ORIGINAL INVESTIGATION. Alcohol Drinking Patterns and Risk of Type 2 Diabetes Mellitus Among Younger Women

ORIGINAL INVESTIGATION Alcohol Drinking Patterns and Risk of Type 2 Diabetes Mellitus Among Younger Women S. Goya Wannamethee, PhD; Carlos A. Camargo, Jr, MD, DrPH; JoAnn E. Manson, MD, DrPH; Walter C. Willett, MD, DrPH; Eric B. Rimm, ScD Objective: To examine the relationship between alcohol consumption and the incidence of type 2 diabetes mellitus among relatively young and middle-aged women. Methods: In a prospective study, 109690 women, aged 25 to 42 years, without a history of coronary heart disease, stroke, cancer, or diabetes mellitus completed a detailed lifestyle and medical history questionnaire in 1989. During 10 years of follow-up, we documented 935 incident cases of type 2 diabetes mellitus. Results: We found a nonlinear relationship between alcohol consumption and risk of type 2 diabetes mellitus after adjustment for multiple confounders, including body mass index, smoking, physical activity, and family history of diabetes mellitus (quadratic trend P=.003). Compared with lifelong abstainers, the adjusted relative risks (95% confidence intervals) were 0.80 (0.66-0.96) for those consuming 0.1 to 4.9 g/d, 0.67 (0.50-0.89) for those consuming 5.0 to 14.9 g/d, 0.42 (0.20-0.90) for those consuming 15.0 to 29.9 g/d, and 0.78 (0.34-1.78) for those consuming 30.0 g/d or more. Further adjustment for dietary factors, including glycemic load, trans-fatty acid, polyunsaturated fat, and total fiber intake, did not appreciably alter these findings. The inverse association with light to moderate drinking was most apparent in women who reported wine or beer drinking. Women who reported 30.0 g/d or more of liquor intake showed a significantly increased risk of diabetes mellitus compared with those who did not report liquor intake (adjusted relative risk, 2.50; 95% confidence interval, 1.00-6.23). Conclusion: Light to moderate alcoholic beverage consumption may be associated with a lower risk of type 2 diabetes mellitus among women aged 25 to 42 years, although this benefit may not persist at higher levels. Arch Intern Med. 2003;163:1329-1336 From the Departments of Nutrition (Drs Wannamethee, Willett, and Rimm) and Epidemiology (Drs Manson, Willett, and Rimm), Harvard School of Public Health, Boston, Mass; the Department of Primary Care and Population Science, Royal Free and University College Medical School, London, England (Dr Wannamethee); the Channing Laboratory, Boston (Drs Camargo, Manson, Willett, and Rimm); the Department of Emergency Medicine, Massachusetts General Hospital and Harvard Medical School, Boston (Dr Camargo); and the Division of Preventive Medicine, the Department of Medicine, Brigham and Women s Hospital and Harvard Medical School, Boston (Dr Manson). The authors have no relevant financial interest in this article. PROSPECTIVE STUDIES of alcohol intake and risk of type 2 diabetes mellitus have produced conflicting results. Some, 1-5 but not all, 6,7 studies have implicated heavy drinking as a risk factor for type 2 diabetes mellitus. 1-4 Conversely, several recent prospective studies, 8-13 mainly conducted in men, have suggested that light to moderate drinking may be inversely associated with the development of type 2 diabetes mellitus. This is consistent with the observations that low to moderate amounts of alcohol intake may increase insulin sensitivity and slow glucose uptake from a meal. 14-17 Few prospective studies have examined the relationship between alcohol intake and diabetes mellitus in women. In an early study of women aged 34 to 59 years from the original Nurses Health Study, low to moderate alcohol consumption was associated with a lower risk of diabetes mellitus, 18 which was confirmed in a later analysis from the same cohort. 13 In the Rancho Bernardo cohort 1 and in the San Antonio Heart Study, 4 alcohol consumption was positively associated with risk of diabetes mellitus in men, but not in women. In the latter study, 4 alcohol consumption was inversely associated with type 2 diabetes mellitus in women. The influence of patterns of drinking or type of drink on the risk of diabetes mellitus has been less studied, but the inverse relationship seems to depend on frequency and pattern of drinking. 12 This study aims to examine the relationship between alcohol consumption and type 2 diabetes mellitus among women aged 25 to 42 years who participated in the Nurses Health Study II, with focus on (1) the role of lifestyle factors that may confound or modify the main association, (2) the influence of beverage choice, and (3) the impact of frequency of alcohol consumption. METHODS The Nurses Health Study II is an ongoing prospective cohort study, patterned after the origi- 1329

