal protrusion of the gluteal muscles, respectively. Intrareader and interreader correlations between repeated waist/hip ratio measures were 0.94 and 0.91, respectively (34). Physical activity was assessed using a modified interviewer-administered version of the questionnaire developed by Baecke et al. (35). Results from the questionnaire concerning physical activity were further condensed to a sport-related physical activity index with scores ranging from 1 to 5, with 1 indicating the lowest level of activity and 5 the highest level. Total caloric intake was derived from an interviewer-administered, modified version of the 61-item food frequency questionnaire developed by Willett et al. (36). Smoking status was categorized into never, former, and current smokers. Hypertension was defined by the presence of any of the following: 1) systolic blood pressure of 140 mmhg, 2) diastolic blood pressure of 90 mmhg, or 3) current use of antihypertensive medication. Outcome assessment Diabetes mellitus was defined as the presence of any one of the following: 1) fasting glucose of 7.0 mmol, 2) nonfasting glucose of 11.1 mmol, 3) current use of diabetic medication, or 4) a positive response to the question, Has a doctor ever told you that you had diabetes (sugar in the blood)? Persons with diabetes or who had unknown diabetes status at baseline were excluded from the prospective analyses. Persons without diabetes at baseline who met any of the these conditions at visit 2 or visit 3 were considered incident cases of diabetes (n 239). Persons who met the criteria for diabetes at visit 2 but not at visit 3 were nonetheless considered to have incident type 2 diabetes. Statistical analysis All analyses were stratified by gender. The means and frequencies of potential confounders assessed at baseline were calculated for each group of alcohol consumption, and analysis of variance and chi-square analysis were used to assess the statistical significance of the differences across consumption groups. For continuous variables, p values for test for trend among current drinkers were also reported. Incidence rates were calculated for each consumption group with the use of a person-years approach, and a Poisson regression model was fitted to determine if the incidence rates for the seven consumption groups differed significantly. Because the outcome was assessed only at 3-year intervals, the time of follow-up for incident cases was assigned to the midpoints between visits, that is, 1.5 and 4.5 years. For example, persons who were nondiabetic at baseline but became diabetic at visit 3 were considered to have been followed for 4.5 years. Nondiabetic persons at both

3 750 Kao et al. visit 1 and visit 2 who were lost to follow-up after visit 2 were censored at 3 years after the baseline visit. The independent association between baseline alcohol consumption and subsequent incident diabetes was examined using logistic regression models. Alcohol consumption was treated both nominally and continuously (for trend among current drinkers). Multiple linear regression analysis was also performed to examine the association between baseline alcohol consumption and fasting glucose levels at visit 2 or visit 3 among persons who did not have diagnosed diabetes by the time of the visit. All statistical analyses were performed using a SAS statistical package (Cary, North Carolina) (37). RESULTS Baseline characteristics As shown in table 1, among the 6,838 female ARIC Study participants, almost one third reported that they were lifetime abstainers, and roughly half identified themselves as current drinkers, of whom roughly half reported an average weekly consumption of 1 drink/week. Both lifetime abstainers and former drinkers tended to be older, less physically active, and more obese; they were more similar to each other than to current drinkers. There was also a greater proportion of Blacks, positive family history of diabetes, and hypertension among lifetime abstainers and former drinkers. Among current drinkers, total caloric intake and prevalence of hypertension were positively associated with alcohol consumption, while educational level, family history of diabetes, and body mass index were inversely associated with alcohol consumption. Persons who drank more than 21 drinks per week had a noticeably higher waist/hip ratio and total caloric intake and less physical activity. As expected, smoking and alcohol consumption were highly associated. Although the fasting serum glucose level was slightly positively associated with alcohol consumption among current drinkers, the fasting serum insulin level was inversely associated with alcohol consumption in this group. Only about 10 percent of the 5,423 male ARIC Study participants were lifetime abstainers, and approximately 70 percent were current drinkers (table 2). Among current drinkers, men were drinking more drinks per week than were women, with the amount consumed by men slightly higher for each consumption group. The most striking male-female difference was among the heavier drinkers: 8 percent of men drank >21 