Flexible Surrogate Marker Evaluation from Several

TECHNICALREPORT0661FLEXIBLESURROGATEMARKEREVALUATIONFROMSEVERALRANDOMIZEDCLINICALTRIALSWITHCONTINUOUSENDPOINTS,USINGRandSASTILAHUNA.,PRYSELEYA.,ALONSOA.,andG.MOLENBERGHS*IAPSTATISTICSNETWORKINTERUNIVERSITYATTRACTIONPOLEfirststudiedbyPrentice(1989),whopresentedadefinitionofasurrogateaswellasasetofcriteria.Freedmanetal(2001)supple-mentedthesecriteriawiththeso-calledproportionexplainedafternotifyingsomedrawbacksinPrentice’sapproach.Buyseetal(2000)framedtheevaluationexercisewithinameta-analyticsetting,therebyovercomingdifficultiesthatnecessarilysurroundevaluationeffortsbasedonasingletrial.Inthispaper,webrieflyreviewthemeta-analyticapproachforcontinuousoutcomes.Advantagesandproblemsarehighlightedbymeansoftwocasestudies,oneinschizophreniaandoneinophthalmology,andasimulationstudy.Oneofthecriticalissuesforthebroadadoptionofmethodologyliketheonepresentedhereistheavailabilityofflexibleimplementationsinstandardstatisticalsoftware.WehavedevelopedgenericallyapplicableSASmacrosandRfunctions,atthereader’sdisposal.SomeKeyWords:Adjustedassociation;Hierarchicalmodel;Meta-analysis;Proportionex-plained;Random-effectsmodel;Relativeeffect;Surrogateendpoint.1IntroductionSurrogateendpointscomeintoplayinanumberofcontextsinplaceoftheendpointofinterest,referredcommonlytoasthetrueormainendpoint.Theuseofsurrogateendpointsispotentiallybeneficial,whentheseendpointscanbemeasuredearlier,leadingtoarapidapprovalofexperimentaldrugs,orcanbeadministeredconveniently,whichcanbeequatedtolessburdenonthesideofboththeexperimenterandthepatients(BuyseandMolenberghs1998).Theuseofsurrogateendpointsinclinicalpracticeisincreasing.Thereareseveralcasesinwhichthereisaneedforanacceleratedapprovalofanexperimentaldrugsothatitsbenefitcanbewitnessedinashortertimespan.Thisisespeciallytrueinthecaseofchronicdiseaseswithhighsocietalcost.1Ideally,thereshouldbeguidelinestodeclareamarkerausefulsurrogateforaclinicalendpoint.Twopossibleviewsarepossiblewhenevaluatingamarker.Thefirstdealswiththeindividualpatientlevelandisconnectedwiththebiologicalpathwayfromthesurrogatetothetrueendpoint.This,however,doesnotnecessarilymeanamarkerisusefultocapturethetreatmenteffectinthesettingofaclinicaltrial.Therefore,asecondview,focusingonthetreatmenteffectisnecessaryandpossible(FlemingandDeMets1996).Precisely,thislevelquantifiestheassociationbetweenthetreatmenteffectsonthemarkerandtheclinicalendpoint.Buyseetal(2000)andBurzykowskyetal(2004),amongothers,havepresentedameta-analyticmodelingframework,withinwhichbothformsofvalidationcanbeundertaken.Akeystumblingblockforthepracticaluseistheavailabilityofflexibleimplementationswithinstandardandcommonlyusedsoftwarepackages.Thepurposeofthispaperistoreviewthevalidationframework,toexemplifythemethodologyintwoclinicaltrialsettings,andtopresentagenericRfunctionandSASmacro.Computationalissues,thatcanbesourcesofconcern,andofwhichthepractitionershouldbeatcurrent,arediscussedindetailandillustratedthroughasimulationstudy.Theseissuescenteraroundthechoiceofunitofanalysis,treatmentcodingschemesandproblemswithnon-positivedefiniteandill-conditionedmatrices.Therestofthispaperisorganizedasfollows.AnintroductiontothemotivatingstudiesisgiveninSection2.ThedifferentvalidationmethodsareoutlinedinSection3withthemethodsthatarebasedonthemeta-analyticapproachreviewedfirstfollowedbynumberofsimplifiedcomputationalstrategies.Computationalissues,arisingwhenusingthemeta-analyticapproach,arediscussedinSection4.Section5containstheresultsofthecasestudiesperformedonthedatasetsintroducedinSection2.2MotivatingCaseStudiesWewillpresentacasestudyinschizophrenia,followedbyoneinophthalmology.22.1AMeta-analysisofFiveclinicalTrialsinSchizopherniaThedatacomefromameta-analysisoffivedouble-blindrandomizedclinicaltrials,comparingtheeffectsofrisperidonetoconventionalantipsychoticagentsforthetreatmentofchronicschizophre-nia.ThetreatmentindicatorforrisperidoneversusconventionaltreatmentwillbedenotedbyZ.Schizophreniahaslongbeenrecognizedasaheterogeneousdisorderwithpatientssufferingfromboth‘negative’and‘positive’symptoms.Negativesymptomsarecharacterizedbydeficitsincogni-tive,affectiveandsocialfunctionsforexamplepovertyofspeech,apathyandemotionalwithdrawal.Positivesymptomsentailmorefloridsymptomssuchasdelusions,hallucinationsanddisorganizedthinking,whicharesuperimposedonmentalstatus(Kay,Fiszbein,andOpler1987).Severalmea-surescanbeconsideredtoassesapatient’sglobalcondition.Clinician’sGlobalimpression(CGI)isgenerallyacceptedasanadmittedlysubjectiveclinicalmeasureofchange.Here,thechangeofCGIfrombaselinewillbeconsideredasthetrueendpoint,denotedbyT.Itisscoredona7-gradescaleusedbythetreatingphysiciantocharacterizehowwellasubjecthasimprovedsincebaseline.AnotherusefulandsufficientlysensitiveassessmentscalesisthePositiveandNegativeSyndromeScale(PANSS)