nal Nurses Health Study, designed to examine associations between lifestyle and nutritional factors and the occurrence of breast cancer and other major illnesses. In 1989, 116671 female registered nurses aged 25 to 42 years who were living in 1 of 14 states in the US responded to a baseline questionnaire about their medical history and lifestyle. Follow-up questionnaires have been mailed every 2 years in 1991, 1993, 1995, 1997, and 1999 to elicit updated information on demographic data and to assess newly diagnosed medical conditions, including type 2 diabetes mellitus. We excluded from the analysis women who reported diabetes mellitus, heart attack or angina, stroke, gestational diabetes, and nonmelanoma cancer at baseline (n=5964), women with missing data on alcohol consumption (n=1015), and women with incomplete data on mortality (n=2), leaving 109690 eligible women who were followed up for incidence of type 2 diabetes mellitus. ALCOHOL CONSUMPTION In 1989, the baseline questionnaires included questions on the average intake of alcoholic beverages (beer, wine, and liquor) during the past year. Nurses responded to the following question: During the past year, what was your usual consumption of these alcoholic beverages none or less than 1/month, 1-3/month, 1/week, 2-4/week, 5-6/week, 7-13/week, 14-24/week, 25-39/ week, and 40+/week. Total alcohol intake was calculated in grams by adding usual intake of alcoholic beverages, assuming the following content: one 360-mL can of beer, 12.8 g; one 120-mL glass of wine, 11.0 g; and one standard drink of liquor, 14.0 g. 19 The reliability and validity of reported alcohol intake were evaluated in a subset of women participating in a similar study 20 ; alcohol intake during the previous year, as assessed in this manner, was highly correlated with alcohol intake as assessed by diet records (r=0.90 in women). 20 In the present study, women were also asked about their past alcohol consumption, when aged 15 to 17, 18 to 22, 23 to 30, and 31 to 40 years. Current nondrinkers who reported drinking during any of these periods were classified as exdrinkers. Thirty-two nondrinkers provided no detail about past drinking habits and were excluded from analyses, separating lifelong abstainers and ex-drinkers. Alcohol intake was also assessed in 1991 and 1995, although the questions were expanded to obtain more detailed information on types of beer and wine. On the 1991 and 1995 questionnaires, alcohol consumption was assessed in a similar manner, except that there were 2 separate questions about consumption of red and white wine and 2 separate questions about regular and light beer. Light beer was estimated to contain 11.3 g per 360-mL serving. DIETARY ASSESSMENT A previously validated, 116-item, semiquantitative food frequency questionnaire was used to assess diet in 1991. 21 For each food, a commonly used unit or portion size (eg, 1 slice of bread) was specified and each participant was asked how often, on average, during the previous year she had consumed that amount. Nine responses were possible, ranging from never to more than 6 per day. Nutrient intake was computed by multiplying the frequency of consumption of each unit of food by the nutrient content, estimated from standard food composition sources. The methods used for calculating nutrient intake and assessing glycemic load and data on questionnaire validity have been reported. 22,23 All dietary factors and glycemic load were adjusted for total calories (energy). Analyses, including dietary factors, were based on total alcohol intake reported in 1991. IDENTIFICATION OF DIABETES MELLITUS CASES A supplementary questionnaire about symptoms, diagnostic tests, and hypoglycemic therapy was mailed to participants who reported a diagnosis of diabetes mellitus on any biennial questionnaire. Before 1997, a case of diabetes mellitus was considered confirmed if at least 1 of the following was reported on the supplementary questionnaire: (1) at least 1 classic symptom (excessive thirst, polyuria, weight loss, hunger, or pruritus) and a fasting plasma glucose concentration of at least 140 mg/dl (7.8 mmol/l) or a random plasma glucose concentration of at least 200 mg/dl (11.1 mmol/l); (2) at least 2 elevated plasma glucose concentrations on different occasions (fasting, 140 mg/dl; random, 200 mg/dl; and/or a concentration 200 mg/dl after 2 hours on glucose tolerance testing) in the absence of symptoms; or (3) treatment with a hypoglycemic medication (insulin or an oral hypoglycemic agent). Women with confirmed diabetes mellitus who began taking insulin within 1 year of the diagnosis and who had a history of ketoacidosis or ketonuria on at least 2 occasions were considered to have type 1 diabetes mellitus, and were excluded during follow-up. The criteria for the classification of diabetes mellitus have been published in detail elsewhere. 24 The diagnostic criteria for type 2 diabetes mellitus changed in June 1996, and a fasting glucose concentration of 126 mg/dl (7.0 mmol/l) is considered the threshold for a diagnosis of diabetes mellitus. 25 This lower threshold for fasting glucose concentration ( 126 mg/dl) was used to define cases that occurred after 1996. STATISTICAL ANALYSIS Person-time for each participant was calculated from the date of return of the 1989 baseline questionnaire to the date of diabetes mellitus diagnosis, death, or June 30, 1999, whichever came first. Women who reported coronary events, stroke, gestational diabetes, or cancer on previous questionnaires or who developed type 1 diabetes mellitus were excluded from the subsequent follow-up; thus, the cohort at risk included only those who remained free from the conditions listed at the beginning of each follow-up interval. Women were divided into categories based on usual alcohol consumption. Pooled logistic regression was used to model risk of diabetes mellitus during each 2-year follow-up period in relation to alcohol intake reported at baseline, with adjustment for other potential confounders. In addition, we excluded (skipped) women who were pregnant at the start of each follow-up interval. These women would not add