OBJECTIVE—This meta-analysis was undertaken to obtain insight regarding the shape and strength of the relationship between alcohol consumption and the risk of type 2 diabetes, the effects of adjustment for confounders, and the effect of modification by type 2 diabetes definition, sex, and BMI.
RESEARCH DESIGN AND METHODS—The 15 original prospective cohort studies that were included comprise 11,959 incident cases of type 2 diabetes in 369,862 individuals who, on average, were followed for 12 years.
RESULTS—After pooling the data, a U-shaped relationship was found. Compared with nonconsumers, the relative risk (RR) for type 2 diabetes in those who consumed ≤6 g/day alcohol was 0.87 (95% CI 0.79–0.95). For the moderate consumption ranges of 6–12, 12–24, and 24–48 g/day, RRs of 0.70 (0.61–0.79), 0.69 (0.58–0.81), and 0.72 (0.62–0.84) were found, respectively. The risk of type 2 diabetes in heavy drinkers (≥48 g/day) was equal to that in nonconsumers (1.04 [0.84–1.29]). In general, nonsignificant trends for larger RR reduction associated with moderate alcohol consumption were observed for women compared with men, for crude compared with multivariate-adjusted analyses, and for studies that used self-reports instead of testing for type 2 diabetes definition. No differences in RR reductions were found between individuals with low or high BMI.
CONCLUSIONS—The present evidence from observational studies suggests an ∼30% reduced risk of type 2 diabetes in moderate alcohol consumers, whereas no risk reduction is observed in consumers of ≥48 g/day.
The prevalence of type 2 diabetes is rising to epidemic proportions. Between 2000 and 2030, a 37% increase in the worldwide prevalence of diabetes is expected (1). Next to the aging of the population, the lack of physical activity and high-energy intake leading to overweight and obesity have been shown to be largely responsible for this epidemic, showing the importance of lifestyle factors for type 2 diabetes risk (2).
Alcohol consumption is a lifestyle factor that also has been suggested to be relevant with respect to the risk of type 2 diabetes. Several studies on the relationship between alcohol consumption and incident type 2 diabetes have been published during the last few years. Two narrative reviews on this topic suggested that moderate alcohol consumption is associated with a decreased incidence of type 2 diabetes but were inconclusive about the magnitude of this decreased incidence and the incidence associated with high levels of alcohol consumption (3,4). Furthermore, these reviews suggested that sex, BMI, and the mode of type 2 diabetes definition (self-report versus objective testing) may be important confounders or modifiers of the relationship. The objective of the present study, therefore, is to perform a meta-analysis on the published data and to calculate pooled estimates with respect to these questions on the relationship between alcohol consumption and the risk of type 2 diabetes.
RESEARCH DESIGN AND METHODS
Articles included were found through a PubMed search of literature published between 1966 and July 2004. The search resulted in articles that included the text words “(alcohol OR ethanol) AND diabetes AND (inciden* OR (new AND cases))” but that did not include the text word “placebo,” or the subject heading “drug therapy.” In addition, randomized controlled trials, editorials, letters to the editor, meta-analyses, and review articles were not included. Titles and abstracts of the resulting publications were screened for articles that are possibly of interest for our meta-analysis. The references in these articles and in relevant reviews were checked for additional studies of interest. Included were original peer-reviewed publications on observational cohort or nested case-control studies on the relationship between alcohol consumption and incident type 2 diabetes. Given the presumed nonlinear relationship, a point estimate and an estimate of variability of type 2 diabetes risk needed to be presented for at least two alcohol consumption categories compared with a third reference category (or data should be given from which this could be calculated). Alcohol consumption categories needed to be quantifiable in grams per day.
For studies that reported results from various analyses including more or fewer covariates, the estimates based on the model that included the most potential con-founders (e.g., age, sex, smoking, physical activity) and that excluded putative intermediate variables (fasting insulin or HDL cholesterol levels) were abstracted.
