a systematic review and meta-analysis of the association...

A Systematic Review andMeta-analysisof the Association Between Depressionand Insulin ResistanceCAROL KAN, MA, MBBS, MRCPSYCH

1

NAOMI SILVA, BSC1

SHERITA HILL GOLDEN, MD, MHS2,3

ULLA RAJALA, MD, PHD4

MARKKU TIMONEN, MD, PHD4

DANIEL STAHL, PHD5

KHALIDA ISMAIL, MRCP, MRCPSYCH, PHD1

OBJECTIVEdDepression is associated with the onset of type 2 diabetes. A systematic reviewand meta-analysis of observational studies, controlled trials, and unpublished data was conduc-ted to examine the association between depression and insulin resistance (IR).

RESEARCH DESIGN AND METHODSdMedline, EMBASE, and PsycINFO weresearched for studies published up to September 2011. Two independent reviewers assessedthe eligibility of each report based on predefined inclusion criteria (study design and measureof depression and IR, excluding prevalent cases of diabetes). Individual effect sizes were stan-dardized, and a meta-analysis was performed to calculate a pooled effect size using randomeffects. Subgroup analyses and meta-regression were conducted to explore any potential sourceof heterogeneity between studies.

RESULTSdOf 967 abstracts reviewed, 21 studies met the inclusion criteria of which 18studies had appropriate data for the meta-analysis (n = 25,847). The pooled effect size (95%CI) was 0.19 (0.11–0.27) with marked heterogeneity (I2 = 82.2%) using the random-effectsmodel. Heterogeneity between studies was not explained by age or sex, but could be partlyexplained by the methods of depression and IR assessments.

CONCLUSIONSdA small but significant cross-sectional association was observed betweendepression and IR, despite heterogeneity between studies. The pathophysiology mechanismsand direction of this association need further study using a purposively designed prospective orintervention study in samples at high risk for diabetes.

Diabetes Care 36:480–489, 2013

Depression is at least twice as com-mon among those with diabetescompared with the general popula-

tion (1) and is associated with adverse ef-fects on diabetes outcomes includingsuboptimal glycemic control (2), compli-cations (3), and higher rates of mortality(4,5). Depression appears to be presenteven at the prediabetes stage of the type2 diabetes (T2DM) continuum withpooled data suggesting that depression

in the nondiabetic population is indepen-dently associated with a 37–60% in-creased prospective risk of developingT2DM (6).

Insulin resistance (IR) is a prediabetesstage. There have now been several stud-ies examining the association betweendepression and IR. These studies havehad mixed findings. The aim of this re-view is to conduct a systematic synthesisand ameta-analysis of the evidence for the

association between depression and IR. Apositive association would increase theplausibility of a biological link betweendepression and diabetes and suggest apotentially modifiable target for the pre-vention of T2DM.

RESEARCH DESIGN ANDMETHODSdOur systematic reviewand meta-analysis was conducted accord-ing to Preferred Reporting Items for Sys-tematic Reviews and Meta-Analyses(PRISMA) guidelines (7).

Data sources and study selectionThe following electronic librariesdMED-LINE (1948 to September 2011), EMBASE(1947 to September 2011), and PsycINFO(1806 to September 2011)dweresearched to identify relevant studies. Thesearch items were based on establishedterminology using Cochrane definitionswhere possible and were “diabetes,” “de-pression,” “insulin resistance,” and “insu-lin sensitivity” (Supplementary Table 1).The titles and/or abstracts were reviewedto exclude any clearly irrelevant studies.The full texts of the remaining studieswere then retrieved and read in full bytwo authors (C.K. and N.S.) indepen-dently to determine whether the studiesmet inclusion criteria. Disagreement wasresolved by a third author (K.I.) who in-dependently examined the studies. Thereference lists of studies that examine thetopic of interest were checked for addi-tional publications while correspondingauthors were contacted for additional in-formation on published and unpublishedstudies.

Criteria for inclusion into the reviewAbstracts were considered eligible for fullmanuscript data extraction if the studymet all the following criteria: a) they re-ported an association between depressionand IR (including its reverse measure, lowinsulin sensitivity); b) sample consisted ofadults ($18 years of age); and c) the de-sign was cross-sectional, observational,or a randomized controlled trial. Studiesthat excluded patients with depression atbaseline or consisted solely of patients

c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c

From the 1Institute of Psychiatry, King’s College London, London, U.K.; the 2Department of Medicine, JohnsHopkins School of Medicine, Baltimore, Maryland; the 3Department of Epidemiology, Johns Hopkins Uni-versity, Bloomberg School of PublicHealth, Baltimore,Maryland; the 4Institute ofHealth Sciences,University ofOulu, Oulu, Finland; and the 5Department of Biostatistics, Institute of Psychiatry, King’s College London,London, U.K.

Corresponding author: Carol Kan, [email protected] 20 July 2012 and accepted 17 September 2012.DOI: 10.2337/dc12-1442This article contains Supplementary Data online at http://care.diabetesjournals.org/lookup/suppl/doi:10

.2337/dc12-1442/-/DC1.The views expressed are those of the authors and not necessarily those of the National Health Service, the

National Institute for Health Research, or the Department of Health.© 2013 by the American Diabetes Association. Readers may use this article as long as the work is properly

cited, the use is educational and not for profit, and thework is not altered. See http://creativecommons.org/licenses/by-nc-nd/3.0/ for details.

480 DIABETES CARE, VOLUME 36, FEBRUARY 2013 care.diabetesjournals.org

M E T A - A N A L Y S I S

http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc12-1442/-/DC1

mailto:[email protected]



http://creativecommons.org/licenses/by-nc-nd/3.0/

http://creativecommons.org/licenses/by-nc-nd/3.0/

with diabetes (or where it was not possi-ble to separate diabetic and nondiabeticparticipants) were not included.