person-time to that period, but reentered the analyses in subsequent follow-up periods. Women pregnant at baseline (n=4773) were excluded from analyses examining baseline characteristics. Time-varying covariates were updated biennially when possible (age, cigarette smoking status, body mass index [BMI] [calculated as weight in kilograms divided by the square of height in meters], oral contraceptive use, history of high blood pressure and high cholesterol, and use of an antihypertensive drug), in 1991, 1993, 1995, and 1997. History of high blood pressure and high cholesterol was based on recall of physician diagnosis. Physical activity was assessed in 1989 and 1991. Family history of diabetes mellitus and history of infertility were assessed at baseline. We also controlled for history of diagnosis of infertility because there is evidence that the gene determining type 2 diabetes mellitus is associated with diminished fertility. 26 When data were missing at the start of the interval, we carried forward the last available data. Average alcohol consumption was assessed from the baseline questionnaire and from the semiquantitative food frequency questionnaires in 1991 and 1995. Because the 1991 and 1995 assessments were modified from those used at baseline and because participants may have reduced their alcohol intake during follow-up because of the development of prediagnostic symptoms, the primary analyses were based on baseline alcohol data. However, in secondary analyses, we updated 1330

Table 1. Age-Standardized Baseline Characteristics According to Alcohol Consumption in 1989 Among 104 885 Women Aged 25 to 42 Years Who Had No History of Cardiovascular Disease, Diabetes Mellitus, or Cancer and Who Were Not Pregnant at Baseline* Alcohol Consumption, g/d Characteristic Lifelong Abstainers Ex-Drinkers 0.1-4.9 5.0-14.9 15.0-29.9 30.0 No. of subjects 14 917 24 125 44 384 18 379 2316 764 Age, mean, y 34.8 34.7 34.3 34.3 34.7 35.4 BMI, mean 24.4 24.8 23.9 23.0 23.2 24.0 Obese (BMI 30) 13.8 15.1 10.5 5.7 6.9 10.3 Current smokers 4.2 13.8 13.0 20.1 27.5 46.9 Inactive 49.9 47.3 40.0 34.9 34.3 42.8 Family history of diabetes mellitus 17.6 17.6 16.4 14.2 14.2 16.8 High BP 5.5 6.3 5.0 4.5 6.2 8.4 Undergoing antihypertensive therapy 3.1 3.1 2.9 2.6 3.6 5.5 High cholesterol 10.0 11.5 10.5 9.0 9.2 10.1 Taking an oral contraceptive 9.9 10.7 14.5 17.3 20.1 16.0 Infertile 18.4 19.6 17.8 16.6 16.0 15.2 Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by the square of height in meters); BP, blood pressure. *Data are given as percentage of women unless otherwise indicated. Fewer than 10.5 metabolic equivalents per week. Father, mother, or sibling has a history of diabetes mellitus. Table 2. Adjusted RRs of Diabetes Mellitus by Alcohol Consumption at Baseline* RR (95% CI) Multivariate Model Variable Person-Years No. of Cases Age Adjusted 1 2 Lifelong abstainers 128 453 181 1.00 1.00 1.00 Ex-drinkers 205 034 334 1.18 (0.98-1.41) 1.11 (0.92-1.33) 1.01 (0.84-1.22) Alcohol consumption, g/d 5.0 371 574 336 0.67 (0.56-0.80) 0.72 (0.59-0.86) 0.80 (0.66-0.96) 5.0-14.9 153 319 70 0.34 (0.25-0.44) 0.40 (0.30-0.53) 0.67 (0.50-0.89) 15.0-29.9 19 257 8 0.29 (0.15-0.60) 0.31 (0.15-0.63) 0.42 (0.20-0.90) 30.0 6502 6 0.63 (0.28-1.42) 0.54 (0.23-1.22) 0.78 (0.34-1.78) P value For linear trend NA NA.001.001.003 For quadratic trend NA NA.001.001.003 Abbreviations: CI, confidence interval; NA, data not applicable; RR, relative risk. *Data are from the 10-year follow-up of 109 705 women in the Nurses Health Study II who were free of baseline cardiovascular disease, cancer, and diabetes mellitus. Adjusted for age, smoking status, physical activity, family history of diabetes mellitus (mother, father, or sibling), use of an oral contraceptive, history of hypertension, use of antihypertensive drugs, elevated cholesterol level, and infertility. Adjusted for all covariates in multivariate model 1 and for body mass index. alcohol data in those periods when it was assessed. Dietary variables were only collected in 1991 and 1995. Analyses adjusting for these dietary variables are based on the 1991 questionnaire, with 8 years of follow-up. The multivariate models included terms for age (5-year categories), cigarette smoking (never, past, and 1-14, 15-24, and 25 cigarettes per day), level of physical activity (metabolic equivalents per week in 5 groups), BMI ( 20.0, 20.0-21.9, 22.0-23.9, 24.0-25.9, 26.0-27.9, 28.0-29.9, 30.0-31.9, 32.0-33.9, 34.0-35.9, and 36.0), family history of diabetes mellitus (yes or no), current use of an oral contraceptive (yes or no), history of hypertension (yes or no), and history of high cholesterol (yes or no). We retained BMI as a covariate, although it may be a confounder and a causal intermediate, to provide maximal control of possible confounding. In additional analyses adjusting for dietary factors, trans-fatty acid, glycemic load, polyunsaturated fat, and total fiber were included as continuous variables. We conducted tests of linear trend across increasing categories of alcohol consumption by assigning the median for the alcohol categories and treating the categories as a continuous variable. To assess departure from linearity, we included linear and quadratic terms (the median and the value squared) in the model. To examine whether the relation of alcohol intake to the risk of diabetes mellitus was modified by important covariates, we conducted analyses stratified by age at enrollment ( 35 and 35 years), categories of BMI ( 30.0 or 30.0), cigarette smoking (never, ex-smoker, and current), family history of diabetes mellitus (no or yes), and physical activity (metabolic equivalents per week, 10.5 or 10.5). The values for smoking and BMI were updated biennially. Because we had less power in our subgroup analyses, women with an alcohol intake of 30.0 g/d or more were excluded and teetotalers and ex-drinkers were combined. We excluded the few women with an intake of 30.0 g/d or more rather than combining them with the 15.0- to 29.9-g/d group because there was some evidence of an upturn in risk in this small group of women. 1331

