Pan)European)Multi6Database)Bladder)Cancer)Risk)Characterisation) Study$

Transcription

1 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Page1of37 Version2.0,20June2013 Version2.0,20June TITLE) Pan)European)Multi6Database)Bladder)Cancer)Risk)Characterisation) Study 2. MARKETING)AUTHORIZATION)HOLDER)AND)SPONSOR) TakedaGlobalResearch&DevelopmentCentre(Europe)Ltd 3. RESPONSIBLE)PARTIES) ) Principal)Investigator:) ) Pasi)Korhonen,)PhD) AdjunctProfessorofBiostatistics EPIDResearchOy Tekniikantie12 FIR02150Espoo Finland Co6Investigators:)) Edith)M.)Heintjes,)PhD) ScientificResearchManager) PharmoInstituteforDrugOutcomesResearch(PHARMO)) VanDeventerlaan30R40) 3528AEUtrecht TheNetherlands) Rachael)Boggon,)MSc) ResearchStatistician TheClinicalPracticeResearchDatalinkGroup(CPRD) TheMedicinesandHealthcareproductsRegulatory Agency 151BuckinghamPalaceRoad LondonSW1W9SZEngland Helle)Kieler,)MD,)PhD)) AssociateProfessor) Head,CentreforPharmacoepidemiology) KarolinskaInstitutet(KI)T2,) KarolinskaUniversityHospital) SE17176StockholmSweden) ) ) Other)co6investigators)who)have)contributed)to)the)design)of)the)study)protocol:)) ShahramBahmanyar,AssociateProfessor,KI;PaulDolin,DirectorEpidemiology,Takeda;SusanEaton,MSPH, CPRD;FabianHoti,PhD,EPIDResearch;LeanneHouweling,MSc,PHARMO;MarieLinder,PhD,KI;Timothy Williams,Dr,CPRD

2 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy 4. ABSTRACT) Title)of)study:)Pan)European)Multi)Database)Bladder)Cancer)Risk)Characterisation)Study Rationale:)TheCommitteeforMedicinalProductsofHumanUse(CHMP)oftheEuropeanMedicinesAgency recentlyfinalizeditsreviewonpioglitazoneuseandoccurrenceofbladdercancer.evidencefromseveral recentepidemiologicalstudiessuggestsasmallincreasedriskofbladdercancerwithpioglitazoneusein patientswithtype2diabetes[lewisetal.2011,cnamts2011].inthe4 th interimanalysis(8ryearsfollowrup) ofthekaiserpermanentenortherncaliforniastudycohort,thehazardratio(hr)ofevervs.neveruseof pioglitazonewas1.07(95%ci0.87r1.30)andanonrsignificantelevatedriskamongpatientswithlongest>4 yearsdurationoftherapy(hr=1.30,95%ci0.91r1.86)[lewisetal.2012].inthefrenchcnamtsstudya similarriskforpioglitazoneusewasobserved(hr=1.22,95%ci1.05r1.43),andasignificantriskwith cumulativedosesover28000mg(hr=1.75,95%ci1.22r2.50),andwithdurationoftherapybetween12to23 months(hr=1.34,95%ci1.02r1.75)orover24months(hr=1.36,95%ci1.04r1.79)[cnamts2011].the CHMPagreedthatfurtherevidenceisstillneeded.Mostpreviousstudieswerelimitedintheircontrolof channellingbias,detectionbiasorcontrolforconfoundingbyimportantbladdercancerriskfactorssuchas BMIandsmoking. Limitedinformationiscurrentlyavailableonwhetherthepotentialincreaseinriskisassociatedwitha particulartypeorstageofbladdercancer.datafromthekaiserpermanentenortherncaliforniastudycohort studyshowedanyincreasedriskwaslargelylimitedtoearlystagetumours[lewisetal.2012]. Thisobservationalstudyisbeingundertakentofurtherassesstheassociationbetweenpioglitazoneuseand bladdercancerriskamongpatientswithtype2diabetesmellitusinfoureuropeancountries:finland, Netherlands,Sweden,andUnitedKingdom. ThisstudyisbeingundertakenattherequestoftheEuropeanMedicinesAgency.Thestudyprotocolhasbeen reviewedbytheeuropeanmedicinesagency. Research)Questions:)) Primary)objective: 1) Toestimateandcomparetheabsoluteandrelativeriskofbladdercancerinpatientswithtype2diabetes whoareeverexposedtopioglitazonevs.neverexposedtopioglitazone. Secondary)objectives: 2) Toestimateandcomparetheabsoluteandrelativeriskofbladdercancerinpatientswithtype2diabetes withincreasingdurationofpioglitazonetreatment, 3) Toestimateandcomparetheabsoluteandrelativeriskofbladdercancerinpatientswithtype2diabetes withincreasingcumulativedoseofpioglitazonetreatment, 4) Toexaminetheassociationbetweenpioglitazoneexposureandriskofbladdercancerwithrespectto possiblechannellingbias,detectionbiasandothersourcesofconfounding. 5) Tocharacterizethestageandgradeofthebladdercancercasesatthetimeofdiagnosisinpatientswith type2diabeteswhoareeverexposedtopioglitazonevs.whoareneverexposedtopioglitazone,and 6) To characterize allrcause mortality and bladder cancerrspecific mortality pattern amongst patients ever exposedtopioglitazonevs.neverexposedtopioglitazone. Version2.0,20June2013 Page2of37