Data extractionUsing a standardized data extractionsheet, the following information (if avail-able) was extracted and recorded fromstudies: authors; year of publication;country of origin; study design; totalsample size of nondiabetic participants;age; sex; methods of IR assessment; meth-ods of depression assessment; and type ofconfounders. Authors were contacted toclarify whether prevalent cases of diabeteswere excluded at baseline. An attempt toretrieve missing or incomplete data in thepublished study was made by e-mail to atleast two coauthors on at least two occa-sions. If multiple risk estimates werepresented in a given manuscript, the un-adjusted estimate was selected for theprimary meta-analysis as some studieswere adjusted for prominent confound-ing variables, such as family history andadiposity, while others were not, re-ndering a direct comparison of estimatesto be questionable. Reporting unadjustedestimates also reduces the bias of selectivereporting of adjusted estimates in primarystudies and the potentiality of overadjust-ment with multiple confounders, whichmay also be on the causal pathway for theeffect of depression on IR, such as obesity(8). Studies written in a foreign languagewere translated by mental health professio-nals fluent in that language.

Quality assessmentThere is no consensus as to the beststandardized method for assessing thequality of observation studies, and thePRISMA guidelines for randomized con-trolled trials (7) and Meta-analysis Of Ob-servational Studies in Epidemiology(MOOSE) guidelines for observationalstudies in epidemiology (9) were used toexamine the quality of the studies. Theseinclude adequacy of study design (pro-spective cross-sectional, observational,and randomized controlled trial with anadequate control group); recruitment ofsample; ascertainment of depression andIR; and control for cofounding variables,such as age, sex, socioeconomic status,and BMI. The quality of the studies wasnot summarized with a score, as this ap-proach has been criticized for allocatingequal weight to different aspects of meth-odology (10), but a formal assessment ofthe risk of bias and strength of evidencesaccording to the Agency for Healthcare

Research and Quality (AHRQ) guidelineswas conducted (11). A study was consid-ered to be of high quality if the study de-sign was prospective in nature; consecutiveor random sampling method was used; theascertainment of depression was through astructural diagnostic interview based onthe ICD (12) or the DSM (13); and co-founders for diabetes and depression (age,sex, ethnicity, BMI/waist circumference,socioeconomic status, physical activity)were accounted for.

Data synthesis and analysisMeta-analyses were carried out usingStata 10.1 and 11.1 (14,15), with user-contributed commands for meta-analyses:metan, metainf, metabias, metatrim andmetareg (16). The Cohen d approachwas used to calculate the primary effectsize, as it allows data from different plat-forms to be combined without the use ofnormalization and can be converted fromdifferent effect sizes. It was calculated forthe majority of the datasets by the meandifference in IR between depressed andnondepressed groups divided by thepooled SD. The SE of each study’s stan-dardized effect estimate was calculatedfrom the estimated effect and the study’sgroup sizes according to a formula pro-vided by Cooper and Hedges (17). If Pear-son correlation coefficient (rr) wasreported instead, it was transformed intoCohen d using Cohen’s conversion for-mula (1988): d ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

4r2=ð12 r2Þp. The

variance was calculated by Vd = 4Vr /(1-r2)3 where Vr is the variance of rp. Thesame conversion was applied to Spearmancorrelation coefficients (rs) since rr isequivalent to rs using rank data or isslightly smaller if the data are binomialdistributed (18). In one study (19), the zstatistic of a Mann-Whitney U test wasused to transform z to r using Fischertransformation r ¼ z=

ffiffiffin

p(20) and then

converting r to Cohen d using the aboveformula. Results reported in odds ratioswere transformed in Cohen d using themethod recommended by Borensteinet al. (21), d ¼ lnðORÞ3 ffiffiffi

3p

=p. The asso-ciated variance of d would then beVðdÞ ¼ VlnðORÞ33=p2and the SE

ffiffiffiffiffiffiffiffiffiVðdÞp

.The effect sizes and SEs of the studies

were pooled using random-effects mod-els. The random-effects meta-analysismodels were chosen as heterogeneity isexpected given the differences in studypopulations and procedures. The as-sumption of homogeneity of true effectsizes was assessed by the Cochran Q test(22), and the degree of inconsistency

across studies was calculated I2 (23). I2

describes the percentage of total variationacross studies that is due to heterogeneityrather than sampling error and ranges be-tween 0% (no inconsistency) and 100%(high heterogeneity) with values of 25,50, and 75% suggesting low, moderate,and high heterogeneity (23). A priorimeta-regression analysis was then per-formed to assess whether conclusionswere sensitive to restricting studies tosubgroups that might modify the effectsize: i) mean age; ii) sex; iii) method ofdepression assessment; and iv) method ofIR assessment. Random-effects modelswere used to allow for the residual hetero-geneity among attrition rates, which werenot modeled by the explanatory variables(24). A secondary analysis of adjusted andcorresponding unadjusted data whenavailable was also conducted using therandom-effects models.

Sensitivity analyses were conductedto weigh up the relative influence of eachindividual study on the pooled effect sizeusing STATA’s user-written function,metainf (16). The presence of publishingbias for the hypothesis of an associationbetween depression and IR was assessedinformally by visual inspections of funnelplots (25) and corroborated by Begg ad-justed rank correlation (26) as implemen-ted in metabias. The nonparametric “trimand fill”method was also used to estimatethe number of hypothetical studies thatwere missing due to possible publicationbias and was implemented in STATA’suser-written command, metatrim (16).It is a sensitivity analysis since it relied onstrong symmetrical assumption and couldbe influenced in the presence of strongbetween-group heterogeneity (27).

RESULTS

Study selectionThe flowchart is shown in Fig. 1. The lit-erature search resulted in 962 studies. Af-ter review of their titles and abstracts, 38studies met the inclusion criteria andwere retrieved for full text. Of these, 21studies were excluded from the system-atic review as they no longer met the in-clusion criteria. The search for additionalstudies among the reference lists of in-cluded articles yielded five more studies,with four meeting inclusion criteria. A to-tal of 21 studies were included in the sys-tematic review, and the extracted data aresummarized in Table 1.

Three studies were excluded from themeta-analysis; one study was published as

care.diabetesjournals.org DIABETES CARE, VOLUME 36, FEBRUARY 2013 481

Kan and Associates

an abstract (28) and did not include anyinformation about sampling method,baseline clinical characteristics of thesample population, and depression mea-sure. The data of two large cohort studieswere presented in quartiles (29,30), andraw data were not available to generate astandardized effect size. This resulted in18 studies being included in the meta-analysis.

