Insulin sensitivity and regular alcohol consumption: large, prospective, cross sectional population study (Bruneck study)

BMJ 1996; 313 doi: (Published 26 October 1996) Cite this as: BMJ 1996;313:1040
  1. Stefan Kiechl, lecturer in neurologya,
  2. Johann Willeit, associate professora,
  3. Werner Poewe, professora,
  4. Georg Egger, consultant cardiologistb,
  5. Friedrich Oberhollenzer, head of hospitalb,
  6. Michele Muggeo, professorc,
  7. Enzo Bonora, associate professorc
  1. a Department of Neurology, University of Innsbruck, Innsbruck, Anichst 35, A-6020 Innsbruck, Austria
  2. b Department of Internal Medicine, Bruneck Hospital, I-39031 Bruneck, Italy
  3. c Department of Endocrinology and Metabolism, Ospedale Civile Maggiore, I-37126 Verona, Italy
  1. Correspondence to: Dr Kiechl.
  • Accepted 6 September 1996


Objectives: To assess the relation between regular alcohol consumption and insulin sensitivity, and to estimate the importance of insulin in the association of alcohol with multiple vascular risk factors and cardiovascular disease.

Design: Prospective and cross sectional study of a large randomly selected population sample.

Setting: Part of the Bruneck study 1990–5 (Bolzano province, Italy).

Subjects: 820 healthy non-diabetic women and men aged 40–79 years.

Main outcome measure: Concentrations of fasting and post-glucose insulin, cholesterol, apolipoproteins, triglycerides, Lp(a) lipoprotein, glucose, fibrinogen, and antithrombin III; blood pressure; insulin resistance estimated by the homeostasis model assessment.

Results: Fasting insulin concentrations in those who did not drink alcohol and subjects reporting low (1–50 g/day), moderate (51–99 g/day), and heavy (>/=100 g/day) alcohol intake were 12.4, 10.0, 8.7, and 7.1 mU/l (P<0.001). Likewise, post-glucose insulin concentrations and estimates for insulin resistance assessed by the homeostasis model assessment decreased significantly with increasing amounts of regular alcohol consumption. These trends were independent of sex, body mass index, physical activity, cigarette smoking, medication, and diet (P<0.001). Regular alcohol intake predicted multiple changes in vascular risk factors over a five year period including increased concentrations of high density lipoprotein cholesterol and apolipoprotein A I; higher blood pressure; and decreased concentration of antithrombin III. These associations were in part attributable to the decrease in insulin concentrations observed among alcohol consumers.

Conclusions: Low to moderate amounts of alcohol, when taken on a regular basis, improve insulin sensitivity. Insulin is a potential intermediate component in the association between alcohol consumption and vascular risk factors (metabolic syndrome).

Key messages

  • Regular alcohol consumption predicted multiple changes of vascular risk factors over a five year period

  • This alcohol associated metabolic syndrome is in part attributable to the decline in insulin concentrations


Insulin resistance and hyperinsulinaemia are prominent predictors of risk for the development of diabetes mellitus1 and may promote atherosclerotic diseases because of the association with multiple vascular risk factors and direct atherogenic effects.2 3 4 5 6 For preventive purposes precise knowledge of environmental determinants of insulin sensitivity is mandatory. Obesity and insufficient physical activity rank among these factors. With regard to alcohol consumption, a further widespread and potentially modifiable behaviour, epidemiological evidence is sparse and restricted to low amounts of alcohol and to women.7 8 We investigated the relation between insulin concentrations and regular alcohol consumption (in the range 0–210 g/day) in a large population sample of men and women.



The town of Bruneck is located in an alpine region in the north of Italy (Bolzano province). The study population was recruited as a random sample stratified for age and sex of all inhabitants aged 40 to 79 years (n = 4793) in such a way that 125 women and 125 men of each decade were invited to participate.9 10 11 The baseline examination was performed from July to November 1990 and the first follow up from July to October 1995. Participation and follow up rates were high at 93.6% and 96.5%. Subjects with cerebrovascular disease, incomplete data collection, unavailable insulin measurements, and diabetes mellitus (n = 60; World Health Organisation criteria12) were considered ineligible, which left 820 (1990) and 771 (1995) men and women for analysis.


