The Effect of Alcohol Consumption on Insulin Sensitivity and Glycemic Status: A Systematic Review and Meta-analysis of Intervention Studies Free

Moderate alcohol consumption, compared with abstaining and heavy drinking, is related to a reduced risk of type 2 diabetes (1,2). Although the risk is reduced with moderate alcohol consumption in both men and women, the association may differ for men and women. In a meta-analysis, consumption of 24 g alcohol/day reduced the risk of type 2 diabetes by 40% among women, whereas consumption of 22 g alcohol/day reduced the risk by 13% among men (1).

The association of alcohol consumption with type 2 diabetes may be explained by increased insulin sensitivity, anti-inflammatory effects, or effects of adiponectin (3). Several intervention studies have examined the effect of moderate alcohol consumption on these potential underlying pathways. A meta-analysis of intervention studies by Brien et al. (4) showed that alcohol consumption significantly increased adiponectin levels but did not affect inflammatory factors. Unfortunately, the effect of alcohol consumption on insulin sensitivity has not been summarized quantitatively. A review of cross-sectional studies by Hulthe and Fagerberg (5) suggested a positive association between moderate alcohol consumption and insulin sensitivity, although the three intervention studies included in their review did not show an effect (6–8). Several other intervention studies also reported inconsistent results (9,10). Consequently, consensus is lacking about the effect of moderate alcohol consumption on insulin sensitivity. Therefore, we aimed to conduct a systematic review and meta-analysis of intervention studies investigating the effect of alcohol consumption on insulin sensitivity and other relevant glycemic measures.

This study was performed according to the PRISMA (Preferred Reporting Items for Systematic Reviews) Statement guidelines for the reporting of systematic reviews and meta-analysis of intervention studies. The PRISMA checklist and the protocol for this study are provided in the Supplementary Data.

A literature search was conducted in PubMed MEDLINE and Embase for relevant intervention studies published up to August 2014. A prespecified search string including search terms on alcohol, consumption, and glycemic measures was used for PubMed and Embase (Supplementary Data). References and related citations of articles were screened to identify other relevant articles. The exposure of interest was (moderate) alcohol consumption and the primary outcome measure, insulin sensitivity. All estimates of insulin sensitivity were included, which were indices from direct measures (e.g., hyperinsulinemic-euglycemic glucose clamp [HEGC]) and indirect measures of insulin sensitivity (e.g., the frequently sampled intravenous glucose tolerance test [FSIVGTT] and oral glucose tolerance test [OGTT]). HOMA of insulin resistance (HOMA-IR) was also included, which is based on fasting insulin and glucose levels and, therefore, primarily reflects hepatic insulin resistance (11). Other relevant outcome measures taken into account were fasting insulin, fasting glucose, and hemoglobin A_1c (HbA_1c). HbA_1c reflects average plasma glucose levels over the past 8–12 weeks and is therefore used as a measure of glycemic status (12).

Relevant studies were selected by two researchers (A.L.J.H., J.W.J.B.) during a multiphase process on the basis of the following inclusion criteria: trials with an alcohol intervention, relevant outcome measures as previously described, intervention period of at least 2 weeks, and written in English or Dutch. We excluded studies of individuals with (a history of) alcoholism or heavy drinkers (individuals consuming ≥60 g alcohol for at least 1 day per week) and animal studies. No publication date or status restrictions were imposed. In the first phase, titles of all retrieved studies were screened to select articles with a relevant subject; the abstracts of these articles were judged on relevance in the next phase. If judged relevant, the full text was studied in the third phase to determine whether the article was eligible for inclusion. When discrepancies occurred about the inclusion of a particular article, a third author (K.J.M. or I.C.S.) was consulted.