Upon further examination, one studywas found to be made up of two separatestudies using two different study popula-tions (31), which were therefore sepa-rated into two different datasets. Thesample for six studies were separatedinto normal/impaired glucose tolerance(32,33) and men/women (19,34–36),yielding an additional six datasets. The totalnumber of datasets in the meta-analysiswas therefore 25.

Qualitative summaryOf the 25 datasets include in the meta-analysis, one was a prospective longitudi-nal cohort study (37), six were case-controlstudies (31,38–41), and 18 were cross-sectional studies (19,32–36,42–47). Fourdatasets were based on clinical diagnosisusing DSM-IV, six used semistructureddiagnostic interviews, and 15 used self-report depressive scales. IR was reportedin 18 datasets while insulin sensitivity wasmeasured in seven datasets. Descriptivedata from the datasets are summarized inTable 1.

Meta-analysisA total of 25 datasets (n = 25,847) pro-vided unadjusted data on the associationbetween depression and IR in adults with-out diabetes. A random-effects meta-analysis revealed a small pooled estimate

of the mean standardized effect sizes (d =0.19 [95% CI: 0.11–0.27]) (Fig. 2), withthe effect sizes ranging from d = -0.56 tod = 1.37. Heterogeneity between the stud-ies was statistically significant (Q (24) =134.83, P , 0.0001) and large in magni-tude (I2 = 82.2%).Subgroup analysis and meta-regression.A series of random-effects subgroup anal-yses and meta-regression was conductedto examine whether the association be-tween depression and IR varied acrossdemographic groups and the methods ofdepression and IR assessments. Age (b =-0.002 per year, t = -0.50; P = 0.62) or sex(b = 0.0006, t = 0.33; P = 0.74) did notsignificantly change the observed associ-ation between depression and IR. Withthe random-effects model, a much greatereffect size was observed for diagnostic in-terviews than self-report measures (0.46[0.22–0.71] vs. 0.13 [0.05–0.21]) and thedifference was statistically significant inthe meta-regression (z = 2.22, P ,0.0001). A larger effect size with insulinsensitivity as an IR measure was found incomparison with studies using homeosta-sis model assessment of insulin resistance(HOMA-IR) or HOMA2-IR test (0.32[0.12–0.53] vs. 0.17 [0.08–0.26]), andthe difference was also significant (z =4.70, P , 0.0001). The observed associ-ation between depression and IR re-mained statistically significant in allsubgroup analyses.Secondary analysis. Of the 17 datasetswith confounders being included, threestudies were excluded from the secondaryanalysis. Depression and IR were not themain outcome of interest in one study(40) and thus, its association was not ad-justed for the confounder being mea-sured, while two studies presented theirdata in quartiles (45,46) and raw datawere not available to generate a standard-ized effect size. This resulted in 14 data-sets being included in the random-effectssecondary meta-analysis. The estimate ofthe mean standardized effect sizes was0.11 (0.04–0.17) for theunadjusteddatasets(n = 22,545) and 0.02 (20.02 to 0.07) forthe adjusted datasets (n = 21,826) (Fig. 3).Sensitivity analysis and publicationbias. The robustness of the estimate wasexamined by sequentially removing eachstudy and reanalyzing the remaining da-tasets. The estimated effect sizes rangedfrom d = 0.14 to d = 0.21, with all effectsizes being significantly different from 0,suggesting that the significant effect size isnot determined by a single study. Sensi-tivity analysis for the secondary analysis

Figure 1dFlowchart of systematic review. *Further information in regards to excluded studiescan be found in Supplementary Table 1.


Depression and insulin resistance

Table

1dSu

mmarytableof

prim

arystud

iesinclud

edin

thesystem

icreview

Firstauthor,

year,cou

ntry

Setting,

study

population

Stud

ydesign

,sample

selection

Sample

size

a

Age

c ,mean(SD)d

orrange

Men/

Wom

enc

Evidenceof

selectionbias

IRmeasure

Depression

assessment

(measure)

Adjustedfor

confoun

ders

Adriaanse,200

6,Netherlands

Com

mun

ity,

IGT

Cross-section

al,

rando

m16

455

–75

84/75

Non

participation

atfollo

w-up

unaccoun

tedfor

HOMA-IR

SR(CES-D)

Non

e

Com

mun

ity,N

GT

Cross-section

al,

rando

m26

055

–75

129/12

9Non

participation

atfollo

w-up

unaccoun

tedfor

HOMA-IR

SR(CES-D)

Non

e

Chiba,2

000,

Japan

Hospital,depressed

andhealthy

control

subjects

Case-control,

consecutiv

e20

846

.8(12.8)

130/78

Non

representative

controlgroup

HOMA-IR

CI

Non

e

Outpatients,NGT

Case-control,

consecutiv

e22

048

.7(8.6)

138/82

Non

participation

atfollo

w-up

unaccoun

tedfor

HOMA-IR

SR(ZSD

S)Non

e

Everson

-Rose,

2004

,U.S.

Com

mun

ity,

wom

enat

midlife

Prospective

longitudinal

coho

rt,rando

msupplem

ented

bysnow

ball

techniqu

e

2,66

246

.4(2.7)

0/2,66

2Non

participation

atfollo

w-up

unaccoun

tedfor

HOMA-IR

SR(CES-D)

Age,ethnicity,w

aist

circumference,

education

,phy

sical

activity,antidepressants,

medications

fornervou

sdisorders,site

Gilb

ey*,

2011

,U.K.

Obesity

clinic,

obese

Cross-section

al,

consecutiv

e53

*NA

NA

Noinform

ationabou

tsamplingmetho

dor

baselin

eclinical

characteristics

HOMA-IR

CI

Non

e

Golden,

2007

,U.S.

Com

mun

ity,local

residents

Cross-section

al,

rando

mand

consecutiv

e,supplem

ented

bysnow

ball

techniqu

e

5,79

061

.7(10.3)

2,68

4/3,10

6HOMA-IR

SR(CES-D,

antid

epressant

use)

Age,sex,ethnicity

,BMI,

SES,

lifestyle,m

etabolic,

inflam

matory,site

Holtb,

2009

,U.K.