Table 3. Alcohol Consumption in 1991 and Age-Adjusted Mean Intake of Dietary Factors Among 72 054 Women Aged 25 to 42 Years at Baseline and Adjusted RRs of Diabetes Mellitus During 8 Years of Follow-up Variable Lifelong Abstainers (n = 9550) Ex-Drinkers (n = 20 312) 0.1-4.9 (n = 28 348) Alcohol Consumption, g/d 5.0-14.9 (n = 10 937) 15.0-29.9 (n = 2070) 30.0 (n = 837) Intake, mean trans-fatty acids, g/d 3.4 3.4 3.3 3.1 3.0 2.8 Total fiber, g/d 18.6 18.3 18.6 18.3 17.3 15.2 Polyunsaturated fats, g/d 11.3 11.4 11.4 11.4 11.1 10.3 Glycemic load, U/d 12 777 12 440 12 105 11 454 10 490 9213 Person-years 76 469 167 231 226 238 85 635 15 986 6469 No. of cases 131 307 214 41 8 6 Multivariate RR (95% CI) Model 1* 1.00 0.94 (0.76-1.16) 0.74 (0.59-0.93) 0.58 (0.40-0.84) 0.52 (0.24-1.11) 0.64 (0.26-1.57) Model 2 1.00 0.94 (0.76-1.16) 0.75 (0.59-0.94) 0.61 (0.42-0.87) 0.55 (0.25-1.20) 0.71 (0.29-1.78) Abbreviations: CI, confidence interval; RR, relative risk. *Adjusted for age, smoking status, physical activity, body mass index, family history of diabetes mellitus (mother, father, or sibling), use of an oral contraceptive, history of hypertension, use of antihypertensive drugs, elevated cholesterol level, and infertility. Adjusted for all covariates in multivariate model 1 and for calorie-adjusted dietary glycemic load, fiber, trans-fatty acids, and polyunsaturated fats. Tests for interaction were conducted to evaluate whether the trends differed significantly by levels of risk factors. RESULTS Table 1 shows the distribution of baseline characteristics across categories of alcohol intake after adjusting for age. At baseline, nearly 40% of the women reported no alcohol consumption and 14.2% were classified as lifelong abstainers. Nondrinkers (lifelong abstainers and exdrinkers) were significantly (P.001) heavier than drinkers. Overall, light drinkers (5.0-14.9 g/d) tended to have the most favorable characteristics. During 10 years of follow-up, there were 935 incident cases of type 2 diabetes mellitus. In age-adjusted analyses, there was a significant inverse relationship between alcohol intake and risk of diabetes mellitus up to levels of 29.9 g/d. Relatively few women consumed 30.0 g/d or more, and among these women, the inverse association was no longer significant (Table 2). The highest risk was seen in ex-drinkers, followed by lifelong abstainers. Adjustment for potential confounders, including cigarette smoking, oral contraceptive use, history of high blood pressure and high cholesterol, use of an antihypertensive drug, physical activity, family history of diabetes mellitus, and history of infertility, made minor differences to the relationship. Further adjustment for BMI somewhat attenuated the risk reductions, but women drinking up to 29.9 g/d still had significantly lower risks than lifelong abstainers. Ex-drinkers showed similar risks as lifelong abstainers. A test for quadratic trend (adding a quadratic term to the linear term in the model) was significant, suggesting the presence of a nonlinear trend. We also examined the relationship between alcohol intake and risk of diabetes mellitus, updating alcohol intake measurements. The pattern was broadly similar to that seen at baseline, with the lowest risks in those consuming 5.0 to 29.9 g/d. The adjusted relative risks (RRs) (95% confidence intervals) were 1.00 for lifelong abstainers, 1.04 (0.86-1.24) for ex-drinkers, 0.75 (0.62-0.91) for those consuming 0.1 to 4.9 g/d, 0.52 (0.38-0.72) for those consuming 5.0 to 14.9 g/d, 0.55 (0.29-1.05) for those consuming 15.0 to 29.9 g/d, and 0.71 (0.33-1.52) for those consuming 30.0 g/d or more. We further assessed the possible role of confounding by dietary factors shown to be associated with risk of type 2 diabetes mellitus in earlier studies 22,23 (dietary glycemic load, total polyunsaturated fat, total transfatty acid, and total fiber). Inverse relationships are seen between alcohol intake and these dietary factors (Table 3); heavy drinkers had the lowest level of total fiber and polyunsaturated fat, but they also had the lowest mean level of trans-fatty acid and glycemic load. To assess confounding by dietary variables, we examined the relationship between alcohol intake and diabetes mellitus using 1991 data and 8 years of follow-up, adjusting first for the factors in Table 2 (P.001 for an inverse linear trend) and then for the dietary variables. Additional adjustment for trans-fatty acid, total fiber, polyunsaturated fat, and glycemic load did not appreciably change the relation. We observed a significant inverse linear trend (P.001), but a test for departure from linearity was not significant in this follow-up period (P=.20). We examined whether the association between alcohol consumption and diabetes mellitus was modified by important risk factors for diabetes mellitus by stratifying according to age at enrollment ( 35 or 35 years), BMI ( 30.0 or 30.0), smoking (never, ex-smoker, and current), family history of diabetes (no or yes), and physical activity (inactive or active). Because lifelong abstainers and ex-drinkers showed similar risks, these 2 groups were combined to form a reference group of nondrinkers. Because of the few women in the heavy drinking categories, we excluded all women who drank more than 29.9 g/d. Nondrinkers had a higher risk than all drinkers up to 29.9 g/d, within all risk factor stratification (Table 4). The inverse association between alcohol consumption and diabetes mellitus was stronger in inactive 1332

