Preventing Bias in Cluster Randomised Trials 英文参考文献.docVIP

Essay PreventingBiasinClusterRandomisedTrials BrunoGiraudeau1,2,3,4*,PhilippeRavaud4,5,6 1INSERM,CIC202, Tours, France, 2Universite′ Franc?ois Rabelais, Tours, France, 3CHRU deTours, Tours, France, 4INSERM,U738, Paris, France,5AssistancePublique– Ho?pitauxdeParis,Ho?pitalBichat,De′partementd’Epide′miologie,BiostatistiqueetRechercheClinique,Paris,France,6Universite′ Paris7–DenisDiderot,Paris,France Inrandomisedtrials,internalvalidityis definedas‘‘theextenttowhichthedesign and conduct of a study are likely to have preventedbias’’[1,2].Inconductingsuch trials,trialiststrytopreventselectionbias [9]. As well, randomisation of clusters rather than individuals could prevent contamination. As an example, to assess the impact of a screening programme, widespreadpublicityisneededtoencour- age participants to undergo screening. Withindividualrandomisation,communi- cation among participants would induce contamination, but with cluster randomi- SummaryPoints N In CRTs, the comparability of groups is challenged because groupsoftrialparticipantsrather than the participants themselves arerandomised. through randomisation and allocation concealment(definedas‘‘theprocessused toensurethatthepersondecidingtoenter aparticipantintoarandomisedcontrolled trialdoesnotknowthecomparisongroup into which that individual will be allocat- ed’’ [2]) and attrition bias through an N The specific chronology of such trialscompromisesallocationcon- cealment(i.e.,clustersarerecruit- ed and randomised and then participants are recruited), which can induce differential recruit- ment and thus quantitative and qualitative imbalance between groups. sation, it may strengthen participant compliance. Use of the CRT has greatly increased over the past 15 years [10] and has motivatedthepublicationofanextension of the CONSORT Statement for this design [11] because of its particular methodological issues. The main issue is thatobservationsfromthesameclusterare more similar than observations