TheEffect ofAlcoholConsumptionon InsulinSensitivity andGlycemicStatus: A Systematic Review andMeta-analysis of InterventionStudiesDiabetes Care 2015;38:723–732 | DOI: 10.2337/dc14-1556

OBJECTIVE

Moderate alcohol consumption is associated with a reduced risk of type 2 di-abetes. This reduced risk might be explained by improved insulin sensitivity orimproved glycemic status, but results of intervention studies on this relation areinconsistent. The purpose of this study was to conduct a systematic review andmeta-analysis of intervention studies investigating the effect of alcohol consump-tion on insulin sensitivity and glycemic status.

RESEARCH DESIGN AND METHODS

PubMed and Embase were searched up to August 2014. Intervention studies onthe effect of alcohol consumption on biological markers of insulin sensitivity orglycemic status of at least 2 weeks’ duration were included. Investigatorsextracted data on study characteristics, outcome measures, and methodologicalquality.

RESULTS

Fourteen intervention studies were included in ameta-analysis of six glycemic endpoints. Alcohol consumption did not influence estimated insulin sensitivity (stan-dardized mean difference [SMD] 0.08 [20.09 to 0.24]) or fasting glucose (SMD0.07 [20.11 to 0.24]) but reduced HbA1c (SMD 20.62 [21.01 to 20.23]) andfasting insulin concentrations (SMD 20.19 [20.35 to 20.02]) compared withthe control condition. Alcohol consumption amongwomen reduced fasting insulin(SMD 20.23 [20.41 to 20.04]) and tended to improve insulin sensitivity (SMD0.16 [20.04 to 0.37]) but not among men. Results were similar after excludingstudies with high alcohol dosages (>40 g/day) and were not influenced by dosageand duration of the intervention.

CONCLUSIONS

Although the studies had small sample sizes and were of short duration, thecurrent evidence suggests that moderate alcohol consumption may decreasefasting insulin and HbA1c concentrations among nondiabetic subjects. Alcoholconsumption might improve insulin sensitivity among women but did not do sooverall.

1The Netherlands Organization for AppliedScientific Research, Zeist, the Netherlands2Division of Human Nutrition, WageningenUniversity, Wageningen, the Netherlands3Julius Center for Health Sciences and PrimaryCare, UniversityMedical Center Utrecht, Utrecht,the Netherlands4Harvard Medical School, Beth Israel DeaconessMedical Center, Boston, MA

Corresponding author: Ilse C. Schrieks, ilse.schrieks@tno.nl.

Received 25 June 2014 and accepted 30November 2014.

This article contains Supplementary Data onlineat http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc14-1556/-/DC1.

© 2015 by the American Diabetes Association.Readers may use this article as long as the workis properly cited, the use is educational and notfor profit, and the work is not altered.

Ilse C. Schrieks,1,2 Annelijn L.J. Heil,3

Henk F.J. Hendriks,1 Kenneth J. Mukamal,4

and Joline W.J. Beulens3

Diabetes Care Volume 38, April 2015 723

META

-ANALYSIS

Moderate alcohol consumption, com-paredwith abstaining and heavy drinking,is related to a reduced risk of type 2 di-abetes (1,2). Although the risk is reducedwith moderate alcohol consumption inboth men and women, the associationmay differ for men and women. In ameta-analysis, consumption of 24 g alco-hol/day reduced the risk of type 2 diabe-tes by 40% among women, whereasconsumption of 22 g alcohol/day reducedthe risk by 13% among men (1).The association of alcohol consump-

tion with type 2 diabetes may be ex-plained by increased insulin sensitivity,anti-inflammatory effects, or effects ofadiponectin (3). Several interventionstudies have examined the effect ofmod-erate alcohol consumption on thesepotential underlying pathways. A meta-analysis of intervention studies by Brienet al. (4) showed that alcohol consump-tion significantly increased adiponectinlevels but did not affect inflammatory fac-tors. Unfortunately, the effect of alcoholconsumptionon insulin sensitivity has notbeen summarized quantitatively. A re-view of cross-sectional studies by Hultheand Fagerberg (5) suggested a positiveassociation between moderate alcoholconsumption and insulin sensitivity, al-though the three intervention studies in-cluded in their review did not show aneffect (6–8). Several other interventionstudies also reported inconsistent results(9,10). Consequently, consensus is lackingabout the effect ofmoderate alcohol con-sumption on insulin sensitivity. There-fore, we aimed to conduct a systematicreview and meta-analysis of interventionstudies investigating the effect of alcoholconsumption on insulin sensitivity andother relevant glycemic measures.

RESEARCH DESIGN AND METHODS

This study was performed according tothe PRISMA (Preferred Reporting Itemsfor Systematic Reviews) Statementguidelines for the reporting of system-atic reviews and meta-analysis of inter-vention studies. The PRISMA checklistand the protocol for this study are pro-vided in the Supplementary Data.

Data Sources and SearchesA literature search was conducted inPubMed MEDLINE and Embase for rele-vant intervention studies published up toAugust 2014. A prespecified search stringincluding search terms on alcohol, con-sumption, and glycemic measures was

used for PubMed and Embase (Supple-mentary Data). References and relatedcitations of articles were screened toidentify other relevant articles. The expo-sure of interest was (moderate) alcoholconsumption and the primary outcomemeasure, insulin sensitivity. All estimatesof insulin sensitivitywere included, whichwere indices from direct measures (e.g.,hyperinsulinemic-euglycemic glucoseclamp [HEGC]) and indirect measures ofinsulin sensitivity (e.g., the frequentlysampled intravenous glucose tolerancetest [FSIVGTT] and oral glucose tolerancetest [OGTT]). HOMA of insulin resistance(HOMA-IR) was also included, which isbased on fasting insulin and glucose lev-els and, therefore, primarily reflects he-patic insulin resistance (11). Otherrelevant outcome measures taken intoaccount were fasting insulin, fasting glu-cose, and hemoglobin A1c (HbA1c). HbA1creflects average plasma glucose levelsover the past 8–12weeks and is thereforeused as ameasure of glycemic status (12).