From the included studies, sample size, participant characteristics, inclusion and exclusion criteria, study design, duration of intervention, and specific outcome measures were extracted on a prespecified form. Detailed information about the alcohol intervention (e.g., dosage, type, frequency, duration) was described. If a study did not report the grams of alcohol per unit, this was calculated based on the amount in milliliters given to the subjects and the alcohol volume of the beverage [g alcohol = (mL × %v/v) × 0.8, where %v/v is the percentage of alcohol volume per total volume]. Authors of included articles were contacted if further information was required (13–15).

To assess the quality of the studies, aspects such as randomization procedures, compliance with the intervention, and dropout rates were extracted. Randomization and the inclusion of an alcohol-free control group were regarded as the most important criteria to decide whether a study had sufficient quality. If these criteria were not met, the studies were excluded from further meta-analyses. Because randomization of crossover studies may be less important than randomization of parallel studies, we also conducted a sensitivity analysis including nonrandomized crossover studies. Because blinding of participants to the alcohol intervention is of uncertain effectiveness, this criterion was not regarded as essential for inclusion. To assess the quality of the included studies, the 5-point Jadad scale was used (16).

The mean and SD of the outcome variables at the end of the alcohol intervention period and control period were extracted from the articles. If SEs were reported, we used the equation SD = SE × square root of the number of subjects. The mean effects of the various studies measuring the insulin sensitivity index (ISI), HOMA-IR, insulin, glucose, or HbA_1c were pooled in a meta-analysis and shown in a forest plot. To combine the studies measuring ISI and HOMA-IR in one meta-analysis, the inverted HOMA-IR (1/HOMA-IR) was calculated using the delta method.

Heterogeneity between studies was tested using χ² and I² statistics. If χ² and I² showed no evidence for heterogeneity (I² < 30%) (17), analyses were conducted using the inverse variance fixed-effects model for pooling the studies. Otherwise, the DerSimonian and Laird random-effects model was used. The mean outcomes for insulin, glucose, and insulin sensitivity were assessed using different methods and needed to be standardized. Therefore, Cohen d was used to calculate the standardized mean difference (SMD), which is the mean difference between the intervention and control group divided by the pooled SD.

In sensitivity analyses, the effect of moderate alcohol consumption on the reported outcomes was determined by excluding studies with high alcohol dosages (>40 g/day). Furthermore, if more than one intervention arm was tested in a study, we combined the outcomes (17). Additionally, analyses were performed excluding studies potentially causing heterogeneity to determine their effect on the results.

In a meta-regression, the influences of alcohol dosage and duration of the intervention on the results were tested. The influence of type of alcoholic beverage was not assessed due to too few studies to stratify by alcoholic beverage. Because only two studies used the gold standard HEGC to estimate insulin sensitivity (11), we tested with a meta-regression whether the effect of alcohol on insulin sensitivity differed between these studies.

Because the association of alcohol consumption with type 2 diabetes differs for men and women, we conducted sex-stratified analyses. Effect modification by sex was tested in a meta-regression for insulin sensitivity.

Potential publication bias was examined by visual inspection of the funnel plot and by the Egger and Begg statistical tests. In case evidence of publication bias was found, we used the trim and fill method by Duval and Tweedie (18) to calculate a pooled SMD based on filled data to adjust for publication bias. The level of significance was set at P < 0.05. Analyses were performed with the STATA meta-procedure (Stata 10.0).

In total, 4,991 titles were found through the database searches and 24 through additional methods (Supplementary Fig. 1). After screening of titles and abstracts, 46 articles remained eligible and the full text was assessed. Finally, 22 articles met criteria for inclusion in the qualitative synthesis.

Descriptive data of the included studies are summarized in Table 1. Of the 22 studies, 15 used a crossover design and 7 a parallel design. The intervention duration of the studies ranged from 2 to 12 weeks, with an average duration of 5.6 weeks for ISI, 4.2 weeks for HOMA-IR, 7.2 weeks for insulin, 5.9 weeks for glucose, and 4.3 weeks for HbA_1c. Two studies did not use an alcohol-free control group (14,19). The dosage of alcohol varied from 10 to 70 g/day of which one study used >40 g/day (20). ISI was measured by six studies, of which two used the gold standard HEGC (10,21) and four used indirect measures of insulin sensitivity (based on OGTT, FSIVGTT, or fasting levels) (8,9,22,23). HOMA-IR was measured by four studies (15,24–26). Seven studies were performed by the same institute (10,21–26), but they were treated as independent because they included different subjects.

