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

SURVEY AREA AND STUDY SUBJECTS

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.

CLINICAL EVALUATION AND LABORATORY METHODS

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.

STATISTICAL ANALYSIS

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.

REGULAR ALCOHOL CONSUMPTION AND INSULINAEMIA

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

ASSOCIATION OF ALCOHOL CONSUMPTION WITH MULTIPLE VASCULAR RISK FACTORS

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

LIMITATIONS AND SELECTED METHODOLOGICAL PROBLEMS IN QUANTITIVE ALCOHOL RESEARCH

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.