drinks/week with a mean intake of 402 g of alcohol per week, whereas less than 1 percent of women drank >21 drinks/week with a mean intake of 347 g/week. Despite the disparity in ethanol intake, the associations between alcohol consumption and most potential confounders were similar in men and women. Men who drank >21 drinks/week had the highest waist/hip ratio and energy intake and the lowest physical activity score. The proportions of current smokers and persons with hypertension were also highest in this group of men. TABLE 1. Means (SDs*) and frequencies of selected baseline characteristics of 6,838 female ARIC* Study participants by alcohol consumption group, p trend p overall Lifetime Former Current drinkers by alcohol consumption group (drinks/week) abstainers drinkers >21 Characteristic 980 (14.3) 2,223 (32.5) (0.8) 347 (107) 53.3 (5.3) (1.9) 202 (22) 53.3 (6.1) (5.8) 117 (25) 52.9 (5.6) ,316 (19.2) 40 (20) 53.2 (5.6) ,736 (25.4) 1 (3) 53.2 (5.6) (5.6) (5.8) (4.73) (0.069) 1,793 (630) 2.16 (0.83) (4.09) (0.072) 1,490 (512) 2.41 (0.85) (4.46) (0.078) 1,545 (502) 2.45 (0.77) (4.89) (0.080) 1,487 (522) 2.48 (0.80) (5.17) (0.078) 1,484 (546) 2.41 (0.76) (6.42) (0.076) 1,549 (598) 2.29 (0.76) (6.14) (0.079) 1,498 (549) 2.24 (0.70) (0.56) (38.58) (0.47) (40.25) (0.49) (37.11) (0.47) (49.00) (0.47) (52.24) (0.51) (70.87) (0.53) (64.94) No. (%) Ethanol consumed (g/week) Age (years) Black (%) Education (%) 11 years or less High school graduate Attended college Family history of diabetes (%) BMI* (kg/m 2 ) Waist/hip ratio Energy intake (kcal/day) Physical activity score Smoking (%) Current Former Never Hypertension (%) Serum glucose (mmol) Serum insulin (pmol) * SD, standard deviation; ARIC, Atherosclerosis Risk in Communities; BMI, body mass index (weight (kg)/height (m) 2 ). Limited to trend among current drinkers. Based on 6,687 fasting participants.

4 Alcohol Consumption and the Risk of Diabetes Mellitus 751 Incident type 2 diabetes and alcohol consumption TABLE 2. Means (SDs*) and frequencies of selected baseline characteristics of 5,423 male ARIC* Study participants by alcohol consumption group, p trend p overall Lifetime Former Current drinkers by alcohol consumption group (drinks/week) abstainers drinkers >21 Characteristic 1,096 (20.2) 669 (12.3) (8.1) 402 (182) 54.0 (5.7) (5.8) 205 (24) 54.4 (5.6) (13.4) 124 (25) 54.3 (5.7) ,366 (25.2) 43 (22) 53.9 (5.8) (15.0) 1 (3) 54.2 (5.8) (5.6) (5.5) (4.30) (0.052) 1,992 (635) 2.46 (0.83) (4.20) (0.050) 1,795 (586) 2.78 (0.85) (3.69) (0.052) 1,723 (589) 2.74 (0.80) (3.67) (0.052) 1,742 (621) 2.68 (0.83) (3.84) (0.052) 1,719 (653) 2.67 (0.82) (4.37) (0.057) 1,749 (661) 2.52 (0.81) (4.10) (0.053) 1,706 (580) 2.47 (0.78) (0.59) (57.23) (0.51) (51.34) (0.48) (47.58) (0.48) (55.58) (0.48) (56.88) (0.51) (58.82) (0.49) (62.52) No. (%) Ethanol consumed (g/week) Age (years) Black (%) Education (%) 11 years or less High school graduate Attended college Family history of diabetes (%) BMI* (kg/m 2 ) Waist/hip ratio Energy intake (kcal/day) Physical activity score Smoking (%) Current Former Never Hypertension (%) Serum glucose (mmol) Serum insulin (pmol) * SD, standard deviation; ARIC, Atherosclerosis Risk in Communities; BMI, body mass index (weight (kg)/height (m) 2 ). Limited to trend among current drinkers. Based on 5,281 fasting participants. Type 2 diabetes incidence rates of the seven alcohol consumption groups differed significantly in women (p < ) and men (p 0.08) (table 3). In women, the highest incidence rates were found in lifetime abstainers and former drinkers (19.88 and per 1,000 person-years, respectively). Among current drinkers, the incidence rates decreased from per 1,000 person-years to 6.41 per 1,000 person-years as alcohol consumption increased. The association between alcohol consumption and diabetes risk appeared inverse and by grade, but the incidence rates of the higher consumption groups (>14 drinks/week) were based on very few cases. In men, although the risks for lifetime abstainers and former drinkers were similar to each other, to the reference group ( 1 drink/week), and to their female counterparts (19.16 and per 1,000 person-years, respectively), the risk pattern of current male drinkers was quite different from that of current female drinkers. Among male current drinkers, the risk of diabetes appeared to be generally flat below 21 drinks/week but increased above the threshold of 21 drinks/week. In fact, men who drank >21 drinks/week had the highest risk of diabetes (26.56 per 1,000 person-years). To determine the independent relation of alcohol consumption and incident type 2 diabetes, we adjusted for age, race, education, family history of diabetes, body mass index, waist/hip ratio, physical activity, total caloric intake, smoking history, and history of hypertension using logistic regression. In women, the inverse association between alcohol intake and incident diabetes persisted (table 4), but the trend was no longer statistically significant. In men, the lack of association between alcohol consumption of 21 drinks/week and diabetes risk persisted after