生物统计学英文课件1

BasicDesignConsiderationsSTAT585EpidemiologyReviewfromSTAT584TypesofstudiesCross-sectionalstudyDataareobtainedfromarandomsampleatonepointintimewhichgivesasnapshotofapopulationProspectivestudyAcohortofindividualsareidentifiedwhoarefreeofaparticulardiseaseunderstudyanddataarecollectedoncertainriskfactorsTheseindividualsarethenfollowedoversomespecifiedperiodoftimetodeterminewhethertheygetdiseaseornotEpidemiologyRetrospectiveorcase-controlstudyIndividualswithdisease(calledcases)andindividualswithoutdisease(calledcontrols)areidentifiedUsingrecordsorquestionnairestheinvestigatorsgobackintimeandascertainexposurestatusandriskfactorsfromtheirpast.Suchdataareusedtoestimatetherelativeriskofdevelopingdiseasebetweenexposedandun-exposed.ClinicaltrialAprospectivestudy(oraninterventionstudy)comparingtheeffectandvalueofinterventionagainstacontrolinhumanbeingsAclinicaltrialisanexperimentwhichinvolvespatientsandisdesignedtoelucidatethemostappropriatetreatmentoffuturepatientsEssentialfeatureisthattheallocationofsubjecttotreatmentisplanned.Contrastthiswithcross-sectionalsurveys,cohortandcase-controlstudieswherein,forexample,thereisnocontroloverwhoisandwhoisnotacigarettesmoker.ClinicalTrialPrerequisitesNeedforthetrialHigherincidenceorpoorprognosisforthediseaseNoexistingtreatmentorlackthereofTheinterventionmusthavepromiseofefficacyTrialquestionmustbeappropriateorunambiguousValidtrialarchitecture(via,randomallocationofpatients,blindedtrial,useofplaceboetc)Appropriateinclusion/exclusioncriteriaFeasibletrialprotocolEffectivetrialadministrationObjectivesandoutcomesofaclinicaltrialPrimaryobjective:Whatistheprimaryquestiontobeanswered?ideallyjustoneimportant,relevanttocareoffuturepatientscapableofbeingansweredPrimaryoutcome(endpoint)ideallyjustonerelativelysimpletoanalyzeandreportshouldbewelldefined;objectivemeasurementispreferredtoasubjectiveone.Forexample,clinicalandlaboratorymeasurementsaremoreobjectivethansayclinicalandpatientimpressionSecondaryobjectivesSecondaryQuestionsotheroutcomesorendpointsofinterestsubgroupanalysessecondaryquestionsshouldbeviewedasexploratorytrialmaylackpowertoaddressthemmultiplecomparisonswillincreasethechanceoffinding“statisticallysignificant”differencesevenifthereisnoeffectavoidexcessiveevaluations;aswellasproblemwithmultiplecomparisons,thismayeffectdataqualityandpatientsupportExample(1)PhysiciansHealthStudy(PHS)Ref:NEnglJMed2005;352:1293-304.Whydothetrial?Randomizedtrialshaveshownthatlow-doseaspirindecreasestheriskofafirstmyocardialinfarctioninmen,withlittleeffectontheriskofischemicstroke.Therearefewsimilardatainwomen.MethodRandomlyassigned39,876initiallyhealthywomen45yearsofageoroldertoreceive100mgofaspirinonalternatedaysorplaceboandthenmonitoredthemfor10yearsforafirstmajorcardiovascularevent(i.e.,nonfatalmyocardialinfarction,nonfatalstroke,ordeathfromcardiovascularcauses).Example(1)PrimaryEndpointTheprimaryendpointwasacombinationofmajorcardiovascularevents,includingnonfatalmyocardialinfarction,nonfatalstroke,anddeathfromcardiovascularcausesThetrialwasinitiallydesignedtohaveastatisticalpowerof86percenttodetecta25percentreductioninthisendpoint.Example(1)SecondaryendpointSecondaryendpointsincludedtheindividualendpointsoffatalornonfatalmyocardialinfarction,fatalornonfatalstroke,ischemicstroke,hemorrhagicstroke,anddeathfromcardiovascularcauses.Example(2)EasternCooperativeOncologygroup(ECOG–1178)Tamoxifen(Soltamox)blocksactionsofestrogenandisusedtotreatandpreventsometypesofbreastcancerTamoxifenvsplaceboPrimary:tumorrecurrence/relapse,disease-freesurvivalSecondary:totalmortalityExample(3)MulticenterInvestigationofLimitationofInfarctionSize(MILIS)Propranololisanon-selectivebetablockerusedinthetreatmentofhypertensionPropranololvsplaceboPrimary:ultimatesizeofanacutemyocardialinfarctionSecondary:leftventricularejectionfractionDealingwithprimaryandsecondaryoutcomesManyadvocatehavingasingleprimaryendpointdrivessamplesizecalculationstestbasedonthisendpointhasa5%typeIerrorrateAllotherendpointsare“secondary”DelineateprimaryandsecondaryoutcomesCanbehardtoadheretoinpracticeForexample,whatifprimaryoutcomeisnotdifferentamonggroups,butallsecondaryoutcomesare?Whattodowithprimaryandsecondaryendpoints?O’Neill,R.(1997)“Secondaryendpointscannotbevalidlyanalyzediftheprimaryendpointdoesnotdemonstrateclearstatisticalsignificance”ControlledClinicalTrials,550–556Davis,C.E.(1997)“Secondaryendpointscanbevalidlyanalyzed,eveniftheprimaryendpointdoesnotprovideclearstatisticalsignificance”ControlledClinicalTrials,557-560O’Neill(1997)Primaryendpointdefinition:“clinicalendpointthatprovidesevidencesufficienttofullycategorizeclinicallytheeffectofatreatmentthatwouldsupportaregulatoryclaimforthetreatment”Secondaryendpoint:“additionalclinicalcharacterizationofatreatmentbutcouldnot,byitself,beconvincingofaclinicallysignificanttreatmenteffect”O’Neill(1997)ArguesthatprimaryandsecondaryoutcomesaregenerallyrelatedAnalysisofsecondaryshouldbeconditionalontheprimaryoutcomeanalysisresultespeciallytruewhen“secondary”outcomesdependdirectlyonprimary(survival)Can’tquantifytheuncertaintyinanalysesdoneafterlookingatresultsDavis(1997)StrictadherencecouldmissimportantandunexpectedresultsArguesthatthemajorproblemismultiplecomparisonissueit’sastatisticalproblem,souseastatisticalsolutionOnesuch“solution”istheBonferroniadjustment(willbediscussedlaterinthecourse)ChoiceofprimaryendpointExample:newtreatmentforHIVpatients;attacksHIVvirus.1.IncreaseinCD4count.2.ViralRNAreduction.Measurestheamountofvirusinthebody3.Timetothefirstopportunisticinfection4.Timetodeathfromanycause5.Timetodeathorfirstopportunisticinfection,whichevercomesfirstOutcomes1&2aregoodforaphaseIItrial(theyareobjectivelymeasurable).Outcome4istheultimateendpoint;mayleadtoobsolescence.Outcome3isnotcomplete.Outcome5isgoodoneinaphaseIIItrial.OpportunisticinfectionsPeoplewithHIVcangetmanyinfections(calledopportunisticinfections,orOIs).Manyoftheseillnessesareveryserious,andtheyneedtobetreated.Somecanbepreventedandmighttakelongtodevelop.Example

