CHAPTER15MultipleEndpoints15.1IntroductionInterimanalysesinclinicaltrialsoftenincludetheexaminationofanumberofpatientoutcomemeasures.Forexample,Pocock,Geller&Tsiatis(1987)describeacrossovertrialofchronicrespiratorydiseaseinwhichactivedrugandplacebowerecomparedwithrespecttothreelungfunctionmeasurements:peakexpiratoryflowrate,forcedexpiratoryvolumeandforcedvitalcapacity.Thisisanexamplewheretheoutcomevariablesmightberegardedasunorderedsinceonedoesnottakepriorityoveranother.AnotherexampleoccurredintheNINDSStrokeTrial(Tilleyetal.,1996)inwhichtreatmenteffectinpatientswithacuteischemicstrokewasmeasuredonfourdifferentratingsscales.Inanextremeexample,O’Brien(1984)describedastudyofdiabetespatientsinwhichimprovementsinnervefunctionweremeasuredon34electromyographicvariables.Inothertrialstherewillbeanorderingofimportanceoftheoutcomevariables.Jennison&Turnbull(1993c)describeaclinicaltrialofananalgesicdrugusedtorelievearthriticpain.Theprimaryendpointwasameasureoftheamountofpainreliefexperiencedbythepatient,butasecondaryendpointconcernedapossibleeffectonthearthriticconditionofthejoint.Itwasthoughtthatthetwooutcomemeasuresmightberelated,iffornootherreasonthanthatthedrug’ssuccessinrelievingpainmayleadthepatienttobelesscarefulinprotectingthejoint.Whentheprimaryoutcomemeasureisarareevent,considerationofadditionalintermediateendpointscanleadtolowersamplesizesorhigherpower.Forinstance,inevaluatingtherapiesforacutemyocardialinfarction(MI),mortalitywillbethemostimportantendpointbutotherclinicallymeaningfulendpointsthatarecorrelatedwithadverselong-termoutcome,suchasrecurrentMIorventriculardysfunction,mayalsobeconsidered(Cannon,1997).Sometimesitisreasonabletoadoptastrategyofassigningacompositescoreorsummaryscalarmeasuretoeachsubjectbasedonallofthatindividual’sresponses.TheproblemisthenreducedtooneofunivariateresponseandthemethodsdescribedinprecedingchaptersformonitoringaZ-statisticoraStudent’st-statisticcanbeappliedtothesummarymeasure.Thisapproachissuitablewhentheoutcomevariablesallat...