Table 4. Alcohol Consumption and Adjusted RRs of Diabetes Mellitus by Levels of Risk Factors During 10 Years of Follow-up* Alcohol Consumption, g/d Variable 0 0.1-4.9 5.0-14.9 15.0-29.9 P Value for Trend Age at enrollment, y 35 (256/396 575) 1.00 0.88 (0.68-1.15) 0.68 (0.42-1.10) 0.27 (0.04-1.92).04 35 (673/481 328) 1.00 0.75 (0.64-0.89) 0.64 (0.47-0.86) 0.45 (0.20-1.00).001 Obese (BMI 30) No (185/730 866) 1.00 0.65 (0.47-0.89) 0.55 (0.34-0.88) 0.17 (0.02-1.24).004 Yes (731/140 767) 1.00 0.83 (0.71-0.98) 0.69 (0.51-0.94) 0.53 (0.24-1.20).005 Cigarette smoking None (566/566 557) 1.00 0.78 (0.65-0.94) 0.68 (0.47-0.98) 0.59 (0.22-1.58).01 Ex (240/200 246) 1.00 0.82 (0.62-1.08) 0.77 (0.51-1.18) 0.32 (0.08-1.30).06 Current (123/11 100) 1.00 0.70 (0.47-1.03) 0.36 (0.18-0.74) 0.22 (0.03-1.58).003 Family history of diabetes mellitus No (575/732 431) 1.00 0.71 (0.59-0.85) 0.64 (0.47-0.88) 0.52 (0.23-1.18).003 Yes (354/145 472) 1.00 0.92 (0.73-1.15) 0.66 (0.43-1.02) 0.17 (0.02-1.23).007 Physical activity Inactive ( 10.5 METS/wk) (492/372 848) 1.00 0.75 (0.62-0.91) 0.48 (0.32-0.73) 0.29 (0.09-0.91).001 Active ( 10.5 METS/wk) (432/501 736) 1.00 0.83 (0.68-1.02) 0.78 (0.55-1.09) 0.58 (0.21-1.57).08 Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by the square of height in meters); METS, metabolic equivalents; RR, relative risk. *Data are given as RR (95% confidence interval) unless otherwise indicated. The RRs are adjusted for age, use of an oral contraceptive, history of hypertension, use of antihypertensive drugs, elevated cholesterol level, infertility, and each of the other factors (smoking status, BMI, family history of diabetes mellitus [mother, father, or sibling], and physical activity). Within nonobese (BMI 30) and obese (BMI 30) women, adjustments were made for BMI levels. Data in parentheses are number of cases/person-years at risk. There were missing data for 13 cases. There were missing data for 5 cases. women than in active women, but a test for interaction was not significant (P=.12). No significant interaction was seen between alcohol consumption and age (P=.77), obesity (P=.20), current smoking status (P=.15), or family history (P=.88). We assessed type of drink and risk of diabetes mellitus (Table 5). The inverse association was clearly seen for beer and wine drinkers. Less risk reduction was seen with liquor intake, and high liquor consumption ( 30.0 g/d) was associated with a higher risk than that seen among women who did not drink liquor. Even compared with nondrinkers, women who reported drinking 30.0 g/d or more of liquor had an increase in risk (RR, 2.28; 95% confidence interval, 0.90-5.74), although this was not significant at the 5% level (P=.08). Although numbers were small, when analysis was confined to women who drank only one type of beverage, using nondrinkers as the reference group, the reduction in risk of diabetes mellitus associated with light to moderate drinking (5.0-29.9 g/d) was more apparent in wine and beer drinkers. Compared with nondrinkers, the adjusted RRs (95% confidence intervals) for those drinking 5.0 to 29.9 g/d were 0.41 (0.10-1.65) for wine, 0.74 (0.39-1.39) for beer, and 0.93 (0.49-1.76) for liquor. We also examined the pattern of alcohol consumption and the risk of diabetes mellitus. The 4 categories of drinkers (0.1-4.9, 5.0-14.9, 15.0-29.9, and 30.0 g/d) were divided further into those who drank regularly (4-7 d/wk) and those who drank less frequently (1-3 d/wk). Within each drinking category, these 2 groups had similar alcohol intake on a weekly basis. Women who consumed on average 5.0 g/d or more of alcohol all showed a lower risk than nondrinkers, irrespective of frequency of drinking. Regular heavy drinkers showed a higher risk than all light to moderate drinkers (5.0-29.9 g/d). The number of episodic heavy drinkers (n=126) was too small to assess the relation of binge drinking to diabetes mellitus risk. Those drinking, on average, 15.0 to 29.9 g/d of alcohol for 4 to 7 d/wk showed a lower risk than those who consumed the same amount on a weekly basis, but consumed less frequently (1-3 d/wk). Compared with nondrinkers, the adjusted RRs (95% confidence intervals) were 0.29 (0.09-0.91) and 0.62 (0.23-1.68), respectively. However, the numbers were small and the result of a test to see whether the relationship between average alcohol consumption and diabetes mellitus differed in regular and irregular drinkers was not statistically significant (P=.30). COMMENT In this cohort of mainly premenopausal women, a nonlinear relationship was seen between alcohol consumption and risk of type 2 diabetes mellitus. We found a linear inverse association up to levels of 29.9 g/d, beyond which risk increased compared with light and moderate drinkers. Light to moderate drinking was associated with a significantly lower risk than lifelong abstention, even after adjustment for potential confounders. The inverse association was not substantially modified by other known risk factors for diabetes mellitus. Additional adjustment for dietary factors, including calorie-adjusted transfatty acid, polyunsaturated fat, fiber, and glycemic load, did not influence the alcohol diabetes mellitus relationship. The lower risk associated with light to moderate drinking was more apparent in beer and wine drinkers. Our findings of an inverse association between regular light to moderate drinking and risk of type 2 diabe- 1333