3 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Explorativeobjective: Toestimateandcomparetheabsoluteandrelativeriskofbladdercancerinpatientswithtype2diabetes withtimesincethelastdoseofpioglitazoneuse. Study)design:)) RetrospectiveCohortStudiesin4countries Individualpatientleveldatametaanalysisoftheretrospectivecohortstudydatasets Matching&Channellingbias Theobservedassociationbetweenlongtermpioglitazoneuseandriskofbladdercancercouldinpartbedue tochannellingbiasbecausepioglitazoneislikelytobegiventopatientswithmoreseverediabetesas comparedtootheroraldiabeticmedications,andpatientswithmoreadvanceddiabetesarelikelytobeat higherriskofbladdercancer. Tominimizechannellingbias,eachpioglitazoneRexposedpatient(exposuregroup)willbematchedwithupto 10pioglitazoneRunexposeddiabeticpatients(referencegroup)basedon: Propensityscorederivedfromthefollowingvariables: Duration(years)oftreateddiabetesmellitusatcohortentry Historyofdiabeticcomplicationsatcohortentry Historyofmyocardialinfarctionorstrokeatcohortentry Historyofcongestiveheartfailureatcohortentry Yearofcohortentry Duration(years)ofmembershipinmedicationdatabasepriortocohortentry Numberofdifferentantidiabeticdrugclassespriortocohortentry Useofotherthiazolidinediones(otherthanpioglitazone)priortocohortentry Typeofantidiabetictreatmentimmediatelypriortocohortentry Type of modification in baseline antidiabetic therapy at cohort entry classified as treatment switch or additionofnewtreatmenttoallpriortreatment(ifanyexisted) Livingingeographicalareaforwhichcancerdataareavailable(yes/no,usedintheNetherlandsonly) Study)population:)) The target population consists of patients with type 2 DM whose antidiabetic treatment at cohort entry is modified to include pioglitazone or another antidiabetic medication. These patients are identified with the followinginclusion/exclusioncriteria. Inclusion)criteria:) Treatmentwithanyoralantidiabeticdrugsatanytimeintheavailablemedicationrecords. Baselineismodified(cohortentrypoint)toincludepioglitazone(exposuregroup)oranotherantidiabetic medication(referencegroup) Age 40yearsatcohortentry Atleast12monthsofmedicationdatabasemembershipduringbaselineperiodpriortocohortentry Exclusion)criteria:)) Diagnosisoftype1diabetes,gestationaldiabetes,orsecondaryandothertypesofdiabetesmellitusprior tocohortentry PatientswhoareentitledtospecialreimbursementfordiabeteswithICDR10diagnosisstartingwith E10, E12, E13, O24.4 indicativeoftype1diabetes,diabeteswithmalnutrition,otherdiabetesorgestational Version2.0,20June2013 Page3of37