adjusting for potential confounders. The increased risk of diabetes seen in men who drank >21 drinks/week also persisted in the fully adjusted model (odds ratio 1.50, 95 percent confidence interval: 1.02, 2.20) (table 4). We tested for effect modification by sex for statistical significance among persons who identified themselves as current drinkers. In the unadjusted model, the interaction term was statistically significant (p 0.04). Although this interaction was no longer statistically significant in the fully adjusted model (p 0.08), the divergence of the fully adjusted relative odds of incident diabetes in the higher consumption groups between women and men remained apparent. Subsidiary analyses To test the robustness of our results, three subsidiary analyses were conducted. The first set of analyses included further adjustment for fasting serum insulin and glucose levels at baseline because both variables were associated with alcohol consumption at baseline (table 4). Adjustment for the fasting serum insulin level at baseline slightly strengthened the association between alcohol consumption and the risk of diabetes in men (male drinkers had higher insulin levels compared with the reference group at baseline) but slightly weakened the association in women (female drinkers at lower insulin levels compared with the reference

5 752 Kao et al. TABLE 3. Incidence of type 2 diabetes in the 12,261 ARIC* Study participants by sex and alcohol consumption groups, Variable Lifetime abstainers Former drinkers Current drinkers by alcohol consumption group (drinks/week) >21 p value Women Incident diabetes (no.) Total person-years Incidence rate Rate ratio , , , , , Men Incident diabetes (no.) Total person-years Incidence rate Rate ratio 69 3, , , , , , , * ARIC, Atherosclerosis Risk in Communities. p value for overall difference. group at baseline). In contrast, adjustment for the fasting serum glucose level at baseline weakened the association between alcohol consumption and the risk of diabetes in men but strengthened the association in women. Because the baseline fasting insulin level is hypothesized to be in the causal pathway and the baseline glucose level is part of the outcome, models 2 and 3 of table 4 likely represent overadjusted models. The second set of analyses examined the association between alcohol consumption at baseline and incident diabetes at visit 2 or at visit 3 separately. This set of analyses was performed to determine if the association between alcohol consumption and diabetes risk was dependent on follow-up time. The increased risk of diabetes for men who consumed >21 drinks/week at baseline was present at both visits 2 and 3; however, neither association was statistically significant at the 0.05 level. The last set of analyses was conducted using continuous serum glucose at either visit 2 or visit 3 as the outcome among current drinkers who did not have diagnosed diabetes, who were not taking antidiabetic medication at the time of the visit, and who were nondiabetic at the previous visit. The fully adjusted mean fasting glucose levels at either visit 2 or visit 3 were similar among the five consumption groups in women; however, the adjusted mean fasting glucose levels were different among the five consumption groups in men. At visit 2, the mean glucose levels (mg/dl) for the five consumption groups in men were ( 1 drink/week), (1.1 7 drinks/week), ( drinks/week), ( drinks/week), and (>21 drinks/week) (p 0.02 for TABLE 4. Adjusted relative odds (95% confidence intervals) of incident type 2 diabetes mellitus in a comparison of alcohol consumption groups in 12,261 ARIC* Study participants by sex, Alcohol consumption (drinks/week) > Model (0.10, 1.77) 0.64 (0.25, 1.64) 0.81 (0.47, 1.37) 1.09 (0.80, 1.49) Women Model 2 Model 3 Model 1 Model 2 Model (0.12, 2.26) 0.77 (0.29, 2.00) 0.88 (0.52, 1.51) 1.14 (0.84, 1.56) 0.33 (0.08, 1.48) 0.54 (0.20, 1.43) 0.63 (0.36, 1.11) 0.98 (0.71, 1.36) 1.50 (1.02, 2.20) 1.07 (0.68, 1.69) 0.80 (0.55, 1.17) 1.12 (0.82, 1.52) Men 1.64 (1.10, 2.42) 1.19 (0.75, 1.88) 0.88 (0.60, 1.28) 1.15 (0.84, 1.57) 1.22 (0.80, 1.85) 0.91 (0.56, 1.48) 0.84 (0.56, 1.25) 1.12 (0.81, 1.56) Former drinkers Lifetime abstainers 1.10 (0.81, 1.49) 1.10 (0.84, 1.43) 1.09 (0.79, 1.48) 1.08 (0.82, 1.43) 1.18 (0.85, 1.65) 1.15 (0.86, 1.54) 1.06 (0.77, 1.47) 1.14 (0.79, 1.65) 1.06 (0.76, 1.46) 1.09 (0.75, 1.58) 1.06 (0.75, 1.51) 1.18 (0.80, 1.76) p for overall p for trend# * ARIC, Atherosclerosis Risk in Communities. Model 1 adjusted for age ( 49 vs vs vs. 59), race (binary), education (11 years or less vs. high school graduate vs. attended college), family history of diabetes (binary), body mass index (quartiles), waist/hip ratio (quartiles), physical activity score (continuous), total energy intake (quartiles), smoking history (never vs. former vs. current), and history of hypertension (binary) using logistic regression. Model 2 = model 1 + fasting serum insulin level (quartiles) at baseline. Model 3 = model 2 + fasting serum glucose level (quartiles) at baseline. p for overall is the p value of the test for overall significance of alcohol consumption groups on a nominal scale. # p for trend is the p value of the test for linear trend among current drinkers