Michalowiczetal(NEJM,Nov2,2006)StudyofperiodontaltherapyandbirthoutcomeSeveraloutcomesofinterest:pretermbirth(before37weeks),birthweight,proportionofinfantswhoaresmallforgestationalage,Apgarscores,admissionstoNICU….What’stheproblem?Iftesteachendpointatthe5%level:overallchanceoffindingatleastoneendpointwherethereisasignificantdifferenceislargerthan5%,evenifthetreatmentsareidenticalPronetodistortedreporting(i.e.pickmostsignificant)Goodreference:Pocock(1997)ControlledClinicalTrials,p530-545MultiplicityissuewillbediscussedlaterDealingwithproblemswithmultipleendpointsHaveapre-definedstrategySomeadvocate:allresultspre-written,withresultsfilledinastrialconcludesAlternativeview:needtobeflexibleneedtoallowforunexpectedfindingsbutrecognizepotentialforproblems:typeIerrorrateisnot5%SurrogateendpointsWhydoweusesurrogateendpoint?CanbemeasuredearlierConvenientorlessinvasiveCanbemeasuredmorefrequentlyCanacceleratetheapprovalprocessAdvantages:MayreducethesizeofclinicaltrialsMayshortenthedurationofclinicaltrialsMayreducethecostofclinicaltrialsEndpointsinclinicaltrialsAsurrogateendpointdoesnotdirectlymeasureanyclinicalbenefittopatient,itonlypredictstheoutcomeAmixedsurrogate/clinicalbenefitendpointdirectlymeasuressignificantbenefittopatientandpredictsanadditional,moresubstantialbenefittopatientAclinicaloutcomedirectlymeasuressubstantialclinicalbenefittopatientEndpointsinappropriatelycharacterizedassurrogatesQualityoflifeItisanoutcomemeasure(notasurrogateendpoint)MorbidityscaleItisaclinicalbenefitendpoint(notasurrogateendpoint)Whenistheuseofsurrogateendpointsjustified?ScreeningForpromisingnewtherapiesEvaluationofbiologicalactivityinphaseI/IItrialsCautioninusingsurrogateendpoints:Usingbiologicalmarkersasasurrogateendpoint,onemayobtainmisleadingfalsepositiveorfalsenegativeconclusionwhenassessingtreatmenteffectsoflongertermclinicaloutcomeRequirements:Beforeasurrogateendpointcanreplaceaprimaryendpoint,itmustbeformallyvalidatedExamplesofFDAapprovalofDrugsusingSurrogatesLowercholesterolwithoutevidenceofsurvivalbenefitLowerbloodpressurewithoutevidenceofbenefitforstroke,MI,congestiveheartfailureorsurvivalIncreasedbonedensitywithoutevidenceofdecreasedfracturesinosteoporosisExamplesofFDAApprovalofDrugsusingSurrogatesIncreasedcardiacfunctionincongestiveheartfailurewithoutevidenceofsurvivalbenefitDecreasedrateofarrhythmiaswithoutevidenceofsurvivalbenefitLowerbloodglucoseandglycosylatedhemoglobinwithoutevidenceaboutdiabeticcomplicationsorsurvivalbenefit.Prentice(1989)definitionofsurrogateendpointSupposef(T|Z)istheconditionaldistributionofthetrueendpointTgiventhetreatmentassignmentZ,andSisthesurrogateendpoint.Allfourofthefollowingconditionsmustbemet:Example