The results of the quality assessment are shown in Supplementary Table 1. Of the 22 studies included in the qualitative synthesis, 4 did not report the measurement of compliance to the intervention (9,14,20,27). Blinding of the researcher was not reported or not conducted in any of the studies. Dropout rates were described in 18 studies. The studies scored between 1 and 3 points on the Jadad scale (range 0–5). Of the 22 studies, 2 were excluded from the meta-analysis because they did not include an alcohol-free control group (14,19), and 4 were excluded because they did not have a randomized design (13,28–30). Because only two studies included subjects with type 2 diabetes, these studies were excluded as well (31,32). One study included both healthy and type 2 diabetic subjects, and from this study, only data from healthy subjects were included (15). Overall, 14 studies were included in the meta-analysis (Table 1 and Supplementary Table 1).

The number of included studies in the analysis was 7 for ISI, 5 for HOMA-IR, 9 for insulin, 10 for glucose, and 3 for HbA_1c. The forest plots on insulin sensitivity and glycemic status are shown in Figs. 1–3.

Pooled analysis showed no difference in ISI after a period of alcohol consumption compared with no alcohol consumption (SMD 0.06 [−0.13 to 0.26], P = 0.53, test for heterogeneity P = 0.76, I² = 0%). For HOMA-IR, both the χ² (P < 0.01) and I² (97%) statistics demonstrated heterogeneity. In a random-effects model, the pooled SMD was 0.35 [−0.90 to 1.59], indicating no effect of alcohol consumption on HOMA-IR (P = 0.59). Similar results were observed when studies measuring ISI and HOMA-IR were combined (SMD −0.12 [−0.61 to 0.39], P = 0.65). A random-effects model was used because heterogeneity was present (P < 0.01, I² = 91%). The funnel plot indicated that the results of the intervention arms (i.e., red wine, gin) of Chiva-Blanch et al. (15) were largely responsible for this heterogeneity. Exclusion of this study resulted in an SMD of 0.08 (−0.09 to 0.24, P = 0.35), with no evidence of heterogeneity (P = 0.90, I² = 0%). Sex-stratified analysis showed different effects in men and women (P_sex = 0.018) (Fig. 1). Alcohol consumption tended to increase insulin sensitivity in women (SMD 0.16 [−0.04 to 0.37], P = 0.12) but not in men (SMD −0.30 [−1.23 to 0.64], P = 0.54). In men, heterogeneity was present (P < 0.01, I² = 95%), and exclusion of the study by Chiva-Blanch et al. resulted in a pooled SMD of −0.07 (−0.34 to 0.20, P = 0.61). However, after exclusion of Chiva-Blanch et al., the pooled SMDs in men and women were no longer significantly different (P = 0.18).

Fasting insulin concentrations were lower after alcohol consumption compared with abstinence, as shown by a pooled SMD of −0.19 (−0.35 to −0.02, P = 0.03) and the test for heterogeneity (P = 0.92, I² = 0%). Sex-stratified analysis showed that alcohol consumption decreased insulin concentrations in women (SMD −0.23 [−0.41 to −0.04], P = 0.02). Only two studies measured insulin concentrations in men, showing a decrease in insulin levels (SMD −0.13 [−0.62 to 0.36], P = 0.59) (Fig. 2A).

For fasting glucose concentrations, the pooled SMD was 0.07 [−0.11 to 0.24], indicating no effect of alcohol consumption on glucose concentration among individuals without diabetes (P = 0.45, P_{heterogeneity} = 0.94, I² = 0%). Similar results were observed when men and women were analyzed separately (Fig. 2B). In women, the SMD was 0.01 (−0.20 to 0.21, P = 0.94); in men, the SMD was 0.14 (−0.24 to 0.53, P = 0.48).