Table 5. Adjusted RRs of Type 2 Diabetes Mellitus in Relation to Type of Beverage During 10 Years of Follow-up RR (95% CI) Multivariate Model Variable Person-Years No. of Cases Age Adjusted 1* 2 Wine Alcohol consumption, g/d 0 460 471 638 1.00 1.00 1.00 5.0 392 350 287 0.53 (0.46-0.61) 0.79 (0.68-0.91) 0.83 (0.71-0.96) 5.0-29.9 31 094 10 0.20 (0.11-0.37) 0.53 (0.28-1.00) 0.57 (0.30-1.07) P value for trend NA NA.001.01.03 Beer Alcohol consumption, g/d 0 620 315 782 1.00 1.00 1.00 5.0 176 359 120 0.59 (0.49-0.72) 0.84 (0.69-1.02) 0.92 (0.75-1.13) 5.0-29.9 84 076 32 0.34 (0.24-0.49) 0.62 (0.43-0.89) 0.68 (0.47-0.98) 30.0 3656 1 0.23 (0.03-1.62) 0.26 (0.04-1.88) 0.27 (0.04-1.95) P value for trend NA NA.001.008.03 Liquor Alcohol consumption, g/d 0 620 767 704 1.00 1.00 1.00 5.0 213 014 197 0.85 (0.72-0.99) 0.87 (0.74-1.03) 0.96 (0.81-1.15) 5.0-29.9 49 040 29 0.51 (0.35-0.74) 0.66 (0.45-0.96) 0.78 (0.53-1.15) 30.0 1584 5 2.48 (1.02-6.00) 1.97 (0.80-4.86) 2.50 (1.00-6.23) P value For linear trend NA NA.35.92.33 For quadratic trend NA NA.001.003.05 Abbreviations: CI, confidence interval; NA, data not applicable; RR, relative risk. *Adjusted for age, smoking status, physical activity, body mass index, family history of diabetes mellitus (mother, father, or sibling), use of an oral contraceptive, history of hypertension, use of antihypertensive drugs, elevated cholesterol level, and infertility. Adjusted for all covariates in multivariate model 1 and for other beverages. The number of women consuming 30.0 g/d or more of wine was small (n = 54), and data are not presented for this group. tes mellitus are consistent with reports from several other prospective studies, mainly conducted among middleaged men. In the US Health Professionals Study, comprising some 42000 men aged 40 to 75 years followed up for 6 years, men who drank 30.0 to 49.9 g/d of alcohol showed a 40% reduction in risk of diabetes mellitus after full adjustment, compared with abstainers; risk increased in the higher-intake group ( 50.0 g/d). 8 In a later report 12 from this study, based on 12 years of follow-up, an inverse relationship was observed between regular alcohol consumption and risk of diabetes mellitus, irrespective of type of beverage. In the British Regional Heart study of more than 7700 men aged 40 to 59 years, a U- shaped relationship was seen, with the lowest risk in moderate drinkers (16-42 drinks per week) (RR, 0.60 [compared with nondrinkers]). 9 In the Osaka Health Survey of more than 6000 Japanese men aged 35 to 61 years followed up for 10 years, a U-shaped relationship was seen, with the lowest risk in those drinking 29.1 to 50.0 g/d (moderate drinking). 10 The Cooper Clinic Study (in Texas) examined more than 8600 men aged 30 to 79 years with 6 years of follow-up and observed a U-shaped relationship, with the lowest risk in those drinking 61.9 to 122.7 g/wk (about 1-2 US drinks per day). 3 In the US Physicians Health Study of 21000 men aged 40 to 84 years followed up for 12 years, those drinking 2 to 4 US drinks per week, 5 to 6 drinks per week, and 1 drink per day or more showed significantly lower risk than those who rarely or never drank, with the lowest risk (RR, 0.57) in the 1 drink per day or more group. 11 Relatively few prospective studies have examined the relationship in women. In 2 small prospective studies, 1,4 light drinking was modestly inversely associated with risk, but the findings were not significant. In the US Nurses Health Study of 85000 women aged 35 to 59 years, followed up for 4 years, the RR of diabetes mellitus decreased progressively from 1.0 in nondrinkers to 0.3 in the heaviest drinking category ( 15.0 g/d). Additional adjustment for BMI attenuated the inverse association, and the possible protective effect of alcohol was considered to be due to residual confounding by obesity and its biological consequences. 18 In a later analysis 13 from the same cohort, based on 16 years of follow-up, the inverse relation persisted after adjustment for BMI. Although BMI accounted for some of the reduction in risk seen in the present cohort, there still remained a significant reduction in risk among light to moderate drinkers ( 15 g/d) compared with lifelong abstainers after adjustment. There has been limited research on patterns and type of drink on risk of diabetes mellitus. In this study, the reduction in risk associated with light and moderate drinking was more apparent among beer and wine drinkers; less benefit was seen in those who consumed liquor. These findings are similar to those of the recent report from the Atherosclerosis Risk in Communities Study, 5 in which 1334