Study SelectionRelevant studies were selected by tworesearchers (A.L.J.H., J.W.J.B.) during amultiphase process on the basis of thefollowing inclusion criteria: trials with analcohol intervention, relevant outcomemeasures as previously described, inter-vention period of at least 2 weeks, andwritten in English or Dutch.We excludedstudies of individuals with (a history of)alcoholism or heavy drinkers (individu-als consuming$60 g alcohol for at least1 day per week) and animal studies. Nopublication date or status restrictionswere imposed. In the first phase, titlesof all retrieved studies were screened toselect articles with a relevant subject;the abstracts of these articles werejudged on relevance in the next phase.If judged relevant, the full text was stud-ied in the third phase to determinewhether the article was eligible for in-clusion. When discrepancies occurredabout the inclusion of a particulararticle, a third author (K.J.M. or I.C.S.)was consulted.

Data Extraction and QualityAssessmentFrom the included studies, sample size,participant characteristics, inclusion andexclusion criteria, study design, durationof intervention, and specific outcomemeasures were extracted on a prespeci-fied form. Detailed information about

the alcohol intervention (e.g., dosage,type, frequency, duration) was de-scribed. If a studydid not report the gramsof alcohol per unit, this was calculatedbased on the amount in milliliters givento the subjects and the alcohol volume ofthe beverage [g alcohol = (mL3 %v/v)30.8, where %v/v is the percentage of alco-hol volume per total volume]. Authors ofincluded articles were contacted if furtherinformation was required (13–15).

To assess the quality of the studies,aspects such as randomization proce-dures, compliance with the interven-tion, and dropout rates were extracted.Randomization and the inclusion of analcohol-free control group were re-garded as the most important criteria todecide whether a study had sufficientquality. If these criteria were not met,the studies were excluded from furthermeta-analyses. Because randomizationof crossover studiesmaybe less importantthan randomization of parallel studies, wealso conducted a sensitivity analysisincluding nonrandomized crossover stud-ies. Because blinding of participants tothe alcohol intervention is of uncertaineffectiveness, this criterion was not re-garded as essential for inclusion. To assessthe quality of the included studies, the5-point Jadad scale was used (16).

Data Synthesis and AnalysisThe mean and SD of the outcome vari-ables at the end of the alcohol interven-tion period and control period wereextracted from the articles. If SEs werereported, we used the equation SD = SE3square root of the number of subjects.The mean effects of the various studiesmeasuring the insulin sensitivity index(ISI), HOMA-IR, insulin, glucose, or HbA1cwere pooled in a meta-analysis andshown in a forest plot. To combine thestudies measuring ISI and HOMA-IR inone meta-analysis, the inverted HOMA-IR (1/HOMA-IR) was calculated using thedelta method.

Heterogeneity between studies wastested using x2 and I2 statistics. If x2

and I2 showed no evidence for hetero-geneity (I2 , 30%) (17), analyses wereconducted using the inverse variancefixed-effects model for pooling the stud-ies. Otherwise, the DerSimonian andLaird random-effects model was used.The mean outcomes for insulin, glucose,and insulin sensitivity were assessed us-ing different methods and needed to be

724 Alcohol Consumption and Insulin Sensitivity Diabetes Care Volume 38, April 2015

standardized. Therefore, Cohen d wasused to calculate the standardized meandifference (SMD), which is the mean dif-ference between the intervention andcontrol group divided by the pooled SD.In sensitivity analyses, the effect of

moderate alcohol consumption on thereported outcomes was determined byexcluding studies with high alcohol dos-ages (.40 g/day). Furthermore, if morethan one intervention arm was testedin a study, we combined the outcomes(17). Additionally, analyses were per-formed excluding studies potentiallycausing heterogeneity to determinetheir effect on the results.In a meta-regression, the influences of

alcohol dosage and duration of the inter-vention on the results were tested. Theinfluence of type of alcoholic beveragewas not assessed due to too few studiesto stratify by alcoholic beverage. Becauseonly two studies used the gold standardHEGC to estimate insulin sensitivity (11),we testedwith ameta-regressionwhetherthe effect of alcohol on insulin sensitivitydiffered between these studies.Because the association of alcohol

consumption with type 2 diabetes dif-fers for men and women, we conductedsex-stratified analyses. Effect modifi-cation by sex was tested in a meta-regression for insulin sensitivity.Potential publication bias was exam-

ined by visual inspection of the funnelplot and by the Egger and Begg statisti-cal tests. In case evidence of publicationbias was found, we used the trim and fillmethod by Duval and Tweedie (18) tocalculate a pooled SMD based on filleddata to adjust for publication bias. Thelevel of significance was set at P, 0.05.Analyses were performed with theSTATA meta-procedure (Stata 10.0).

RESULTS

In total, 4,991 titles were found throughthe database searches and 24 throughadditional methods (Supplementary Fig.1). After screening of titles and ab-stracts, 46 articles remained eligibleand the full text was assessed. Finally,22 articles met criteria for inclusion inthe qualitative synthesis.

Study CharacteristicsDescriptive data of the included studiesare summarized in Table 1. Of the 22studies, 15 used a crossover designand 7 a parallel design. The interventionduration of the studies ranged from 2 to

12 weeks, with an average duration of5.6 weeks for ISI, 4.2 weeks for HOMA-IR, 7.2 weeks for insulin, 5.9 weeks forglucose, and 4.3 weeks for HbA1c. Twostudies did not use an alcohol-free con-trol group (14,19). The dosage of alcoholvaried from 10 to 70 g/day of which onestudy used.40 g/day (20). ISI was mea-sured by six studies, of which two usedthe gold standard HEGC (10,21) and fourused indirect measures of insulin sensi-tivity (based on OGTT, FSIVGTT, or fast-ing levels) (8,9,22,23). HOMA-IR wasmeasured by four studies (15,24–26).Seven studies were performed by thesame institute (10,21–26), but theywere treated as independent becausethey included different subjects.