4 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy diabetes,respectively(specifictofindatasetonly),and DiagnosisorhistoryofbladdercancerorinRsitubladdercancerpriortocohortentryusingavailabledata sources Data)Sources:) Thecohortstudieswillbeundertakeninthe4Europeancountries(FIN,NL,SWE,UK)usinglinkageofdrug prescribing/dispensingdatabasestocountryrspecificselecteddatabasesincluding(i)cancerregistries,(ii)gp recordsand/orhospitaldischargerecords,(iii)deathrecords,(iv)reimbursementdecisions,and(v) immigrationandemigrationrecords.) AssomecountriesonlyhavehospitalRdischargemorbiditydatabaseswhileotherhaveGPRbasedrecords,the datasetsarecategorizedaccordingtosourceofmorbiditycovariatedata: 4hospitalRbasedmorbiditydatasets(FINdataset,SWEdataset,NLPharmacyRHospitaldataset,UKCPRD GOLDRHESdataset) 2GPRbasedmorbiditydatasets(NLGPdataset,UKCPRDGOLDdataset) Throughouttheprotocol,definitionsofoutcomeandotherexplanatoryvariablesareharmonizedacrossthe datasetswithineachcategory. DefinitionoffollowRuptimeandoutcome Periodcoveredbythedatasets Dataset Startofcoverage EndofFollowRup FIN 01January June2011 NLPharmacyRHospital 01January June2011 NLGP 01January June2011 SWE 01July June2011 UKCPRDGOLD 01January June2011 UKCPRDGOLDRHES 01January December2010 Cohortentry/StartofFollowRup CohortentryandstartoffollowRupisdefinedasthedatewhenthebaselinetherapywasmodifiedtoinclude pioglitazoneoranotherdiabeticmedicine. Baselineobservationperiodbeforecohortentry Eachpatientmusthaveatleast1yearofmembershipintherelevantmedicationdatabasepriortocohort entry.thisperiodisusedtocharacterisebaselinetherapy,baselinemedicalhistory,andbaselinecovariates, andminimisetheriskofleftrcensoringofexposureinformation. EndofFollowRup Inthebladdercanceranalyses,eachpatientwillbefollowedRupfromcohortentryuntilthedateofdiagnosis ofthefirstincidentbladdercancerorcensoredatendofmembershipofthedatabase,death,startofother thiazolidinediones(otherthanpioglitazone)atoraftercohortentry,endofdatabasecoverageor30june 2011whicheveroccursfirst. Version2.0,20June2013 Page4of37

5 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Inthemortalityanalyses,eachpatientwillbefollowedRupfromcohortentryuntildateofdeath,startofother thiazolidinediones(otherthanpioglitazone)atoraftercohortentry,endofmembershipofthedatabase,end ofdatabasecoverageor30june2011,whicheveroccursfirst. DiagnosisofBladderCancer In the FIN and SWE and CPRD GOLDRHES datasets, cancer registryrbased information will be used for all patients. TheNLPharmacyRHospitalandNLGPdatasetsonlyhavepartiallinkagetothecancerregistrydata(20R30%in NL datasets). It is not possible to use cancer registryrbased diagnosis for all subjects. Diagnosis of bladder cancerwillbebasedoncancerregistrydatawhereavailable.forpracticeswithnolinkage,gprecordsdata (NLGPdataset)andHospitaldischargedata(NLPharmacyRHospitaldataset)willbeusedtoidentifybladder cancers. In the NL PharmacyRHospital, NL GP, UK CPRD GOLD and UK CPRD GOLDRHES datasets, the occurrence of bladdercancerincancerregistrydataandgp/hospitalrecordsdatawillbecomparedtovalidateusinggpor hospitaldischargedataforbladdercancerdiagnosis. TheCPRDGOLDdatasetisnotlinkedtothecancerregistrydata.Diagnosisofbladdercancerwillbebasedon GPrecordsdataonly. Allcausemortalityandbladdercancermortality In the FIN, SWE and UK CPRD GOLDRHES datasets, there is full linkage to death registries. Bladder cancer mortalitywillonlyincludedeathswherebladdercancerwasrecordedastheunderlyingcauseofdeath.date ofdeathbutnotcausesofdeathareavailableforthetwonldatasetsandukcprdgolddataset. Definitionofexposure:) Pioglitazone)exposure) variable)(t)=)time) dependent)) Evervs.neveruse(T) )Description) Takingthevalue1(ever)assoonasoneprescriptionwithpioglitazonehas beenpurchased,and0(never)otherwise. Durationofexposure(T) Cumulativedose(T) Timesincelastdose(T) Otherexplanatoryvariables TimeRdependentcumulativesumofdurationsofthepreviouspioglitazone exposureperiods.thedurationofexposurecategorieswillbetakenidentical tothoseusedinthekpncfinal(10ryearfollowrup)analysis. TimeRdependentcumulativesumofdrugconsumptionbasedondailydosage ofpioglitazonecontainingdispensingsorprescriptionssinceentryintothe cohort.thecumulativedosecategorieswillbeidenticaltothoseusedinthe KPNCfinal(10RyearfollowRup)analysis. Currenttimeminusthetimeoftheendoflastcurrentexposuretopioglitazone containingdispensingorprescriptionsinceentryintothestudycohort. CategorisedintoCurrent,<1year,1R2years,2R3years,3R4years,4+years, Never Version2.0,20June2013 Page5of37

