Alcohol Consumption and the Incidence of Type 2 Diabetes

A 20-year follow-up of the Finnish Twin Cohort Study

  1. Sofia Carlsson, PHD12,
  2. Niklas Hammar, PHD2,
  3. Valdemar Grill, MD, PHD3 and
  4. Jaakko Kaprio, MD, PHD4
  1. 1Division of Epidemiology, Stockholm Centre of Public Health, Karolinska University Hospital, Stockholm, Sweden
  2. 2Department of Epidemiology, Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden
  3. 3Department of Molecular Medicine, Karolinska Institutet, Stockholm, Sweden
  4. 4Department of Public Health, University of Helsinki and Department of Mental Health, National Public Health Institute, Helsinki, Finland
  1. Address correspondence and reprint requests to Sofia Carlsson, Division of Epidemiology, Karolinska University Hospital, S-171 76 Stockholm, Sweden. E-mail: sofia.carlsson{at}


OBJECTIVE—The aim of this study was to investigate alcohol consumption in relation to the incidence of type 2 diabetes.

RESEARCH DESIGN AND METHODS—The study population consisted of 22,778 twins of the Finnish Twin Cohort. This cohort was compiled in 1975 and includes all same-sexed twins born in Finland before 1958. Information on alcohol, smoking, diet, physical activity, medical, and social conditions was obtained by questionnaires administered in 1975, 1981, and 1990. By record linkage to national registers of hospital discharge and prescribed medication, 580 incident cases of type 2 diabetes were identified during 20 years of follow-up.

RESULTS—Moderate alcohol consumption (5–29.9 g/day in men and 5–19.9 g/day in women) tended to be associated with a reduced incidence of type 2 diabetes compared with low consumption (<5 g/day). The estimates were lower in overweight (BMI ≥25.0 kg/m2) subjects (relative risk 0.7, 95% CI 0.5–1.0 [men]; 0.6, 0.3–1.1 [women]). High alcohol consumption (≥20 g/day) was associated with an increased incidence of type 2 diabetes in lean women (2.9, 1.1–7.5) but not in overweight women or in men. In women, binge drinking was associated with an increased incidence of type 2 diabetes (2.1, 1.0–4.4). Analyses of alcohol-discordant twin pairs supported a reduced risk in moderate consuming twins compared with their low-consuming cotwins (odds ratio 0.5, 95% CI 0.2–1.5).

CONCLUSIONS—The results of this study suggested that moderate alcohol consumption may reduce the risk of type 2 diabetes. On the other hand, binge drinking and high alcohol consumption may increase the risk of type 2 diabetes in women.

In recent years, increasing attention has been paid to the role of alcohol in the etiology of type 2 diabetes. A number of studies have investigated this issue, but with contradictory results. Seven studies report an increased risk in high consumers (17) that was not confirmed by others (815), and nine report a protective effect (3,4,612) primarily at moderate levels of consumption. In addition, a lack of association between alcohol and type 2 diabetes has been reported (1315). The increased risk has been confined to men (17), whereas in women, the results indicate no relationship (1,5,14,15) or a protective effect (7,12) of alcohol on type 2 diabetes.

Diverging results from previous studies may have several explanations. Positive and negative effects of alcohol on glucose homeostasis have been documented. Alcohol has been demonstrated to enhance insulin sensitivity (16,17). On the other hand, alcohol may exert a diabetogenic influence by a toxic effect on the pancreas (18) or indirectly by increasing adiposity (19,20). Differences in drinking patterns may also have contributed to diverging results. For example, binge drinking could affect glucose homeostasis differently than consumption that is more evenly distributed. Finally, misclassification of self-reported alcohol consumption may have diluted the association between high alcohol intake and type 2 diabetes. Hence, there is a need for additional studies in which the intake of alcohol can be analyzed prospectively in more detail in relation to the incidence of type 2 diabetes.

Against this background, we investigated alcohol consumption and incidence of type 2 diabetes in a 20-year follow-up of 11,501 twin pairs of the Finnish Twin Cohort. The long follow-up, large study population, repeated measures of alcohol intake, and possibilities to analyze the association between alcohol intake and type 2 diabetes in alcohol discordant twin pairs makes this study unparalleled.