Com

mun

ity,m

enBirthcohort,

consecutiv

e98

665

.7(2.9)

986/0

Non

participation

atfollo

w-upun

accoun

ted

for;lownu

mberof

subjectswith

depression

(17/1,57

8b)

HOMA-IR

SR(H

AD-D

)Age,B

MI,SE

S,sm

oking,alcoho

l

Com

mun

ity,

wom

enBirthcohort,

consecutiv

e62

566

.6(2.7)

0/62

5Non

participation

atfollo

w-up

unaccoun

tedfor;

lownu

mberof

subjectswith

depression

(20/1,41

7b)

HOMA-IR

SR(H

AD-D

)Age,B

MI,SE

S,sm

oking,alcoho

l

Contin

uedon

p.484


Kan and Associates

Tab

le1d

Con

tinu

ed

Firstauthor,

year,cou

ntry

Setting,

stud

ypo

pulation

Stud

ydesign

,sample

selection

Sample

size

a

Age

c ,mean(SD)d

orrange

Men/

Wom

enc

Evidenceof

selectionbias

IRmeasure

Depression

assessment

(measure)

Adjusted

for

confou

nders

Hung,2

007,

Taiwan

Inpatients

andstaffs

Case-control,

consecutive

3523

.3(0.12)

35/0

Non

representative

controlgroup

Minim

almod

elCI

Non

e

Kahl,20

05,

Germany

Patientsand

university

recruited

Case-control,

consecutive

3828

.8(5.4)

0/38

Non

representative

controlgroup

HOMA-IR

DI(SCID

I/II)

Age

Kauffman,

2005

,U.S.

Outpatients,

gynecological

wom

en

Case-control,

consecutive

2118

–45

0/21

HOMA-IR

CI

Non

e

Krishnam

urthy,

2008

,U.S.

Com

mun

ity,

prem

enop

ausal

wom

en

Cross-sectio

nal,

consecutive

8035

.6(7.0)

0/80

HOMA-IR

DI(SCID

)Non

e

Lawlor*,

2003

,U.K.

Com

mun

ity,

wom

enCross-sectio

nal,

consecutive

4,28

6*60

–79

0/4,28

6Rationalefor

presenting

results

inqu

artileswas

notprovided

HOMA-IR

SR(EQ5D

,past

depression

,antidepressant

use)

Age,B

MI,WHR,

SES,

physicalactiv

ity,

smok

ing,alcoho

l

Lawlor*,

2005

,U.K.

Com

mun

ity,men

Prospective

coho

rt,

consecutive

2,51

2*45

–59

2,51

2/0

Rationaleforpresenting

resultsin

quartiles

was

notprovided

HOMA-IR

SR(GHQ)

Age,SES,

physical

activ

ity,smok

ing,

alcoho

l,samplingtime

Okamura,

2000

,Japan

Patientsand

staffs

Case-control,

consecutive

3344

.8(13.2)

20/13

Non

representative

controlgroup

Minim

almod

elCI

Non

e

Pan,

2008

,China

Com

mun

ity,

localresidents

Cross-sectio

nal,

consecutive

2,83

858

.4(6.0)

1,23

6/1,60

2HOMA2-IR

SR(CES-D)

Age,sex,B

MI,

comorbidity,education,

physicalactivity,

smok

ing,alcoho

l,geographicallocation,

residentialregion

Pearson,

2010

,Australia

Com

mun

ity,men

Cross-sectio

nal,

consecutive

833

31.1

(2.5)

833/0

Highno

nparticipation

atfollo

w-up(25%

oforiginalcoho

rtsampled);

lownu

mberof

subjects

withdepression

(45/83

3)

HOMA-IR

DI(CID

I)Age,education,

physical

activ

ity,smok

ing,

alcoho

l,antid

epressants,

fish

consumption

Com

mun

ity,

wom

enCross-sectio

nal,

consecutive

899

30.9

(2.7)

0/89

9Highno

nparticipation

atfollo

w-up(25%

oforiginalcoho

rtsampled)

HOMA-IR

DI(CID

I)Age,education,

physical

activ

ity,smok

ing,

alcoho

l,antid

epressants,

oralcontraceptives,fi

shconsum

ptio

n,PC

OS

Roo

s,20

07,

Sweden

Com

mun

ity,

wom

enwith

T2D

Mrisk

factors

Cross-sectio

nal,

consecutive

1,04

756

.0(5.1)

0/1,04

7Rationaleforpresenting

resultsin

quartiles

was

notprovided

HOMA-IR

SR(SRSD

)Age,B

MI,WHR,

smok

ing,alcoho

l,ph

ysicalactivity,

timedelay

Contin

uedon

p.485



Tab

le1d

Con

tinu

ed

Firstauthor,

year,cou

ntry

Setting,

stud

ypo

pulation

Study

design

,sample

selection

Sample

size

a

Age

c ,mean(SD)d

orrange

Men/

Wom

enc

Evidenceof

selectionbias

IRmeasure

Depression

assessment

(measure)

Adjusted

for

confou

nders

Schlotzb,

2007

,U.K.

Com

mun

ity,

localresidents

Cross-section

al,

consecutive

1,19

664

.4(2.6)

618/57

8HOMA-IR

SR(H

RQoL

-MH)

Age,B

MI,CHD,

depression

,SES,

smok

ing,alcoho

lSh

en,2

011,

U.S.