For studies that only reported ranges of alcohol consumption for the categories used, the mean of the lower and upper limits was used as the average consumption for that range, and 1.5 times the lower limit was assigned as the average consumption in the highest category (where no upper limit was given). The factor 1.5 was based on the information in included studies that reported the average consumption in addition to the lower limit of the highest consumption category. The averages of the consumption ranges that were used in a study were assigned to one of six alcohol categories that were used in this meta-analysis: 0 (reference), <6, 6–12, 12–24, 24–48, and ≥48 g/day. These alcohol categories were chosen because it resulted in about an equal number of independent observations within each stratum. A standard drink contains ∼12 g alcohol in the U.S./Canada, 10 g in Europe/Australia, and 21 g in Japan. Stratified analyses were performed because we were unable to find software to perform a meta-regression analysis that takes into account the fact that multiple observations from one study are not independent and that can incorporate the weight (inverse of variance) of the observations properly. Within the strata chosen, multiple observations from a single study were rare.
Odds ratios, relative risks (RRs), and hazard ratios that are used in the original publications were interpreted to reflect RRs. Natural logarithms of the RR estimates for each level of alcohol consumption from the individual studies were used to normalize the data. Natural logarithms of the reported CI limits were used to calculate the standard errors of the log RR estimates of the included studies. If the original study did not use the nonconsumers as a reference, the RRs were recalculated relative to the nondrinkers, assuming equal variance of the RR estimates when using the reported category or the nondrinker category as a reference.
Pooled RR estimates were calculated within each of the alcohol categories. Because the risk estimates differed more across studies than expected from the sampling error within studies, random-effects models were used to combine the reported data from these studies (5). If a study reported more than one RR within one of our six relatively small alcohol categories, one pooled RR was calculated for the study in question per alcohol category before pooling the RRs of the studies in that category.
To examine potential publication bias, separate funnel plots were drawn for the alcohol categories because the method of assessing funnel plot asymmetry assumes one overall RR. In this meta-analysis, however, it was expected that different RRs would be found for different alcohol consumption categories. For each funnel plot, the degree of asymmetry was visually inspected and quantitatively tested with a method modified from Egger et al. (6) that uses the slope instead of the intercept of a fitted regression line.
In addition to the multivariate-adjusted RRs, crude RRs were calculated for studies that reported the number of incident type 2 diabetic cases and noncases (or person-years) per alcohol category. To investigate the effect of confounding, the RRs from these crude data were pooled and compared with the pooled multivariate-adjusted RRs from the same studies. The RRs reported in studies that used self-reports to determine a subject’s type 2 diabetes status were compared with the RRs reported in studies that used biochemical testing of type 2 diabetes status. Sex, BMI, and type of beverage have been mentioned as putative modifiers of the relationship between alcohol consumption and the risk of type 2 diabetes. We therefore calculated and compared pooled RRs for men and women. From studies that reported RR estimates for more than one BMI stratum, the RRs reported for the highest strata were pooled and compared with the pooled RRs that are reported for the lowest BMI strata. Modification by type of beverage was not analyzed because only two studies reported alcohol consumption category–specific RR estimates for the different types of beverages (7,8). Student’s t tests were used to study differences in pooled RRs between the crude and adjusted data, between men and women, and between high and low BMI.
RESULTS
Twenty-eight publications on the relationship between alcohol consumption and the incidence of type 2 diabetes were found after reading titles and abstracts of the 482 hits from the PubMed search and after reference checking. Independent assessment of these publications by L.L.J.K. and J.M.D. with respect to the eligibility for the meta-analysis did not result in disagreement on any publication. For three studies on which multiple reports were published, the most recent publication was included while the others were excluded (9–12). Nine studies were not included because their alcohol consumption data were not quantifiable in grams per day (13–15) or because of incompleteness with respect to alcohol category–specific risk estimates (16–21). The search did not yield any nested case-control study that met the inclusion criteria, whereas 15 relevant cohort studies were identified (7,8,22–34).