The study population consisted of twin pairs of the Finnish Twin Cohort. The compilations of the Finnish Twin Cohort and baseline and follow-up data collections have been described previously in detail (21,22). In total, it consists of ∼16,000 same-sexed twin pairs who were born in Finland before 1958 and in which both twins of the pair were alive in 1967. A baseline questionnaire was administered in 1975 with an overall response rate of 89%. The first follow-up questionnaire was mailed in 1981 with a response rate of 84% and a second follow-up in 1990 with a response rate of 77% among those asked to participate. All three questionnaires included questions on lifestyle factors such as alcohol intake, smoking, and physical activity along with questions on psychosocial and medical conditions and questions for classification of zygosity. The present study consisted of 22,778 twins (14,389 dizygotic twins, 6,694 monozygotic twins, and 1,695 of unknown zygosity) free of diabetes at baseline and with complete information on alcohol intake from baseline.


Intake of alcohol was measured with separate questions on quantity of beer, wine, and spirits used during an average week (beer, wine) or month (spirits). The validity and reliability of these self-reports of alcohol use has been reported previously (23). Quantity was measured on three seven-point scales, with the upper limits defined as consuming ≥48 bottles of beer per week, ≥10 bottles of wine per week, or ≥20 bottles of spirits per month. For each type of beverage, the reported consumption was converted into grams of absolute alcohol and summed to yield an estimate of total alcohol consumption in grams per day. The midpoint of each response category was used for calculations except for the highest consumption category for which we used the lower limit in the calculations to obtain a conservative estimate. To assess binge drinking, subjects were asked if they had consumed more than five bottles of beer, one bottle of wine, or four drinks (≥18 cl of spirits) on the same occasion at least once a month during the preceding year. In addition, the questionnaire in 1981 included a question on pass outs from alcohol consumption during the preceding year, which was measured on a five-point scale with more than seven times as the highest category.

Self-reported weight and height were used to calculate BMI as weight in kilograms divided by the square of height in meters. Several questions were used to measure leisure time physical activity, including frequency, duration, and intensity. Based on answers to these questions, an activity MET was calculated by assigning an MET index to each activity and by calculating the product of intensity × duration × frequency of activity (24). Based on a detailed smoking history, subjects were classified as current, former, never, and occasional smokers. Classification by occupational group was based on job title according to the Central Statistical Office of Finland (25). Zygosity of twins (monozygotic, dizygotic, or unclassified) was classified on the basis of questions about strangers confusing the twins in childhood and similarity in childhood appearance, which is a method that has been shown to be highly accurate (26).

Type 2 diabetes

Incident cases of type 2 diabetes during the period 1972–1995 were identified through record linkage using information from the Finnish national hospital discharge register together with information from a national drug register in Finland using procedures described by Kaprio et al. (27). In Finland, patients with diabetes are provided with antidiabetic drug therapy fully or nearly free of charge. The prerequisite for such free medication is a patient application that is based on a detailed medical certificate, which is registered at the Social Insurance Institution. This register was linked with the Finnish Twin Cohort for the years 1972–1995 using a unique personal identification number assigned to each Finnish citizen. With regard to type 1 diabetes, the register is virtually complete, but not with regard to type 2 diabetes, since a large number of these patients are treated with diet only, at least for some time after diagnosis. The National Agency for Welfare and Health has kept a national register of hospital discharges since 1968. This register was linked to the Finnish Twin Cohort for the years 1964–1987, and all subjects with the diagnosis code 250 (diabetes) (ICD 8 and later ICD 9) were identified. For the years 1988–1995, the hospital records were only used to verify unclear diabetes diagnoses based on the drug register. The reason for this change in the classification procedure was the high coverage and reliability of the drug register. We excluded subjects with diabetes at baseline either based on register information or according to the 1975 questionnaire from the analyses (n = 417).