Com

mun

ity,

men

Stratified

cross-

sectional,

consecutive

279

29.6

(0.0)

279/0

Selectionof

subsam

plewas

not

described;

low

numberof

subjects

withdepression

(16/27

9)

HOMA-IR

DI(CID

I)Age,ethnicity,w

aist

circum

ference,

smok

ing,sBP,

triglycerides

Com

mun

ity,

wom

enStratified

cross-

sectional,

consecutive

358

29.7

(0.0)

0/35

8Selectionof

subsam

ple

was

notdescribed;

lownu

mberof

subjectswith

depression

(18/35

8)

HOMA-IR

DI(CID

I)Age,ethnicity,w

aist

circum

ference,

smok

ing,

sBP,

triglycerides

Tim

onen,

2005

,Finland

Com

mun

ity,

NGT

Cross-section

al,

consecutive

336

61.8

(0.7)

130/20

6Non

participation

atfollo

w-up

unaccoun

tedfor

QUICKI

SR(BDI)

Sex,

BMI,education,

physicalinactivity,

smok

ing,alcoho

lCom

mun

ity,

IGT

Cross-section

al,

consecutive

9261

.7(0.7)

35/57

Non

participation

atfollo

w-up

unaccoun

tedfor

QUICKI

SR(BDI)

Sex,

BMI,education,

physicalinactivity,

smok

ing,alcoho

lTim

onen,

2006

,Finland

Com

mun

ity,

men

Cross-section

al,

consecutive

2,74

831

.3(0.4)

2,74

8/0

QUICKI

SR(H

SCL-25

)BM

I,comorbidity,S

ES,

physicalinactivity,

smok

ing,alcoho

l,CRP,

cholesterol

Com

mun

ity,

wom

enCross-section

al,

consecutive

3,01

3e31

.3(0.4)

0/3,01

3QUICKI

SR(H

SCL-25

)BM

I,comorbidity,S

ES,

physicalinactivity,

smok

ing,alcoho

l,CRP,

cholesterol

Tim

onen,

2007

,Finland

Military

conscripts,m

enCross-section

al,

consecutive

1,08

619

.2(1)

1,08

6/0

Potentialh

ealthy

workerseffect

QUICKI

SR(R-BDI)

Waistcircumference,

education,

physical

inactivity,alcoh

ol,

smok

ing

BDI:Beck

DepressionInventory;CES-D:C

enterforEpidemiologicStud

iesDepressionScale;CHD:coron

aryheartd

isease;C

I:clinicalinterview;C

IDI:Com

positeInternationalD

iagnostic

Interview;C

RP:

C-reactive

protein;D

I:diagno

sticinterview;E

Q5D

:EuroQ

ualityof

Life5Dim

ension

s;GHQ:G

eneralHealthQuestionn

aire;H

AD-D

:HospitalA

nxiety

andDepressionScale-Su

bscaleforDepression;

HAM-D

:Ham

ilton

Rating

ScaleforDepression;HRQoL

-MH:H

ealth-Related

QualityofLife-M

entalH

ealth;HSC

L-25

:Hop

kins

SymptomChecklist-25

;IGT:impaired

glucosetolerance;NA:notavailable;NGT:normalglucosetolerance;PCOS,

polycysticovarysynd

rome;R-BDI:Revised-BeckDepressionInventory;SC

ID:S

tructuredClin

icalInterviewforDiagnostic

andStatisticalManualo

fMentalD

isorders;S

ES:

socioecono

micstatus;S

R:self-repo

rted;

SRSD

:Self-Rated

SymptomsofDepression;sBP

:systolic

bloo

dpressure;W

HR:w

aist-to-hipratio;ZSD

S:Zun

gSelf-RatingDepressionScale.*Studiesexclud

edfrom

meta-analysis.aSamplesize

used

inthemeta-analysis

andcanbe

asubgroupwithinastudy

thatfulfills

theinclusioncriteria(e.g.,individualswith

outthe

diagno

sisofdiabetes).

b(Personalcom

mun

icationwith

K.Jam

eson

ofMRCLifecourseEpidemiology

Unitrevealedthat

theHertfordshireCoh

ortStudy

wasdo

neinph

asesandbasedon

different

geograph

icalregion

s).cThe

numberofparticipantsinanalysesforthesevariablesvariesdu

etomissing

data.dWeigh

tedmeans

andvariancesare

reported.

e Raw

data.


Kan and Associates

also revealed that no single study has sub-stantial influence on the effect size for theadjusted and unadjusted datasets.

There was some evidence of publish-ing bias regarding studies of depressionand IR from visual inspection of thefunnel plot (Supplementary Fig. 1) andthe Begg coefficient (z = 2.22, P =0.027). The trim and fill sensitivitymethod imputed estimates from nine hy-pothesized negative unpublished studies.The “publication bias” corrected effectsize attenuated to 0.07 (20.02 to 0.16),and this was not significant (P = 0.117).

Risk of bias and strength of evidenceGiven that most studies were cross-sectional and all were observational, theoverall risk of bias was medium to high

and the study quality was fair. The overallmagnitude of association was small andthere was substantial heterogeneity be-tween studies, but the estimate was pre-cise as reflected by the narrow CIs and themagnitude of association for diagnosticcriteria for depression was larger than forself-report depression measures. Thissuggests that the strength of evidence islow to moderate.

CONCLUSIONS

Main findingsTo our knowledge, this study representsone of the first systematic reviews andmeta-analysis of the evidence for an asso-ciation between depression and IR usingdata from observational studies, controlled

trials, and unpublished data. A small butsignificant association between depressionand IR was observed that was attenuated inanalyses adjusted for body weight andother confounders. The magnitude of theassociation increased when a diagnosticinterview for depression was used to definedepression or insulin sensitivity was a mea-sure of IR.

Strengths and limitationsThe primary strength of this meta-analy-sis is the expansive literature search but ithas several limitations, mainly stemmingfrom the quality of the included studies assummarized in Table 1. There was sub-stantial evidence of heterogeneity and po-tential publication bias. The observedfunnel plot asymmetry could be partly

Figure 2dForest plots showing the effect size of the association between depression and insulin resistance. Estimates are at the center of the boxesand drawn in proportion to the SEs. Lines indicate 95% CIs. Diamond shows the pooled effect size at its center and 95% CI at its horizontal points.IGT, impaired glucose tolerance; NGT, normal glucose tolerance.




explained by the heterogeneity in depres-sion measure as clinical/diagnostic inter-views were the depression assessments infive out of six datasets with a sample sizebelow 100. The random-effects modelwas chosen to account for heterogeneityand the association remained significantin all subgroup analyses. There was sub-stantial evidence of heterogeneity and po-tential publication bias.