Table 1 shows characteristics of the 15 included studies. In total, our meta-analysis comprises 11,959 incident cases of type 2 diabetes in 369,862 individuals who, on average, were followed for 12.0 years. All studies were published in peer-reviewed journals. Four studies included both male and female subjects, eight studies included only men, and three studies only women. In seven studies, all participants were tested for the presence of type 2 diabetes; seven others relied on self-reported diagnoses, while one study used record linkage with national registers. The included studies differed considerably with respect to the number of confounders adjusted for (range 0–15 confounders). In 8 of the 15 studies (7,8,22,24,29–32), various models including more or fewer confounders were used to calculate the RR of type 2 diabetes for the alcohol categories used. The confounders incorporated in the models that were chosen to be included in this meta-analysis are reported in Table 1.
Figure 1 shows the scatter plot of the RRs of type 2 diabetes by alcohol consumption (abstinence is the reference) as reported in the 15 included studies. Figure 1 indicates considerable heterogeneity between the observations. The areas of the circles differ substantially, indicating that certain estimates add more statistical power than others. The random-effect pooled RRs that are also shown in Fig. 1 indicate a U-shaped relationship between alcohol consumption and the risk of type 2 diabetes. The lowest risks were observed in alcohol drinkers of 6–48 g/day. The risk of type 2 diabetes in drinkers of ≥48 g/day was equal to that of nonconsumers. No publication bias was suspected after visual inspection or statistical testing of the funnel plots drawn for the alcohol consumption categories (graphs not shown).
In 10 of the 15 included studies (7,8,22–24,26,29–32), the numbers of incident type 2 diabetic cases and noncases (or person-years) were reported per alcohol category in addition to multivariate-adjusted RR estimates. The pooled RRs from these crude data and the pooled multivariate-adjusted RRs are reported in Table 2. The effect of confounding was significant for the lowest alcohol category. Table 2 further indicates that lower RR estimates were found in the studies that used self-reports to assess a subject’s type 2 diabetes status (8,22,24,26,28,31,34) than in studies that used glucose testing to define type 2 diabetes (7,23,25,27,29,30,33).
We further investigated possible modification of the relationship between alcohol consumption and the risk of type 2 diabetes by sex and BMI (Table 2). All but one study (25) gave sex-specific estimates. In women, observations >24 g/day were pooled because only one (imprecise) observation >48 g/day was reported. A significant sex difference was found only for the alcohol consumption range between 6 and 12 g/day. The RRs for women that were found in the other drinking categories were not significantly smaller than those for men. These stratified analyses show that the lower risk of type 2 diabetes of moderate alcohol consumption is significant in both men and women.
CONCLUSIONS
To our knowledge, this is the first meta-analysis on the relationship between alcohol consumption and the risk of type 2 diabetes. It shows a U-shaped relationship with a highly significant ∼30% reduced risk of type 2 diabetes in alcohol consumers of 6–48 g/day compared with heavier consumers or abstainers. Fifteen cohort studies on the relationship between alcohol consumption and the risk of type 2 diabetes were retrieved. In the absence of long-term randomized intervention studies, these can be considered the best available evidence.
The lower type 2 diabetes risk in moderate drinkers was consistent over most included studies. Despite this consistency, the risk estimates differed more across studies than was expected from the sampling error within studies. We therefore studied whether multiple adjustment, the type 2 diabetes definition used, and differences between men and women or individuals with low and high BMI could explain the observed heterogeneity. In most cases, this did not turn out to be the case. Numerous other factors, however, may also have caused the observed heterogeneity, such as differences in alcohol assessment methods, period of follow-up, and all 30 confounders that were adjusted for in some but not all included studies.