Quality AssessmentThe results of the quality assessment areshown in Supplementary Table 1. Of the22 studies included in the qualitativesynthesis, 4 did not report the measure-ment of compliance to the intervention(9,14,20,27). Blinding of the researcherwas not reported or not conducted inany of the studies. Dropout rates weredescribed in 18 studies. The studiesscored between 1 and 3 points on theJadad scale (range 0–5). Of the 22 stud-ies, 2 were excluded from the meta-analysis because they did not includean alcohol-free control group (14,19),and 4 were excluded because they didnot have a randomized design (13,28–30). Because only two studies includedsubjects with type 2 diabetes, thesestudies were excluded as well (31,32).One study included both healthy andtype 2 diabetic subjects, and from thisstudy, only data from healthy subjectswere included (15). Overall, 14 studieswere included in the meta-analysis (Ta-ble 1 and Supplementary Table 1).

Meta-analysisThe number of included studies in theanalysis was 7 for ISI, 5 for HOMA-IR, 9for insulin, 10 for glucose, and 3 forHbA1c. The forest plots on insulin sensi-tivity and glycemic status are shown inFigs. 1–3.

Pooled analysis showed no differencein ISI after a period of alcohol consump-tion compared with no alcohol con-sumption (SMD 0.06 [20.13 to 0.26],P = 0.53, test for heterogeneity P =0.76, I2 = 0%). For HOMA-IR, both thex2 (P, 0.01) and I2 (97%) statistics dem-onstrated heterogeneity. In a random-

effects model, the pooled SMD was0.35 [20.90 to 1.59], indicating no ef-fect of alcohol consumption on HOMA-IR (P = 0.59). Similar results wereobserved when studies measuring ISIand HOMA-IR were combined (SMD20.12 [20.61 to 0.39], P = 0.65). Arandom-effects model was used becauseheterogeneity was present (P , 0.01,I2 = 91%). The funnel plot indicated thatthe results of the intervention arms (i.e.,red wine, gin) of Chiva-Blanch et al. (15)were largely responsible for this hetero-geneity. Exclusion of this study resulted inan SMD of 0.08 (20.09 to 0.24, P = 0.35),with no evidence of heterogeneity (P =0.90, I2 = 0%). Sex-stratified analysisshowed different effects in men andwomen (Psex = 0.018) (Fig. 1). Alcohol con-sumption tended to increase insulin sen-sitivity in women (SMD 0.16 [20.04 to0.37], P = 0.12) but not in men (SMD20.30 [21.23 to 0.64], P = 0.54). Inmen, heterogeneity was present (P ,0.01, I2 = 95%), and exclusion of the studyby Chiva-Blanch et al. resulted in a pooledSMD of 20.07 (20.34 to 0.20, P = 0.61).However, after exclusion of Chiva-Blanchet al., the pooled SMDs in men andwomen were no longer significantly dif-ferent (P = 0.18).

Fasting insulin concentrations werelower after alcohol consumption com-pared with abstinence, as shown by apooled SMD of 20.19 (20.35 to 20.02,P = 0.03) and the test for heterogeneity(P = 0.92, I2 = 0%). Sex-stratified analysisshowed that alcohol consumption de-creased insulin concentrations in women(SMD20.23 [20.41 to20.04], P = 0.02).Only two studies measured insulin con-centrations in men, showing a decreasein insulin levels (SMD 20.13 [20.62 to0.36], P = 0.59) (Fig. 2A).

For fasting glucose concentrations,the pooled SMD was 0.07 [20.11 to0.24], indicating no effect of alcoholconsumption on glucose concentrationamong individuals without diabetes (P =0.45, Pheterogeneity = 0.94, I

2 = 0%). Similarresults were observed when men andwomen were analyzed separately (Fig.2B). In women, the SMD was 0.01(20.20 to 0.21, P = 0.94); in men, theSMD was 0.14 (20.24 to 0.53, P = 0.48).

For HbA1c, a random-effects modelwas used because the I2 statistic indi-cated evidence for some heterogeneity(I2 = 30%). The pooled SMD was 20.62(21.01 to20.23), showing lower HbA1c

care.diabetesjournals.org Schrieks and Associates 725

Table

1—Characteristicsofstudiesincludedin

this

system

aticreview

andmeta-analysisontheeffect

ofalcoholco

nsu

mptiononinsu

linse

nsitivity

Studyreference

Design

Participan

tsParticipan

tcharacteristics

Interven

tion

Alcohol

dosage

(g/day)

Interven

tion

period

(weeks)

Outcomemeasure

Inmeta-an

alysis

Bantle20

08(31)

Randomized

crossover

17diabeticmen

and

women

Age

64(45–

82)years

BMI31

.7(21.3–41

.2)

kg/m

2

Abstinenceor

white/red

winedu

ring

dinn

er18

4Insulin,glucose,H

bA1c

No*

Beu

lens20

06(21)

Randomized

crossover

17healthymen

with

waistcircumference

.94

cm

Age

53(9)years

BMI29

.1(4.2)kg/m

2

Insulin

10.7

(5.6)units/L

Red

wineor

dealcoholized

redwinewithdinner

404

ISI(HEG

C)

Yes

Beu

lens20

07(23)

Randomized

crossover

19healthylean

or

overw

eigh

tmen

Lean

:Age

21(2)years

BMI21

.4(2.0)kg/m

2

Insulin

4.7(1.2)units/L

Overw

eight:

Age

28(6)years

BMI30

.1(3.4)kg/m

2

Insulin

11.0

(5.4)units/L

Whiskyor

mineralwater

324

ISI(OGTT),HbA1c

Yes

Beu

lens20

08(22)

Randomized

crossover

20healthylean

or

overw

eigh

tmen

Lean

:Age

19(2)years

BMI20

.1(1.0)kg/m

2

Overw

eight:

Age

21(2)years

BMI31

.3(3.9)kg/m

2

Beeror

alcohol-freebeer

during

dinner

403

ISI(OGTT)

Yes

Bhathen

a19

95(27)

Randomized

crossover

37healthy

premen

opausal

women

Age

30(7)years

BMI24

.4(4.6)kg/m

2Ethano

lmixed

withfruit

juiceor

softdrinkafter

dinn

er

3012

Insulin

Yes

Cesen

a20

11(28)

Parallel

(onearm)

42healthymen

andwomen

Age

46(9)years

BMI25

.1(2.8)kg/m

2Abstinen

ceorredwine

duringdinner

242

Glucose

No†

Chiva-Blanch

2013

(15)

