Académique Documents
Professionnel Documents
Culture Documents
trials
withmultiplecoprimaryendpoints
TakashiSozu,Ph.D.
KyotoUniversitySchoolofPublicHealth
TomoyukiSugimoto,Ph.D.
HirosakiUniversityGraduateSchoolofScienceandTechnology
ToshimitsuHamasaki,Ph.D.
OsakaUniversityGraduateSchoolofMedicine
December16,2011
JointMeetingofthe2011TaipeiInternationalStatisticalSymposium
and7thConferenceoftheAsianRegionalSectionoftheIASC
Outline
Backgroundandobjectives
Statisticalsettings
Derivationofpowerformula
Behaviorofsamplesizes
Conclusion
Background
Clinicaltrialsoftenemploytwoormoreprimary
endpoints.
Amajorconcerniswhetherornotclinicaltrials
shouldachievestatisticalsignificanceonallofthe
multipleprimaryendpoints (i.e.,coprimaryendpoints).
StatisticalPrinciplesforClinicalTrials
ICH(1998)
MultipleEndpointExpertTeam(PhRMA)listed20
diseaseswhereregulatoryagencieshaverequired
coprimaryendpoints.
Offenetal.(2007)
3
Common solutionsforthemultiplicity
Compositevariable ICH(1998)
Thisapproachaddressesthemultiplicityproblem
withoutrequiringadjustmenttomultiplicity.
Aclinicallymeaningfulandvalidatedvariableisnot
alwaysavailable.
Interpretationofthevariableisnoteasy.
Assumingtheindependencyamongendpointsand
increasingthepowerforeachendpoint
Thepowerissimplydefined. EatonandMuirhead(2006)
Thepowerwouldbeunderevaluated.
(Thesamplesizeisoversized.)
4
Existingapproaches forpowerand
samplesizedetermination
Continuous (Normal)
Xiongetal.(2005)
Sozuetal.(2006,2011)
Eaton,Muirhead(2006)
Sugimotoetal. (2011)
Binary
IBC2010
(2010.12.6)
Sozuetal.
(submitted)
Song(2009)
Sozuetal.(2010,2011)
Hamasakietal.
(submitted)
MCP2011(2011.8.31)
MCP2011(2011.8.31)
Hamasakietal.
(submitted)
Ordinal
Timetoevent
5
Objective
Wediscussthepowerandsamplesizedetermination
forsuperioritycomparativeclinicaltrialswithmultiple
coprimaryendpoints(forachievingstatistical
significanceforalloftheendpoints).
Continuous
(Normal)
Cont.and Binary
Binary
6
Commonstepsofourresearch
Definetheresponsevariablesandassociation
measures(correlations)amongthem
Calculatethecorrelationcoefficientsamongthe
teststatisticstoderiveapowerformula
Evaluatethebehaviorsofsamplesizes
Associationmeasuresamongendpoints
Scale
Distribution
Cont.
Multivariate
(Normal) Normal(MN)
Associationmeasure(s)
Correlationcoefficient(r)
Binary
Multivariate
Bernoulli(MB)
Cont.
and
Binary
(1)rofMBdist.(MB.
corr.)
(2)Oddratio(OR)
(3)rofalatentMNdist.
(Tetrachoriccorr.)
MN
(latentdistribution)
rofalatentMNdist.
(Biserialcorr.)
8
Relationshipsamongassociation
measuresforbinaryvariables
(1)
MB.corr.
Joint
Prob.
(2)
OR
Bahadur(1961)
leCessie(1994)
Dale(1986)
Emrich,
Piedmonte(1991)
Prentice(1988)
(3)
T.corr.
Ajointprobabilitycanbeestimated
iftheindividualdataofendpointsareavailable.