The registers used to identify cases of diabetes do not include computerized information on type of diabetes. Therefore, to determine type of diabetes, copies of original records from the two registers were reviewed together with copies of death certificates for the deceased. The classification was based on World Health Organization criteria from 1985 (28). Patients with an abrupt onset of the disease that required insulin from the time of diagnosis were classified as having type 1 diabetes. Patients who only received oral medication or dietary therapy were classified as having type 2 diabetes, as were subjects in whom insulin therapy had been implemented as an adjuvant treatment to hypoglycemic drugs. A physician subjected all cases with ambiguous information on type of therapy. The diagnosis of type 1 diabetes was accepted if the development of insulin dependence was clearly demonstrated. In all, 807 cases of diabetes were identified 1976–1995, and of these, information from the medical registers were obtained for 701 cases (87%). Based on the review of these records, 589 cases of type 2 diabetes, 35 cases of type 1 diabetes, 38 cases of gestational diabetes, 19 cases of secondary diabetes, 15 cases of impaired glucose tolerance, and 5 cases who were unclassifiable or belonged to other types of diabetes were identified. Nine cases of type 2 diabetes could not be included because of missing data on alcohol, yielding a total of 580 cases of type 2 diabetes. In addition, a history of diabetes was also asked for in all three questionnaires mailed to the Finnish Twin Cohort. Subjects who stated that they had diabetes but who were not identified through record linkage were excluded due to lack of information on onset year and type of diabetes (n = 168).


Person-years of follow-up were accumulated starting 1 January 1976. The follow-up was terminated at the date of diagnosis of diabetes, at the date of death, at the date of emigration from Finland, or on 31 December 1995, whichever came first. Alcohol consumption was analyzed categorically with different cutoff levels in men and women because men in our sample consumed substantially more alcohol than women. To analyze the association between alcohol consumption and type 2 diabetes, we calculated relative risk (RR) estimates together with 95% CIs using Cox proportional hazards models (SAS PHREG) adjusted for age and BMI. Time-dependent analyses were performed using baseline information on alcohol together with information from 1981 and 1990. Correspondingly, information on BMI was updated during follow-up. The incidence of type 2 diabetes was also assessed during the time period 1982–1995, and the incidence among subjects reporting the same consumption level in 1975 and 1981 (abstaining, moderate, or high alcohol consumption) was compared with the incidence in subjects reporting low alcohol consumption on both occasions.

The methods for analyzing our data were based on the statistical assumption of independent observations, which was not strictly true because the study population to a large extent consisted of complete twin pairs. To take the intrapair correlations into account, some results were recalculated using γ frailty models (29) analyzed by S-PLUS (30). These analyses gave basically similar results and will therefore not be displayed in tables. To investigate the association between alcohol consumption and type 2 diabetes taking hereditary and childhood factors into account, we calculated odds ratios of type 2 diabetes in alcohol-discordant twin pairs.



Few women of the study population reported consumption of even moderate amounts of alcohol. More than three-fourths of the women consumed <5 g alcohol per day, and only ∼3% consumed ≥20 g alcohol per day (Table 1). Female abstainers and women reporting low alcohol consumption were older on average and tended to have slightly higher BMI than women consuming more alcohol. High consumers of alcohol tended to be smokers, less physically active, and have lower socioeconomic status.

Alcohol consumption and incidence of type 2 diabetes

In both men and women, moderate consumers of alcohol tended to have a lower incidence of type 2 diabetes than low consumers (Table 2). A slightly increased incidence of diabetes was suggested in women with comparatively high alcohol consumption, but the CIs were wide. Abstaining from alcohol compared with low consumption was not associated with type 2 diabetes in either men or women. From the abstainer group, we excluded subjects who reported prior consumption of alcohol. This resulted in an RR of 1.0 (95% CI 0.7–1.4) in men and 1.2 (1.0–1.6) in women.

Some of those counted as incident cases in our study may have had undiagnosed diabetes already at baseline. To avoid this possible bias, we made separate analyses in which cases of diabetes occurring during the first 6 years of follow-up were excluded. The results were very similar, indicating an RR of 0.7 (95% CI 0.4–1.0) for men and 1.4 (0.6–3.1) for women in moderate consumers. Corresponding estimates in high consumers were 1.0 (0.5–1.9) for men and 1.4 (0.6–3.3) for women.