A borderline significant difference between the crude and multivariate-adjusted pooled RRs was found for the lowest consumption level. Adjustment for confounders such as age, family history of diabetes, and BMI appears to have some attenuating effect on the estimated RRs. In the seven studies that were based on self-reported type 2 diabetes status, the RRs associated with moderate drinking (6–48 g/day) were 0.57–0.65, whereas in the seven studies that used population testing of type 2 diabetes, the RRs were 0.76–0.89. This difference may be due to the higher frequency of alcohol abstainers attending a general practitioner (35), resulting in relatively more diagnosed cases in this group. The difference may also be caused by a stronger relationship of alcohol consumption with the more severe cases of diabetes that likely are overrepresented in studies relying soley on self-report.
The apparent sex difference was due to the relatively low RRs in the two Nurses’ Health Studies and the Iowa Women’s Health Study. The trend of lower RRs in women than in men was not found within the three studies that gave results for both men and women (7,28,32). Possibly, the trend for lower RRs in women than in men is caused by a study-related factor other than sex.
Within-study analyses showed that the reduced risk associated with moderate alcohol consumption was present both in individuals with a relatively low BMI and in those with a relatively high BMI. The alcohol consumption level associated with the lowest type 2 diabetes risk appears to be lower in individuals with a relatively low BMI (6–12 g/day) than in those with a higher BMI (12–24 g/day).
Several studies were not included because their data were not presented in a way that could be used in our meta-analysis. The unadjusted RRs that can be calculated from the Women’s Health Study show a convincing decrease in type 2 diabetes risk with the increase in drinking frequency (14). Also in line with the present findings, Hodge et al. (18), Lynch et al. (13), and Watanabe et al. (21) showed trends of a decreased risk of type 2 diabetes associated with alcohol consumption. In three other excluded studies, logistic regression analyses were performed with alcohol as a continuous determinant of type 2 diabetes risk (17,19,20). These studies suggest a positive relationship for men, which is in contrast with the first part but in line with the second part of the U-shaped relationship that was found for men in our meta-analysis.
Our finding of a U-shaped relationship between alcohol consumption and type 2 diabetes risk is analogous with the previously demonstrated relationship with cardiovascular diseases (36,37) and may partly share underlying mechanisms. Moderate alcohol consumption is known to increase HDL cholesterol concentration (38), whereas, at higher consumption levels, body weight, triglyceride concentration, and blood pressure increase (39–42). Another plausible mechanism is through the anti-inflammatory effect of alcohol (43,44). Enhanced insulin sensitivity with lower plasma insulin concentrations is another (and more type 2 diabetes–specific) plausible mechanism because inverse and U-shaped relationships between alcohol consumption and insulin levels have been shown (41). In a recent randomized controlled trial in women without diabetes, alcohol consumption of 30 g/day was shown to have beneficial effects on insulin and triglyceride concentrations (45).
Our study has several potential limitations. First, its quality fully depends on that of the original studies included. Validity threats of those studies are directly inherited. One of these threats is the measurement error that is inevitable when using relatively simple methods to assess alcohol consumption (46). Both differential and nondifferential misclassification of the amount of alcohol consumed at baseline may have resulted in bias. The misclassification, on average, is expected to be an underestimation of the amount of alcohol consumed (47). Therefore, the amount of alcohol consumption associated with the lowest risk of type 2 diabetes in reality may be higher than reported. In addition, one alcohol assessment at baseline may not be a precise representation of the average alcohol amount consumed over the complete time at risk. Because moderate consumers may have changed to abstainers or heavy consumers, and vice versa, it is likely that the magnitude of the U-shaped relationship is underestimated, as is shown to be the case for the relationship between alcohol and mortality (48). In addition, the meta-analysis does not give information on the importance of a stable pattern of alcohol consumption on type 2 diabetes risk. However, as has been demonstrated for coronary heart disease before (49), Conigrave et al. (24) presumed that drinking frequency is inversely associated with type 2 diabetes risk. Because detailed information on the influence of the pattern of consumption, including binge drinking, was missing in all other studies, this remains an important topic for further investigations.
Second, 7 of the 15 studies did not originally use nonconsumers as the reference category. In recalculating the RRs from these studies relative to the category of nonconsumers, the variance had to be reestimated. Since the category of nonconsumers overall comprised 50% more cases of type 2 diabetes than the reference categories that were used in the original publications, it is not likely that we underestimated the variance.