Biserialmodelformixedcont.andbinary
variables
X2
Latentbivariate
normaldistribution
1
Y2:observedas
abinaryvariable
cutoff
point
0 BinaryPrimaryVariable
X1
Y1:observedasanormalvariable
NormalPrimaryVariable
Associationmeasure
=corr(X1,X2)
Pearson(1909)
Tate(1955a,1955b)
10
Hypothesistesting
DifferenceinmeansforNormalPrimaryVariable
Differenceinsuccessprobabilitiesfor
BinaryPrimaryVariable
Parameterspace
meansanimprovement
fortwoendpoints
foratlestone
forall
11
Testingmethods
NormalPrimaryVariable
Xiongetal.(2005)
(1)Ztest:Knownvariance
(2)Ttest:Unknownvariance Sozuetal.(2006,2011)
BinaryPrimaryVariable
(1)(2)Chisquaretestwithout/withcc
(3)(4)Arcsinetransformationwithout/withcc
(cc:continuitycorrection)
(5)Fishersexacttest
(6)Testbasedonlogtransformedrelativerisks
(15)Sozuetal.(2010) (6)Hamasakietal.(submitted)
12
Powerformula:
Asymptoticnormaltest(exceptforTtest)
Overallpower=
:Teststatistic
:Criticalvalue(:Significancelevel)
Transformintothestandardizedstatistics.
diagonal=1,offdiagonalisgivenbyafunction
oftheassociationmeasureoftheresponsevariables.
ThepowercanbecalculatedfromtheCDFof
Sozuetal.(2010,2011)
13
Powerformula: Ttest
Overallpower=
:Teststatistic
:Criticalvalue
CalculatethepowerusingaMonteCarlo
integration
GenerateWishartrandomnumbersfor
variancecovariancematrix
Calculatetheconditionalpower
Repeattheabovestepsandcalculatethemean
oftheconditionalpower
Sozuetal.(2006,2011)
14
Powerformula:Fishersexacttest
Overallpower=
:onesidedpvalue
CalculatethepowerusingaMonteCarlo
integration.
Generaterandomnumbersfortheresponse
variablesfromtheassumeddistribution
Sozuetal.(2010,2011)
15
Behaviorofsamplesizesfor
NormalPrimaryVariables:Ztest
Decreaseinsamplesize=
Power=80%
samplesizeateachr
samplesizeatr=0.0
Power=90%
K=2
K=3
11%
8%
11%
16%
r:correlationcoefficient(compoundsymmetry)
16
Behaviorofsamplesizes
Sozuetal.(2006,2011)
forNormalPrimaryVariables
Thesamplesizethecorrelation
whentheeffectsizesareapproximatelyequalamong
endpoints
ThedecreaseinSSisdeterminedbythefollowing
fourdesignparameters:
(1)Significancelevelofalpha
(2)Targetpower
(3)Effectsizeratios ofendpoints
(4)Correlationsamongendpoints
Sugimotoetal.(2011)
Thesamplesize(SS)basedonTtest(unknownvariance) +1
theSSbasedonZtest(knownvariance).
17
Behaviorofsamplesizes
forotherPrimaryVariables
BinaryPrimaryVariables:
Sozuetal.(2010,2011)
Thefeaturesofthetestingmethodsaresimilarto
thecaseofsingleendpoint(K=1).
e.g.TheArcsinetransformationwithCCprovides
samplesizesapproximatelyequaltothose
obtainedbyFishersexacttest.
MixedNormalandBinaryVariables:
Thedecreaseinthesamplesizeisrelativelysmallas
comparedtothecaseofNormalPrimaryVariable.
18
Convenientformulaforsamplesize
SamplesizeformulaforNormalPrimaryVariables
Single
endpoints(K=1)
isthefunctionofthefourdesignparameters
AValuecanbeobtainedfromthenumericaltable.
:Thenumberofsubjectsofeachgroup
Arequiredsamplecanbecalculatedwithoutusing
astatisticalsoftware.
Sugimotoetal.(2011)
19
Annumericalexample:K=2
(Correlationbetweentwoendpoints)
Numerical
table
20
Exampleofnumericaltablefor
Curveof
22
Conclusions
Weintroducedthemethodofpowerandsample
sizecalculationsformultiplecoprimaryendpoints
forachievingstatisticalsignificanceforallofthe
endpoints.
Itisimportanttoconsiderassociationsamong
endpointsintosamplesizecalculationwhen
theendpointsare(positively)correlatedand
theeffectsizes(i.e.,thecorrespondingindividualpowers)
areapproximatelyequalamongtheendpoints.
23
Thankyouverymuchforyourattention
Search IBC2012
ThisresearchispartiallysupportedbytheMinistryofEducation,Culture,Sports,
ScienceandTechnology,GrantinAidforYoungScientists(B),No.22700290.