Osteoporosis(Riggsetal,NEJM,1990)Bonelossinpost-menopausalwomenleadstoincreaseriskoffractureSodiumFluoridestimulatesboneformationandincreasedbonemass(double)HypothesisWillFluoridetreatmentdecreaserateofvertebralfractures?DesignRandomized,doubleblind,placebo-controlled202postmenopausalwomenrandomizedAllreceivedcalciumsupplementOsteoporosisFluoridetrialresultsFluorideincreasedbonedensityby35%35%(p=0.0001)inspine12%(p=0.0001)infemoralneckFluoridedecreasedbonedensityby4%inwrist(p=0.02)VertebralfractureshigheronFluoride(F163,P136,p<0.05)Non-vertebralfractureshigheronFluoride(72vs24;p=0.01)Fluoridewasconcludedtobenoteffectiveasatreatmentforpost-menopausalosteoporosisThoughincreaseinbonedensitywasnoticed,vertebralfractureriskremained.CommentsAllfourconditionsimplythatthesurrogatecapturesthefullofeffectoftreatmentontheprimaryendpoint,verystringentanddifficulttoverify.Canfitamodel andtestthenullH0:b1=0toverifythelastconditionP4AsignificanttreatmenteffectontheprimaryendpointafteradjustmentforSleadstotheconclusionthatSisapoorsurrogate.Ontheotherhand,failingtorejectisinadequatetovalidateSasagoodsurrogate