6 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Covariatetobeassessed(whereavailableinadataset)inthecohortstudiesinclude: Gender BMI Smoking HbA1C Historyandconcurrentrelevantcomorbidities,includingcardiovasculardisease,CHF,othercancers, chronicrenaldisease HistoryandconcurrentrenalcoRmorbidity,includingrenaldialysis,transplant Historyandconcurrentbladdercomorbidities,includingincontinence,UTI,pyelonephritis,urolithiasis, haematuria,retention,catheterization Concurrentdiabeticcomplications,includingretinopathy,peripheralneuropathy,nephropathy, proteinuria,microalbuminuria Historyandconcurrentuseofothermedications,includingstatins,ARBs,ACEs,BPHdrugs Inthe4HospitalRbasedmorbiditydatasetssomecovariateswillnotbeassessable,whileothersmaybebased onhospitalizations(i.e.,moreseverecases)oronuseofspecificmedications. Inthe2GPRbasedmorbiditydatasetscovariateswillbebasedonGPrecordsoruseofspecificmedications. Where possible identical definitions are used across all six datasets. Where this is not possible, a single definitionisusedwithinthehospitalrbasedmorbiditydatasets,andaseparatesingledefinitionisusedinthe GPRbasedmorbiditydatasets. Statistical)methods:) AnalysisofindividualcohortdatasetsIdenticalstatisticalmethods,analysesanddatacategorizationswillbe used in each cohort study. The hazard ratio (HR) estimates with 95% CIs for each pioglitazone exposure definition will be estimated using the conventional Cox s proportional hazards model with adjustments for relevantbaselineandtimerdependentcovariates.analysiswillbeperformedforeachcountryseparately. MetaanalysisofindividualpatientdataTheindividualpatientdatafromthe6cohortswillbepooledintoa singledataset,andtheprimaryanddoseresponseanalysesrepeatedusingthepooleddataset.thehazard ratio (HR) estimates with 95% CIs for each pioglitazone exposure definition will be estimated using Cox s proportionalhazardsmodelwithcohortdatasetincludedasacategoricalcovariate.thehazardmodellingwill includeadjustmentsforrelevantbaselineandtimerdependentcovariates. Sample)size:) ) ) Finland) Sweden) Version2.0,20June2013 NL) Pharmacy6 Hospital) UK)CPRD) GOLD6HES) subset) NL)GP) UK)CPRD) GOLD) Everexposedto pioglitazone Neverexposedto Page6of37