Third, the so-called “sick quitter effect” may have caused bias (individuals with preexisting disease stop drinking and thereby cause the observed lower disease risk in drinkers [50]). However, several large studies investigated this, and the concerns are largely allayed (37). Likewise, the two studies included in the present meta-analysis that reported results for former drinkers and lifetime nonconsumers separately did not find results in accordance with the “sick quitter hypothesis” (7,8).
Fourth, although most original studies adjusted for multiple potential confounders, residual confounding may have resulted in the present findings. A meta-analyses of observational studies has the limitation of all observational data that causal relationships cannot be established, even when experimental work on biological mechanisms is supporting the hypotheses. Only long-term intervention studies can determine the true benefits or adverse effects of alcohol consumption.
With the expectation of >100 million new cases of type 2 diabetes in the coming two decades, and with the prevention of type 2 diabetes now being recognized as an urgent priority, attaining prevention is the central challenge. There are calculations that 91% of new type 2 diabetic cases could be attributed to the lack of adherence to five lifestyle behaviors, among which is the moderate consumption of alcohol (26).
In conclusion, the present study supports the evidence of a considerably reduced risk of type 2 diabetes associated with moderate but not with heavy alcohol consumption in men and women with low or high BMI.
Scatterplot of the RR estimates of type 2 diabetes reported in the 15 included studies, and the pooled RR estimates with corresponding 95% CIs for five alcohol consumption categories with the nonconsumers as reference category. Each study provides more than one RR estimate. The area of each circle is proportional to the precision of the RR estimate (inverse of its variance).
Scatterplot of the RR estimates of type 2 diabetes reported in the 15 included studies, and the pooled RR estimates with corresponding 95% CIs for five alcohol consumption categories with the nonconsumers as reference category. Each study provides more than one RR estimate. The area of each circle is proportional to the precision of the RR estimate (inverse of its variance).
Characteristics of the included studies
| Reference | Study name, country | Population | Age range at baseline (years) | Sex | No. at risk with follow-up | No. of incident cases of diabetes | Follow-up (years) | Diabetes definition | Confounders adjusted for |
|---|---|---|---|---|---|---|---|---|---|
| Lee et al. (34) | Iowa Women’s Health Study, U.S. | Population based | 55–69 | Women | 35,698 | 1,921 | 9.3 | Baseline and four follow-up self reports | None |
| Sawada et al. (33) | Tokyo Gas Company Study, Japan | Employees of one company | 20–40 | Men | 4,745 | 280 | 13.6 | Annual testing | 1–6 |
| Carlsson et al. (32) | Finnish Twin Cohort, Finland | Population based | ≥18 | Men and women | 10,856 and 11,664 | 277 and 297 | 20.0 | Record linkage with national hospital and drug registers | 1 and 3 |
