Biases in Studies of Screening Programs在篩選程序研究的偏見

BiasesinStudiesofScreeningProgramsThomasB.Newman,MD,MPHJune10,2021OverviewIntroductionTNBiasesDefintionsProblemswithobservationalstudiesVolunteerbiasLeadtimebiasLengthbiasStagemigrationbiasPseudodiseaseScreeningtests:TNBiases“Whenyouronlytoolisahammer,youtendtoseeeveryproblemasanail.〞Clinicalcareaccountsfor95%ofspendingbutonly20%ofdeterminantsofhealth*BiggestthreatsarepublichealththreatsInterventionsaimedatindividualsareoveremphasizedbecausetheyaremoreprofitableandweknowhowtodo/sellthem*TeutschSM,FieldingJE.Comparativeeffectiveness:lookingunderthelamppost.JAMA2021;305:2225-6Culturalcharacteristics"Weliveinawasteful,technologydriven,individualisticanddeath-denyingculture."--GeorgeAnnas,NewEnglJMed,1995Whatisscreening?Commondefinition:testingtodetectasymptomaticdiseaseBetterdefinition*:applicationofatesttodetectapotentialdiseaseorconditioninpeoplewithnoknownsignsorsymptomsofthatdiseaseorcondition.Diseasevs.conditionAsymptomaticvs.noknownsignsorsymptoms*Commonscreeningtests.DavidM.Eddy,editor.Philadelphia,PA:AmericanCollegeofPhysicians,1991

ScreeningtestsmaybehistoryquestionsScreeningSpectrumRiskfactorRecognizedsymptomaticdiseasePresymp-tomaticdiseaseUnrecognizedsymptomaticdiseaseDecreasingnumberslabeledandtreatedDecreasingdifficultydemonstratingbenefitExamplesandoverlapUnrecognizedsymptomaticdisease:visionandhearingproblemsinyoungchildren;irondeficiencyanemia,depressionPresymptomaticdisease:neonatalhypothyroidism,syphilis,HIVRiskfactor:hypercholesterolemia,hypertensionSomewherebetween:prostatecancer,ductalcarcinomainsituofthebreast,moreseverehypertensionScreenedNotscreenedMortalityafterRandomizationRD+D-D-D+MortaltiyafterRandomizationEvaluatingStudiesofScreeningIdealStudy:RandomizepatientstobescreenedornotCompareoutcomesinENTIREscreenedgrouptoENTIREunscreenedgroupObservationalstudies:PatientsarenotrandomizedCompareoutcomesinscreenedvs.unscreenedpatientsOramongpatientswithdisease:Compareoutcomesinthosediagnosedbyscreeningvs.thosediagnosedbysymptomsComparestage-specificsurvivalwithandwithoutscreeningKEYDIFFERENCE:Mortalityvs.SurvivalMortality:denominatorisapopulation,mostofwhomnevergetthediseaseSurvival:denominatorispatientswiththediseaseBewareofanystudiesevaluatingscreeningtestsusingsurvivalPossibleBiasesinObservationalStudiesofScreeningTestsVolunteerbiasLeadtimebiasLengthtimebiasStagemigrationbiasPseudodiseaseVolunteerBiasPeoplewhovolunteerforscreeningdifferfromthosewhodonotExamplesHIPMammographystudy:WomenwhovolunteeredformammographyhadlowerheartdiseasedeathratesMulticenterAneurysmScreeningStudy(MASS;Problem6.3)Menaged65-74wererandomizedtoeitherreceiveaninvitationforanabdominalultrasoundscanornot.MASSWithinGroupsResultinInvitedGroupAvoidingVolunteerBiasRandomizepatientstoscreenedandunscreenedOtherwise,trytocontrolforfactors(confounders)associatedwithbothscreeningandoutcomeExamples:familyhistory,levelofhealthconcern,otherhealthbehaviors,baselinehealth/illnessesLeadTimeBias(zero-timebias)Screeningidentifiesdiseaseduringalatentperiod

beforeitbecomessymptomaticIfsurvivalismeasuredfromtimeofdiagnosis,screeningwillalwaysimprovesurvivaleveniftreatmentisineffectiveLeadtimebiasSource:EDITORIAL:FindingandRedefiningDisease.