Randomized

crossover

52healthyand

15diabeticmen

Age

60(8)years

BMI29

.6(3.9)kg/m

2Gin,red

wine,

or

dealcoholized

redwine

304

HOMA-IR,insulin,glucose

Yes

Contaldo19

89(20)

Randomized

crossover

8healthymen

BMI25

.4(1.4)kg/m

2Abstinen

ceorredwine

duringdinner

752

Insulin,glucose

Yes

Cordain19

97(13)

Randomized

crossover

14healthymen

Age

32(9)years

Abstinen

ceorwine

286

Insulin,glucose

No†

Cordain20

00(8)

Randomized

crossover

20seden

tary

and

overw

eigh

tpremen

opausal

women

BMI29

.8(2.2)kg/m

2

Insulin

8.6(3.3)units/L

Abstinen

ceorredwine

2010

ISI(FSIVGTT),insulin,glucose

Yes

Davies20

02(9)

Randomized

crossover

51healthy

postmen

opau

sal

women

Age

60(8)years

BMI27

.4(5.7)kg/m

2

Insulin

6.5(5.7)units/L

Alcoholo

risocaloric

beverage

15or30

8ISI(M

FFM),insulin,glucose

Yes

Flechtner-M

ors

2004

(52)

Randomized

parallel

40overw

eightmen

andwomen

Age

48(11)

years

BMI34

.2(6.4)kg/m

2Grapejuiceorwhite

wineduringmeals

1712

Insulin,glucose

Yes

Con

tinuedon

p.72

7

726 Alcohol Consumption and Insulin Sensitivity Diabetes Care Volume 38, April 2015

Table

1—Continued

Studyreference

Design

Participan

tsParticipan

tcharacteristics

Interven

tion

Alcohol

dosage

(g/day)

Interven

tion

period

(weeks)

Outcomemeasure

Inmeta-an

alysis

Joosten

2008(24)

Randomized

crossover

36healthy

postmen

opau

sal

women

Age

57(4)years

BMI25

.4(3.3)kg/m

2

Insulin

37.4

(12.6)

pmol/L

Whitewineorwhite

grap

ejuicedaily

duringdinner

256

HOMA-IR,insulin,glucose,HbA1c

Yes

Joosten

2011and

2014

(25,26

)Randomized

crossover

24healthy

premen

opausal

women

Age

24(4)years

BMI22

.2(1.6)kg/m

2

Insulin

41.7

(16.0)

pmol/L

Beeroralcohol-free

beerduringdinner

263

HOMA-IR,insulin,glucose,HbA1c

Yes

Joosten

2012and

2014

(53,26

)Randomized

crossover

24healthymen

Age

26(3)years

BMI24

(3)kg/m

2Vod

kaandorange

juiceor

orange

juice

during

dinn

er

304

HOMA-IR,insulin,glucose

Yes

Kim

2009

(29)

Parallel

20nondiabetic,

insulin-resistant

men

andwomen

Age

54(7)years

BMI32

(5)kg/m

2Abstinen

ceorvodka

or

redwineduringdinner

308

Steady-stateplasm

aglucose,

glucose

No†

Lavy

1994(14)

Randomized

parallel

20healthymen

dRed

orwhitewine

402

Glucose

No‡

Queipo-Ortu~ no20

12(33)

Randomized

crossover

10healthymen

Age

48(2)years

BMI27

.6(3.2)kg/m

2Gin,red

wine,

or

dealcoholized

redwine

303

Glucose

Yes

Romeo

2008

(30)

Parallel

(onearm)

57healthywomen

andmen

Women

:Age

38(9)years

BMI24

.4(3.5)kg/m

2

Men

:Age

35(6)years

BMI25

.5(2.4)kg/m

2

Abstinen

ceorbeer

duringthemeal

11(w

omen

),22

(men

)

4Glucose

No†

Shai20

07(32)

Randomized

parallel

(multicen

ter)

91diabeticmen

and

women

Age

62(6)years

BMI30

.1(4.6)kg/m

2Wineornonalcoholic

beerduringdinner

1312

Glucose,H

bA1c

No*

Sierksma20

04(10)

Randomized

crossover

23healthymen

Age

52(5)years

BMI26

.7(3.0)kg/m

2

Insulin

8.9(8.8)units/L

Whisky

ortapwater

duringdinner

402.5

ISI(HEG

C)

Yes

Zhen

g20

12(19)

Randomized

parallel

45healthymen

and

women

TFL:

Age

24(2)years

BMI21

.3(1.6)kg/m

2

TCL:

Age

24(1)years

BMI21

.1(2.2)kg/m

2

TFLorTC

L10

4HOMA-IR,insulin,glucose

No‡

Dataaremean(SD)or(range)unlessotherwiseindicated

.MFFM,w

hole-bodyglucose

disposalratenorm

alized

tofat-free

mass;TC

L,trad

itionalChineseliquor;TFL,tea-flavorliquor.Reasonforexclusionfrom

meta-an

alysis:*p

articipan

tswithtype2diabetes,†norandomized

design,‡nocontrolgroup.

care.diabetesjournals.org Schrieks and Associates 727

concentrations after alcohol consump-tion compared with no alcohol con-sumption (P , 0.01) (Fig. 3).

Sensitivity Analyses andMeta-regressionOnly the study by Contaldo et al. (20)used a high alcohol dosage (70 g/day)

and measured insulin and glucose con-centrations. Exclusion of this study fromthe meta-analysis resulted in generallysimilar results for insulin (SMD 20.18[20.36 to 20.01]) and glucose (SMD0.06 [20.12 to 0.23]).

Combining the two intervention armsof the studies by Davies et al. (9) with 15

and 30 g alcohol/day and those ofQueipo-Ortu~no et al. (33) with redwine and gin resulted in generally simi-lar outcomes. The pooled SMD for insu-lin sensitivity (ISI and HOMA-IR) was0.06 (20.11 to 0.24) overall and 0.15(20.08 to 0.38) in women. For insulin,SMD was20.18 (20.38 to20.01) over-all and 20.22 (20.43 to 20.02) inwomen. Including the nonrandomizedcrossover study by Cordain et al. (13)resulted in generally similar results forinsulin (SMD20.17 [20.33 to 0.00]) andglucose (SMD 0.08 [20.09 to 0.25]).