| Lee et al. (27) | GGT-study, Korea | Employees of one company | 25–55 | Men | 4,055 | 83 | 4.0 | Annual testing | None |
| Nakanishi, Suzuki, and Tatara (28) | Building Company Study, Japan | Office workers of one company | 35–59 | Men | 2,953 | 370 | 7.0 | Annual testing* | 1–4 and 7 |
| Wannamethee et al. (8) | Nurses’ Health Study 2, U.S. | Nurses | 25–42 | Women | 109,705 | 935 | 10.0 | Biennial self-report | 1–4 and 7–11 |
| de Vegt et al. (25) | Hoorn Study, the Netherlands | Population based | 50–75 | Men and women | 1,322 | 131 | 7.6 | Baseline and follow-up testing | 1 and 12 |
| Meisinger et al. (28) | MONICA, Germany | Population based | 35–74 | Men and women | 3,052 and 3,114 | 128 and 85 | 7.6 | Baseline and follow-up self-report | 1, 4, and 13 |
| Wannamethee et al. (31) | British Regional Heart Study, U.K. | General practitioners’ register | 40–59 | Men | 5,221 | 198 | 16.8 | Baseline and follow-up self-report | 1, 2, 4, 7, 14, and 15 |
| Conigrave et al. (24) | Health Professionals Follow-up Study, U.S. | Health professionals | 40–75 | Men | 46,892 | 1,571 | 12.0 | Biennial self-report | 1–4, 7, 9, 10, and 15–22 |
| Hu et al. (26) | Nurses’ Health Study, U.S. | Nurses | 30–55 | Women | 82,393 | 3,296 | 16.0 | Biennial self-report | 1, 3, and 23–25 |
| Kao et al. (7) | ARIC Study, U.S. | Population based | 45–64 | Men and women | 5,423 and 6,838 | 547 and 569 | 6.0 | Baseline and follow-up testing | 1–4, 7, 9, 20, and 26–28 |
| Ajani et al. (22) | Physicians’ Health Study, U.S. | Physicians | 40–85 | Men | 20,951 | 666 | 12.1 | Baseline and follow-up self-report | 1–3, 7, 9, and 10 |
| Wei et al. (23) | Cooper Clinic Study, U.S. | Population based | 30–79 | Men | 8,633 | 149 | 6.0 | Baseline and follow-up testing | 1, 3, and 29 |
| Tsumura et al. (30) | Osaka Health Survey, Japan | Employees of one company | 35–61 | Men | 6,362 | 456 | 9.7 | Biennial testing | 1–4, 7, and 30 |
| Reference | Study name, country | Population | Age range at baseline (years) | Sex | No. at risk with follow-up | No. of incident cases of diabetes | Follow-up (years) | Diabetes definition | Confounders adjusted for |
|---|---|---|---|---|---|---|---|---|---|
| Lee et al. (34) | Iowa Women’s Health Study, U.S. | Population based | 55–69 | Women | 35,698 | 1,921 | 9.3 | Baseline and four follow-up self reports | None |
| Sawada et al. (33) | Tokyo Gas Company Study, Japan | Employees of one company | 20–40 | Men | 4,745 | 280 | 13.6 | Annual testing | 1–6 |
| Carlsson et al. (32) | Finnish Twin Cohort, Finland | Population based | ≥18 | Men and women | 10,856 and 11,664 | 277 and 297 | 20.0 | Record linkage with national hospital and drug registers | 1 and 3 |
| Lee et al. (27) | GGT-study, Korea | Employees of one company | 25–55 | Men | 4,055 | 83 | 4.0 | Annual testing | None |
| Nakanishi, Suzuki, and Tatara (28) | Building Company Study, Japan | Office workers of one company | 35–59 | Men | 2,953 | 370 | 7.0 | Annual testing* | 1–4 and 7 |