A further limitation of the study is theinconsistent reporting of results, makingit necessary to convert different effectsizes into a common one. The conversionof correlation and odds ratio into Cohend rely on the assumptions that the distri-bution of the underlying trait is continu-ous (21). Association also does not implycausation and the temporal relationshipbetween depression and IR could not bedelineated since the present meta-analysisis mainly based on cross-sectional data.Nearly every individual study was a sec-ondary analysis of a study designed totest a different primary hypothesis andthis inevitably will result in some mea-surement bias and residual confounding.The strength of systematic reviews is thatby systematically identifying these limita-tions, future designs can be improved.

InterpretationThese results suggest that the associationbetween depression and diabetes may

start at an earlier stage since prevalentcases of diabetes were excluded from thisstudy and IR is on the casual pathway ofdeveloping T2DM. This is supported byother recent reviews (6,48) and a recentmeta-analysis, which provides evidencefor a bidirectional association betweendepression and metabolic syndrome(49) with hyperglycemia being one ofthe diagnostic criteria of metabolic syn-drome. These results indicate a smallbut significant association between de-pression and prediabetes or other bio-markers of glucose dysregulation.

There are several possible pathophys-iological mechanisms that may explainthe observed relationship. Depression is asso-ciated with disruption to the hypothalamic-pituitary adrenal axis, causing an increasein cortisol and catecholamine, hormonesresponsible for antagonizing the hypogly-cemic effects of insulin and resulting in IR(50). People with diagnostic depressionhave increased levels of inflammation(51), and psychological stresses havebeen shown to activate the innate inflam-matory response with chronic cytokinae-mia leading to IR and b-cell apoptosis,antecedents to the development ofT2DM (52). Depression can also have in-fluences on lifestyle behaviors associatedwith diabetes risk factors such as dietaryintake, exercise, and medication adher-ence (53,54). Findings from the secondary

analysis using data adjusted for confound-ers, such as obesity, might explain some ofthe observed associations between depres-sion and IR, although it should be inter-preted with caution as body weight hasbeen postulated to be on the casual path-way for depression and IR, raising the po-tentiality of overadjustment.

The type of depression assessmentmakes a substantive difference in theobserved association between depressionand IR, which may in part reflect a greatersensitivity of clinical interviews in detect-ing depression. Self-report measures suchas the Center for Epidemiologic StudiesDepression Scale (CES-D) have been val-idated in epidemiological studies (55) butuncertainty remains in regards to theirrelation to clinically diagnosed depres-sion. Estimates of depression have beensuggested to differ depending on the useof dimensionally verses categoricallybased depression assessment tools (56).The method at which IR is being mea-sured also has an impact upon the find-ing. There are several different methodsto measure depression and IR. Thehyperinsulinemic-euglycemic clamp iscurrently the gold standard but is un-suitable for large-scale cross-sectionalstudies for practical reasons. Good cor-relation has been demonstrated betweenestimates from HOMA and euglycemicclamp (Rs = 0.88, P , 0.0001) (57),

Figure 3dForest plots of the unadjusted and adjusted association between depression and insulin resistance for studies with confounders included.Confounders adjusted for A: age, ethnicity, waist circumference, education, physical activity, antidepressants, medications for nervous conditions,site; B: age, sex, ethnicity, BMI; C: weight, BMI, waist-to-hip ratio; D: weight, BMI, waist-to-hip ratio; E: age, sex, BMI, comorbidity, education,physical activity, smoking, alcohol, geographical location, residential region; F: age, education, physical activity, smoking, alcohol, anti-depressants, fish consumption; G: age, education, physical activity, smoking, alcohol, antidepressants, oral contraceptives, fish consumption,polycystic ovary disease; H: age, ethnicity, waist circumference, smoking, systolic blood pressure, triglyceride; I: age, ethnicity, waist circumference,smoking, systolic blood pressure, triglyceride; J: sex, BMI, education, physical inactivity, smoking, alcohol; K: sex, BMI, education, physicalinactivity, smoking, alcohol; L: BMI, comorbidity, socioeconomic status, physical inactivity, smoking, alcohol, C-reactive protein, cholesterol level;M: BMI, comorbidity, socioeconomic status, physical inactivity, smoking, alcohol, C-reactive protein, cholesterol level; and N: waist circumference,education, physical inactivity, alcohol, smoking. IGT, impaired glucose tolerance; NGT, normal glucose tolerance.


Kan and Associates

whereas the quantitative insulin sensi-tivity check index (QUICKI) has beensuggested to be superior to HOMA-IR(58). Some studies have, however,shown that minimal model analysisfrom frequently sampled intravenousglucose tolerance tests underestimatesinsulin sensitivity (59).

ImplicationsThis review suggests that it is now time tomove from repeating cross-sectionalstudies to studies examining causal rela-tionships. The ideal study design couldeither be a prospective design of patientspotentially at high risk for T2DM (e.g.,positive family history of diabetes); withor without diagnostic depression;matched for at least age, sex, obesity,and change in IR measured over time;or a randomized controlled trial in a sam-ple of depressed patients testing whetherintensive treatment of depression (phar-macological or psychological) leads toimprovements in IR and other markersof glucose dysregulation. Further sec-ondary analyses are unlikely to con-tribute further to the field unless thereis adequate assessment of potentialconfounding.

ConclusionThis systematic review and meta-analysiscontributes to the growing evidence of asmall but persistent association betweendepression and the onset of T2DM.

AcknowledgmentsdC.K. receives salarysupport from the National Institute for HealthResearch Mental Health Biomedical ResearchCentre at South London, Maudsley NationalHealth Service Foundation Trust, and King’sCollege London.No potential conflicts of interest relevant to

this study were reported.C.K., N.S., S.H.G., and K.I. designed the

protocol with contributions from U.R. andM.T. C.K. and N.S. extracted the data whileC.K. wrote the manuscript. D.S. supervisedthe statistical analysis. All authors reviewedand edited the manuscript.The authors would like to thank An Pan,

National University of Singapore; Jari Joke-lainen, University of Oulu, Finland; HollySyddall and Karen Jameson, MRC LifecourseEpidemiology Unit, University of South-ampton, UK, for providing further data tocomplete the meta-analysis, and GiovanniCizza, National Institute of Diabetes and Di-gestive and Kidney Diseases; MichaelDeuschle, Central Institute of Mental Health;Susan Everson-Rose, University of Min-nesota; and Sue Pearson, University of

Tasmania, for clarifying the methodology oftheir studies.