6 Alcohol Consumption and the Risk of Diabetes Mellitus 753 overall difference). The corresponding numbers from visit 3 were 100.0, 101.2, 101.2, 101.2, and (p 0.06 for overall difference). At both visits 2 and 3, men who drank >21 drinks/week at baseline had the highest fasting serum glucose level at subsequent visits. Consumption of specific alcoholic beverages and the risk of type 2 diabetes To assess whether the discrepancy between women and men might have been explained by differences in alcoholic beverage preference, we characterized the beverage preference of current drinkers. Figure 1 shows the mean percentage of ethanol contributed by specific alcoholic beverages for each consumption group (total alcohol intake) in women and men. In the reference groups ( 1 drink/week), although all persons identified themselves as current drinkers, only 11 percent of the women and 8 percent of the men reported regular weekly alcohol consumption (the remaining persons in the reference group identified themselves as current alcohol drinkers but reported no regular weekly alcohol consumption), and in those persons who reported regular use of alcohol, wine accounted for 100 percent for ethanol consumed. In general, the contribution to total alcohol intake from wine was roughly two times higher in women than in men, the contribution from beer was about two times higher in men than in women, and the contribution from spirits was similar between women and men. Next, we examined the association between incident type 2 diabetes and each of the three alcoholic beverages, spirits, beer, and wine, separately. Since the amount of alcohol consumed by female ARIC Study participants was relatively small, analysis by alcohol type was limited to the male participants. In each beverage analysis, the reference group was composed of male participants who identified themselves as current drinkers but reported no regular weekly consumption of wine, beer, or spirits. Table 5 shows the adjusted relative odds of diabetes from multiple logistic regression, which adjusted for intakes of other types of alcoholic beverages; that is, nine dummy variables were included in the regression to adjust for intake of other alcoholic beverages, in addition to the potential confounders mentioned previously. In the fully adjusted model, persons who drank >14 drinks of spirits per week were still at higher risk for developing diabetes (odds ratio 1.82, 95 percent confidence interval: 1.14, 2.92). There was no statistically significant association between wine intake and the risk of diabetes at the conventional 0.05 level. One possible explanation for the excess risk associated with spirits might be differences in underlying health behaviors as compared with wine drinkers, particularly with regard to diet. To investigate this possibility, we examined the correlation between the percentage of ethanol contributed by each alcoholic beverage and measures that were thought to indicate healthy behavior. In general, the increased percentage of ethanol from wine was associated with healthier behaviors, such as increased physical activity and total intake of dietary fiber and decreased total fat intake. Although these associations were statistically significant, the magnitude of the correlations was small ( 0 r 0 < 0.2). Additional adjustment for dietary intakes of carbohydrate, total fat, and protein did not alter the association between a high intake of spirits (>14 drinks/week) and the risk of diabetes in men (results not shown). DISCUSSION These data support the following conclusions. First, after adjustment for potential confounders, middle-aged women FIGURE 1. Mean percentage of ethanol contributed by specific alcoholic beverage in 12,261 participants by sex and consumption group, Atherosclerosis Risk in Communities Study,

7 754 Kao et al. TABLE 5. Adjusted* relative odds (95% confidence intervals) of incident type 2 diabetes mellitus in 5,423 male ARIC Study participants by alcoholic beverage type and alcohol consumption groups, Servings of alcohol per week >14 servings/week 8 14 servings/week 1 7 servings/week None Spirits Beer Wine 1.82 (1.14, 2.92) 1.02 (0.67, 1.56) 1.00 (0.77, 1.29) 1.19 (0.77, 1.85) 0.74 (0.48, 1.14) 0.88 (0.68, 1.13) 0.63 (0.08, 4.95) 0.84 (0.29, 2.44) 1.10 (0.82, 1.50) p for overall * Adjusted for age ( 49 vs vs vs. 59), race (binary), education (11 years or less vs. high school graduate vs. attended college), family history of diabetes (binary), body mass index (quartiles), waist/hip ratio (quartiles), physical activity score (continuous), total energy intake (quartiles), smoking history (never vs. former vs. current), history of hypertension (binary), spirits consumption (none vs. 1 7 servings/week, 8 14 servings/week, >14 servings/week), wine consumption (none vs. 1 7 servings/week, 8 14 servings/week, >14 servings/week), and beer consumption (none vs. 1 7 servings/week, 8 14 servings/week, >14 