Inclusion of moderate consumers reporting frequent binge drinking into the high consumption group indicated an RR of 0.8 (95% CI 0.6–1.2) in men and 1.3 (0.7–2.5) in women. Moderate consumers (excluding binge drinkers) indicated an RR of 0.5 (0.2–1.1) in men and 0.6 (0.3–1.2) in women.

The results were adjusted for age and BMI. Additional adjustment for smoking, socioeconomic status, and physical activity had essentially no effect on the RR associated with alcohol consumption. For example, the RR of type 2 diabetes associated with moderate alcohol consumption was 0.8 (95% CI 0.6–1.1) for men and 0.7 (0.4–1.2) for women after this adjustment. For high alcohol consumption, corresponding estimates were 0.9 (0.6–1.3) for men and 1.5 (0.7–3.2) for women. When intrapair correlation of twins was taken into account, the RR associated with moderate consumption was 0.8 (0.6–1.1) for men and 0.7 (0.4–1.1) for women. Corresponding estimates in high alcohol consumers were 0.9 (0.6–1.4) for men and 1.6 (0.8–3.5) for women.

Alcohol consumption and BMI

Moderate alcohol consumption was associated with a 30–40% reduced risk of type 2 diabetes in overweight (BMI ≥25.0 kg/m2) men and women with no corresponding reduction in lean or normal weight men (BMI <25.0 kg/m2) (Table 3). Among lean or normal weight women, high alcohol consumption was associated with an almost threefold increased RR of type 2 diabetes. There was also a tendency for an increased risk in lean and normal weight abstainers.

Duration of alcohol consumption

We also compared the incidence of diabetes in subjects reporting the same level of consumption in 1981 as in 1975. They were considered to have stable alcohol consumption over 5 years. Women with a stable high consumption displayed a more than four times (95% CI 1.5–12.1) increased risk of diabetes compared with women with low consumption, but these results were based on only four exposed cases. In men, stable high alcohol consumption was not associated with an increased RR. Stable moderate consumption was associated with a 30% reduced incidence of type 2 diabetes (0.4–1.1) in men but not in women (RR 1.2, 0.6–2.4). These results were similar in separate analyses of overweight and lean to normal weight subjects. In neither women nor men was stable abstaining from alcohol compared with low consumption associated with type 2 diabetes.

Binge drinking

Binge drinking was not associated with type 2 diabetes in men. In women, a doubled incidence was seen in women reporting binge drinking (RR 2.1, 95% CI 1.0–4.4) or passing out from alcohol intoxication (2.4, 1.1–5.0).

Incidence of type 2 diabetes in alcohol discordant twin pairs

Analyses of alcohol discordant twin pairs suggested a 50%, nonsignificant, reduced odds ratio of type 2 diabetes in moderate consumers of alcohol (15–29.9 g/day) compared with their low-consuming (<5 g/day) siblings (Table 4). No reduced risk was seen in high-consuming twins compared with their low-consuming siblings. Because of small numbers, separate analyses in monozygotic and dizygotic twins were not meaningful.


A number of previous studies has found that moderate alcohol consumers experience a 30–40% reduced incidence of type 2 diabetes (3,4,612). The findings of this study are in line with these observations. With regard to high alcohol consumption, we find no support for an increased risk in men as suggested previously (17) or for a protective effect, as indicated by others (9). Lean to normal weight women who consume comparatively high amounts of alcohol were found to have an increased incidence of type 2 diabetes. However, these results must be interpreted with caution as very few women consumed large amounts of alcohol, and there may be confounding from social, dietary, or other factors that we could no fully take into account.

To our knowledge, this is the first study to investigate the association between alcohol consumption and type 2 diabetes in a prospective study of twins allowing us to control for hereditary and early childhood factors. Small numbers seriously hampered the twin/cotwin analyses. However, the notion of a protective effect of moderate drinking was supported by a comparison of moderate alcohol-consuming twins with their low alcohol-consuming sibling.