PurposeofcontrolgroupToallowdiscriminationofpatientoutcomescausedbyexperimentalinterventionfromthosecausedbyotherfactorsNaturalprogressionofdiseaseObserver/patientexpectationsOthertreatmentFaircomparisonsNecessarytobeinformativeChoiceofcontrolgroupGoalsofControlledClinicalTrialsTypesofControlGroupsSignificanceofControlGroupAssaySensitivityICHE-10,ChoiceofcontrolgroupsinclinicaltrialsConsiderationsinChoiceofControlGroupAvailablestandardtherapiesAdequacyofthecontrolevidenceforthechosendesignEthicalconsiderationsSignificanceofcontrolgroupInferencedrawnfromthetrialEthicalacceptabilityofthetrialDegreetowhichbiasisminimizedTypeofsubjectsKindofendpointsthatcanbestudiedCredibilityoftheresultsAcceptabilityoftheresultsbyregulatoryauthoritiesOtherfeaturesofthetrial,itsconduct,andinterpretationTypesofcontrolsExternalHistoricalConcurrent,notrandomizedInternalandconcurrentNotreatmentPlaceboDose-responseActive(Positive)controlMultipleBothanActiveandPlaceboMultipledosesoftestdrugandofanactivecontrolUseofplacebocontrolThe“placeboeffect”iswelldocumentedCouldbeNotreatment+placeboStandardcare+placeboMatchedplacebosarenecessarysopatientsandinvestigatorscannotdecodethetreatmentE.g.VitaminCtrialforcommoncoldPlacebowasused,butwasdistinguishableManyonplacebodroppedoutofstudyThosewhoknewtheywereonvitaminCreportedfewercoldsymptomsanddurationthanthoseonvitaminwhodidn'tknowHistoricalControlsAnewtreatmentusedinaseriesofsubjectsOutcomecomparedwithpreviousseriesofcomparablesubjectsNon-randomized,non-concurrentRapid,inexpensive,goodforinitialtestingofnew treatmentsTwosourcesofhistoricalcontroldata:Literature

SubjecttopublicationbiasDatabaseHistoricalcontrolVulnerabletobiasChangesinoutcomeovertimemaycomefromchangein:underlyingpatientpopulationscriteriaforselectingpatientspatientcareandmanagementperipheraltotreatmentdiagnosticorevaluatingcriteriaqualityofdataavailableChangeindefinitionTimetrendAge-adjustedDeathRatesforSelectedCauses:UnitedStates,1950-76HistoricalControlsTendtoexaggeratethevalueofanewtreatmentLiteraturecontrolsparticularlypoorEvenhistoricalcontrolsfromaprevioustrialinthesameinstitutionororganizationmaystillbeproblematicPocock(1977,BritMedJ)In19studieswherethesametreatmentwasusedintwoconsecutivetrials,differencesinsurvivalrangedfrom46to24,withfourdifferencesbeingstatisticallysignificantAdjustmentforpatientselectionmaybemade,butallotherbiaseswillremainConcurrentcontrolsNotrandomizedOnechosenfromthesamepopulationasthetestgroupPatientscompared,treatedbydifferentstrategies,sameperiodAdvantageEliminatetimetrendDataofcomparablequalityDisadvantageSelectionBiasTreatmentgroupsnotcomparableCovarianceanalysisnotadequateBiasinconcurrentcontrolstudyTypesMagnitudeofeffectsFalsepositiveSourcesPatientselection

ReferralpatternsRefusalsDifferenteligibilitycriteriaExperimentalenvironment

Diagnosis/stagingSupportivecareEvaluationmethodsDataqualityRandomizedcontrolstudyReference:Byaretal.(1976) NewEnglandJournalofMedicinePatientsassignedatrandomtoeithertreatment(s)orcontrolConsideredtobe“GoldStandard”AdvantagesofRandomized