no benefit was seen for men who consumed liquor. However, a recent report 12 from the Health Professionals Study observed benefit for all beverage types after mutual adjustment. Light to moderate drinking was significantly inversely associated with risk of diabetes mellitus only when consumed frequently ( 5 d/wk). Those who consumed alcohol 2 d/wk or fewer did not have a lower risk, even though their weekly consumption was similar. 12 In the present study, we observed that all women who drank 5 g or more a day had a lower risk of diabetes mellitus than nondrinkers, irrespective of the pattern of drinking, although moderate intake of alcohol (5.0-29.9 g/d), when taken regularly (4-7 d/wk), was associated with a lower risk than when the same amount was consumed over 1 to 3 d/wk. It is unlikely that our results are biased by preexisting disease, because we excluded all women with cardiovascular disease or cancer at baseline and censored all women who developed these diseases during followup. Furthermore, we separated lifelong abstainers from ex-drinkers who may have given up because of ill health, and light to moderate drinkers still showed lower risk than lifelong abstainers. Although we relied on selfreported consumption of alcohol, the validity and reliability of these questions have been documented. 20 While some misclassification may occur among drinking groups and subjects change drinking habits over time, the findings based on baseline data and updated alcohol data showed consistently lower risk among regular drinkers, up to 29.9 g/d, than lifelong abstainers. Although the inverse relationship between light to moderate drinking and diabetes mellitus persisted after adjustment and stratification by BMI, there remains the possibility of residual confounding arising from imprecise measurements of adiposity, but this is not likely to be substantial. The finding that light to moderate drinking is associated with lower risk of type 2 diabetes mellitus is biologically plausible. It is established that insulin resistance and hyperinsulinemia play an important role in the cause of type 2 diabetes mellitus, 27,28 and light to moderate alcohol intake may be associated with enhanced insulin sensitivity. 14-17 This is consistent with the findings of a stronger inverse association seen among inactive women who are more likely to be insulin resistant. 29 Heavy drinking has been implicated as a risk factor for type 2 diabetes mellitus. 1-5 In most prospective studies, heavy drinkers, variously defined from as little as 2 drinks to 6 or more drinks a day, have a higher risk than light or moderate drinkers; and in many studies, heavy drinkers have the highest risk. In 2 cross-sectional population-based studies, 30,31 high alcohol consumption was associated with increased prevalence of type 2 diabetes mellitus in men. Most studies 1,4,5 that have included men and women have shown high alcohol consumption to be a risk factor in men only. However, this may relate to the limited number of women who drank at higher levels. In the Rancho Bernardo Study, 1 high alcohol use was defined as just more than 15 g/d; and in the San Antonio Heart Study, 4 comparisons were made between those drinking just more than 10 g/wk and those drinking less. In the present study, the number of women who reported heavy drinking ( 30.0 g/d) was small, but this group did have a higher risk of diabetes mellitus than the light and moderate drinkers, especially among those consuming 30.0 g/d or more of liquor. The increased risk seen in heavy liquor drinkers, but not in heavy beer drinkers, may be because of their higher overall total alcohol consumption rather than beverage per se. Women who reported drinking 30.0 g/d or more of liquor had a substantially higher total average consumption (average, 54 g/d) than those reporting drinking 30.0 g/d or more of beer (average, 44 g/d). However, our findings are consistent with the recent findings from the Atherosclerosis Risk in Communities Study, 5 in which the increased risk among men who drank more than 21 drinks per week was predominantly related to liquor rather than to beer or wine. In conclusion, we found a nonlinear relationship between alcohol consumption and risk of type 2 diabetes mellitus. These data from the Nurses Health Study II of younger women support the hypotheses that light to moderate drinking is inversely associated with risk of type 2 diabetes mellitus and that heavy drinking does not confer benefit; high levels of liquor intake may confer increased risk. Further studies are needed to determine whether the increased risk associated with heavy drinking is dependent on beverage type. Despite the consistent association between light to moderate drinking and lower risk of diabetes mellitus, the potential harmful effects of drinking on other aspects of health outcome need to be considered. 32 Light to moderate alcohol intake has not been shown to be associated with reduced all-cause mortality in younger women 33 ; thus, there seems little justification to encourage those who do not drink regularly to do so for health benefits. Accepted for publication September 16, 2002. This study was supported by grants HL-03804, AA- 11181, and CA50385 from the National Institutes of Health, Bethesda, Md. Dr Wannamethee was a visiting research scholar at Harvard School of Public Health, Boston, Mass, during the analyses presented in this article. We thank Maureen Ireland and Karen Corsano for their help in data preparation, and Yan Liu for computing assistance. Corresponding author and reprints: S. Goya Wannamethee, PhD, Department of Primary Care and Population Science, Royal Free and University College Medical School, Rowland Hill Street, London NW3 2PF, England (e-mail: goya@pcps.ucl.ac.uk). REFERENCES 1. Holbrook TL, Barrett-Connor E, Wingard DL. A prospective population-based study of alcohol and non insulin-dependent diabetes mellitus. Am J Epidemiol. 