| Wannamethee et al. (8) | Nurses’ Health Study 2, U.S. | Nurses | 25–42 | Women | 109,705 | 935 | 10.0 | Biennial self-report | 1–4 and 7–11 |
| de Vegt et al. (25) | Hoorn Study, the Netherlands | Population based | 50–75 | Men and women | 1,322 | 131 | 7.6 | Baseline and follow-up testing | 1 and 12 |
| Meisinger et al. (28) | MONICA, Germany | Population based | 35–74 | Men and women | 3,052 and 3,114 | 128 and 85 | 7.6 | Baseline and follow-up self-report | 1, 4, and 13 |
| Wannamethee et al. (31) | British Regional Heart Study, U.K. | General practitioners’ register | 40–59 | Men | 5,221 | 198 | 16.8 | Baseline and follow-up self-report | 1, 2, 4, 7, 14, and 15 |
| Conigrave et al. (24) | Health Professionals Follow-up Study, U.S. | Health professionals | 40–75 | Men | 46,892 | 1,571 | 12.0 | Biennial self-report | 1–4, 7, 9, 10, and 15–22 |
| Hu et al. (26) | Nurses’ Health Study, U.S. | Nurses | 30–55 | Women | 82,393 | 3,296 | 16.0 | Biennial self-report | 1, 3, and 23–25 |
| Kao et al. (7) | ARIC Study, U.S. | Population based | 45–64 | Men and women | 5,423 and 6,838 | 547 and 569 | 6.0 | Baseline and follow-up testing | 1–4, 7, 9, 20, and 26–28 |
| Ajani et al. (22) | Physicians’ Health Study, U.S. | Physicians | 40–85 | Men | 20,951 | 666 | 12.1 | Baseline and follow-up self-report | 1–3, 7, 9, and 10 |
| Wei et al. (23) | Cooper Clinic Study, U.S. | Population based | 30–79 | Men | 8,633 | 149 | 6.0 | Baseline and follow-up testing | 1, 3, and 29 |
| Tsumura et al. (30) | Osaka Health Survey, Japan | Employees of one company | 35–61 | Men | 6,362 | 456 | 9.7 | Biennial testing | 1–4, 7, and 30 |
Impaired fasting glucose also included in diabetes definition. 1, age; 2, smoking; 3, family history of diabetes; 4, BMI; 5, blood pressure; 6, fitness; 7, physical activity; 8, oral contraceptive use; 9, hypertension; 10, hypercholesterolemia; 11, infertility; 12, sex; 13, survey; 14, social class; 15, preexisting coronary heart disease; 16, dietary glycemic load; 17, fiber; 18, trans fat; 19, polyunsaturated fat; 20, energy intake; 21, profession; 22, cancer; 23, time; 24, menopausal status; 25, hormone therapy; 26, waist-to-hip ratio; 27, race; 28, education; 29, years of follow-up; 30, fasting plasma glucose.
RR (95% CI) of type 2 diabetes for alcohol consumption categories for all data, crude and adjusted data, diabetes definition tested and self-reported, men and women, and low and high BMI, with two-sided P values for differences between subgroups
| Number of studies | Alcohol consumption categories (g/day) | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 (ref.) | <6 | 6–12 | 12–24 | 24–48 | >48 | |||||||
| All data | 15 | 1 | 0.87 (0.79–0.95) | 0.70 (0.61–0.79) | 0.69 (0.58–0.81) | 0.72 (0.62–0.84) | 1.04 (0.84–1.29) | |||||
| Crude | 10 (w) | 1 | 0.70 (0.59–0.83) | 0.59 (0.40–0.89) | 0.50 (0.36–0.69) | 0.71 (0.59–0.85) | 0.98 (0.78–1.23) | |||||
| Adjusted | 10 (w) | 1 | 0.88 (0.80–0.95) | 0.73 (0.62–0.86) | 0.66 (0.59–0.75) | 0.74 (0.63–0.88) | 0.93 (0.74–1.18) | |||||
| P | 0.04 | 0.36 | 0.12 | 0.75 | 0.77 | |||||||