24
References1
BahadurRR.Arepresentationofthejointdistributionofresponsestondichotomousitems.In
StudiesinItemAnalysisandPrediction,Vol.VI,StanfordMathematicalStudiesintheSocial
Sciences, SolomonH(ed.).StanfordUniversityPress:Stanford,CA,1961;158168.
BartlettMS.Theuseoftransformations.Biometrics 1947;3:3952.
BlochDA,LaiTL,SuZ,TubertBitterP.Acombinedsuperiorityandnoninferiorityapproachto
multipleendpointsinclinicaltrials.StatisticsinMedicine 2007;26:11931207.
BreedveldFC,WeismanMH,KavanaughAF,CohenSB,PavelkaK,van VollenhovenR,SharpJ,
PerezJL,SpencerGreenGT.ThePREMIERstudy:Amulticenter,randomized,doubleblind
clinicaltrialofcombinationtherapywithadalimumabplusmethotrexateversusmethotrexate
aloneoradalimumabaloneinpatientswithearly,aggressiverheumatoidarthritiswhohadnot
hadpreviousmethotrexatetreatment.Arthritis&Rheumatism2006;51,2637.
CapizziT,ZhangJ.Testingthehypothesisthatmattersformultipleprimaryendpoints.Drug
InformationJournal 1996;30:949956.
CasagrandeJT,PikeMC,SmithPG.Animprovedapproximateformulaforcalculatingsample
sizesforcomparingtwobinomialdistributions.Biometrics 1978;34:483486.
ChowSC,ShaoJ,WangH.SampleSizeCalculationsinClinicalResearch (2nd~edn).Chapman&
Hall/CRC,BocaRaton,FL,2007.
25
References2
ChuangSteinC,StryszakP,DmitrienkoA,OffenW.Challengeofmultiplecoprimary
endpoints:Anewapproach.StatisticsinMedicine 2007;26:11811192.
CommitteeforMedicinalProductsforHumanUse(CHMP).GuidelineonMedicinalProducts
fortheTreatmentofAlzheimer'sDiseaseandotherDementias (CPMP/EWP/553/95).EMEA:
London,2008.
CommitteeforProprietaryMedicinalProducts(CPMP).PointstoConsideronMultiplicityIssues
inClinicalTrials (CPMP/EWP/908/99).EMEA:London,2002.
leCessieS,vanHouwelingenJC.Logisticregressionforcorrelatedbinarydata.Applied
Statistics 1994;43:95108.
ChowSC,ShaoJ,WangH.SampleSizeCalculationsinClinicalResearch (2ndedn).Chapman
andHall/CRC:BocaRaton,FL,2007.
DaleJR.Globalcrossratiomodelsforbivariate,discrete,orderedresponses.Biometrics 1986;
42:909917.
EatonML,MuirheadRJ.Onamultipleendpointstestingproblem.JournalofStatistical
PlanningandInference 2007;137:34163429.
EmrichLJ,PiedmonteMR.Amethodforgeneratinghighdimensionalmultivariatebinary
variates.TheAmericanStatistician 1991;45:302304.
26
References3
FoodandDrugAdministration.DaftGuidanceforIndustry.IrritableBowelSyndrome:Clinical
EvaluationofProductsforTreatment.CenterforDrugEvaluation andResearch,FoodandDrug
Administration,Rockville,MD,March2010.
GenzA.Numericalcomputationofmultivariatenormalprobabilities.JournalofComputational
andGraphicalStatistics 1992;1:141150.
HoTW,FerrariMD,DodickDW,GaletV,KostJ,FanX,LeibenspergerH,FromanS,AssaidC,
LinesC,KoppenH,WinnerPK.EfficiencyandtolerabilityofMK0974(telcagepant)aneworal
antagonistofcalcitoningenerelatedpeptidereceptor,comparedwithZolmitriptanforacute
migraine:arandomised,placebocontrolled,paralleltreatmenttrial.TheLancet 2008;37:
21152123.
HungHMJ,WangSJ.Somecontroversialmultipletestingproblemsinregulatoryapplications.
JournalofBiopharmaceuticalStatistics 2009;19:111.