The results of this study suggested that the protective effect of alcohol may primarily concern overweight subjects and that the increased risk may primarily concern normal weight or lean women. This is in line with findings by Tsmura et al. (4), who found that alcohol is associated with an increased risk of type 2 diabetes in lean men but a protective effect in overweight men. On the other hand, other studies indicate that alcohol is associated with a reduced risk of diabetes independent of BMI (3,10,12). A protective effect of alcohol consumption primarily in overweight subjects seems biologically plausible because alcohol is associated with enhanced insulin sensitivity (16,17), which could counteract obesity-induced insulin resistance.

The Nordic drinking pattern is traditionally characterized by weekend binge drinking and consumption of beer and spirits rather than wine. Our results suggest that binge drinking will increase the risk of type 2 diabetes in women but not in men. Binge drinking in men is almost normative in Finland (40% of subjects) and, hence, only weakly indicative of alcohol dependence or abuse compared with the situation in women. However, we do not have the data to identify subjects with alcoholism in the sample.

Studies on alcohol and type 2 diabetes are difficult to compare in part because of the use of different methods for identification of cases. Previous studies reporting an increased risk of diabetes in high consumers have all investigated diabetes diagnosed by screening (17). In contrast, the majority of studies demonstrating a lack of association between high alcohol consumption and diabetes-investigated self-reported diabetes (812) with three exceptions: the studies by Hodge et al. (14) and Monterrosa et al. (15), which included four high-risk populations, and a study by Todoroki et al. (13). The present study is the first to use register information for identification of cases. The method has been shown to detect ∼80% of all incident cases in a population (27) but will miss cases treated only by diet and not requiring hospital care, as well as subjects with undiagnosed diabetes. Screening will detect milder forms of type 2 diabetes. If high alcohol consumption primarily promotes mild diabetes, this may explain why only studies based on screening have shown an increased risk of diabetes in high consumers. Furthermore, if the propensity to see a doctor or have routine health examinations is related to alcohol consumption, the results of our study and studies based on self-reports may be biased. On the other hand, it does not seem probable that the lack of association between high alcohol intake and type 2 diabetes results from an effect of diabetes symptoms on alcohol consumption because the association was similar when we only used baseline information on alcohol and excluded the first 6 years of follow-up.

Self-reported alcohol information is generally afflicted with underreporting (31), and this could have contributed to the lack of association between high alcohol consumption and type 2 diabetes among men in the present study. A previous study based on the Finnish Twin Cohort (32) indicated that classifying moderate consumers reporting binge drinking as high consumers could improve sensitivity for identification of high consumers. Application of this classification in the present study did, however, only have a minor effect on the risk estimates. Further misclassification and, hence, dilution of the RR will occur if changes in alcohol consumption that occur during follow-up are not taken into account. We tried to minimize this bias by updating baseline alcohol information from 1975 in 1981 and 1990.

In conclusion, our results suggested that moderate alcohol consumption may reduce the risk of type 2 diabetes. For women, prolonged high alcohol consumption and binge drinking may increase the risk of type 2 diabetes.

Table 1—

Baseline characteristics of subjects according to reported alcohol consumption

Table 2—

RR of type 2 diabetes associated with alcohol intake

Table 3—

RR of type 2 diabetes associated with alcohol consumption by BMI

Table 4—

Odds ratio of type 2 diabetes in alcohol discordant twin pairs


This study was supported by grants from the Swedish Council for Working Life and Social Research and the Sigrid Juselius foundation.

The authors thank Tomas Andersson for valuable statistical advice, as well as Drs. Giti Khalighi-Sikaroudi and Mikko Lehtovirta for reviewing the diabetes records.


  • N.H. is now employed as a senior Epidemiological Scientist at Astra Zeneca R&D. This study was designed and carried out before N.H. began his employment with Astra Zeneca R&D. N.H. also works part time with the Institute of Environmental Medicine, Karolinska Institutet.

  • A table elsewhere in this issue shows conventional and Système International (SI) units and conversion factors for many substances.

    • Accepted June 26, 2003.
    • Received January 29, 2003.


| Table of Contents