Themeta-regression showed no influ-ence of duration (all Ptrend . 0.60) and/or alcohol dosage (all Ptrend . 0.67) onthe pooled SMD of ISI and HOMA-IR andof insulin and glucose. Additionally, themeta-regression showed no differencesbetween results from the studies usingthe HEGC tomeasured insulin sensitivityand the other studies (SMD 20.03 forHEGC studies vs. 0.09 for other studies,P = 0.64).

Publication BiasResults of the Egger and Begg testsshowed publication bias for the out-comes of ISI, ISI and HOMA-IR, and glu-cose (Supplementary Table 2). Visualinspection of the funnel plots showedsome asymmetry, which was due tomissing results in favor of alcohol treat-ment from smaller studies (Supplemen-tary Fig. 2). For ISI and HOMA-IR, we

Figure 1—Forest plot of meta-analysis of the effect of alcohol consumption on insulin sensitivity.Data are pooled SMDs with 95% CIs and are calculated with exclusion of the results of the twostudy arms of Chiva-Blanch et al. (15) because they induced heterogeneity.

Figure 2—Forest plots of meta-analysis of the effect of alcohol consumption on fasting insulin (A) and fasting glucose (B). Data are pooled SMDswith95% CIs. RW, red wine.

728 Alcohol Consumption and Insulin Sensitivity Diabetes Care Volume 38, April 2015

calculated an adjusted pooled SMD byusing the trim and fill approach by Duvaland Tweedie (18). This resulted in fourextra study estimates (linear methodused) and an adjusted pooled SMD of0.17 (0.02–0.31, P = 0.03). The trim andfill method shows that without publica-tion bias, the pooled SMD would proba-bly indicate a positive effect of alcoholconsumption on insulin sensitivity,whereas the unadjusted SMD did notshow an effect (SMD 0.08 [20.09 to0.24], P = 0.35). The adjusted resultsand funnel plot are shown in Supplemen-tary Table 2 and Supplementary Fig. 2.

CONCLUSIONS

This meta-analysis shows that moderatealcohol consumption did not affect esti-mates of insulin sensitivity or fastingglucose levels, but it decreased fastinginsulin concentrations and HbA1c. Sex-stratified analysis suggested that mod-erate alcohol consumptionmay improveinsulin sensitivity and decrease fastinginsulin concentrations in women butnot in men. The meta-regression sug-gested no influence of dosage and dura-tion on the results. However, thenumber of studies may have been toolow to detect influences by dosage andduration.

Comparison With Other StudiesThe primary finding that alcohol con-sumption does not influence insulin sen-sitivity concords with the intervention

studies included in the review of Hultheand Fagerberg (5). This is in contrastwith observational studies suggesting asignificant association between moder-ate alcohol consumption and improvedinsulin sensitivity (34,35). However,the results of these studies might be bi-ased through residual confounding be-cause of their observational nature.Moreover, in contrast to interventionstudies, observational studies are notdesigned to detect a causal relationship.On the other hand, we cannot excludethe possibility that the interventionstudies in this review had an insufficientsample size or too short a duration todetect an effect of alcohol consumptionon insulin sensitivity (10,21,23,24).

We found lower fasting insulin levelsafter alcohol consumption. This findingagrees with the inverse relation be-tween alcohol consumption and insulinlevels observed in observational studies(36–39). However, in the DESIR (Datafrom an Epidemiological Study on theInsulin Resistance syndrome) cohort, alongitudinal study, no relation betweenthe average or a change in alcohol con-sumption and fasting insulin levels wasfound, but this may be a result of theinclusion of subjects with type 2 diabe-tes (40). Fasting insulin level is a surro-gate marker of insulin sensitivity inhealthy subjects, with lower insulin lev-els indicating higher insulin sensitivity(11,41). Conversely, low insulin levelsare a common phenomenon in subjects

with type 2 diabetes due to impairedinsulin secretion by b-cells. Becausewe excluded studies in subjects withtype 2 diabetes, the results of lowerfasting insulin levels may indicate higherinsulin sensitivity. Additionally, we ob-served no change in glucose levels byalcohol consumption, and lower insulinlevels coinciding with unchanged glu-cose levels suggest an improved insulinsensitivity.

The current meta-analysis suggeststhat men andwomenmight respond dif-ferently to a period of alcohol consump-tion with regard to insulin sensitivity.Subgroup analysis showed that the ef-fect of alcohol consumption on insulinsensitivity was only present amongwomen, but the pooled effects in menandwomenwere not significantly differ-ent. These results generally concordwith observational studies showing alarger risk reduction of moderate alco-hol consumption on risk of type 2 diabe-tes in women than in men (40% vs. 13%)(1) and with the study by Beulens et al.(42). The studies included in the reviewby Hulthe and Fagerberg (5), which weremainly cross-sectional, did not find sexdifferences in alcohol effects.

We observed lower levels of HbA1c insubjects consuming moderate amountsof alcohol compared with abstainers.This has also been shown in several ob-servational studies (39,43,44). Alcoholmay decrease HbA1c by suppressingthe acute rise in blood glucose after ameal and increasing the early insulin re-sponse (45). This would result in lowerglucose concentrations over time and,thus, lower HbA1c concentrations. Un-fortunately, the underlying mechanismof glycemic control by alcohol is notclearly understood.