For HbA_1c, a random-effects model was used because the I² statistic indicated evidence for some heterogeneity (I² = 30%). The pooled SMD was −0.62 (−1.01 to −0.23), showing lower HbA_1c concentrations after alcohol consumption compared with no alcohol consumption (P < 0.01) (Fig. 3).

Only the study by Contaldo et al. (20) used a high alcohol dosage (70 g/day) and measured insulin and glucose concentrations. Exclusion of this study from the meta-analysis resulted in generally similar results for insulin (SMD −0.18 [−0.36 to −0.01]) and glucose (SMD 0.06 [−0.12 to 0.23]).

Combining the two intervention arms of the studies by Davies et al. (9) with 15 and 30 g alcohol/day and those of Queipo-Ortuño et al. (33) with red wine and gin resulted in generally similar outcomes. The pooled SMD for insulin sensitivity (ISI and HOMA-IR) was 0.06 (−0.11 to 0.24) overall and 0.15 (−0.08 to 0.38) in women. For insulin, SMD was −0.18 (−0.38 to −0.01) overall and −0.22 (−0.43 to −0.02) in women. Including the nonrandomized crossover study by Cordain et al. (13) resulted in generally similar results for insulin (SMD −0.17 [−0.33 to 0.00]) and glucose (SMD 0.08 [−0.09 to 0.25]).

The meta-regression showed no influence of duration (all P_trend > 0.60) and/or alcohol dosage (all P_trend > 0.67) on the pooled SMD of ISI and HOMA-IR and of insulin and glucose. Additionally, the meta-regression showed no differences between results from the studies using the HEGC to measured insulin sensitivity and the other studies (SMD −0.03 for HEGC studies vs. 0.09 for other studies, P = 0.64).

Results of the Egger and Begg tests showed publication bias for the outcomes of ISI, ISI and HOMA-IR, and glucose (Supplementary Table 2). Visual inspection of the funnel plots showed some asymmetry, which was due to missing results in favor of alcohol treatment from smaller studies (Supplementary Fig. 2). For ISI and HOMA-IR, we calculated an adjusted pooled SMD by using the trim and fill approach by Duval and Tweedie (18). This resulted in four extra study estimates (linear method used) and an adjusted pooled SMD of 0.17 (0.02–0.31, P = 0.03). The trim and fill method shows that without publication bias, the pooled SMD would probably indicate a positive effect of alcohol consumption on insulin sensitivity, whereas the unadjusted SMD did not show an effect (SMD 0.08 [−0.09 to 0.24], P = 0.35). The adjusted results and funnel plot are shown in Supplementary Table 2 and Supplementary Fig. 2.

This meta-analysis shows that moderate alcohol consumption did not affect estimates of insulin sensitivity or fasting glucose levels, but it decreased fasting insulin concentrations and HbA_1c. Sex-stratified analysis suggested that moderate alcohol consumption may improve insulin sensitivity and decrease fasting insulin concentrations in women but not in men. The meta-regression suggested no influence of dosage and duration on the results. However, the number of studies may have been too low to detect influences by dosage and duration.

The primary finding that alcohol consumption does not influence insulin sensitivity concords with the intervention studies included in the review of Hulthe and Fagerberg (5). This is in contrast with observational studies suggesting a significant association between moderate alcohol consumption and improved insulin sensitivity (34,35). However, the results of these studies might be biased through residual confounding because of their observational nature. Moreover, in contrast to intervention studies, observational studies are not designed to detect a causal relationship. On the other hand, we cannot exclude the possibility that the intervention studies in this review had an insufficient sample size or too short a duration to detect an effect of alcohol consumption on insulin sensitivity (10,21,23,24).