servings/week) using logistic regression. ARIC, Atherosclerosis Risk in Communities. One serving of spirits is one 1.5-ounce glass, one serving of beer is one 12-ounce can, and one serving of wine is one 4-ounce glass. One ounce = ml. and men who consumed a moderate amount of alcohol (1 14 drinks/week) were not at higher risk of developing type 2 diabetes compared with their counterparts who drank little alcohol ( 1 drink/week). Second, after adjustment for potential confounders and established diabetes risk factors, middle-aged men who consumed a substantial amount of alcohol (>21 drinks/week) were about 50 percent more likely to develop type 2 diabetes compared with their counterparts who drank 1 drink/week. Third, this increased risk of diabetes among men who drank >21 drinks/week was predominantly related to spirits consumption. The risk of diabetes among middle-aged men who drank >14 drinks of spirits per week was about 80 percent higher than the risk of men who were current drinkers but did not drink any spirits. There are two major limitations to this study. First, our analyses of heavy consumption and alcohol beverage types were limited to men only. This study did not have sufficient power to estimate the relative odds of diabetes at higher levels of alcohol intake in women, specifically consumption of 14 drinks/week. Second, like most previous studies in this field, our assessment of alcohol was likely suboptimal for several reasons. The assessment was done at a single time point and did not evaluate drinking pattern. Participants, especially those in the heavier drinking groups, may have underreported their consumption level because the questions regarding alcohol use were administered by an interviewer. Nevertheless, this study had several strengths. First, this is the only community-based, prospective study of the association between alcohol and the risk of type 2 diabetes with a sample that included Whites, Blacks, women, and men. Second, the present analyses distinguished among lifetime abstainers, former drinkers, and current drinkers who drink minimally. It has been suggested that persons who are not currently drinking may be abstaining from drinking alcohol because of health reasons, and that the use of such persons as the reference group could potentially lead to an artifactual protective effect of alcohol use (38). Third, to our knowledge, this study is the first study to present a more comprehensive examination of the associations between different alcoholic beverages and the risk of diabetes. Both the present study and a previous ARIC Study publication by Vitelli et al. (39) reported an inverse crosssectional relation between the baseline alcohol consumption and the baseline fasting serum insulin level in both men and women, and yet the present study also found a direct crosssectional association of the baseline alcohol consumption with the baseline fasting serum glucose level and a direct prospective relation with the risk of diabetes during follow-up in men. It is not clear why these contradictory associations exist. One hypothesis for these apparently contradictory results in men is that the inverse association between the baseline alcohol consumption and the baseline fasting serum insulin level could be indicative of the beginning of the deterioration of beta-cell function experienced by prediabetic persons rather than insulin sensitivity; hence, a direct association of the baseline alcohol consumption and the baseline fasting serum glucose level and incident diabetes during follow-up was observed. The divergent results from epidemiologic studies of alcohol consumption and the risk of diabetes are reflected in nonepidemiologic studies, which have also reported both diabetogenic and nondiabetogenic effects of ethanol on physiologic endpoints associated with diabetes. In vitro studies indicate that exposure of beta cells to alcohol is associated with decreased insulin secretion (40, 41). On the other hand, hepatic oxidation of ethanol results in an increased ratio of reduced nicotinamide adenine dinucleotide to nicotinamide adenine dinucleotide, which can lead to impairment in gluconeogenesis (42). In physiologic studies, acute ethanol ingestion has been shown to improve glucose tolerance in normal subjects and insulin sensitivity in diabetic subjects (11, 16); however, acute ethanol administration has been shown to cause a reduction in the glucose disposal rate, cause acute insulin resistance in nondiabetic