Strengths and Weaknesses of theStudyA major strength of this meta-analysis isthe inclusion of studies with a random-ized controlled design and the inclusionof several complementary end points,providing a comprehensive overview ofthe evidence on this topic. There arealso limitations that warrant consider-ation. As in any meta-analysis, thestrength of the current study is largelydetermined by the quality and numberof the included studies. The results ofthe quality assessment show that thelarger part of the included studies did

Figure 3—Forest plot of meta-analysis of the effect of alcohol consumption on HbA1c. Data arepooled SMDs with 95% CIs.

care.diabetesjournals.org Schrieks and Associates 729

not report or did not take into accountsome important aspects, such as blind-ing. Nevertheless, randomization andthe inclusion of an alcohol-free controlgroup were the most important qualityfactors for this review, and only six stud-ies did not satisfy those criteria. Compli-ance was measured in most studies (17of 22) but was only reported in 13. How-ever, of these 13 studies, 11 reportedgood or excellent compliance, suggest-ing that low compliance did not influ-ence the results of the studies. Second,the analysis of several different out-comes resulted in inclusion of a smallnumber of studies for certain endpoints, such as HbA1c. Third, only twostudies used the gold standard HEGCto estimate insulin sensitivity (11). Be-cause this may lead to inconsistency inthe results, we standardized the resultsof the different studies using Cohen d.However, the results from the studiesusing HEGC were similar to the otherintervention studies, and no significantheterogeneity was present except forthe combined meta-analysis of ISI andHOMA-IR. This was due to the study ofChiva-Blanch et al. (15), who reported arelatively small variation in HOMA-IR,causing a relatively large SMD. Exclusionof this study removed heterogeneitywithout changing the effect. Fourth, be-cause most studies used a crossoverdesign, a carryover effect might haveinfluenced the outcomes. Another limi-tation was the short duration and smallsample sizes of the included studies. Theaverage duration of 5.4 weeks may nothave been long enough to show detect-able differences in insulin sensitivity orglucose status. In addition, effects maychange after longer-term intake of alco-hol. Therefore, the short-term nature ofthe included studies does not allow us todraw conclusions on longer-term alco-hol consumption.It is important to note evidence for

publication bias for certain outcomesin the current study. The publicationbias unexpectedly suggests that smallerstudies with positive results are missing.After adjustment for publication bias us-ing the trim and fill method, even a sig-nificant increase in insulin sensitivity byalcohol consumption was shown. How-ever, statistical tests for publication biasmay yield biased results with small num-bers of studies and are prone to hetero-geneity (17).

Finally, the results of this researchmay not be generalizable to all healthysubjects because the selected studies in-cluded mainly light to moderate alcoholconsumers. Therefore, the period of ab-staining from alcohol might also be seenas an intervention, and subjects mighthave responded differently than alcoholabstainers.

ImplicationsTo draw implications from the current re-search, the findings need to be placed in aclinical context. In this meta-analysis, weobserved that alcohol consumption de-creased fasting insulin levels by 0.19,which translates to an ;11% decreasein insulin (220 pmol/L) in people withimpaired glucose tolerance, as calculatedfromdata of theDiabetes Prevention Pro-gram study (46), and a 13% decrease ininsulin (25.2 pmol/L) in normoglycemicpeople, as calculated from data of theMultiethnic Study of Atherosclerosis(MESA) (47). For comparison, metformintreatment results in a 14% decrease infasting insulin levels and a 40% lowerrisk of diabetes versus a control group(48). An 11% reduction of fasting insulinlevels after alcohol consumption wouldresult in an ;30% reduced risk of diabe-tes, which is in line with the 40% risk re-duction observed among women.

The reduced HbA1c concentrationfound in the current study by alcohol con-sumption (SMD 20.62) is equal to a 5%reduction in HbA1c concentration in boththe MESA and the Diabetes Preven-tion Program studies (from 5.4% [36mmol/mol] to 5.1% [33 mmol/mol] andfrom 5.9% [41 mmol/mol] to 5.6% [38mmol/mol], respectively) (46,49). The Di-abetes PreventionProgram study showedthat 4 years of metformin medicationand a lifestyle intervention both resultedin a reduction in HbA1c of ;3% (50). Be-cause type 2 diabetes is characterized byhyperglycemia, HbA1c could be seen as asurrogate end point of the disease ratherthan an intermediate factor in the path-way toward type 2 diabetes. The WorldHealth Organization indeed suggeststhat a level .6.5% (48 mmol/mol) beused as a cutoff point for diagnosing di-abetes (51). In this respect, the currentresults for HbA1c match with the reducedrisk of type 2 diabetes with moderate al-cohol consumption. Results of alcohol in-take on HbA1c should be carefullyinterpreted because we included only

three intervention studies in the analysis.However, the results suggest that drink-ing a moderate amount of alcohol is notharmful with regard to insulin sensitivityand glycemic status in healthy adultswithout type 2 diabetes.

ConclusionThis systematic review andmeta-analysisshowed that moderate alcohol consump-tion decreased fasting insulin and HbA1cconcentrations among nondiabetic sub-jects. Alcohol consumption might im-prove insulin sensitivity among womenbut did not do so overall. These resultsmay partly explain the lower risk of type 2diabetes with moderate alcohol con-sumption found in observational studies.However, more intervention studieswith a longer intervention period are nec-essary to confirm the results.

Funding. I.C.S. and H.F.J.H. were supported byboth the Dutch Ministry of Economic Affairs,Agriculture and Innovation and the DutchFoundation for Alcohol Research, representingDutch producers of and traders in beer, wine,and spirits and The Netherlands Organizationfor Applied Scientific Research. Their joint aim isto independently study the health effects ofmoderate alcohol consumption.

The funding sources had no role in conductingthe study, in analyzing or interpreting the studyresults, or in the decision to submit the manu-script for publication.Duality of Interest. No potential conflicts ofinterest relevant to this article were reported.Author Contributions. I.C.S. contributed tothe data extraction and analysis, data interpre-tation, and writing of the manuscript. A.L.J.H.contributed to the study design, article searchand data extraction, data analysis, and datainterpretation. H.F.J.H. contributed to the datainterpretation. K.J.M. contributed to the studydesign and data interpretation. J.W.J.B. contrib-uted to the study design, search and dataextraction, data analysis, data interpretation,and writing of the manuscript.Prior Presentation. Parts of this study werepresented in abstract form at the Epidemiologyand Prevention/Nutrition, Physical Activity andMetabolism 2014 Scientific Sessions of theAmerican Heart Association, San Francisco, CA,18–21 March 2014, and were presented orally atthe 7th European Beer and Health Symposium,Brussels, Belgium, 30 September 2014.