7 ) pioglitazone (10:1matched) Totalmatched cohort PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Statistical)power)of)individual)patient)data)meta6analysis) Pooleddataset FollowRuptime 5.5years Cumulativeprobability ofdevelopingbladder cancer(%) Power(%) RR=1.2 RR=1.3 RR=1.4 RR=1.5 RR= ) 34.4) 58.0) 76.3) 87.8) 99.7 Ethics:)Alltheinformationisobtainedwithoutanypotentialharmtopatients.ThisisfullyaregisterRbased study and patients will not be contacted in any phase of the study. All patient data is anonymous which ensuresthefulldataprotectionofpatients.approvalfromrelevantethicalreviewboardswillberequested beforeconductingthestudy. TheinvestigatorsandtheSponsorandindividualsactingontheirbehalfcommittoadheretotherulesofthe ENCePP Code of Conduct in their entirety. The study has been registered into the ENCePP s ERregister of studiesandwillbesubmittedfortheenceppsealofapproval.) Results) and) publications:) The principal) investigator and corinvestigators will jointly write the pharmacoepidemiologic analysis reports. The reports are delivered to Sponsor. Based on these results the investigators will corauthor scientific manuscript(s) of the results to be published. A summary of the main resultsofthestudy,whetherpositiveornegativeandincludingresultsfromprematurelyterminatedstudies, willalwaysbemadeavailabletothepublic.anabstractofthestudyfindingswillbeprovidedthroughthe ENCePPERregisterofstudieswithinthreemonthsfollowingthefinalstudyreport.Theprincipalinvestigators mayasktheenceppsecretariattodelaythepublicationofthisabstractforalimitedperiodpendingresponse to peerrreview comments. Sponsor is entitled to view the final results and interpretations thereof prior to submissionforpublicationandtocommentinadvanceofsubmissionasagreedintheresearchcontractand withoutunjustifiablydelayingthepublication.thefinalstudyreportmaybesharedwithregulatoryagencies asrequired.) Version2.0,20June2013 Page7of37

8 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Page8of37 Version2.0,20June2013 Version2.0,20June2013 LIST)OF)ABBREVIATIONS) ACEinhibitor Angiotensinconvertingenzymeinhibitor ARB Angiotensinreceptorblockers ATCcode Anatomicaltherapeuticchemicalclassificationsystemcode BPH Benignprostatichypertrophy CHF CongestiveHeartFailure CHMP CommitteeforMedicinalProductsofHumanUse CI Confidenceinterval COPD Chronicobstructivepulmonarydisease) CPRD ClinicalPracticeResearchDatalinkGroup DDD Defineddailydose ENCePP EuropeanNetworkofCentresforPharmacoepidemiologyandPharmacovigilance FIN Finland HR Hazardratio ICDR9 Internationalclassificationofdiseases,9 th revision ICDR10 Internationalclassificationofdiseases,10 th revision ICDROR3 Internationalclassificationofdiseasesforoncology,3rdEdition ICPC Internationalclassificationofprimarycare ISPE InternationalSocietyforPharmacoepidemiology NL TheNetherlands NLGP PHARMOGeneralPracticeDataset PGJ2 ProstaglandinJ2 PID Personalidentificationnumber PPAR PeroxisomeproliferatorsRactivatedreceptors PUNLMP Papillaryurothelialneoplasiaoflowmalignantpotential RXR RetinoidXreceptor SID Studyidentificationnumber SWE Sweden TZD Thiazolidinediones UK UnitedKingdom UTI Urinarytractinfection Vnr Nordicarticlenumber