We found lower fasting insulin levels after alcohol consumption. This finding agrees with the inverse relation between alcohol consumption and insulin levels observed in observational studies (36–39). However, in the DESIR (Data from an Epidemiological Study on the Insulin Resistance syndrome) cohort, a longitudinal study, no relation between the average or a change in alcohol consumption and fasting insulin levels was found, but this may be a result of the inclusion of subjects with type 2 diabetes (40). Fasting insulin level is a surrogate marker of insulin sensitivity in healthy subjects, with lower insulin levels indicating higher insulin sensitivity (11,41). Conversely, low insulin levels are a common phenomenon in subjects with type 2 diabetes due to impaired insulin secretion by β-cells. Because we excluded studies in subjects with type 2 diabetes, the results of lower fasting insulin levels may indicate higher insulin sensitivity. Additionally, we observed no change in glucose levels by alcohol consumption, and lower insulin levels coinciding with unchanged glucose levels suggest an improved insulin sensitivity.

The current meta-analysis suggests that men and women might respond differently to a period of alcohol consumption with regard to insulin sensitivity. Subgroup analysis showed that the effect of alcohol consumption on insulin sensitivity was only present among women, but the pooled effects in men and women were not significantly different. These results generally concord with observational studies showing a larger risk reduction of moderate alcohol consumption on risk of type 2 diabetes in women than in men (40% vs. 13%) (1) and with the study by Beulens et al. (42). The studies included in the review by Hulthe and Fagerberg (5), which were mainly cross-sectional, did not find sex differences in alcohol effects.

We observed lower levels of HbA_1c in subjects consuming moderate amounts of alcohol compared with abstainers. This has also been shown in several observational studies (39,43,44). Alcohol may decrease HbA_1c by suppressing the acute rise in blood glucose after a meal and increasing the early insulin response (45). This would result in lower glucose concentrations over time and, thus, lower HbA_1c concentrations. Unfortunately, the underlying mechanism of glycemic control by alcohol is not clearly understood.

A major strength of this meta-analysis is the inclusion of studies with a randomized controlled design and the inclusion of several complementary end points, providing a comprehensive overview of the evidence on this topic. There are also limitations that warrant consideration. As in any meta-analysis, the strength of the current study is largely determined by the quality and number of the included studies. The results of the quality assessment show that the larger part of the included studies did not report or did not take into account some important aspects, such as blinding. Nevertheless, randomization and the inclusion of an alcohol-free control group were the most important quality factors for this review, and only six studies did not satisfy those criteria. Compliance was measured in most studies (17 of 22) but was only reported in 13. However, of these 13 studies, 11 reported good or excellent compliance, suggesting that low compliance did not influence the results of the studies. Second, the analysis of several different outcomes resulted in inclusion of a small number of studies for certain end points, such as HbA_1c. Third, only two studies used the gold standard HEGC to estimate insulin sensitivity (11). Because this may lead to inconsistency in the results, we standardized the results of the different studies using Cohen d. However, the results from the studies using HEGC were similar to the other intervention studies, and no significant heterogeneity was present except for the combined meta-analysis of ISI and HOMA-IR. This was due to the study of Chiva-Blanch et al. (15), who reported a relatively small variation in HOMA-IR, causing a relatively large SMD. Exclusion of this study removed heterogeneity without changing the effect. Fourth, because most studies used a crossover design, a carryover effect might have influenced the outcomes. Another limitation was the short duration and small sample sizes of the included studies. The average duration of 5.4 weeks may not have been long enough to show detectable differences in insulin sensitivity or glucose status. In addition, effects may change after longer-term intake of alcohol. Therefore, the short-term nature of the included studies does not allow us to draw conclusions on longer-term alcohol consumption.

It is important to note evidence for publication bias for certain outcomes in the current study. The publication bias unexpectedly suggests that smaller studies with positive results are missing. After adjustment for publication bias using the trim and fill method, even a significant increase in insulin sensitivity by alcohol consumption was shown. However, statistical tests for publication bias may yield biased results with small numbers of studies and are prone to heterogeneity (17).