References1. Baliunas DO, Taylor BJ, Irving H, et al. Alcoholas a risk factor for type 2 diabetes: a systematicreview and meta-analysis. Diabetes Care 2009;32:2123–21322. Koppes LL, Dekker JM, Hendriks HF, BouterLM, Heine RJ. Moderate alcohol consumptionlowers the risk of type 2 diabetes: a meta-analysis of prospective observational studies.Diabetes Care 2005;28:719–725

730 Alcohol Consumption and Insulin Sensitivity Diabetes Care Volume 38, April 2015

3. Hendriks HFJ. Moderate alcohol consump-tion and insulin sensitivity: observations andpossible mechanisms. Ann Epidemiol 2007;17:S40–S424. Brien SE, Ronksley PE, Turner BJ, Mukamal KJ,Ghali WA. Effect of alcohol consumption on bi-ological markers associated with risk of coronaryheart disease: systematic review and meta-analysis of interventional studies. BMJ 2011;342:d6365. Hulthe J, Fagerberg B. Alcohol consumptionand insulin sensitivity: a review. Metab SyndrRelat Disord 2005;3:45–506. Zilkens RR, Burke V, Watts G, Beilin LJ,Puddey IB. The effect of alcohol intake on in-sulin sensitivity in men: a randomized con-trolled trial. Diabetes Care 2003;26:608–6127. Flanagan DE, Pratt E, Murphy J, et al. Alcoholconsumption alters insulin secretion and car-diac autonomic activity. Eur J Clin Invest 2002;32:187–1928. Cordain L, Melby CL, Hamamoto AE, et al.Influence of moderate chronic wine consump-tion on insulin sensitivity and other correlates ofsyndrome X in moderately obese women. Me-tabolism 2000;49:1473–14789. Davies MJ, Baer DJ, Judd JT, Brown ED,Campbell WS, Taylor PR. Effects of moderatealcohol intake on fasting insulin and glucoseconcentrations and insulin sensitivity in post-menopausal women: a randomized controlledtrial. JAMA 2002;287:2559–256210. Sierksma A, Patel H, Ouchi N, et al. Effect ofmoderate alcohol consumption on adiponectin,tumor necrosis factor-a, and insulin sensitivity.Diabetes Care 2004;27:184–18911. Muniyappa R, Lee S, Chen H, Quon MJ. Cur-rent approaches for assessing insulin sensitivityand resistance in vivo: advantages, limitations,and appropriate usage. Am J Physiol EndocrinolMetab 2008;294:E15–E2612. Nathan DM, Turgeon H, Regan S. Relation-ship between glycated haemoglobin levels andmean glucose levels over time. Diabetologia2007;50:2239–224413. Cordain L, Bryan ED, Melby CL, Smith MJ.Influence of moderate daily wine consumptionon body weight regulation and metabolism inhealthy free-living males. J Am Coll Nutr 1997;16:134–13914. Lavy A, Fuhrman B, Markel A, et al. Effect ofdietary supplementation of red or white wineon human blood chemistry, hematology and co-agulation: favorable effect of red wine onplasma high-density lipoprotein. Ann NutrMetab 1994;38:287–29415. Chiva-Blanch G, Urpi-Sarda M, Ros E, et al.Effects of red wine polyphenols and alcohol onglucose metabolism and the lipid profile: a ran-domized clinical trial. Clin Nutr 2013;32:200–20616. Jadad AR,Moore RA, Carroll D, et al. Assess-ing the quality of reports of randomized clinicaltrials: is blinding necessary? Control Clin Trials1996;17:1–1217. Higgins JPT, Green S, Eds. Cochrane Hand-book for Systematic Reviews of Interventions.Version 5.1.0. Oxford, U.K., The Cochrane Col-laboration, 201118. Duval S, Tweedie R. Trim and fill: a simplefunnel-plot-basedmethod of testing and adjust-ing for publication bias in meta-analysis. Bio-metrics 2000;56:455–463

19. Zheng J, Yang J, Huang T, Hu X, Luo M, Li D.Effects of Chinese liquors on cardiovascular dis-ease risk factors in healthy young humans. Sci-entificWorldJournal 2012;2012:37214320. Contaldo F, D’Arrigo E, Carandente V, et al.Short-term effects of moderate alcohol consump-tion on lipid metabolism and energy balance innormal men. Metabolism 1989;38:166–17121. Beulens JW, van Beers RM, Stolk RP,Schaafsma G, Hendriks HF. The effect of moderatealcohol consumption on fat distribution and adi-pocytokines.Obesity (Silver Spring) 2006;14:60–6622. Beulens JW, de Zoete EC, Kok FJ, SchaafsmaG, Hendriks HF. Effect of moderate alcohol con-sumption on adipokines and insulin sensitivityin lean and overweight men: a diet interventionstudy. Eur J Clin Nutr 2008;62:1098–110523. Beulens JW, van Loon L J, Kok FJ, et al. Theeffect of moderate alcohol consumption on adi-ponectin oligomers and muscle oxidative capac-ity: a human intervention study. Diabetologia2007;50:1388–139224. Joosten MM, Beulens JW, Kersten S,Hendriks HF. Moderate alcohol consumption in-creases insulin sensitivity and ADIPOQ expres-sion in postmenopausal women: a randomised,crossover trial. Diabetologia 2008;51:1375–138125. Joosten MM, Witkamp RF, Hendriks HF. Al-terations in total and high-molecular-weightadiponectin after 3 weeks of moderate alcoholconsumption in premenopausal women. Me-tabolism 2011;60:1058–106326. Joosten MM, Schrieks IC, Hendriks HF. Ef-fect of moderate alcohol consumption onfetuin-A levels in men and women: post-hocanalyses of three open-label randomized cross-over trials. Diabetol Metab Syndr 2014;6:2427. Bhathena SJ, Berlin E, Judd JT, et al. Selec-tive responses of hormones involved in carbo-hydrate and lipid metabolism and properties oferythrocyte membranes during the menstrualcycle in premenopausal women consumingmoderate amounts of alcohol. Am J Clin Nutr1995;62:751–75628. Cesena FHY, Coimbra SR, Andrade ACM, daLuz PL. The relationship between body mass in-dex and the variation in plasma levels of triglyc-erides after short-term red wine consumption.J Clin Lipidol 2011;5:294–29829. Kim SH, Abbasi F, Lamendola C, Reaven GM.Effect of moderate alcoholic beverage con-sumption on insulin sensitivity in insulin-resistant, nondiabetic individuals. Metabolism2009;58:387–39230. Romeo J, Gonzalez-Gross M, Warnberg J,Dıaz LE, Marcos A. Effects of moderate beerconsumption on blood lipid profile in healthySpanish adults. Nutr Metab Cardiovasc Dis2008;18:365–37231. Bantle AE, Thomas W, Bantle JP. Metaboliceffects of alcohol in the form of wine in personswith type 2 diabetesmellitus. Metabolism 2008;57:241–24532. Shai I, Wainstein J, Harman-Boehm I, et al.Glycemic effects of moderate alcohol intakeamong patients with type 2 diabetes: a multi-center, randomized, clinical intervention trial.Diabetes Care 2007;30:3011–301633. Queipo-Ortu~noMI, Boto-Ordo~nezM,MurriM,et al. Influence of red wine polyphenols and etha-nol on the gutmicrobiota ecology and biochemicalbiomarkers. Am J Clin Nutr 2012;95:1323–1334