9 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy 5. AMENDMENTS)&)UPDATES) Version1.0(dated22Jan2013) None Version2.0 Currentversion(dated20June2013) Objectives o Theoriginalsevenobjectivesclassifiedasoneprimaryobjectiveandsixsecondaryobjectives CoRinvestigators o ThelistofcoRinvestigatorsbroadenedtoincludeallpersonsinvolvedinthedesignofthestudy protocol. Inclusionandexclusioncriteria o Criteriaconsolidatedacrossdatabasestoincludepatientsover40yearsusinganyantidiabetic medication and further exclude patients with diagnosis of type 1 diabetes, gestational diabetes,ordruginduceddiabetespriortocohortentry. o Theuseofotherthiazolidinedionesbutpioglitazonepriortocohortentryallowed.Preliminary dataonexposurefrompharmoandcprdrecordssuggestthatupto30%ofpatientsstarting pioglitazonehavebeenexposedearliertoothertdzs.toincreasethepowerofthestudythese patientswillbeincluded,andthehistoryofpriortdzuseusedincohortmatching.stratified analysiswillbeperformedtoexploretheeffectofpriortdzuseontheassociationbetween pioglitazoneexposureandbladdercancerincidence.followruptimewillbecensoredatstartof othertdzsaftercohortentry Bladdercancer o Bladder cancer as an underlying cause of death added as source of information on bladder cancer o Diagnosiscodesspecifiedindetailinthedefinitionofbladdercancer Matching o Variables used for the matching within the propensity score and outside of the propensity scorereorganisedanddefined. Otherantidiabeticdrugs o Categoriesgroupedintomostcommonregimens CoRmorbidities,drugsandexposurealgorithm o Codelistsupdated o Detailsmovedintoappendices Analysis o Definitionsofthecrude,baseandadjustedmodelsdefinedindetail.Fullyadjustedmodelnot usedasatermanymore. o Stratifiedanalysesadded. o ThesensitivityanalysisoftheimpactofexclusionofunmatchedpioglitazoneRexposedpatients described in more detail. Sensitivity analysis with cumulative duration of insulin added. Sensitivity analysis where pioglitazone exposure group is changed from at least one prescriptionordispensingofpioglitazonetoatleast2prescriptionsordispensingwithina6 month period added. Analysis of the association of ever vs. never use of pioglitazone with bladdercancerincidencewheninsulinuseisincludedasacumulativedurationintheadjusted model. Sensitivity analysis for incidence of neoplasm of uncertain or unknown behaviour added. Version2.0,20June2013 Page9of37

10 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy o A separate metaranalysis added by using a subset of the pooled dataset including those datasets where information on smoking, BMI, HbA1C and urinary/renal markers is also available. Datasharing o o StudytoberegisteredfortheENCePPSealofapproval Criteria and process for sharing the analytical country specific datasets and metaranalysis datasetgiveninappendix 6. MILESTONES) Activity) ) FinalupdatedprotocolsubmittedtoEMA ResearchEthicsapprovalfordatalinkage&access 1 Datasetsavailable 2 NL,UK FIN,SWE Pooleddatasetforindividualpatientdata levelmetaranalysis AnalysesCompleted 3 NL,UKcohorts FIN,SWEcohorts IndividualpatientdatalevelmetaRanalysis DraftStudyReporttoSponsor 6Individualcohortstudies IndividualpatientdatalevelmetaRanalysis Version2.0,20June2013 Date) 31Jan Apr July Oct Dec Apr Apr Jun Jun Aug Sep2014 StudyreportstoEMA 1 ResearchEthicsapprovaltypicallytakes3monthsineachcountry.Allcountriesrequirethefinalisedprotocol beforesubmissionforresearchethicsapproval. 2. Datasetavailabilitylargelydependentondeliveryoflinkeddatafromcancerregistriesandothersourcesin eachcountry.bestestimatesarethatresearchdatasetshouldbeavailable3monthsafterresearchethics approvalinnlanduk,and6monthsforfinandswe. 3. AnalysisofNLandUKdatasets,includingsensitivityanalyses(1R3,5R7)isestimatedat9monthsfrom availabilityofresearchdataset.analysisofsweandfindatasets,withsensitivityanalyses(1r4)isestimatedat 6monthsfromavailabilityofresearchdataset. Page10of37