1990; 132:902-909. 2. Balkau B, Randrianjohany A, Papoz L, Eschwege E. A prospective study of alcohol use and non insulin-dependent diabetes mellitus. Am J Epidemiol. 1991; 134:1469-1470. 3. Wei M, Gibbons LW, Mitchell TL, Kampert JB, Blair SN. Alcohol intake and incidence of type 2 diabetes in men. Diabetes Care. 2000;23:18-22. 4. Monterrosa AE, Haffner SM, Stern MP, Hazuda HP. Sex difference in lifestyle factors predictive of diabetes in Mexican-Americans. Diabetes Care. 1995;18:448-456. 5. Kao WHL, Puddey IB, Boland LL, Watson RL, Brancati FL. Alcohol consumption 1335

and the risk of type 2 diabetes mellitus: Atherosclerosis Risk in Communities Study. Am J Epidemiol. 2001;154:748-757. 6. Feskens EJM, Kromhout D. Cardiovascular risk factors and the 25-year incidence of diabetes mellitus in middle-aged men. Am J Epidemiol. 1989;130:1101-1108. 7. Hodge AM, Dowse GK, Collins VR, Zimmet PZ. Abnormal glucose tolerance and alcohol consumption in three populations at high risk of non insulin-dependent diabetes mellitus. Am J Epidemiol. 1993;137:178-189. 8. Rimm EB, Chan J, Stampfer MJ, Colditz GA, Willett WC. Prospective study of cigarette smoking, alcohol use, and the risk of diabetes in men. BMJ. 1995;310: 555-559. 9. Perry IJ, Wannamethee SG, Walker MK, Thomson AG, Whincup PH, Shaper AG. Prospective study of risk factors for development of non insulin diabetes in middleaged British men. BMJ. 1995;310:560-564. 10. Tsumara K, Hayashi T, Suetmatsu C, Endo G, Fujii S, Okada K. Daily alcohol consumption and the risk of type 2 diabetes in Japanese men. Diabetes Care. 1999; 22:1432-1437. 11. Ajani UA, Hennekens CH, Spelsberg A, Manson JE. Alcohol consumption and risk of type 2 diabetes mellitus among US male physicians. Arch Intern Med. 2000;160:1025-1030. 12. Conigrave KM, Hu FB, Camargo CA, Stampfer MJ, Willett W, Rimm EB. A prospective study of drinking patterns in relation to risk of type 2 diabetes among men. Diabetes. 2001;50:2390-2395. 13. Hu FB, Manson JE, Stampfer MJ, et al. Diet, lifestyle, and the risk of type 2 diabetes mellitus in women. N Engl J Med. 2001;345:790-797. 14. Mayer EJ, Newman B, Quesenberry CP, Friedman GD, Selby JV. Alcohol consumption and insulin concentrations: role of insulin in associations of alcohol intake with high-density lipoprotein cholesterol and triglycerides. Circulation. 1993; 88:2190-2197. 15. Facchini FS, Chen YD, Reaven GM. Light-to-moderate alcohol intake is associated with enhanced insulin sensitivity. Diabetes Care. 1994;17:115-119. 16. Kiechl S, Willeit J, Poewe W, et al. Insulin sensitivity and regular alcohol consumption: large, prospective, cross sectional population study (Bruneck Study). BMJ. 1996;313:1040-1044. 17. Lazarus R, Sparrow D, Weiss ST. Alcohol intake and insulin levels: the Normative Aging Study. Am J Epidemiol. 1997;145:909-916. 18. 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;128:549-558. 19. USDA Nutrition Database for Standard Reference, Standard Release 12. Washington, DC: Agricultural Research Service, US Dept of Agriculture; 1998. 20. Giovannuci E, Colditz G, Stampfer MJ, et al. The assessment of alcohol consumption by a simple self-administered questionnaire. Am J Epidemiol. 1991; 133:810-817. 21. Willett WC, Sampson L, Stampfer MJ, et al. Reproducibility and validity of a semiquantitative food frequency questionnaire. Am J Epidemiol. 1985;122:51-65. 22. Salmeron J, Manson JE, Stampfer MJ, Colditz GA, Wing AL, Willett WC. Dietary fiber, glycemic load, and risk of non insulin-dependent diabetes mellitus in women. JAMA. 1997;277:472-477. 23. Salmeron J, Hu FB, Manson JE, et al. Dietary fat intake and risk of type 2 diabetes in women. Am J Clin Nutr. 2001;73:1019-1026. 24. Manson JE, Rimm EB, Stampfer MJ, et al. Physical activity and incidence of non insulin-dependent diabetes mellitus in women. Lancet. 1991;338:774-778. 25. Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Diabetes Care. 1997;20:1183-1197. 26. Fox R. The thrifty genotype and non insulin dependent diabetes [letter]. BMJ. 1993;306:933. 27. Haffner SM, Stern MP, Mitchell BD, et al. Incidence of type II diabetes in Mexican Americans predicted by fasting insulin and glucose levels, obesity and body fat distribution. Diabetes. 1990;39:283-288. 28. Saad MF, Knowler WC, Pettitt DJ, Nelson RG, Charles MA, Bennett PH. A twostep model for development of non insulin-dependent diabetes. Am J Med. 1991; 90:229-235. 29. Henriksson J. Influence of exercise on insulin sensitivity. J Cardiovasc Risk. 1995; 2:303-309. 30. Carlsson S, Hammar N, Efendic S, Persson PG, Ostenson CG, Grill V. Alcohol consumption, type 2 diabetes mellitus and impaired glucose tolerance in middleage Swedish men. Diabet Med. 2000;17:776-781. 31. Pomerleau J, McKeigue PM, Chaturvedi N. Relationships of fasting and postload glucose levels to sex and alcohol consumption: are American Diabetes Association criteria biased against detection of diabetes in women? Diabetes Care. 1999;22:430-433. 32. Friedman GD, Klatsky AL. Is alcohol good for your health [comment]? N Engl J Med. 1993;329:1882-1883. 33. Fuchs CS, Stampfer MJ, Colditz GA, et al. Alcohol consumption and mortality among women. N Engl J Med. 1995;332:1245-1250. 1336