8 Alcohol Consumption and the Risk of Diabetes Mellitus 755 subjects, and worsen insulin resistance in diabetic patients, while chronic alcohol consumption has also been implicated to induce beta-cell dysfunction (10, 14, 15). Similarly contradictory results arise from animal studies. In rats, acute alcohol administration can attenuate basal and hormoneinduced glucose utilization by skeletal muscles while, in obese mice, chronic ethanol consumption can lower blood glucose and insulin (17, 18). The exact relation between alcohol and diabetes-related endpoints may depend on the nature of alcohol exposure, that is, acute versus chronic and alcohol administration versus alcohol ingestion. Since 1965, eight prospective studies have examined the association between alcohol consumption and the risk of type 2 diabetes (25, 26, 28 32, 43, 44). Five are comparable with the present study, while significant differences in the study methodology and study population exist between the other three studies and the present study. Of these three, two showed no association between alcohol intake and the risk of diabetes in men even at >21 drinks/week (29, 43). We believe that two methodological issues could explain the discrepancy between our study and these two studies. First, both previous studies lacked information on potential confounders, such as the waist/hip ratio, education, and hypertension, all of which were significantly associated with both alcohol and diabetes in our study. Second, these studies had data on current drinking status only and so had to combine former drinkers, lifetime abstainers, and occasional drinkers into a single reference group. In the present study, former drinkers and lifetime abstainers had the highest unadjusted risk of diabetes; thus, such misclassification of exposure status may lead to an overestimation of the risk of diabetes in the reference group and produce a bias toward the null. The last of these three studies found a protective association of moderate alcohol among the elderly (44). Differences in the age distribution of the two populations and, hence, in the inherent risks for developing diabetes and differences in the characterization of alcohol use (moderate use was defined as 0.5 <1 ounce/day) may explain the discrepancy with our results. Of the five studies that were comparable with the present study regarding methodology and population, three studies presented results regarding women (25, 28, 31, 32), and in these studies, either an inverse or no association between alcohol intake and diabetes risk was reported. The Nurses Health Study demonstrated a significant inverse association between light-to-moderate alcohol consumption, irrespective of beverage type, and the risk of type 2 diabetes (28, 32). The remaining two smaller studies showed that increased alcohol intake was not associated with diabetes risk but was strongly associated with a decreased body mass index (25, 31). This strong inverse association, as indicated by previous ARIC studies (34), may account for the lack of association between alcohol and the risk of diabetes. In our study, women with moderate alcohol consumption were not more likely to develop diabetes than were their counterparts who drank occasionally, even after adjustment for body mass index and the waist/hip ratio. Of the five studies that were comparable with the present study regarding methodology and population, four presented results regarding men (25, 26, 30, 31). The increased risk of diabetes seen in ARIC Study male participants who drank >21 drinks/week is consistent with results reported by two of these studies, the Rancho Bernardo Study and the San Antonio Heart Study (25, 31). However, an inverse association was found in the other two studies, the Health Professionals Follow-up Study and the British Regional Heart Study (26, 30). Given the unique nature of the Health Professionals cohort, it is conceivable that these persons had different beverage preference and drinking patterns from those of ARIC Study men or that they had healthier behaviors, which were not measured and were associated with increased consumption of alcohol. Furthermore, the association between the waist/hip ratio and alcohol consumption, which was positive in the ARIC Study, was unknown in the Health Professionals Study and therefore was not included in their multivariate analyses. The main difference between the present study and the British Regional Heart Study may lie in the populations studied. The prevalence of coronary heart disease was reported to be 24 percent in the British Regional Heart Study but only about 8 percent in male ARIC Study participants at baseline. Because the lower risk of diabetes seen in moderate drinkers was more apparent and significant only in men with preexisting coronary heart disease versus men without evidence of coronary heart disease in the British Regional Heart Study, one cannot exclude the possibility that the reference group in the British Regional Heart Study (current occasional drinkers) was contaminated with persons with prevalent coronary heart disease who may have been advised to reduce their alcohol consumption. Several reasons can possibly explain the differences in risk noted between women and men in this study. First, women and men may have differing responses to dietary questionnaires. Previous studies have shown differences between men and women in their attitudes toward alcohol use (45). If women who consumed high amounts of alcohol were more likely to develop diabetes but were also more likely to underreport the amount of alcohol they consumed, then this could result in an apparent protective effect of alcohol on diabetes