34. Kawamoto R, Kohara K, Tabara Y, et al. Al-cohol consumption is associatedwith decreasedinsulin resistance independent of body mass in-dex in Japanese community-dwelling men. To-hoku J Exp Med 2009;218:331–33735. Englund Ogge L, Brohall G, Behre CJ,Schmidt C, Fagerberg B. Alcohol consumptionin relation to metabolic regulation, inflamma-tion, and adiponectin in 64-year-old Caucasianwomen: a population-based study with a focuson impaired glucose regulation. Diabetes Care2006;29:908–91336. Mayer EJ, Newman B, Quesenberry CP Jr,Friedman GD, Selby JV. Alcohol consumptionand insulin concentrations. Role of insulin inassociations of alcohol intake with high-densitylipoprotein cholesterol and triglycerides. Circu-lation 1993;88:2190–219737. Kiechl S, Willeit J, Poewe W, et al. Insulinsensitivity and regular alcohol consumption:large, prospective, cross sectional populationstudy (Bruneck study). BMJ 1996;313:1040–104438. Lazarus R, Sparrow D, Weiss ST. Alcohol in-take and insulin levels. The Normative AgingStudy. Am J Epidemiol 1997;145:909–91639. Kroenke CH, Chu NF, Rifai N, et al. A cross-sectional study of alcohol consumption patternsand biologic markers of glycemic control among459 women. Diabetes Care 2003;26:1971–197840. Vernay M, Balkau B, Moreau J, Sigalas J,Chesnier M, Ducimetiere P; Desir Study Group.Alcohol consumption and insulin resistance syn-drome parameters: associations and evolutionsin a longitudinal analysis of the French DESIRcohort. Ann Epidemiol 2004;14:209–21441. Laakso M. How good a marker is insulinlevel for insulin resistance? Am J Epidemiol1993;137:959–96542. Beulens JW, van der Schouw YT, BergmannMM, et al.; InterAct Consortium. Alcohol con-sumption and risk of type 2 diabetes in Euro-pean men and women: influence of beveragetype and body size The EPIC-InterAct study[published correction appears in J Intern Med2013;273:422]. J Intern Med 2012;272:358–37043. Harding AH, Sargeant LA, Khaw KT, et al.Cross-sectional association between total leveland type of alcohol consumption and glycosy-lated haemoglobin level: the EPIC-NorfolkStudy. Eur J Clin Nutr 2002;56:882–89044. GullifordMC, Ukoumunne OC. Determinantsof glycated haemoglobin in the general popula-tion: associations with diet, alcohol and cigarettesmoking. Eur J Clin Nutr 2001;55:615–62345. McMonagle J, Felig P. Effects of ethanolingestion on glucose tolerance and insulinsecretion in normal and diabetic subjects.Metabolism 1975;24:625–63246. Kitabchi AE, Temprosa M, Knowler WC,et al.; Diabetes Prevention Program ResearchGroup. Role of insulin secretion and sensitivityin the evolution of type 2 diabetes in the di-abetes prevention program: effects of lifestyleintervention and metformin. Diabetes 2005;54:2404–241447. Bertoni AG, Wong ND, Shea S, et al. Insulinresistance, metabolic syndrome, and subclinicalatherosclerosis: the Multi-Ethnic Study of Ath-erosclerosis (MESA). Diabetes Care 2007;30:2951–295648. Salpeter SR, Buckley NS, Kahn JA, SalpeterEE. Meta-analysis: metformin treatment in

care.diabetesjournals.org Schrieks and Associates 731

persons at risk for diabetes mellitus. Am J Med2008;121:149–15749. McNeely MJ, McClelland RL, Bild DE, et al.The association between A1C and subclinical car-diovascular disease: themulti-ethnic study of ath-erosclerosis. Diabetes Care 2009;32:1727–173350. Knowler WC, Barrett-Connor E, FowlerSE, et al.; Diabetes Prevention Program Re-search Group. Reduction in the incidence of

type 2 diabetes with lifestyle intervention ormetformin. N Engl J Med 2002;346:393–40351. World Health Organization. Use of GlycatedHaemoglobin (HbA1c) in the Diagnosis of Diabe-tes Mellitus. Geneva, World Health Org., 2011(Rep. no. WHO/NMH/CHP/CPM/11.1)52. Flechtner-Mors M, Biesalski HK,Jenkinson CP, Adler G, Ditschuneit HH. Effectsof moderate consumption of white wine on

weight loss in overweight and obese subjects.Int J Obes Relat Metab Disord 2004;28:1420–142653. Joosten MM, van Erk MJ, Pellis L, WitkampRF, Hendriks HFJ. Moderate alcohol consump-tion alters both leucocyte gene expression pro-files and circulating proteins related to immuneresponse and lipid metabolism in men. Br J Nutr2012;108:620–627

732 Alcohol Consumption and Insulin Sensitivity Diabetes Care Volume 38, April 2015