| Type 2 diabetes tested | 7 (b) | 1 | 0.91 (0.79–1.05) | 0.89 (0.72–1.10) | 0.86 (0.64–1.15) | 0.76 (0.63–0.93) | 1.14 (0.89–1.45) | |||||
| Type 2 diabetes self-reported | 7 (b) | 1 | 0.85 (0.75–0.97) | 0.65 (0.57–0.73) | 0.57 (0.50–0.65) | 0.59 (0.49–0.71) | 0.88 (0.55–1.39) | |||||
| P | 0.51 | 0.04 | 0.03 | 0.10 | 0.36 | |||||||
| Men | 11 (w and b) | 1 | 0.93 (0.82–1.04) | 0.80 (0.71–0.90) | 0.75 (0.60–0.95) | 0.71 (0.60–0.83) | 1.06 (0.86–1.32) | |||||
| Women | 6 (w and b) | 1 | 0.81 (0.75–0.88) | 0.59 (0.54–0.64) | 0.55 (0.47–0.65) | 0.78 (0.49–1.23)* | ||||||
| P | 0.14 | 0.01 | 0.21 | |||||||||
| BMI lowest strata | 6 (w) | 1 | 0.76 (0.65–0.88) | 0.64 (0.44–0.92) | 0.73 (0.53–1.01) | 0.85 (0.49–1.46) | 1.28 (0.37–4.40) | |||||
| BMI highest strata | 6 (w) | 1 | 0.85 (0.79–0.92) | 0.75 (0.63–0.90) | 0.67 (0.57–0.78) | 0.71 (0.55–0.91) | 0.92 (0.73–1.16) | |||||
| P | 0.53 | 0.45 | 0.53 | 0.79 | 0.65 | |||||||
| Number of studies | Alcohol consumption categories (g/day) | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 (ref.) | <6 | 6–12 | 12–24 | 24–48 | >48 | |||||||
| All data | 15 | 1 | 0.87 (0.79–0.95) | 0.70 (0.61–0.79) | 0.69 (0.58–0.81) | 0.72 (0.62–0.84) | 1.04 (0.84–1.29) | |||||
| Crude | 10 (w) | 1 | 0.70 (0.59–0.83) | 0.59 (0.40–0.89) | 0.50 (0.36–0.69) | 0.71 (0.59–0.85) | 0.98 (0.78–1.23) | |||||
| Adjusted | 10 (w) | 1 | 0.88 (0.80–0.95) | 0.73 (0.62–0.86) | 0.66 (0.59–0.75) | 0.74 (0.63–0.88) | 0.93 (0.74–1.18) | |||||
| P | 0.04 | 0.36 | 0.12 | 0.75 | 0.77 | |||||||
| Type 2 diabetes tested | 7 (b) | 1 | 0.91 (0.79–1.05) | 0.89 (0.72–1.10) | 0.86 (0.64–1.15) | 0.76 (0.63–0.93) | 1.14 (0.89–1.45) | |||||
| Type 2 diabetes self-reported | 7 (b) | 1 | 0.85 (0.75–0.97) | 0.65 (0.57–0.73) | 0.57 (0.50–0.65) | 0.59 (0.49–0.71) | 0.88 (0.55–1.39) | |||||
| P | 0.51 | 0.04 | 0.03 | 0.10 | 0.36 | |||||||
| Men | 11 (w and b) | 1 | 0.93 (0.82–1.04) | 0.80 (0.71–0.90) | 0.75 (0.60–0.95) | 0.71 (0.60–0.83) | 1.06 (0.86–1.32) | |||||
| Women | 6 (w and b) | 1 | 0.81 (0.75–0.88) | 0.59 (0.54–0.64) | 0.55 (0.47–0.65) | 0.78 (0.49–1.23)* | ||||||
| P | 0.14 | 0.01 | 0.21 | |||||||||
| BMI lowest strata | 6 (w) | 1 | 0.76 (0.65–0.88) | 0.64 (0.44–0.92) | 0.73 (0.53–1.01) | 0.85 (0.49–1.46) | 1.28 (0.37–4.40) | |||||
| BMI highest strata | 6 (w) | 1 | 0.85 (0.79–0.92) | 0.75 (0.63–0.90) | 0.67 (0.57–0.78) | 0.71 (0.55–0.91) | 0.92 (0.73–1.16) | |||||
| P | 0.53 | 0.45 | 0.53 | 0.79 | 0.65 | |||||||
The single observation in women in the >48-g/day category is pooled with the observations in the 24- to 48-g/day category. w, within-study comparison; b, between-study comparison.
Article Information
This work was supported by an unconditional grant from the Alcohol Task Force of the European branch of the International Life Sciences Institute (ILSI Europe). Industry members of this task force are Allied Domecq, Brasseries Kronenbourg, Diageo, Heineken, and Moët et Chandon.
The opinions expressed in this article are those of the authors and do not necessarily represent the views of ILSI Europe. We thank Dr. Rob J.P.M. Scholten, Dutch Cochrane Centre, for his assistance with the data analysis.