Alcohol consumption was assessed by a standardised interview based on a questionnaire.11 Subjects were instructed to indicate their customary frequency of drinking (days a week) and the average amount of alcoholic beverages ingested on a typical occasion or during a typical day (500 ml bottles of beer and 250 ml glasses of wine equivalent to 25 g alcohol, standard drinks of spirits 8–10 g alcohol each). Additionally, self administered prospective diet records including the same alcohol items were collected over an extended period of seven days in a representative subsample (n = 404; participation rate 91%11). In both evaluations average alcohol consumption was quantified in terms of grams a day (g/day) and classified in four categories: abstainers (0 g/day) and light (1–50 g/day), moderate (51–99 g/day), and heavy drinkers (>/=100 g/day). Advantages and validation of this categorisation, including the high reproducibility ((kappa) coefficient 0.87), have been described previously.11

Subjects were coded as non-smokers or light (1–10 cigarettes), moderate (11–20), or heavy (>20) smokers on the basis of usual smoking habits. Social class was defined by the education of the subject and the occupation of the person with the highest income in the household.9 Systolic and diastolic blood pressures were assessed as means of two independent measurements, each of which was taken after at least 10 minutes of rest. The activity score consisted of the average of the scores for work (three categories) and sports or leisure activities (0, </=2, >2 hours a week). Blood samples were collected between 7 30 and 9 30 am after 12 hours of fasting and abstinence from smoking and two hours after a standardised oral glucose challenge.9 Concentrations of laboratory variables were determined as follows: triglycerides (interassay coefficient of variation 4.3-5.4% for different standards) and total and high density lipoprotein cholesterol were determined enzymatically (by the cholesterol oxidase-4-phenyl-2, 3-dimethyl-4-amino-5-pyrazolone and the glycerol phosphate oxidase-4-phenyl-2, 3-dimethyl-4-amino-5-pyrazolone (CHOD-PAP and GPO-PAP) methods, Merck; coefficient of variation 2.2-2.4%); Lp(a) lipoprotein with an enzyme linked immunosorbent assay (ELISA) (Immuno, Vienna; 3.5-6.3%), insulin according to Hales and Randle (3.2-4.8%)13 and by a human insulin specific radioimmunoassay (Linco Research, 3.9%); apolipoproteins by a nephelometric fixed time method (apolipoprotein A I 5.7%; apolipoprotein B 2.4%); and antithrombin III with a chromogenic assay (3.9-4.9%). Concentrations of low density lipoprotein cholesterol were calculated with the Friedewald formula and corrected for Lp(a) lipoprotein cholesterol. Insulin resistance and ß cell deficit were estimated by the homoeostasis model assessment (coefficients of variation 31% and 32%).14 Normal weight, healthy subjects aged <35 who did not drink alcohol or smoke, who had normal glucose tolerance, and who were not taking any drug treatment were assumed to have 100% ß cell function and an insulin resistance of 1.


Multiple regression analysis was applied to assess relations of alcohol consumption with insulin and vascular risk attributes (SPSS-X version 4.015 and BMDP software16). Levels of alcohol intake were modelled as a set of categories or as a trend. Confidence intervals and P values presented were mainly derived from the latter approach as most results showed clear (linear) trends with alcohol intake. To evaluate the consistency of these associations by sex first order interaction terms were added to the models. Variables with a skewed distribution of serum concentrations were normalised by ln (natural log) transformation. Quantitive changes in vascular risk factors expected for the variation of insulin concentration across alcohol categories were estimated from linear regression equations (of the variable on insulin and covariates) and expressed as proportions of the differences actually observed.


A total of 114 women (28%) and 314 men (76%) reported regular alcohol consumption. In the population as a whole, fasting and post-glucose insulin concentrations steadily decreased with increasing daily consumption, as did ratios of insulin resistance estimated by the homeostasis model assessment (table 1). When alcohol intake was modelled as a continuous variable an increase in daily alcohol consumption of 50 g predicted changes in insulin resistance and fasting and post-glucose insulin of −10% (95% confidence interval −5% to −14%), −14% (−4% to −22%), and −10% (−3% to −15%), respectively. These findings were consistent by sex and independent of age, body mass index, smoking, social status, and physical activity.