risk. We have no knowledge of previous studies using questionnaires similar to ours that indicated that such underreporting occurs; however, it has been reported that the sensitivity of instruments designed to detect problem drinking, such as the CAGE questionnaire (a four-item test with questions on Cutting down, Annoyance at criticism, Guilty feelings, and use of Eye openers), is lower in college women than men (46). Second, a different beverage preference between women and men may explain some of the observed interaction. Our data showed that the increased diabetes risk in men in the highest alcohol intake group was related to the consumption of spirits. In this study, women were about as likely to consume spirits as men, but women were also more likely to consume wine and less likely to consume beer. Alcoholic beverage preference is associated with demographic and health behavior-related characteristics. Persons who prefer wine are likely to be women, temperate, nonsmokers, better educated, and free of symptoms or risk of illness while persons who prefer spirits are likely to be men, heavier drinkers, less educated, and

9 756 Kao et al. afflicted with symptoms or risk factors of major illnesses, and persons who prefer beer had intermediate traits (24). In addition, we demonstrated in our study that wine consumption was statistically associated with healthier behaviors while spirits and beer consumption were not. Therefore, we cannot exclude the possibility that the differences in beverage preference may simply represent differences in lifestyles, which we were not able to adequately measure and adjust in our regression model. Third, this difference could simply be a type II error due to the small number of women who drank >14 drinks/week. The results of the present study support the hypothesis that high alcohol intake ( 21 drinks/week) predicts type 2 diabetes mellitus among middle-aged men, specifically men who drink more than 14 drinks of spirits per week. However, more moderate levels of consumption (<21 drinks/week) do not appear to increase the risk of type 2 diabetes in middle-aged men and women. Our results, along with those of previous studies, suggest that strategies for the prevention of type 2 diabetes need not target moderate alcohol consumption. In contrast, men who drink >14 drinks of spirits per week should be advised of the increased risk of diabetes associated with heavy alcohol consumption. Further research should investigate the potential effect modification by sex and alcoholic beverages in the association between alcohol and the risk of type 2 diabetes. ACKNOWLEDGMENTS The ARIC Study is carried out as a collaborative study supported by contracts N01-HC-55015, N01-HC-55016, N01-HC-55018, N01-HC-55019, N01-HC-55020, N01-HC , and N01-HC from the National Heart, Lung, and Blood Institute. F. L. B. was supported by an Established Investigator Grant from the American Heart Association (Dallas, Texas). W. H. L. K. was supported by NIH training grant T32HL The authors thank the staff and participants in the ARIC Study for their important contributions. REFERENCES 1. Harris MI, Flegal KM, Cowie CC, et al. Prevalence of diabetes, impaired fasting glucose, and impaired glucose tolerance in U.S. adults. The Third National Health and Nutrition Examination Survey, Diabetes Care 1998;21: Wingard DL, Barrett-Connor E. Heart disease and diabetes. In: National Diabetes Data Group, ed. Diabetes in America. Bethesda, MD: National Institutes of Health, 1995: Geiss LS, Herman WH, Smith PJ. Mortality in non-insulindependent diabetes. In: National Diabetes Data Group, ed. Diabetes in America. Bethesda, MD: National Institutes of Health, 1995: Berkelman RL, Ralston M, Herndon J, et al. Patterns of alcohol consumption and alcohol-related morbidity and mortality. MMWR CDC Surveill Summ 1986;35:1SS 5SS. 5. Avogaro A, Tiengo A. Alcohol, glucose metabolism and diabetes. Diabetes Metab Rev 1993;9: Manolio TA, Savage PJ, Burke GL, et al. Association of fasting insulin with blood pressure and lipids in young adults. The CARDIA Study. Arteriosclerosis 1990;10: Perry IJ, Wannamethee SG, Shaper AG. Prospective study of serum gamma-glutamyltransferase and risk of NIDDM. Diabetes Care 1998;21: Phillips GB, Safrit HF. Alcoholic diabetes. Induction of glucose intolerance with alcohol. JAMA 1971;217: Dornhorst A, Ouyang A. Effect of alcohol on glucose tolerance. Lancet 1971;2: Avogaro A, Valerio A, Miola M, et al. Ethanol impairs insulinmediated glucose uptake by an indirect mechanism. J Clin Endocrinol Metab 1996;81: McMonagle J, Felig P. Effects of ethanol ingestion on glucose tolerance and insulin secretion in normal and diabetic subjects. Metabolism 1975;24: Yki-Jarvinen H, Koivisto VA, Ylikahri R, et al. Acute effects of ethanol and acetate on glucose kinetics in normal subjects. Am J Physiol 1988;254(2 Pt 1):E Yki-Jarvinen H, Nikkila EA. Ethanol decreases glucose utilization in healthy man. J Clin Endocrinol Metab 1985;61: Shelmet JJ, Reichard GA, Skutches CL, et al. Ethanol causes acute inhibition of carbohydrate, fat, and protein