analys

Measuringinconsistencyinmeta-analys

JulianPTHiggins,SimonGThompson,JonathanJDeeks,DouglasGAltman

CochraneReviewshaverecentlystartedincludingthequantityI2tohelpreadersasssthe

esthisnewquantitymean,andwhyis

asssmentofheterogeneitysoimportanttoclinicalpractice?

Systematicreviewsandmeta-analyscanprovide

convincingandreliableevidencerelevanttomany

aspectsofmedicineandhealthcare.1Theirvalueis

especiallyclearwhentheresultsofthestudiesthey

includeshowclinicallyimportanteffectsofsimilar

r,theconclusionsarelessclear

attempttoestablishwhetherstudiesareconsistent,

reportsofmeta-analyscommonlyprentastatisti-

tekstodetermine

whethertherearegenuinedifferencesunderlyingthe

resultsofthestudies(heterogeneity),orwhetherthe

variationinfindingsiscompatiblewithchancealone

(homogeneity).However,thetestissusceptibletothe

developedanewquantity,I2,whichwebelievegivesa

bettermeasureoftheconsistencybetweentrialsina

meta-analysis.

Needforconsistency

Asssmentoftheconsistencyofeffectsacrossstudies

weknow

howconsistenttheresultsofstudiesare,wecannot

determinethegeneralisabilityofthefindingsofthe

,veralhierarchicalsystemsfor

gradingevidencestatethattheresultsofstudiesmust

beconsistentorhomogeneoustoobtainthehighest

grading.2–4

Testsforheterogeneityarecommonlyudto

decideonmethodsforcombiningstudiesandforcon-

cludingconsistencyorinconsistencyoffindings.56But

whatdoesthetestachieveinpractice,andhowshould

theresultingPvaluesbeinterpreted?

Testingforheterogeneity

Atestforheterogeneityexaminesthenullhypothesis

al

teststatistic(Cochran’sQ)iscomputedbysummingthe

squareddeviationsofeachstudy’stimatefromthe

overallmeta-analyticestimate,weightingeachstudy’s

contributioninthesamemannerasinthemeta-

analysis.7Pvaluesareobtainedbycomparingthe

statisticwitha