Table 1

Mean values of surrogates for insulin resistance and haemoglobin A1c, and glucose concentrations by daily alcohol consumption (n = 820)

View this table:

Of the variability of serum insulin concentration in the entire population sample, 3.9% was attributable to alcohol consumption as compared with 9.4% for obesity and 4.1% for differences in physical activity. According to the high incidence of regular alcohol consumption, the attributable fraction for men amounted to 5.8% (25% of totally explained variability of insulin).

The prevalence of impaired glucose tolerance was low and consistent in the four categories of alcohol consumption (10.2, 8.8, 10.1, and 7.9%). Exclusion of these subjects did not affect changes in insulin concentration with increasing alcohol consumption (insulin 11.7 mU/l (0 g/day) to 7.1 mU/l (>/=100 g/day); P<0.001). Separate analyses that excluded or recategorised former drinkers11 yielded similar results (insulin 12.1 mU/l (0 g/day) to 7.1 mU/l (>/=100 g/day); P<0.001), as did models that adjusted for dietary differences, drugs known to interfere with insulin and glucose metabolism (corticosteroids, thyroid hormones, diuretics, oral contraceptives, ß blockers, etc17), and markers of liver function including (gamma)-glutamyl-transferase, the transaminases, prothrombin time, and albumin concentrations.

Finally, all analyses were supplemented and consistently confirmed by calculations on the basis of alcohol quantities derived from the prospective diet records (data not presented).

Regular alcohol consumption may produce considerable changes in lipoprotein metabolism and blood clotting. In our population sample concentrations of high density lipoprotein cholesterol and apolipoprotein A I linearly increased across categories of alcohol consumption, whereas antithrombin III, fibrinogen, and triiodothyronine concentrations exhibited opposite trends (table 2). The pattern of related risk factors did not differ between sexes except for a preferential enhancement of concentration of high density lipoprotein cholesterol in the men (P<0.05 for effect modification). In a second stage we prospectively assessed changes of risk factors over five years, dependent on the average amount of daily alcohol consumption in this period. Except for the association between alcohol intake and fibrinogen concentration the trends observed correspond with the cross sectional data (table 3). Varying concentrations of insulin and insulin resistance (homoeostasis model assessment) among drinkers might explain 9–35% of changes in concentrations of high density lipoprotein cholesterol, apolipoprotein A I, and antithrombin III associated with alcohol.

Table 2

Mean values of potential cardiovascular risk factors and selected laboratory variables by daily alcohol consumption (n = 820)

View this table:
Table 3

Changes in cardiovascular risk factors and laboratory variables between 1990 and 1995 by daily alcohol consumption (n = 771)

View this table:



In both sexes we found a strong tendency of serum insulin concentrations to decrease with increasing chronic alcohol consumption. This trend emerged independent of obesity, physical activity, smoking habits, social status, and drug use. The few previous reports on the relation between regular alcohol consumption and insulin concentrations were restricted to low amounts of average alcohol intake in women. In the Kaiser permanent women twins study median daily alcohol consumption amounted to 4 g a day.7 An increment of 12 g a day (highest category of drinking) was associated with 8% lower post-glucose insulin concentrations. In a study of women in Bristol, those consuming up to 30 g a day had lower fasting insulin concentrations than did non-drinkers.8 Consistent epidemiological evidence of lower insulin concentrations among drinkers7 8 18 19 was further substantiated by results from an experimental animal model. In obese rats prone to atherosclerosis insulin concentration and ß cell hyperplasia were reduced after long term alcohol intake.20