References1. Anderson RJ, Freedland KE, Clouse RE,

Lustman PJ. The prevalence of comorbiddepression in adults with diabetes: a meta-analysis. Diabetes Care 2001;24:1069–1078

2. Lustman PJ, Clouse RE. Depression indiabetic patients: the relationship betweenmood and glycemic control. J DiabetesComplications 2005;19:113–122

3. de Groot M, Anderson R, Freedland KE,Clouse RE, Lustman PJ. Association ofdepression and diabetes complications:a meta-analysis. Psychosom Med 2001;63:619–630

4. Ismail K, Winkley K, Stahl D, Chalder T,Edmonds M. A cohort study of peoplewith diabetes and their first foot ulcer: therole of depression on mortality. DiabetesCare 2007;30:1473–1479

5. Katon WJ, Rutter C, Simon G, et al. Theassociation of comorbid depressionwith mortality in patients with type 2diabetes. Diabetes Care 2005;28:2668–2672

6. Mezuk B, Eaton WW, Albrecht S, GoldenSH. Depression and type 2 diabetes overthe lifespan: a meta-analysis. DiabetesCare 2008;31:2383–2390

7. Moher D, Liberati A, Tetzlaff J, AltmanDG; PRISMA Group. Preferred reportingitems for systematic reviews and meta-analyses: the PRISMA statement. PLoSMed 2009;6:e1000097

8. Morris AA, Ahmed Y, Stoyanova N, et al.The association between depression andleptin is mediated by adiposity. Psycho-som Med 2012;74:483–488

9. Stroup DF, Berlin JA, Morton SC, et al.Meta-analysis of observational studies inepidemiology: a proposal for reporting.Meta-analysis of observational studies inepidemiology (MOOSE) group. JAMA2000;283:2008–2012

10. J€uni P, Altman DG, Egger M. Systematicreviews in health care: assessing thequality of controlled clinical trials. BMJ2001;323:42–46

11. OwensDK, LohrKN,AtkinsD, et al. AHRQseries paper 5: grading the strength of abody of evidence when comparing medicalinterventionsdagency for healthcare re-search and quality and the effective health-care program. J Clin Epidemiol 2010;63:513–523

12. World Health Organization. The ICD-10classification of mental and behaviouraldisorders: clinical descriptions and diagnosticguidelines. World Health Organization,1992

13. American Psychiatric Association. Di-agnostic and statistical manual of mentaldisorders: DSM-IV-TR. Washington, DC,American Psychiatric Association, 2000

14. StataCorp, Stata Statistical Software: Re-lease 11. 2009

15. StataCorp, Stata Statistical Software: Re-lease 10. 2007

16. Sterne JAC, Newton HJ, Cox NJ. Meta-Analysis in Stata: An Updated Collectionfrom the Stata Journal. College Station, TX,Stata Press, 2009

17. Cooper HM, Hedges LV. The Handbook ofResearch Synthesis. New York, Russell SageFoundation, 1994

18. Gilpin AR. Table for conversion ofKendall’s Tau to Spearman’s Rho withinthe context of measures of magnitude ofeffect for meta-analysis. Educ PsycholMeas 1993;53:87–92

19. Shen Q, Bergquist-Beringer S, Sousa VD.Major depressive disorder and insulin re-sistance in nondiabetic young adults inthe United States: the National Health andNutrition Examination Survey, 1999-2002. Biol Res Nurs 2011;13:175–181

20. Rosenthal, R.Meta-analytic Procedures ForSocial Research. Newbury Park, CA, SagePublications, 1991

21. Borenstein M, Hedges LV, Higgins JPT,Rothstein HR. Introduction to Meta-Analysis.New York, John Wiley & Sons, 2009

22. Cochran WG. The comparison of per-centages in matched samples. Biometrika1950;37:256–266

23. Higgins JP, Thompson SG, Deeks JJ,Altman DG. Measuring inconsistency inmeta-analyses. BMJ 2003;327:557–560

24. Thompson SGS, Sharp SJ. Explainingheterogeneity in meta-analysis: a compar-ison of methods. Stat Med 1999;18:2693–2708

25. Egger M, Davey Smith G, Schneider M,Minder C. Bias in meta-analysis detectedby a simple, graphical test. BMJ 1997;315:629–634

26. Begg CB, Mazumdar M. Operating char-acteristics of a rank correlation test forpublication bias. Biometrics 1994;50:1088–1101

27. Higgins JPT, Green S. Cochrane Handbookfor Systematic Reviews of Interventions. NewYork, John Wiley & Sons, 2011

28. Gilbey MP, Casale C, Lei S, et al. Preva-lence and treatment of depression aredifferentially associated with insulin re-sistance and obesity. 1st Joint Meeting ofthe International Society for AutonomicNeuroscience and the American Auto-nomic Society (ISAN/AAS 2011), Buzios,Rio de Janeiro, Brazil, September 12–16,2011( Abstracts). Clin Auton Res 2011;21:215–308

29. Lawlor DA, Smith GD, Ebrahim S; BritishWomen’s Heart and Health Study. Asso-ciation of insulin resistance with de-pression: cross sectional findings from theBritish Women’s Heart and Health Study.BMJ 2003;327:1383–1384

30. Lawlor DA, Ben-Shlomo Y, Ebrahim S,et al. Insulin resistance and depressivesymptoms in middle aged men: findings



from the Caerphilly prospective cohortstudy. BMJ 2005;330:705–706

31. Chiba M, Suzuki S, Hinokio Y, et al.Tyrosine hydroxylase gene microsatellitepolymorphism associated with insulinresistance in depressive disorder. Metab-olism 2000;49:1145–1149

32. Timonen M, Laakso M, Jokelainen J,Rajala U, Meyer-Rochow VB, Keinänen-Kiukaanniemi S. Insulin resistance anddepression: cross sectional study. BMJ2005;330:17–18

33. Adriaanse MC, Dekker JM, Nijpels G,Heine RJ, Snoek FJ, Pouwer F. Associa-tions between depressive symptoms andinsulin resistance: the Hoorn Study. Dia-betologia 2006;49:2874–2877