EffectiveClinicalPractice,March/April1999.Availableat:ACP-Onlineaccessed8/30/02AvoidingLeadTimeBiasOnlyoccurswhensurvivalfromdiagnosisiscomparedbetweendiseasedpersonsScreenedvs.notscreenedDiagnosedbyscreeningvs.bysymptomsAvoidingleadtimebiasMeasuremortality,notsurvivalCountfromdateofrandomizationFollowpatientsforalongtime(20years?)andusetotal,note.g.5-yearsurvivalLengthBias(Differentnaturalhistorybias)ScreeningpicksupprevalentdiseasePrevalence=incidencexdurationSlowlygrowingtumorshavegreaterdurationinpresymptomaticphase,thereforegreaterprevalenceTherefore,casespickedupbyscreeningwillbedisproportionatelythosethatareslowgrowingLengthbiasSource:EDITORIAL:FindingandRedefiningDisease.

EffectiveClinicalPractice,March/April1999.Availableat:ACP-OnlineLengthBiasEarlydetectionHighercurerateSlowergrowingtumorwithbetterprognosis?AvoidingLengthBiasOnlypresentwhensurvivalfromdiagnosisiscomparedANDdiseaseisheterogeneousLeadtimebiasusuallypresentaswellAvoidinglengthbias:ComparemortalityintheENTIREscreenedgrouptotheENTIREunscreenedgroupStudydiseasesubgroupswithauniformnaturalhistoryStagemigrationbiasOldtestsNewtestsStagemigrationbiasAlsocalledthe"WillRogersPhenomenon""WhentheOkiesleftOklahomaandmovedtoCalifornia,theyraisedtheaverageintelligencelevelinbothstates." --WillRogersDocumentedwithcoloncanceratYaleOtherexamplesabound–themoreyoulookfordisease,thehighertheprevalenceandthebettertheprognosisBestreferenceonthistopic:BlackWCandWelchHG.Advancesindiagnosticimagingandoverestimationofdiseaseprevalenceandthebenefitsoftherapy.NEJM1993;328:1237-43.AmoregeneralexampleofStageMigrationBiasVLBW(<1500g),LBW(1500-2499g)andNBW(>2500g)newbornsexposedtoFactorXinuterohavedecreasedmortalitycomparedwiththosenotexposedIsfactorXgood?Maybenot!FactorXcouldbecigarettesmoking!SmokingmovesbabiestolowerbirthweightstrataComparedwithothercausesofLBW(i.e.,prematurity)itisnotasbadStageMigrationBiasLBWVLBWNBWNBWLBWVLBWUnexposedtosmokeExposedtosmokeAvoidingStageMigrationBiasTheharderyoulookfordisease,andthemoreadvancedthetechnologythehighertheprevalence,thehigherthestage,andthebetterthe(apparent)outcomeforthestageBewareofstagemigrationinanystratifiedanalysisCheckOVERALLsurvivalinscreenedvs.unscreenedgroupMoregenerally,donotstratifyonfactorsdistalinacausalpathwaytothefactoryouwishtoevaluate!PseudodiseaseAconditionthatlooksjustlikethedisease,butneverwouldhavebotheredthepatientTypeI:DiseasewhichwouldnevercausesymptomsTypeII:PreclinicaldiseaseinpeoplewhowilldiefromanothercausebeforediseasepresentsInanindividualtreatedpatientitisimpossibletodistinguishpseudodiseasefromsuccessfullytreatedasymptomaticdiseaseTheProblem:TreatingpseudodiseasewillalwayslooksuccessfulTreatingpseudodiseasewillalwaysbeharmfulExample:MayoLungProjectRCToflungcancerscreeningEnrollment1971-769,211malesmokersrandomizedtotwostudyarmsIntervention:chestx-rayandsputumcytologyevery4monthsfor6years(75%compliance)Control:Testsattrialentry,thenarecommendationtoreceivethesametestsannually*Marcusetal.,JNCI2000;92:1308-16MayoLungProjectExtendedFollow-upResults*Amongthosewithlungcancer,interventiongrouphadmorecancersdiagnosedatearlystageandbettersurvival*Marcusetal.,JNCI2000;92:1308-16MLPExtendedFollow-upResults*Interventiongroup:slightincreaseinlung-cancermortality(P=0.09by1996)*Marcusetal.,JNCI2000;92:1308-16Whathappened?After20yearsoffollowup,therewasasignificantincrease(29%)inthetotalnumberoflungcancersinthescreenedgroupExcessoftumorsinearlystageNodecreaseinlatestagetumorsOverdiagnosis(pseudodisease)BlackW.Overdiagnosis:anunderrecognizedcauseofconfusionandharmincancerscreening.JNCI2000;92:1308-16LookingforPseudodiseaseAppreciatethevaryingnaturalhistoryofdisease,andlimitsofdiagnosisImpossibletodistinguishfromsuccessfulcureof(asymptomatic)diseaseinindividualpatie