Fasting and post-glucose insulin are well established surrogate measures for insulin resistance in nondiabetic subjects,21 22 with low concentrations of insulin as observed in alcohol consumers indicating higher insulin sensitivity. Theoretically, impaired secretion or enhanced hepatic degradation of insulin could interfere with this assumption: The latter mechanism, however, may not apply to alcohol drinking given the experimental observation of an unchanged or even reduced liver uptake of insulin after acute or chronic intake of alcohol.23 24 Likewise, inhibition of ß cell secretion was observed only after administration of high doses of ethanol,25 26 27 whereas low to moderate intake of alcohol tended to enhance insulin release (“priming effect”) in an experimental setting.25 28 The homoeostasis model assessment permits differentiation between insulin resistance and ß cell deficit in epidemiological surveys with the limitations of high variability in measurement and uncertain estimates of ß cell function at normal glucose concentrations.14 When we used this method we observed a significant decrease in insulin resistance across all alcohol categories. Besides, ß cell deficiency as indicated by the coincidence of fasting hyperglycaemia (>6.4 mmol/l) with clearly reduced ß cell activity and insulin concentrations was demonstrated in 13% of heavy drinkers.

Direct experimental evaluations of insulin sensitivity in alcohol consumers have dealt mainly with immediate effects of acute alcohol loading25 29 30 rather than consequences of long term alcohol consumption.31


Our population study confirmed previous reports in that it assessed an increase in concentrations of high density lipoprotein cholesterol and apolipoprotein A I and higher blood pressure among alcohol consumers, extended the metabolic complex associated with drinking by a decline in antithrombin III concentration, and provided first prospective confirmation of these findings. Several mechanisms have been proposed that link regular alcohol intake and changes in vascular risk attributes. Enhanced insulin sensitivity as postulated for alcohol consumers in our study may rank among these pathways because of the broad overlap of the alcohol associated metabolic complex with components of the “syndrome X.”32 33 This hypothesis could partly explain alcohol associated variations in concentrations of high density lipoprotein cholesterol, apolipoprotein A I, and antithrombin III. With regard to systolic blood pressure, however, changes across alcohol categories (+3.6 mm Hg) were contrary to those expected (−1.1 mm Hg), probably because of an alcohol induced activation of the autonomous nervous system or enhancement of renal tubular sodium reabsorption.25


The insulin assay used in the current study cross reacts with pro-insulin and pro-insulin split products. In a subsample of the Bruneck cohort (n = 200), however, insulin concentration was additionally measured with a specific assay (cross reactivity <0.2%). Both assessments were strongly correlated (r = 0.85; mean difference −8.4%; −0.8 to −15.8%) and showed an analogous inverse association with regular alcohol intake (“specific” insulin 11.0 mU/l (0 g/day) to 6.6 mU/l (>/=100 g/day); P<0.001). Likewise, Haffner and coworkers reported that both types of tests perform equally well for the purpose of analysing the associations between vascular risk factors and insulin.34

The quality of alcohol research primarily relies on the accuracy of assessing alcohol consumption. In our survey, the non-differential response error which is believed to be the major source of incorrect ascertainment of alcohol consumption35 was dealt with by the application of diet records in a representative subsample. Prospective assessment of daily food records over an extended period is the best possible method to minimise error in recall and overcome fluctuation in drinking behaviour and inadequate self estimation of alcohol intake.11 36 Analysis of these data further substantiated results derived from point estimates of alcohol intake. The second important type of error, the deliberate denial of alcohol use or selective non-responding by heavy drinkers, may also be regarded as low because of a participation rate over 93% and the fairly high boundaries of socially accepted drinking in the survey area.

Alcohol is an important constituent of the European and American diet. Our survey identified regular alcohol intake as a further determinant of insulin sensitivity in a general healthy population, even though its explanatory contribution to insulin variability was only half that of obesity. Light to moderate drinking emerged as safe regarding its effects on insulin. Conclusions on severe alcohol consumption require particular caution given the associated social and health risks, potential impairment of ß cell function, and occurrence of diabetes due to advanced pancreatitis induced by alcohol.


  • Funding Pustertaler Verein zur Pravention der Herz und Hirngefaßerkrankungen, Sanitatseinheit Ost, Assessorat fur Gesundheit, province of Bozen, Italy.

  • Conflict of interest None.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
  32. 32.
  33. 33.
  34. 34.
  35. 35.
  36. 36.