oxidation and insulin resistance. J Clin Invest 1988;81: Patto RJ, Russo EK, Borges DR, et al. The enteroinsular axis and endocrine pancreatic function in chronic alcohol consumers: evidence for early beta-cell hypofunction. Mt Sinai J Med 1993;60: Facchini F, Chen YD, Reaven GM. Light-to-moderate alcohol intake is associated with enhanced insulin sensitivity. Diabetes Care 1994;17: Spolarics Z, Bagby GJ, Pekala PH, et al. Acute alcohol administration attenuates insulin-mediated glucose use by skeletal muscle. Am J Physiol 1994;267(6 Pt 1):E Alqatari M, Shih MF, Taberner PV. Chronic ethanol consumption ameliorates the maturity-onset diabetes-obesity syndrome in CBA mice. Alcohol Alcohol 1996;31: Mayer EJ, Newman B, Quesenberry CP Jr, et al. Alcohol consumption and insulin concentrations. Role of insulin in associations of alcohol intake with high-density lipoprotein cholesterol and triglycerides. Circulation 1993;88(5 Pt 1): Barrett-Connor E, Schrott HG, Greendale G, et al. Factors associated with glucose and insulin levels in healthy postmenopausal women. Diabetes Care 1996;19: Jose GM, Klatsky AL, Siegelaub AB, et al. Serum glucose levels and alcohol-consumption habits in a large population. Diabetes 1977;26: Razay G, Heaton KW, Bolton CH, et al. Alcohol consumption and its relation to cardiovascular risk factors in British women. BMJ 1992;304: Selby JV, Newman B, King MC, et al. Environmental and behavioral determinants of fasting plasma glucose in women. A matched co-twin analysis. Am J Epidemiol 1987;125: Klatsky AL, Friedman GD, Armstrong MA. The relationships between alcoholic beverage use and other traits to blood pressure: a new Kaiser Permanente study. Circulation 1986;73: Holbrook TL, Barrett-Connor E, Wingard DL. A prospective population-based study of alcohol use and non-insulindependent diabetes mellitus. Am J Epidemiol 1990;132: Perry IJ, Wannamethee SG, Walker MK, et al. Prospective study of risk factors for development of non-insulin dependent diabetes in middle aged British men. BMJ 1995;310: Medalie JH, Papier CM, Goldbourt U, et al. Major factors in the development of diabetes mellitus in 10,000 men. Arch Intern Med 1975;135: Stampfer MJ, Colditz GA, Willett WC, et al. A prospective study of moderate alcohol drinking and risk of diabetes in women. Am J Epidemiol 1988;128: Feskens EJ, Kromhout D. Cardiovascular risk factors and the

10 Alcohol Consumption and the Risk of Diabetes Mellitus year incidence of diabetes mellitus in middle-aged men. The Zutphen Study. Am J Epidemiol 1989;130: Rimm EB, Chan J, Stampfer MJ, et al. Prospective study of cigarette smoking, alcohol use, and the risk of diabetes in men. BMJ 1995;310: Monterrosa AE, Haffner SM, Stern MP, et al. Sex difference in lifestyle factors predictive of diabetes in Mexican-Americans. Diabetes Care 1995;18: Colditz GA, Manson JE, Stampfer MJ, et al. Diet and risk of clinical diabetes in women. Am J Clin Nutr 1992;55: The Atherosclerosis Risk in Communities (ARIC) Study: design and objectives. The ARIC investigators. Am J Epidemiol 1989; 129: Duncan BB, Chambless LE, Schmidt MI, et al. Association of the waist-to-hip ratio is different with wine than with beer or hard liquor consumption. Atherosclerosis Risk in Communities Study investigators. Am J Epidemiol 1995;142: Baecke JA, Burema J, Frijters JE. A short questionnaire for the measurement of habitual physical activity in epidemiological studies. Am J Clin Nutr 1982;36: Willett WC, Sampson L, Stampfer MJ, et al. Reproducibility and validity of a semiquantitative food frequency questionnaire. Am J Epidemiol 1985;122: SAS Institute, Inc. SAS user s guide: statistics, version 6.11 ed. Cary, NC: SAS Institute, Inc, Shaper AG, Wannamethee G, Walker M. Alcohol and mortality in British men: explaining the U-shaped curve. Lancet 1988;2: Vitelli LL, Folsom AR, Shahar E, et al. Association of dietary composition with fasting serum insulin level: the ARIC Study. Nutr Metab Cardiovasc Dis 1996;6: Tiengo A, Valerio A, Molinari M, et al. Effect of ethanol, acetaldehyde, and acetate on insulin and glucagon secretion in the perfused rat pancreas. Diabetes 1981;30: Holley DC, Bagby GJ, Curry DL. Ethanol-insulin interrelationships in the rat studied in vitro and in vivo: evidence for direct ethanol inhibition of biphasic glucose-induced insulin secretion. Metabolism 1981;30: Arky RA, Veverbrants E, Abramson EA. Irreversible hypoglycemia. A complication of alcohol and insulin. JAMA 1968; 206: Hodge AM, Dowse GK, Collins VR, et al. Abnormal glucose tolerance and alcohol consumption in three populations at high risk of non-insulin-dependent diabetes mellitus. Am J Epidemiol 1993;137: Gurwitz JH, Field TS, Glynn RJ, et al. Risk factors for noninsulin-dependent diabetes mellitus requiring treatment in the elderly. J Am Geriatr Soc 1994;42: Neve RJ, Lemmens PH, Drop MJ. Gender differences in alcohol use and alcohol problems: mediation by social roles and gender-role attitudes. Subst Use Misuse 1997;32: Werner MJ, Walker LS, Greene JW. Screening for problem drinking among college freshmen. J Adolesc Health 1994;15: