A-H Harding 1, LA Sargeant 2, K-T Khaw 1, A Welch 1, S Oakes 1, RN Luben 1, S Bingham 1, NE Day 1 and NJ Wareham 1 *

Transcription

1 (2002) 56, ß 2002 Nature Publishing Group All rights reserved /02 $ ORIGINAL COMMUNICATION Cross-sectional association between total level and type of alcohol consumption and glycosylated haemoglobin level: the EPIC-Norfolk Study A-H Harding 1, LA Sargeant 2, K-T Khaw 1, A Welch 1, S Oakes 1, RN Luben 1, S Bingham 1, NE Day 1 and NJ Wareham 1 * 1 Department of Public Health and Primary Care, Institute of Public Health, University of Cambridge, Cambridge, UK; and 2 Epidemiology Research Unit, Tropical Medicine Research Institute, University of the West Indies, Kingston, Jamaica Objective: To investigate the association between total level and type of alcohol consumed and glycaemia. Design: Cross-sectional study. Setting: The EPIC-Norfolk Study, a population-based cohort study of diet and chronic disease. Subjects and methods: Non-diabetic men (n ¼ 2842) and women (n ¼ 3572), aged y. Alcohol intake was assessed by self-reported questionnaire, and glycaemia measured by glycosylated haemoglobin (HbA 1c ). Results: Ten percent of men and 18% of women reported drinking no alcohol. Among drinkers, median alcohol intake was 8 units=week for men and 3 units=week for women. In analyses stratified by sex and adjusted for age, total energy intake, education, fruit and vegetable intake, smoking, family history of diabetes, physical activity, body mass index and waist:hip ratio, alcohol intake was inversely associated with HbA 1c in men and women, although the association was stronger in women. A 1 unit=week increase in alcohol intake was associated with % (s.e. ¼ ; P-value ¼ 0.028) and 0.017% (s.e. ¼ ; P-value < 0.001) reduction in HbA 1c in men and women respectively. In similar multivariate analyses, wine intake was inversely associated with HbA 1c in men, and wine, spirits and beer intake were inversely associated with HbA 1c in women. When also adjusted for total alcohol intake, only the association between wine intake and HbA 1c in men remained significant. Conclusion: Alcohol intake was associated with lower HbA 1c level, an association not explained by confounding. The distinction between type of alcohol consumed was particularly important in men. Sponsorship: NJW is an MRC Clinician Scientist Fellow. (2002) 56, doi: /sj.ejcn Keywords: alcohol; glycosylated haemoglobin; diabetes mellitus; epidemiology Introduction Hyperglycaemia, as measured by glycosylated haemoglobin (HbA 1c ), is related to all-cause and cardiovascular disease *Correspondence: NJ Wareham, Department of Public Health and Primary Care, University of Cambridge, Institute of Public Health, Robinson Way, Cambridge CB2 2SR, UK. njw1004@medschl.cam.ac.uk Guarantor: NJ Wareham. Contributors: A-HH, LAS, K-TK and NJW contributed to the study concept, analysis and interpretation of the data and the drafting of the manuscript. K-TK, SB, NED and NJW are principal investigators of EPIC-Norfolk. AW, RNL and SO contributed to the study design and data collection and management. All authors contributed to the revision and drafting of the final version. Received 18 July 2001; revised 11 December 2001; accepted 14 December 2001 (CVD) mortality. In the EPIC-Norfolk study, HbA 1c was positively related to all-cause and coronary heart and cardiovascular disease mortality in men across the normal range of HbA 1c (Khaw et al, 2001). The Rancho Bernardo (Park et al, 1996) and Framingham Heart Studies (Singer et al, 1992) found a positive association between HbA 1c and CVD and coronary heart disease (CHD) only in women. Other continuous measures of glycaemia (fasting=2 h glucose) have also been associated with increased risk of mortality (Coutinho et al, 1999). Understanding the population determinants of continuous measures of glycaemia may contribute to understanding the aetiology and designing populationlevel interventions. Previous studies indicate that light to moderate alcohol intake may have beneficial effects on glycaemia, although

2 the results are inconsistent. In the EPIC-Potsdam cohort (Boeing et al, 2000) HbA 1c was inversely related to alcohol intake, and in the Hoorn Study (Mooy et al, 1995) 2 h plasma glucose was negatively associated with alcohol intake, if alcohol intake was less than 30 g=day. However, there was a positive association between plasma glucose and alcohol intake in the Hoorn Study for women drinking more than 30 g=day. Other studies have reported positive associations between fasting plasma glucose and alcohol intake (Godsland et al, 1998; Selby et al, 1987). Studies investigating the effect of alcohol consumption on the risk of developing type 2 diabetes are similarly inconsistent. Several report that moderate alcohol intake significantly decreases the risk of type 2 diabetes (Ajani et al, 2000; Conigrave et al, 2001; Rimm et al, 1995; Stampfer et al, 1988) or describe a U-shaped association between alcohol intake and risk of diabetes (Perry et al, 1995; Wei et al, 2000). However others report no association (Hodge et al, 1993; Levitt et al, 1999) or a positive association (Holbrook et al, 1990; Monterrosa et al, 1995). Studies including men and women in their sample population have found sex differences (Holbrook et al, 1990; Kao et al, 2001; Monterrosa et al, 1995), suggesting a positive association with risk of diabetes in men and a negative or no association in women. To our knowledge, no studies have investigated the effect of total and type of alcohol consumed on HbA 1c. We therefore undertook an analysis of the cross-sectional association between total and type of alcohol consumed and HbA 1c,ina Caucasian population of men and women aged y. Methods Study population The people in this study were participants in the Norfolk arm of the European Prospective Investigation into Cancer (EPIC). EPIC-Norfolk is part of an international multicentre cohort study designed to investigate the relationship between diet and cancer (Riboli & Kaaks, 1997). The Norfolk study broadened its scope to include chronic diseases other then cancer, and lifestyle exposures other than diet. Approval for the study was obtained from the Norfolk Local Research Ethics Committee. EPIC-Norfolk is a population-based cohort study, which recruited volunteers from March 1993 to the end of 1997 (Day et al, 1999). General practices in the city of Norwich, England, and surrounding small towns were invited to participate. All individuals in the age range in each general practice were invited to take part. Those who consented were invited to attend for a health check. Of the individuals contacted, 39% consented to take part and (33%) attended the health check, which was close to the target recruitment figure of In November 1995, mid-way through recruitment of the cohort, the measurement of HbA 1c commenced. The sub-cohort selected for this analysis comprises all individuals who had HbA 1c measured and whose data had been processed by July Data collection Volunteers completed a detailed health and lifestyle questionnaire. It included questions on personal and family history of diabetes, physical activity, smoking, education, diet and alcohol intake. Three questions addressed the respondent s personal history of diabetes. The questions asked whether they had ever been told by a doctor that they had diabetes; whether they had modified their diet in the past year due to diabetes; and whether they followed a diabetic diet. A positive response to any of these questions was taken as an indication of prevalent diabetes. Family history of diabetes was covered in a question asking whether any of the respondent s immediate family had diabetes. The age at which diabetes was diagnosed in mother, father and=or siblings was recorded. A four-point physical activity index was used combining level of occupational activity (Wareham et al, 1999) and recreational physical activity (hours per week in cycling and other recreational activity). Smoking was derived from questions asking whether they had ever smoked as much as one cigarette a day, whether they were current smokers, and, if they smoked, the number of cigarettes smoked each day. Participants were put into one of five smoking categories: never smoked, former smoker, and three categories of current smoker ( 7 15, 7 30, > 30 cigarettes per day). The weekly consumption of fresh fruit and vegetables and alcoholic drinks was reported. Respondents were asked At present, about how many alcoholic drinks do you have each week? Separate responses were given for four categories of alcoholic drink: beer, lager or cider, wine, spirits (whisky, gin, brandy, liqueurs, etc), and fortified wine. Respondents were requested to put 0 if they drank none of the beverage, or to tick a separate box labelled occasional if they had less than one drink a week. Otherwise, they were asked to enter the number of pints of beer, cider or lager, and the number of glasses of wine, spirits and fortified wine drunk each week. Responses were converted to units of alcohol per week, taking 0.5 pints beer, lager or cider, and one glass of wine, spirits or fortified wine to be equivalent to one unit of alcohol (8 g by weight). Occasional drinkers were assumed to drink 0.25 units of alcohol per week. The health and lifestyle questionnaire asked nondrinkers if they were lifelong teetotallers. Total energy intake was estimated from a semi-quantitative food frequency questionnaire (Bingham et al, 1997), which volunteers completed before attending a health check. Research nurses at the EPIC-Norfolk clinic carried out standardised health checks. Anthropometric measurements were taken with participants dressed in light clothing and without shoes. Height was measured to the nearest 0.1 cm using a stadiometer, and weight was measured to the nearest 100 g using Salter scales. Body mass index (BMI) was calculated as weight (kg)=height 2 (m). Of those attending the health check, 95% consented to have blood taken. A sample of EDTA-anticoagulated blood was taken for HbA 1c measurement. The blood was stored in a refrigerator at 4 7 C until it was transported at ambient temperature to be 883

3 884 assayed, within one week of sampling. The HbA 1c assays were undertaken using high-performance liquid chromatography on a Bio-Rad Diamat (Richmond, CA, USA; Standing & Taylor, 1992). The coefficient of variation was 3.6% at the lower end of the range (mean 4.9%) and 3.0% at the upper end of the range (mean 9.8%). Statistical analysis Individuals with self-reported diabetes were excluded from the analysis since it is probable that they would have changed their alcohol intake and diet after diagnosis, or would have altered how they reported it. Data for the two sexes were analysed separately. Further education was used as an index of socio-economic status. The associations between potential confounding variables, total energy intake, age, education, fruit consumption, vegetable consumption, cigarette smoking, physical activity, family history of diabetes, waist-to-hip ratio (WHR) and BMI, and the exposures and outcome were explored by correlation analysis or analysis of variance. Exploratory data analysis indicated that the observed relationships between alcohol consumption and HbA 1c were similar when alcohol consumption was included as a category and when included as a continuous variable. Different boundaries for the categories of alcohol intake were required for men and women, and for total alcohol intake and the individual types of alcohol, thereby reducing the comparability of the analyses. Consequently, alcohol intake was analysed as a continuous variable. Linear regression methods were used to investigate the relationship between HbA 1c and alcohol intake in a series of models. A quadratic term for alcohol intake was included in all models, to allow for a non-linear relationship between HbA 1c and alcohol intake. Total alcohol, and the four main types of alcohol (beer, lager or cider, wine, spirits, and fortified wine) were analysed. The simplest model included only alcohol intake, adjusted for total energy intake and age. The final model included alcohol and total energy intake, age, education, fruit consumption, vegetable consumption, cigarette smoking, physical activity, family history of diabetes, WHR and BMI. The coefficient for alcohol in this model can be interpreted as the effect on HbA 1c of a unit per week change in alcohol consumption at a constant level of total energy intake. A similar model in which former drinkers were excluded was also investigated. Similar models were developed for the individual types of alcohol. The final model for the types of alcohol consumed included all the potential confounding variables and total alcohol intake. Results A total of 7056 men and women, who had HbA 1c measured at the baseline health check and whose data had been processed before July 1998, were eligible for this study. Of these, 82 women and 113 men reported having diabetes at baseline and were excluded from the analysis. A further 319 women and 323 men had missing covariates, leaving 6414 men and women in the study. Nearly 10% of men and 18% of women were non-drinkers, and of these 79 and 60%, respectively, were former drinkers. Amongst drinkers, the median total alcohol intake was 7.75 units=week (1 unit ¼ 8 g alcohol) for men and three units=week for women. The 90th percentiles of intake were 26 units=week and 13.5 units=week respectively. Distinct Table 1 Characteristics of the study population: men in EPIC-Norfolk (n ¼ 2842) Categories of alcohol intake (units=week) > 0and > 2.0 and > 7.75 and Zero > 15.5 (n ¼ 279) (n ¼ 687) (n ¼ 599) (n ¼ 643) (n ¼ 634) P-value HbA 1c (%) 5.39 (0.614) 5.38 (0.665) 5.41 (0.785) 5.36 (0.785) 5.26 (0.594) Total energy intake (kj) 9103 (2742) 9300 (2683) 9353 (2684) 9272 (2571) 9333 (2397) Body mass index (kg=m 2 ) 26.8 (3.57) 26.4 (3.20) 26.5 (3.21) 26.5 (3.23) 27.0 (3.44) Waist:hip ratio (0.0595) (0.0574) (0.0556) (0.0558) (0.0549) Age (y) 62.4 (9.28) 60.1 (8.97) 59.7 (9.39) 59.2 (9.30) 56.9 (8.70) < Family history of diabetes positive a 34 (12%) 73 (11%) 74 (12%) 91 (14%) 57 (9%) Further education: yes a 89 (33%) 225 (33%) 233 (39%) 299 (47%) 279 (44%) < Physical activity level a 1 (lowest) 129 (46%) 279 (41%) 230 (38%) 237 (37%) 201 (32%) < (27%) 213 (31%) 214 (36%) 199 (31%) 218 (34%) 3 38 (14%) 126 (18%) 103 (17%) 130 (20%) 129 (20%) 4 37 (13%) 69 (10%) 52 (9%) 77 (12%) 86 (14%) Smoking status a Never smoked 94 (34%) 270 (39%) 237 (40%) 222 (34%) 144 (23%) < Ex-smoker 145 (52%) 322 (47%) 289 (48%) 357 (56%) 389 (61%) Current smoker 40 (14%) 95 (14%) 73 (12%) 64 (10%) 101 (16%) Categories are non-drinkers, and the four quartiles of alcohol drinkers; data are means, with standard deviations in parentheses; P-values refer to F-tests, or chisquared tests indicating differences between categories for the count data; the 5th and 95th percentile points of the male distribution of HbA 1c are 4.4 and 6.3%, respectively. a Data are counts with percentages in parentheses.

4 trends in HbA 1c level and age were observed. In men and women, those with higher alcohol intakes were younger and had lower HbA 1c (Tables 1 and 2). There was a tendency for those with further education, those who were more physically active and those who smoked to be in the higher categories of alcohol consumption. In women, BMI and WHR decreased significantly with increasing alcohol intake while in men there was a non-significant trend of increasing BMI and WHR with increasing alcohol intake. The type of alcohol consumed differed between men and women (Table 3). Men drank relatively more beer than women; the median percentage of total alcohol that came from beer was 38% for men and 0% for women. On the other hand, women drank relatively more wine than men; the median intake from wine was 50% for women and 25% for men. The proportion of alcohol derived from spirits was similar for men and women. Although neither men nor women consumed much fortified wine, women drank relatively more fortified wine. The correlation coefficients (Table 4) suggest that, unadjusted for any potential confounding variables, in women the inverse associations between total alcohol and wine intake and HbA 1c were approximately twice as large as in men. The associations with beer and fortified wine were similar in men and women. There was no evidence that spirits intake was associated with HbA 1c in men, although there was in women. Fruit and vegetable intakes tended to be negatively associated with beer and spirits, and positively associated with wine consumption. Alcohol intake was positively associated with BMI and WHR in men, and negatively associated with BMI and WHR in women. In the basic linear regression model (model 1, Table 5), alcohol intake adjusted for total energy intake and age was not associated with HbA 1c in men (b ¼ ; P ¼ 0.121). There was no evidence that the association between alcohol and HbA 1c was non-linear (quadratic term, b ¼ ; P ¼ 0.625). When the potential confounders, further education, fruit and vegetable intakes, cigarette smoking, physical activity and family history of diabetes, were included in the regression (model 2), the effect of alcohol was strengthened (b ¼ ; P ¼ 0.046). Including BMI and WHR into this model (model 3) and excluding former drinkers from the analysis (model 4), strengthened the association between alcohol and HbA 1c further. In women, alcohol intake adjusted for total energy intake and age (model 1, Table 5) was significantly associated with HbA 1c (b ¼ ; P < 0.001), and there was evidence that the association was non-linear (quadratic term, b ¼ ; Table 3 Sources of alcohol: percentage of total alcohol intake accounted for by each of the four types of alcohol, stratified by sex; EPIC-Norfolk (n ¼ 6414) Men (n ¼ 2842) Women (n ¼ 3572) Median Interquartile range Median Interquartile range Beer, lager and cider (%) 38 (10 73) 0 (0 20) Wine (%) 25 (1 48) 50 (25 80) Spirits (%) 9 (0 32) 6 (0 33) Fortified wine (%) 0 (0 6) 3 (0 33) Total alcohol (units=week) 7.8 ( ) 3.0 ( ) 885 Table 2 Characteristics of the study population: women in EPIC-Norfolk (n ¼ 3572) Categories of alcohol intake (units=week) > 0 > 0.75 and > 3.0 and Zero and > 7.25 (n ¼ 630) (n ¼ 863) (n ¼ 699) (n ¼ 687) (n ¼ 723) P-value HbA 1c (%) 5.42 (0.620) 5.39 (0.694) 5.32 (0.655) 5.27 (0.678) 5.15 (0.537) < Total energy intake (kj) 7978 (2303) 8218 (2348) 8286 (2399) 8013 (2247) 8289 (2193) Body mass index (kg=m 2 ) 26.6 (4.50) 26.6 (4.64) 26.4 (4.46) 25.6 (3.96) 25.7 (3.74) < Waist:hip ratio (0.0634) (0.0619) (0.0613) (0.0593) (0.0594) < Age (y) 61.5 (8.99) 58.8 (9.20) 58.7 (9.18) 58.1 (9.14) 56.9 (9.28) < Family history of diabetes positive a 73 (12%) 132 (15%) 103 (15%) 75 (11%) 97 (13%) Further education: yes a 147 (24%) 250 (30%) 215 (32%) 241 (35%) 335 (47%) < Physical activity level a 1 (lowest) 321 (51%) 372 (43%) 282 (42%) 233 (34%) 257 (36%) < (32%) 310 (36%) 252 (38%) 296 (43%) 292 (40%) 3 65 (10%) 117 (14%) 103 (15%) 117 (17%) 110 (15%) 4 46 (7%) 64 (7%) 32 (5%) 41 (6%) 64 (9%) Smoking status a Never smoked 390 (62%) 533 (62%) 396 (59%) 366 (53%) 320 (44%) < Ex-smoker 157 (25%) 239 (28%) 208 (31%) 247 (36%) 314 (44%) Current smoker 83 (13%) 91 (10%) 65 (10%) 74 (11%) 89 (12%) Categories are non-drinkers, and the four quartiles of drinkers; data are means, with standard deviations in parentheses; P-values refer to F-tests for trend, or chisquared tests indicating differences between categories for the count data; the 5th and 95th percentile points of the female distribution of HbA 1c are 4.4 and 6.2% respectively. a Data are counts with percentages in parentheses.

5 886 Table 4 Pearson correlation coefficients, stratified by sex, relating exposure and outcome variables; EPIC-Norfolk (n ¼ 6414) Alcohol intake Wine intake Beer intake Spirits intake Fortified wine intake Men (n ¼ 2842) HbA 1c *** *** * Age *** *** *** 0.041* 0.053** BMI 0.057** *** WHR 0.066*** * 0.102*** Total energy intake Fruit intake *** *** *** Vegetable intake ** Physical activity 0.053** *** Women (n ¼ 3572) HbA 1c *** *** *** ** Age *** *** *** *** BMI *** *** ** ** WHR *** Total energy intake *** Fruit intake ** Vegetable intake 0.070*** 0.103*** Physical activity 0.069*** 0.090*** ***P 0.001; **P 0.01; *P Table 5 Multiple regression models, stratified by sex, predicting glycosylated haemoglobin level with alcohol intake as the explanatory variable; EPIC-Norfolk (n ¼ 6414) Model 1 Model 2 Model 3 Model 4 Men (n ¼ 2842) Alcohol Linear (0.121) (0.046) (0.028) (0.014) Quadratic (0.625) (0.702) (0.677) (0.496) Women (n ¼ 3572) Alcohol Linear ( < 0.001) ( < 0.001) ( < 0.001) ( < 0.001) Quadratic (0.003) (0.007) (0.031) (0.020) Alcohol consumption measured in units per week. Data are regression coefficients with P-values in parentheses. Model 1: adjusted for age and total energy intake. Model 2: adjusted for total energy intake, age, further education, leafy vegetables, fresh fruit, smoking, family history of diabetes and physical activity level. Model 3: adjusted for total energy intake, age, further education, leafy vegetables, fresh fruit, smoking, family history of diabetes, physical activity level, waist:hip ratio and body mass index. Model 4: as Model 3, but excluding former drinkers. P ¼ 0.003). Adjusting for the potential confounders, further education, fruit and vegetable intakes, cigarette smoking, physical activity and family history of diabetes (model 2) had little effect on the association of alcohol with HbA 1c (b ¼ ; P ¼ < 0.001). Including BMI and WHR into the regression (model 3) attenuated this association, while excluding former drinkers from the analysis (model 4) strengthened the association. In the analysis investigating the effect on HbA 1c of the type of alcohol consumed (Table 6), in men there was an association between HbA 1c level and wine intake (b ¼ ; P ¼ 0.002). There was no evidence of nonlinearity in the relationship (quadratic term; b ¼ ; P ¼ 0.130). The association remained when former drinkers were excluded (not shown). Although attenuated when total alcohol was included in the model, the association with wine intake remained significant (b ¼ ; P ¼ 0.021). There was no evidence of any association between beer, spirits and fortified wine intake and HbA 1c level. Glycosylated haemoglobin (Table 6) was associated with intake of wine (b ¼ ; P < 0.001), beer (b ¼ ; P ¼ 0.004) and spirits (b ¼ ; P < 0.001) in women. There was some evidence that the relationships were nonlinear. The effects were not materially changed when former drinkers were excluded, but all the associations were strongly attenuated when total alcohol intake was included in the regression and were no longer statistically significant (P > 0.300).

6 Table 6 Multiple regression models, stratified by sex, predicting glycosylated haemoglobin level with type of alcohol intake as the explanatory variables; EPIC-Norfolk (n ¼ 6414) 887 Wine Beer Spirits Fortified wines Men (n ¼ 2842) Linear (0.002) (0.586) (0.226) (0.278) Quadratic (0.130) (0.589) (0.643) (0.500) Women (n ¼ 3572) Linear ( < 0.001) (0.004) ( < 0.001) (0.350) Quadratic (0.102) (0.0613) (0.0396) (0.570) Alcohol consumption measured in units per week. Data are regression coefficients with P-values in parentheses. All analyses adjusted for total energy intake, age, further education, leafy vegetables, fresh fruit, smoking, family history of diabetes, physical activity level, waist: hip ratio and body mass index. Discussion In this study, we have demonstrated an inverse association between total alcohol intake and HbA 1c, in a population of men and women without self-reported diabetes. The association was stronger in women than in men. In women, the effect of alcohol was not specific to any type of drink, with the effects of wine, beer and spirits similar to that of total alcohol. The association between wine intake and HbA 1c was similar in men and women, although in women it was strongly attenuated when adjusted for total alcohol intake. There was no evidence of an association with fortified wine, which may be due to the small proportion of the population drinking fortified wine ( < 40% of men and < 50% of women) and the limited range of intake (99th percentile among drinkers was 10 units=week in men and women). It is unlikely that chance was an explanation for the consistent associations between HbA 1c and alcohol intake observed in this large study. The EPIC-Norfolk study was designed as a population-based cohort study, which would allow comparisons to be made within the cohort, both crosssectionally and over time. In terms of anthropometric measurements, blood pressure and serum lipids, the EPIC-Norfolk cohort is similar to the population studied in the Health Survey of England (Day et al, 1999). However, the EPIC- Norfolk cohort has a smaller proportion of current smokers than was observed in the Health Survey of England (Day et al, 1999). This may be a reflection of selection bias. Since cigarette smoking is associated with an increased risk of various diseases, the EPIC-Norfolk cohort may be at reduced risk of smoking-related diseases compared to the general population. Although this selection may represent recruitment of a relatively healthy cohort, it is unlikely to affect the cross-sectional relationship between HbA 1c and alcohol consumption. The similarity of the EPIC-Norfolk cohort to nationally representative samples with respect to other variables suggests that any bias resulting from selection is likely to be small. There was no evidence of any difference in age and BMI between the study group defined for this analysis and the entire EPIC-Norfolk cohort (Sargeant et al, 2000). A definition of diabetes was chosen which would exclude those who may have changed their diet following a diagnosis of diabetes. Of those who reported on the health and lifestyle questionnaire that they had modified their diet because of diabetes, 72% were also taking diabetic medication or had an HbA 1c measurement of > 7% (Sargeant et al, 2000). Confounding is the major issue affecting the inferences that can be drawn from this study. Tables 1 and 2 suggest a complex pattern of confounding which may differ between men and women. The observed associations were independent of obesity, as measured by BMI and WHR. Whether obesity is a true confounder or not depends on whether it is part of the causal pathway between alcohol intake and HbA 1c. Alcohol intake is inversely related to body weight in women, but the relationship is less clear in men (Istvan et al, 1995). In the Lung Health Study, total alcohol intake was related to a lower BMI whereas a greater modal intake was related to a higher BMI, in both men and women. The Rancho Bernado Study reported a positive relationship between WHR and alcohol consumption in men and women (Laws et al, 1990). In our study, alcohol intake was inversely associated with BMI and WHR in women, and positively associated with BMI and WHR in men. If obesity is on the causal pathway, then adjustment for obesity will lead to over-adjustment. We considered age, further education, fruit and vegetable intakes, smoking, and physical activity to be potential confounders since they were related to HbA 1c in our population, and previous studies have suggested that lifestyle factors may be associated with choice and pattern of alcohol intake (Burke et al, 1995; Goldberg et al, 2001; Tjonneland et al, 1999). Relative imprecision in the assessment of diet, smoking and physical activity leaves the possibility of residual confounding. Measurement error may also affect the association between exposure and outcome, and will generally attenuate any observed associations. Usual alcohol intake was assessed by questionnaire, a method which is likely to recall modal quantities and thus underestimate the long-term average intake (Poikolainen, 1995). A comparison of alcohol intake assessed by food frequency questionnaire (FFQ) and by diet records indicated that the FFQ gives lower absolute amounts of alcohol (Bingham et al, 2001). Differences in the size of units reported would introduce some random error into our study, which would tend to attenuate relationships between variables. Alcohol intake is usually underestimated by study

7 888 participants (Poikolainen, 1995). If this underestimate is not linked to actual intake, then the shape of the observed association between HbA 1c and alcohol intake would not be affected but the distribution would be shifted horizontally away from the origin. Heavy drinkers are reported to underestimate their intake more than light drinkers (Poikolainen, 1995). If this is the case, then the shape of the association will be changed, and the effects associated with heavy drinking will be exaggerated. The shape of the association between alcohol and glycaemia or risk of diabetes observed in previous studies has been both linear and non-linear, and in our study the association was linear in men and there was evidence of non-linearity in the association in women. The differences may result from incomplete adjustment for confounding, and from differences in reported alcohol intake. The majority of studies indicate that light to moderate alcohol intake is protective against hyperglycaemia and diabetes. In the context of studies investigating the effect of alcohol consumption on the risk of mortality, considerable debate has revolved around the possibility that the protective effect of alcohol may be attributable to the presence of former drinkers with pre-existing disease amongst non-drinkers (Marmot, 2001; Poikolainen, 1995; Shaper et al, 1988). In our study, we addressed this possibility by repeating the analysis with former drinkers excluded. The exclusion resulted in a modest strengthening of the observed associations (Table 5, models 3 and 4). Consequently, our data do not provide any support for this explanation of the protective effect of moderate alcohol intake. Although the biological mechanisms underlying the apparently protective effect of alcohol need clarification, some plausible mechanisms have been proposed. Studies have reported that light to moderate alcohol consumption is associated with lower insulin resistance (Facchini et al, 1994; Flanagan et al, 2000; Lazarus et al, 1997; Mayer et al, 1993). Acute insulin secretion and glucose disposal rates are enhanced after moderate alcohol intake (McMonagle & Felig, 1975; Metz et al, 1969). However, heavy alcohol drinkers have decreased glucose tolerance and are at higher risk of diabetes (Lindegard & Langman, 1985). Alcohol withdrawal improved glucose tolerance in alcoholics (Sereny & Endrenyi, 1978). Our study clearly demonstrated a stronger relationship between HbA 1c and total alcohol consumption in women than in men. In women, the beneficial relationship appeared to be independent of the type of alcohol consumed. On the other hand, only the intake of wine had any association with HbA 1c in men, and it appears that, for men, wine intake accounted for the beneficial effect of total alcohol intake on HbA 1c. The effect of drinking wine was very similar in men and women, and for women the effect of beer and spirits were similar to that of wine. Why, in men, the effect of beer and spirits was different to the effect of wine and was different to the effect of beer and spirits in women, is not clear. It may be due to the drinking patterns associated with each type of alcohol, and there may be sex differences in drinking patterns. Carlsson et al (2000) undertook a cross-sectional study investigating the association between alcohol consumption and impaired glucose tolerance and type 2 diabetes in Swedish men. High alcohol consumption was positively associated with diabetes, and moderate alcohol intake was negatively associated with impaired glucose tolerance. The positive association with diabetes was ascribed mainly to drinking beer and spirits, and not to wine. Stampfer et al (1988) found that, in women, the effect of specific alcoholic drinks on the risk of diabetes was similar to that of total alcohol. Generally, there has been little debate in the diabetes literature about the possible effect of the type of alcohol. In the literature relating to CVD and total mortality, considerable attention has been given to whether it is the effect of alcohol or the effect of specific alcoholic drinks that is important. Although individual studies have reported a differential effect between types of alcoholic drink (Gronbaek et al, 1995), reviews of the effect of the types of alcohol on total mortality (Poikolainen, 1995) and CHD (Rimm et al, 1996) concluded that all alcoholic drinks are linked with lower risk. The relatively greater benefit which has been found in wine drinkers has been attributed to other lifestyle factors (Wannamethee & Shaper, 1999). The choice of alcoholic drink may be correlated with lifestyle and personality factors. Australian (Burke et al, 1995) men who preferred beer had a higher rate of smoking, chose a less healthy diet and drank larger volumes, and had higher scores for extraversion, resentment and verbal hostility, than those who drank wine. The Atherosclerosis Risk in Communities Study (Kao et al, 2001) found that wine consumption was associated with healthy behaviours whereas drinking beer or spirits was not. In a Danish study, people who preferred wine were younger, had a smaller proportion of smokers, and were more educated and less obese than all other groups of drinkers (Tjonneland et al, 1999). Wine drinking was also associated with healthy dietary habits. However, the beneficial effects of wine drinking may simply be related to the context in which the alcohol is consumed, since wine tends to be drunk with meals more than beer and spirits (Truelsen et al, 1998). The US Health Professionals Follow-up Study (Conigrave et al, 2001) reported that an increased frequency of alcohol consumption reduced the risk of developing type 2 diabetes, irrespective of quantity, type of alcohol, or whether the alcohol was consumed with a meal. From the Insulin Resistance and Atherosclerosis Study, Bell et al (2000) reported that, although insulin sensitivity did not vary according to type of alcohol consumed, typical patterns of alcohol consumption predicted insulin sensitivity. Sensitivity was greater for those who drank the same amount each day, followed by those who drank more at weekends. The lowest sensitivity was found for those who only drank on special occasions. The possibility that there may be healthier drinking patterns has also emerged from CVD research

8 (Marmot, 2001). A study investigating the effect of binge drinking of beer (Kauhanen et al, 1997) on CVD mortality suggested that such episodic drinking has an effect on health that is independent of total alcohol intake. However, we could not address this issue since information on the pattern of alcohol intake was not available. Little research into drinking patterns has been undertaken (Poikolainen, 1995), and this is an area that needs further study. The current UK recommendations for safe drinking suggest a maximum weekly consumption of 21 units for men and 14 units for women. The alcohol consumption of the majority of our study population lay well within these limits, and our findings provide further evidence that moderate alcohol consumption may be one component of a healthy lifestyle. Further research into patterns of drinking may suggest that recommendations regarding the frequency of drinking alcohol should also be considered. The association between alcohol consumption and HbA 1c is complex, and may depend on the quantity and the context in which it is drunk. At the same time, residual confounding and inaccurate reporting of intake may conceal the true nature of the relationship. If the weight of evidence from prospective studies confirms the inverse association between alcohol consumption and HbA 1c observed in this study, then, by shifting the population distribution of HbA 1c and reducing the risk of CVD, moderate alcohol intake may have benefits that extend beyond the association we observed with HbA 1c. References Ajani UA, Hennekens CH, Spelsberg A & Manson JE (2000): Alcohol consumption and risk of type 2 diabetes mellitus among US male physicians. Arch. Intern. Med. 160, Bell RA, Mayer-Davis EJ, Martin MA, D Agostino RB & Haffner SM (2000): Associations between alcohol consumption and insulin sensitivity and cardiovascular disease risk factors the insulin resistance and atherosclerosis study. Diabetes Care 23, Bingham SA, Gill C, Welch A, Cassidy A, Runswick SA, Oakes S, Lubin R, Thurnham DI, Key TJ, Roe L, Khaw KT & Day NE (1997): Validation of dietary assessment methods in the UK arm of EPIC using weighed records, and 24-hour urinary nitrogen and potassium and serum vitamin C and carotenoids as biomarkers. Int J Epidemiol 26(Suppl 1), S Bingham SA, Welch AA, McTaggart AA, Mulligan A, Runswick SA, Luben R, Oakes S, Khaw K-T, Wareham N & Day NE (2001): Nutritional methods in the European Prospective Investigation into Cancer in Norfolk. Publ. Health Nutr. 4, Boeing H, Weisgerber UM, Jeckel A, Rose HJ & Kroke A (2000): Association between glycated hemoglobin and diet and other lifestyle factors in a nondiabetic population: cross-sectional evaluation of data from the Potsdam cohort of the European Prospective Investigation into Cancer and Nutrition Study. Am J Clin Nutr 71, Burke V, Puddey IB & Beilin LJ (1995): Mortality associated with wines, beers, and spirits. Australian data suggest that choice of beverage relates to lifestyle and personality. Br. Med. J. 311, Carlsson S, Hammar N, Efendic S, Persson PG, Ostenson CG & Grill V (2000): Alcohol consumption, Type 2 diabetes mellitus and impaired glucose tolerance in middle-aged Swedish men. Diabetic Med. 17, Conigrave KM, Hu BF, Camargo CA, Stampfer MJ, Willett WC & Rimm EB (2001): A prospective study of drinking patterns in relation to risk of type 2 diabetes among men. Diabetes 50, Coutinho M, Gerstein HC, Wang Y & Yusuf S (1999): The relationship between glucose and incident cardiovascular events. A metaregression analysis of published data from 20 studies of 95,783 individuals followed for 12.4 y. Diabetes Care 22, Day N, Oakes S, Luben R, Khaw KT, Bingham S, Welch A & Wareham N (1999): EPIC-Norfolk: study design and characteristics of the cohort. European Prospective Investigation of Cancer. Br. J. Cancer 80(Suppl 1), Facchini F, Chen YD & Reaven GM (1994): Light-to-moderate alcohol intake is associated with enhanced insulin sensitivity. Diabetes Care 17, Flanagan DEH, Moore VM, Godsland IF, Cockington RA, Robinson JS & Phillips DI (2000): Alcohol consumption and insulin resistance in young adults. Eur. J. Clin. Invest. 30, Godsland IF, Leyva F, Walton C, Worthington M & Stevenson JC (1998): Associations of smoking, alcohol and physical activity with risk factors for coronary heart disease and diabetes in the first follow-up cohort of the Heart Disease and Diabetes Risk Indicators in a Screened Cohort study (HDDRISC-1). J. Intern. Med. 244, Goldberg IJ, Mosca L, Piano MR & Fisher EA (2001): Wine and your heart. Circulation 103, Gronbaek M, Deis A, Sorensen T, Becker U, Schnohr P & Jensen G (1995): Mortality associated with moderate intakes of wine, beer, or spirits. Br. Med. J. 310, Hodge AM, Dowse GK, Collins VR & Zimmet PZ (1993): Abnormal glucose tolerance and alcohol consumption in three populations at high risk of non-insulin-dependent diabetes mellitus. Am. J. Epidemiol. 137, Holbrook TL, Barrett Connor E & Wingard DL (1990): A prospective population-based study of alcohol use and non-insulin-dependent diabetes mellitus. Am. J. Epidemiol. 132, Istvan J, Murray R & Voelker H (1995): The relationship between patterns of alcohol consumption and body weight. Lung Health Study Research Group. Int. J. Epidemiol. 24, Kao WH, Puddey IB, Boland LL, Watson RL & Brancati FL (2001): Alcohol consumption and the risk of type 2 diabetes mellitus: atherosclerosis risk in Communities study. Am. J. Epidemiol. 154, Kauhanen J, Kaplan GA, Goldberg DE & Salonen JT (1997): Beer binging and mortality: results from the Kuopio ischaemic heart disease risk factor study, a prospective population based study. Br. Med. J. 315, Khaw K-T, Wareham N, Luben R, Bingham S, Oakes S, Welch A & Day N (2001): Glycated haemoglobin, diabetes, and mortality in men in Norfolk cohort of European Prospective Investigation of Cancer and Nutrition (EPIC-Norfolk). Br. Med. J. 322, 15. Laws A, Terry RB & Barrett Connor E (1990): Behavioral covariates of waist-to-hip ratio in Rancho Bernardo. Am. J. Public Health 80, Lazarus R, Sparrow D & Weiss ST (1997): Alcohol intake and insulin levels the Normative Aging Study. Am. J. Epidemiol. 145, Levitt NS, Steyn K, Lambert EV, Reagon G, Lombard CJ, Fourie JM, Rossouw K & Hoffman M (1999): Modifiable risk factors for Type 2 diabetes mellitus in a peri-urban community in South Africa. Diabetic Med. 16, Lindegard B & Langman MJS (1985): Marital state, alcohol-consumption, and liability to myocardial-infarction, stroke, diabetes-mellitus, or hypertension in men from Gothenburg. Br. Med. J. 291, Marmot MG (2001): Reflections on alcohol and coronary heart disease. Int. Epidemiol. Assoc. 30, Mayer EJ, Newman B, Quesenberry CP Jr, Friedman GD & Selby JV (1993): Alcohol consumption and insulin concentrations. Role of insulin in associations of alcohol intake with high-density lipoprotein cholesterol and triglycerides. Circulation 88,

9 890 McMonagle J & Felig P (1975): Effects of ethanol ingestion on glucose tolerance and insulin secretion in normal and diabetic subjects. Metabolism 24, Metz R, Berger S & Mako M (1969): Potentiation of the plasma insulin response to glucose by prior administration of alcohol. An apparent islet-priming effect. Diabetes 18, Monterrosa AE, Haffner SM, Stern MP & Hazuda HP (1995): Sex difference in lifestyle factors predictive of diabetes in Mexican- Americans. Diabetes Care 18, Mooy JM, Grootenhuis PA, de Vries H, Valkenburg HA, Bouter LM, Kostense PJ & Heine RJ (1995): Prevalence and determinants of glucose intolerance in a Dutch caucasian population. The Hoorn Study. Diabetes Care 18, Park S, Barrett Connor E, Wingard DL, Shan J & Edelstein S (1996): GHb is a better predictor of cardiovascular disease than fasting or postchallenge plasma glucose in women without diabetes. The Rancho Bernardo Study. Diabetes Care 19, Perry IJ, Wannamethee SG, Walker MK, Thomson AG, Whincup PH & Shaper AG (1995): Prospective study of risk factors for development of non-insulin dependent diabetes in middle aged British men. Br. Med. J. 310, Poikolainen K (1995): Alcohol and mortality: a review. J. Clin. Epidemiol. 48, Riboli E & Kaaks R (1997): The EPIC Project: rationale and study design. European Prospective Investigation into Cancer and Nutrition. Int. J. Epidemiol. 26(Suppl 1), S6 14. Rimm EB, Chan J, Stampfer MJ, Colditz GA & Willett WC (1995): Prospective study of cigarette smoking, alcohol use, and the risk of diabetes in men. Br. Med. J. 310, Rimm EB, Klatsky A, Grobbee D & Stampfer MJ (1996): Review of moderate alcohol consumption and reduced risk of coronary heart disease: is the effect due to beer, wine, or spirits. Br. Med. J. 312, Sargeant LA, Wareham NJ, Bingham S, Day NE, Luben RN, Oakes S, Welch A & Khaw KT (2000): Vitamin C and hyperglycemia in the European Prospective Investigation into Cancer Norfolk (EPIC- Norfolk) study a population-based study. Diabetes Care 23, Selby JV, Newman B, King MC & Friedman GD (1987): Environmental and behavioral determinants of fasting plasma glucose in women. A matched co-twin analysis. Am. J. Epidemiol. 125, Sereny G & Endrenyi L (1978): Mechanism and significance of carbohydrate intolerance in chronic alcoholism. Metabolism 27, Shaper AG, Wannamethee G & Walker M (1988): Alcohol and mortality in British men: explaining the U-shaped curve. Lancet 2, Singer DE, Nathan DM, Anderson KM, Wilson PW & Evans JC. (1992): Association of HbA1c with prevalent cardiovascular disease in the original cohort of the Framingham Heart Study. Diabetes 41, Stampfer MJ, Colditz GA, Willett WC, Manson JE, Arky RA, Hennekens CH & Speizer FE (1988): A prospective study of moderate alcohol drinking and risk of diabetes in women. Am. J. Epidemiol 128, Standing SJ & Taylor RP (1992): Glycated haemoglobin: an assessment of high capacity liquid chromatographic and immunoassay methods. Ann. Clin. Biochem. 29, Tjonneland A, Gronbaek M, Stripp C & Overvad K (1999): Wine intake and diet in a random sample of Danish men and women. Am. J. Clin. Nutr. 69, Truelsen T, Gronbaek M, Schnohr P & Boysen G (1998): Intake of beer, wine, and spirits and risk of stroke: the Copenhagen city heart study. Stroke 29, Wannamethee SG & Shaper AG (1999): Type of alcoholic drink and risk of major coronary heart disease events and all-cause mortality. Am. J. Publ. Health 89, Wareham NJ, Rennie KL, Hennings SJ, Mitchell J & Day NE (1999): Comparison of EPIC physical activity questionnaire with repeated assessment of energy expenditure and fitness. Med. Sports Sci. Exerc. 31(Suppl), S232. Wei M, Gibbons LW, Mitchell TL, Kampert JB & Blair SN (2000): Alcohol intake and incidence of type 2 diabetes in men. Diabetes Care 23,