InternationalConferenceonHarmonization(ICH)ofTechnicalRequirementsforRegulationsof
PharmaceuticalsforHumanuse.ICHTripartiteGuidelineE9Documents,StatisticalPrinciples
forClinicalTrials,5February1998.
JohnsonME.MultivariateStatisticalSimulation.NewYork:JohnWiley&Sons,1987.
27
References4
JohnsonNL,KotzS.DistributionsinStatistics:ContinuousMultivariateDistributions.NewYork:
JohnWiley&Sons,1972.
JuliousSA.Designingclinicaltrialswithuncertainestimatesofvariability.Pharmaceutical
Statistics2004;3:261268.
JuliousSA.SampleSizesforClinicalTrials.BocaRaton,FL:Chapman&Hall,2009.
KordzakhiaG,SiddiquiO,HuqueMF.Methodofbalancedadjustmentintestingcoprimary
endpoints.StatisticsinMedicine 2010;29:20552066.
LevJ.Thepointbiserialcoefficientofcorrelation.TheAnnalsofMathematicalStatistics 1949;
20:125126.
MachinD,CampbellM,TanSB,TanSH.SampleSizeTablesforClinicalStudies.3rded.
Chichester,UK:WilyBlackwell,2009.
MiwaA,HayterJ,KurikiS.Theevaluationofgeneralnoncentredorthantprobabilities.Journal
oftheRoyalStatisticalSociety,SeriesB2003;65:223234.
MolenberghsG,GeysH,BuyseM.Evaluationofsurrogateendpointsinrandomized
experimentswithmixeddiscreteandcontinuousoutcomes.StatisticsinMedicine 2001;20,
30233038.
28
References5
OffenW,ChuangSteinC,DmitrienkoA,LittmanG,MacaJ,MeyersonL,MuirheadR,Stryszak
P,BoddyA,ChenK,CopleyMerrimanK,DereW,GivensS,HallD,HenryD,JacksonJD,Krishen
A,LiuT,RyderS,SankohAJ,WangJ,YehCH.Multiplecoprimaryendpoints:medicaland
statisticalsolutions.DrugInformationJournal 2007;41:3146.
PatelHI.Comparisonoftreatmentsinacombinationtherapytrial.Journalof
BiopharmaceuticalStatistics 1991;1:171183.
PearsonK.Mathematicalcontributionstothetheoryofevolution.VII.Onthecorrelationof
charactersnotquantitativelymeasurable.PhilosophicalTransactionsoftheRoyalSociety,
SeriesA 1900;195:147.
PearsonK.OnanewmethodfordeterminingthecorrelationbetweenameasuredcharacterA
andacharacterB.Biometrika1909; 7:96105.
PeskindER,PotkinSG,PomaraN,OttBR,GrahamSM,OlinJT,McDonaldS.Memantine
treatmentinmildtomoderateAlzheimerdisease:A24weekrandomized,controlledtrial.
AmericanJournalofGeriatricPsychiatry 2006;14:704715.
PhillipsA,EbbuttA,FranceL,MorganD.TheInternationalconferenceonharmonization
guidelineStatisticalPrinciplesforClinicalTrials:Issueinapplyingtheguidelineinpractice.
DrugInformationJournal 2000;34:337348.
29
References6
PlackettRL.Aclassofbivariatedistributions.JournaloftheAmericanStatisticalAssociation
1965;60:516522.
PocockSJ.Clinicaltrialswithmultipleoutcomes:astatistical perspectiveontheirdesign,
analysisandinterpretation.ControlledClinicalTrials1997;18:530545.
PongA,SheinChungC.Statistical/practicalissuesinclinicaltrials.DrugInformationJournal
1997;31:11671174.
PrenticeRL.Correlatedbinaryregressionwithcovariatesspecifictoeachbinaryobservation.
Biometrics1988;44:10331048.
RgersSL,FarlowMR,DoodyRS,MohsR,FriedhoffLT,TheDonepezilStudyGroup.A24week,
doubleblind,placebocontrolledtrialofdonepezilinpatientswithAlzheimersdisease.
Neurology1998;50:136145.
RoslerM,AnandR,CicinSainA,GauthierS,AgidY,DalBiancoP,StahelinHB,HartmanR,
GharabawiM.EfficacyandsafetyofrivastigmineinpatientswithAlzheimersdisease:
Internationalrandomisedcontrolledtrial.BritishMedicalJournal1999;318:633640.