ControlClinicalTrial1. Randomization"tends"toproducecomparablegroupsDesign

SourcesofImbalanceRandomized ChanceConcurrent Chance&SelectionBias (Non-randomized)Historical Chance,SelectionBias, (Non-randomized) &TimeBias2. RandomizationproducesvalidstatisticaltestsReference:Byaretal(1976)NEJMDisadvantagesofRandomizedControlClinicalTrial1. GeneralizableResults?Subjectsmaynotrepresentgeneralpatientpopulation–volunteereffect2. RecruitmentTwiceasmanynewpatients3. AcceptabilityofRandomizationProcessSomephysicianswillrefuseSomepatientswillrefuse4. AdministrativeComplexityBiasofNon-RCT’sExample-Peto(1979)Biomedicine TrialsofanticoagulanttherapyDesign #Patients P<0.05 ObservedEffect18Historical 900 15/18 50%8Concurrent 3000 5/8 50%6Randomized 3000 1/6 20%BiasesFalsepositivesMagnitudeofeffectEthicsofrandomizationStatistician/clinicaltrialspecialistmustsellbenefitsofrandomizationEthics

ÞMDshoulddowhathethinksisbestforhispatientTwoMD'smightethicallytreatsamepatientquitedifferentlyChalmers&Shaw(1970)AnnalsNewYorkAcademyofScience

1. IfMD"knows"besttreatment,shouldnotparticipateintrial 2. Ifindoubt,randomizationgiveseachpatientequalchanceto receiveoneoftherapies(i.e.best) 3. MoreethicalwayofpracticingmedicineEthicsofrandomizationByaretal.(1976)NEJM 1. RCT

Þhonestadmissionbestisnot known! 2. RCTisbestmethodtofindout! 3. Reducesriskofbeingoninferior treatment 4. ReducesriskforfuturepatientsEthicsofrandomizationClassicExample- Reference:Silverman(1977)ScientificAmerican 1. Highdoseoxygentoprematureinfantswas commonpractice

2. Suspicionaboutfrequencyofblindness 3. RCTshowedhighdosecauseofblindnessComparingtreatmentsFundamentalprincipleGroupsmustbealikeinallimportantaspectsandonlydifferinthetreatmenteachgroupreceivesInpracticalterms,“comparabletreatmentgroups”means

“alikeontheaverage”RandomizationEachpatienthasthesamechanceofreceivinganyofthe

treatmentsunderstudyAllocationoftreatmentstoparticipantsiscarriedoutusingachancemechanismsothatneitherthepatientnorthephysicianknowinadvancewhichtherapywillbeassignedBlindingAvoidanceofpsychologicalinfluenceFairevaluationofoutcomesRandomizedPhaseIIIExperimentalDesignsAssume:PatientsenrolledintrialhavesatisfiedeligibilitycriteriaandhavegivenconsentBalancedrandomization:eachtreatmentgroupwillbeassignedanequalnumberofpatientsIssueDifferentexperimentaldesignscanbeusedtoanswerdifferenttherapeuticquestionsCommonlyUsedPhaseIIIDesignsParallelWithdrawalGroup/ClusterRandomizedConsentCrossOverFactorialLargeSimpleEquivalence/Non-inferioritySequentialStudyPopulationWhatisthestudypopulation?Subsetofthegeneralpopulationdeterminedbytheeligibilitycriteria GeneralPopulation

eligibilitycriteria StudyPopulation enrollment StudySample observedEligibilitycriteriaDefineinadvanceNeedtodescribewhoweintendtostudyPrecisionrelatedtoimportanceCharacterizepopulationImpactofresultsReplicationofstudyEaseofrecruitmentRiskoreventratesBiasedsampledoesnotimplybiasedtrialClinicalmeasurementsinPhaseII/IIItrialsEfficacySafetyAdverseeventsLaboratorymeasurementsVitalSignsECGsParalleldesignScreen

TrtARandomize- TrtBH0:Avs.BAdvantageSimple,GeneralUseValidComparisonDisadvantageFewQuestions/StudyExample–Table8.2.1Consideraclinicaltrialconductedinordertoevaluatethesafetyofaninjectabledosageformofadrugproduct(denotedbytreatmentA)thatiscomparedtoaplacebo(denotedbytreatmentB)insubjectsundergoingstressechocardiography.Thistrialwasamulticenter,single-blind,andrandomizedstudy.Theprimarysafetyvariablesincludedincidenceofadverseeventsandchangesinlaboratoryparameters.Table8.2.1liststhepartialdataofpre-a