Finally, the results of this research may not be generalizable to all healthy subjects because the selected studies included mainly light to moderate alcohol consumers. Therefore, the period of abstaining from alcohol might also be seen as an intervention, and subjects might have responded differently than alcohol abstainers.

To draw implications from the current research, the findings need to be placed in a clinical context. In this meta-analysis, we observed that alcohol consumption decreased fasting insulin levels by 0.19, which translates to an ∼11% decrease in insulin (−20 pmol/L) in people with impaired glucose tolerance, as calculated from data of the Diabetes Prevention Program study (46), and a 13% decrease in insulin (−5.2 pmol/L) in normoglycemic people, as calculated from data of the Multiethnic Study of Atherosclerosis (MESA) (47). For comparison, metformin treatment results in a 14% decrease in fasting insulin levels and a 40% lower risk of diabetes versus a control group (48). An 11% reduction of fasting insulin levels after alcohol consumption would result in an ∼30% reduced risk of diabetes, which is in line with the 40% risk reduction observed among women.

The reduced HbA_1c concentration found in the current study by alcohol consumption (SMD −0.62) is equal to a 5% reduction in HbA_1c concentration in both the MESA and the Diabetes Prevention Program studies (from 5.4% [36 mmol/mol] to 5.1% [33 mmol/mol] and from 5.9% [41 mmol/mol] to 5.6% [38 mmol/mol], respectively) (46,49). The Diabetes Prevention Program study showed that 4 years of metformin medication and a lifestyle intervention both resulted in a reduction in HbA_1c of ∼3% (50). Because type 2 diabetes is characterized by hyperglycemia, HbA_1c could be seen as a surrogate end point of the disease rather than an intermediate factor in the pathway toward type 2 diabetes. The World Health Organization indeed suggests that a level >6.5% (48 mmol/mol) be used as a cutoff point for diagnosing diabetes (51). In this respect, the current results for HbA_1c match with the reduced risk of type 2 diabetes with moderate alcohol consumption. Results of alcohol intake on HbA_1c should be carefully interpreted because we included only three intervention studies in the analysis. However, the results suggest that drinking a moderate amount of alcohol is not harmful with regard to insulin sensitivity and glycemic status in healthy adults without type 2 diabetes.

This systematic review and meta-analysis showed that moderate alcohol consumption decreased fasting insulin and HbA_1c concentrations among nondiabetic subjects. Alcohol consumption might improve insulin sensitivity among women but did not do so overall. These results may partly explain the lower risk of type 2 diabetes with moderate alcohol consumption found in observational studies. However, more intervention studies with a longer intervention period are necessary to confirm the results.

Funding. I.C.S. and H.F.J.H. were supported by both the Dutch Ministry of Economic Affairs, Agriculture and Innovation and the Dutch Foundation for Alcohol Research, representing Dutch producers of and traders in beer, wine, and spirits and The Netherlands Organization for Applied Scientific Research. Their joint aim is to independently study the health effects of moderate alcohol consumption.

The funding sources had no role in conducting the study, in analyzing or interpreting the study results, or in the decision to submit the manuscript for publication.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. I.C.S. contributed to the data extraction and analysis, data interpretation, and writing of the manuscript. A.L.J.H. contributed to the study design, article search and data extraction, data analysis, and data interpretation. H.F.J.H. contributed to the data interpretation. K.J.M. contributed to the study design and data interpretation. J.W.J.B. contributed to the study design, search and data extraction, data analysis, data interpretation, and writing of the manuscript.

Prior Presentation. Parts of this study were presented in abstract form at the Epidemiology and Prevention/Nutrition, Physical Activity and Metabolism 2014 Scientific Sessions of the American Heart Association, San Francisco, CA, 18–21 March 2014, and were presented orally at the 7th European Beer and Health Symposium, Brussels, Belgium, 30 September 2014.