34. Holt RI, PhillipsDI, JamesonKA,CooperC,Dennison EM, Peveler RC; HertfordshireCohort Study Group. The relationship be-tween depression and diabetes mellitus:findings from the Hertfordshire CohortStudy. Diabet Med 2009;26:641–648

35. Timonen M, Rajala U, Jokelainen J,Keinänen-Kiukaanniemi S, Meyer-RochowVB, Räsänen P. Depressive symptoms andinsulin resistance in young adult males: re-sults from the Northern Finland 1966 birthcohort. Mol Psychiatry 2006;11:929–933

36. Pearson S, Schmidt M, Patton G, et al. De-pression and insulin resistance: cross-sectionalassociations in young adults. Diabetes Care2010;33:1128–1133

37. Everson-Rose SA, Meyer PM, Powell LH,et al. Depressive symptoms, insulin re-sistance, and risk of diabetes in women atmidlife. Diabetes Care 2004;27:2856–2862

38. Hung YJ, Hsieh CH, Chen YJ, et al.Insulin sensitivity, proinflammatorymarkers and adiponectin in young maleswith different subtypes of depressivedisorder. Clin Endocrinol (Oxf) 2007;67:784–789

39. Kauffman RP, Castracane VD, White DL,Baldock SD, Owens R. Impact of the se-lective serotonin reuptake inhibitor cit-alopram on insulin sensitivity, leptin andbasal cortisol secretion in depressed andnon-depressed euglycemic women of re-productive age. Gynecol Endocrinol2005;21:129–137

40. Kahl KG, Bester M, Greggersen W, et al.Visceral fat deposition and insulin sensi-tivity in depressed women with and

without comorbid borderline personalitydisorder. Psychosom Med 2005;67:407–412

41. Okamura F, Tashiro A, Utumi A, et al.Insulin resistance in patients with de-pression and its changes during the clin-ical course of depression: minimal modelanalysis. Metabolism 2000;49:1255–1260

42. Golden SH, Lee HB, Schreiner PJ, et al.Depression and type 2 diabetes mellitus:the multiethnic study of atherosclerosis.Psychosom Med 2007;69:529–536

43. Krishnamurthy P, Romagni P, Torvik S,et al.; P.O.W.E.R. (Premenopausal, Osteo-porosis Women, Alendronate, Depression)Study Group. Glucocorticoid receptor genepolymorphisms in premenopausal womenwith major depression. Horm Metab Res2008;40:194–198

44. Pan A, Ye X, Franco OH, et al. Insulinresistance and depressive symptoms inmiddle-aged and elderly Chinese: find-ings from the Nutrition and Health ofAging Population in China Study. J AffectDisord 2008;109:75–82

45. Roos C, Lidfeldt J, Agardh CD, et al. In-sulin resistance and self-rated symptomsof depression in Swedish womenwith riskfactors for diabetes: the Women’s Healthin the Lund Area study. Metabolism 2007;56:825–829

46. Schlotz W, Ambery P, Syddall HE, et al.;Hertfordshire Cohort Study Group. Spe-cific associations of insulin resistance withimpaired health-related quality of life inthe Hertfordshire Cohort Study. Qual LifeRes 2007;16:429–436

47. Timonen M, Salmenkaita I, Jokelainen J,et al. Insulin resistance and depressivesymptoms in young adult males: findingsfrom Finnish military conscripts. Psy-chosom Med 2007;69:723–728

48. Nouwen A, Winkley K, Twisk J, et al.;European Depression in Diabetes (EDID)Research Consortium. Type 2 diabetesmellitus as a risk factor for the onset ofdepression: a systematic review and meta-analysis. Diabetologia 2010;53:2480–2486

49. Pan A, Sun Q, Czernichow S, et al. Bi-directional association between de-pression and obesity in middle-aged andolder women. Int J Obes (Lond) 2012;36:595–602

50. Musselman DL, Betan E, Larsen H,Phillips LS. Relationship of depression todiabetes types 1 and 2: epidemiology, bi-ology, and treatment. Biol Psychiatry2003;54:317–329

51. Howren MB, Lamkin DM, Suls J. Associ-ations of depression with C-reactive pro-tein, IL-1, and IL-6: a meta-analysis.Psychosom Med 2009;71:171–186

52. Black PH. The inflammatory response isan integral part of the stress response:implications for atherosclerosis, insulinresistance, type II diabetes and metabolicsyndrome X. Brain Behav Immun 2003;17:350–364

53. DiMatteo MR, Lepper HS, Croghan TW.Depression is a risk factor for non-compliance with medical treatment:meta-analysis of the effects of anxiety anddepression on patient adherence. ArchIntern Med 2000;160:2101–2107

54. Marcus MD, Wing RR, Guare J, Blair EH,Jawad A. Lifetime prevalence of majordepression and its effect on treatmentoutcome in obese type II diabetic patients.Diabetes Care 1992;15:253–255

55. McDowell I. Measuring Health: A Guide toRating Scales and Questionnaires. Newell C,Ed. New York, Oxford University Press,1987

56. Kessler RC. The categorical versus di-mensional assessment controversy in thesociology of mental illness. J Health SocBehav 2002;43:171–188

57. Matthews DR, Hosker JP, Rudenski AS,Naylor BA, Treacher DF, Turner RC. Ho-meostasis model assessment: insulin re-sistance and beta-cell function fromfasting plasma glucose and insulin con-centrations inman. Diabetologia 1985;28:412–419

58. Rabasa-Lhoret R, Bastard JP, Jan V, et al.Modified quantitative insulin sensitivitycheck index is better correlated to hy-perinsulinemic glucose clamp thanother fasting-based index of insulinsensitivity in different insulin-resistantstates. J Clin Endocrinol Metab 2003;88:4917–4923

59. Cobelli C, Caumo A, Omenetto M. Mini-mal model SG overestimation and SI un-derestimation: improved accuracy by aBayesian two-compartment model. Am JPhysiol 1999;277:E481–E488


Kan and Associates

a systematic review and meta-analysis of the association...

Documents