30
References7
RudnickMR,DavidsonC,LaskeyW,StaffordJL,SherwinPF.VALOR TrialInvestigators.
Nephrotoxicityofiodixanolversusioversolinpatientswithchronickidneydisease:The
VisipaqueAngiography/InterventionswithLaboratoryOutcomesinRenalInsufficiency
(VALOR)Trial.AmericanHeartJournal2008;156:776782.
SahaiH,KhurshidA.Formulaeandtablesforthedeterminationofsamplesizesandpowerin
clinicaltrialsfortestingdifferencesinproportionsforthetwosampledesign:areview.
StatisticsinMedicine1996;15:121.
SankohAJ,DAgostinoRB,HuqueMF.Efficacyendpointselectionandmultiplicityadjustment
methodinclinicaltrialswithinherentmultipleendpointissues.StatisticsinMedicine2003;22:
31333150.
SASInstitute,Inc.,SASOnlineDoc9.1.3.SASInstitute,Inc.:Cary,NC,2006.
SongJM.Samplesizeforsimultaneoustestingofratedifferencesinnoninferioritytrialswith
multipleendpoints.ComputationalStatistics&DataAnalysis2009;53:12011207.
SozuT,KanouT,HamadaC,YoshimuraI.Powerandsamplesizecalculationsinclinicaltrials
withmultipleprimaryvariables.JapaneseJournalofBiometrics2006;27:8396.
31
References8
SozuT,SugimotoT,HamasakiT.Samplesizedeterminationinclinicaltrialswithmultipleco
primarybinaryendpoints.StatisticsinMedicine2010;29:21692179.
SugimotoT,SozuT,HamasakiT.Aconvenientformulaforsamplesizecalculationsinclinical
trialswithmultiplecoprimarycontinuousendpoints.PharmaceuticalStatistics,toappear.
TariotPN,SolomonPR,MorrisJC,KershawP,LilienfeldS,DingC,TheGalantamineUSAStudy
Group.A5month,randomized,placebocontrolledtrialofgalantamineinAD.Neurology
2000;54:22692276.
TateRF.Correlationbetweenadiscreteandacontinuousvariable.Pointbiserialcorrelation.
TheAnnalsofMathematicalStatistics1954;25:603607.
TateRF.Thetheoryofcorrelationbetweentwocontinuousvariablewhenoneisdichotomized.
Biometrika1955a.42:205216.
TateRF.Applicationsofcorrelationmodelsforbiserialdata.JournaloftheAmericanStatistical
Association1955b.50:10781095.
UryHK.Continuitycorrectedapproximationstosamplesizeorpowerwhencomparingtwo
proportions:Chisquaredorarcsine?TheStatistician1981;30:199203.
WaltersDE.Indefenceofthearcsineapproximation.TheStatistician1979;28:219222.
32
References9
WinbladB,KilanderL,ErikssonS,MinthonL,BatsmanS,WetterholmfAL,JanssonBlixtC,
HaglundA.DonepezilinpatientswithsevereAlzheimer'sdisease:doubleblind,parallelgroup,
placebocontrolledstudy.TheLancet2006;367:10571065.
YatesF.Contingencytablesinvolvingsmallnumbersandthe2test.JournaloftheRoyal
StatisticalSociety,Supplement1934;1:217235.
XiongC,YuK,GaoF,YanY,ZhangZ.Powerandsamplesizeforclinicaltrialswhenefficacyis
requiredinmultipleendpoints:applicationtoanAlzheimerstreatmenttrial.ClinicalTrials
2005;2:387393.
33
Backup
34
Whymultipleendpointsarerequired?
(1) Lackofaconsensusonasinglemostimportant
variablefromthemedicalperspective
(2) Noclearaetiologyofdiseases
(3)Adiseaseconditionischaracterizedinmulti
dimensionalways
Pocock(1997)
PongandSheinChung(1997)
Sankohetal.(2003)
ChuangSteinetal.(2007)
35
StatisticalPrinciplesforClinicalTrials
2.2.2PrimaryandSecondaryVariables
Thereshouldgenerallybeonlyoneprimaryvariable.
2.2.5MultiplePrimaryVariables
Itmaysometimesbedesirabletousemorethanone
primaryvariable,eachofwhich(orasubsetof
which)couldbesufficienttocovertherangeof
effectsofthetherapies.
ICH(1998)
36
Examplesofcoprimaryendpoints
MultipleEndpointExpertTeam(PhRMA)listed20
diseaseswhereregulatoryagencieshaverequiredco
primaryendpoints.
Frequency
Offenetal.(2007)
Thenumberofendpoints
37
Exampleofaclinicaltrial
Arandomized,paralleltreatment,placebo
controlled,doubleblindtrial
Acutemigraine
Fivecoprimaryendpointsareconsidered.
(1)painfreedom
(2)painrelief
(3)phonophobia
(4)photophobia
(5)nausea
Hoetal.(2008)
38
Exampleofguidance
IrritableBowelSyndrome
Primaryendpoint
FDA(2010)
(1)Abdominalpain (11points:0to10)
(2)IBSC(Constipation):stoolfrequency
(2)IBSD(Diarrhea):stoolconsistency
TheBristolstoolFormScale(sevenlevels)
http://www.e
chiken.com/shikkan/kabinsei.htm
39
Rsler etal.(1999;BMJ)
40
Threepossibleclinicalscenarios
Showingsignificancefor
Sankohetal.(2003)
(1)all primaryvariables
coprimaryendpoints
Offenetal.(2007)
reversemultiplicityproblem
(2)majority oftheprimaryvariables
(Theconstitutionofthemajorityisdefinedintheprotocol)
(3)oneormoreoftheprimaryvariables
alternativeprimaryendpoints
Offenetal.(2007)
41
Featureofthreeassociation
measures
Pros
Parameterofthe
B.corr
assumeddistribution
Norestrictedrange
from0toinfinity
OR
Adirectextensionto
aglobalcrossratio
L.corr
Norestrictedrange
from1to1
Cons
Restrictedrange
Ascaledepends
stronglyonresponse
probabilities
Iterativecalculations
arenecessarytospecify
thevaluefromavalue
ofothermeasure
42
Assumptionsforresponsevariables
Biserialmodel
NPV(X)isobservedasX(X=X)
BPV(Y)isobtainedbyadichotomizedofX(X>Y)
Pointbiserialmodel
Pearson(1903)
NPV(X)isdistributedasamixednormaldistribution
BPV(Y)isdistributedasaBernoullidistribution
43
Standardizedteststatistics
44
Correlationbetweenstandardized
teststatistics
45
Samplesizecalculation
Specifythevalueofparameters.
NPV:meanandvariance
BPV:successprobability
Specifythevalueof .
(Considerequalsamplesizes:)
Chooseastartingvalueofandcalculatethepower.
Repeattheabovestepsbygraduallyincreasing.
Endtheoperationwhenthecalculatedpowerexceeds
thedesired,andselectastheminimumvalue
ofsamplesize.
46
Behaviorofoverallpower(mixed)
onecontinuousandonebinaryvariable(K=2)
individualpower=0.8
47
Behaviorsofsamplesizes
48
Behaviorsofsamplesizes(cont.)
49
Summaryoftheresults
Thebehaviorsofsamplesizesareinagreement
withtheresultsinthecontinuouscase.
isnotcloseto1
iscloseto1
AS<AN<<ASc= Fi<ANc
ThebehaviorsofachievedsamplesizesforK=3
aresimilartothoseforK=2.
Theempiricalpowerattainsaprespecifiedpower.
TheperformanceofAScandFiisbetter.
50
Anillustration:PREMIERstudy
Earlyaggressiverheumatoidarthritis
ACR50
ThepercentageofpatientsinwhomanACR50
responsewasachieved
mTSS
Themeanchangefrombaselineinthemodified
totalSharpscore.
Adalimumab+Methotrexatev.s.Methotrexate
Breedveld(2006)
51
Anillustration:PREMIERstudy
Samplesizespergroupwith
E1,E2:Samplesizeseparatelycalculatedfor
eachendpointsothattheindividualpowerisatleast0.8.
Breedveld(2006)
52
Jointdensityfunctionofteststatistics
Rejection
region
.
53