11 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy 7. RATIONALE)AND)BACKGROUND) Peroxisome proliferatorsractivated receptors (PPARs) are members of the nuclear hormone receptor super familyoftranscriptionfactorswhoseactivitiesareregulatedbyhighraffinitybindingofsmall,lipophilicligands suchassteroidhormones,vitamind,retinoids,andthyroidhormone.pparalpha,deltaandathirdsubtype calledgammaarerelatedsufficientlytobeconsideredmembersofasubfamily,andhavesimilarproperties including DNA binding specificity and heterodimerization with retinoid X receptor (RXR), whose ligands also activate the PPAR/RXR heterodimers. Ligands selective for PPAR gamma include prostaglandin J2 (PGJ2) derivatives, such as 15RdeoxyRΔ12,14RPGJ2, and antirdiabetic thiazolidinediones (TZD) compounds, including troglitazone,rosiglitazone,andpioglitazone. TheCommitteeforMedicinalProductsforHumanUse(CHMP)oftheEuropeanMedicinesAgencyrecently finalizeditsreviewonpioglitazoneuseandoccurrenceofbladdercancer.evidencefromseveralrecent epidemiologicalstudiessuggestsasmallincreasedriskofbladdercancerwithpioglitazoneuseinpatientswith type2diabetes[lewiset#al.2011;cnamts2011].inthe4 th interimanalysis(8ryearsfollowrup)ofthekaiser PermanenteNorthernCaliforniaStudycohort,thehazardratio(HR)ofevervs.neveruseofpioglitazonewas 1.07(95%CI0.87R1.30)andanonRsignificantelevatedriskamongpatientswithlongest>4yearsdurationof therapy(hr=1.30,95%ci0.91r1.86)[lewisetal.2012].inthefrenchcnamtsstudyasimilarriskfor pioglitazoneusewasobserved(hr=1.22,95%ci1.05r1.43),andasignificantriskwithcumulativedosesover 28000mg(HR=1.75,95%CI1.22R2.50),andwithdurationoftherapybetween12to23months(HR=1.34,95% CI1.02R1.75)orover24months(HR=1.36,95%CI1.04R1.79)[CNAMTS2011].TheCHMPagreedthatfurther evidenceisstillneeded.previousobservationalstudieswerelimitedintheirabilitytocontrolforpotential channellingbias,detectionbiasandconfoundingbyimportantbladdercancerriskfactorssuchasbmiand smoking. Limitedinformationisalsocurrentlyavailableonwhetherthepotentialincreaseinriskisassociatedwitha particulartypeorstageofbladdercancer.datafromthekpnccohortstudyshowedanyincreasedriskwas largelylimitedtoearlystagetumours(lewisetal2012).thisfindinghasyettobeconfirmedinanotherstudy. 8. RESEARCH)QUESTION)/)STUDY)HYPOTHESIS)) Theobjectivesofthestudyare: Research)Questions:)) Primary)objective:) 1) Toestimateandcomparetheabsoluteandrelativeriskofbladdercancerinpatientswithtype2diabetes whoareeverexposedtopioglitazonevs.neverexposedtopioglitazone. Secondary)objectives:) 2) Toestimateandcomparetheabsoluteandrelativeriskofbladdercancerinpatientswithtype2diabetes withincreasingdurationofpioglitazonetreatment, 3) Toestimateandcomparetheabsoluteandrelativeriskofbladdercancerinpatientswithtype2diabetes withincreasingcumulativedoseofpioglitazonetreatment, 4) Toexaminetheassociationbetweenpioglitazoneexposureandriskofbladdercancerwithrespectto possiblechannellingbias,detectionbiasandothersourcesofconfounding, 5) Tocharacterizethestageandgradeofthebladdercancercasesatthetimeofdiagnosisinpatientswith type2diabeteswhoareeverexposedtopioglitazonevs.whoareneverexposedtopioglitazone,and Version2.0,20June2013 Page11of37

12 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy 6) To characterize allrcause mortality and bladder cancerrspecific mortality pattern amongst patients ever exposedtopioglitazonevs.neverexposedtopioglitazone. Inadditionanexploratoryobjectiveofthestudyis: 7) Toestimateandcomparetheabsoluteandrelativeriskofbladdercancerinpatientswithtype2diabetes withtimesincethelastdoseofpioglitazoneuse. 9. RESEARCH)METHODS)) 9.1 Study)design) RetrospectiveobservationalcohortstudieswillbeundertakenusingsixdatasetsinfourEUcountries:Finland, Sweden,theNetherlands,andUnitedKingdom. Each dataset will be analysed separately, then pooled into a single data set and this pooled anonymous individualpatientleveldatasetwillbeusedinmetaranalysis. The)cohort)studies) Thepreviouslyobservedweakassociationbetweenlongtermpioglitazoneuseandriskofbladdercancercould inpartbeduetochannellingbiasbecausepioglitazoneismorelikelytobeprescribedtopatientswithmore advanced diabetes or who are obese. Advanced diabetes and obesity are both independent risk factors for developingbladdercancer, and the preferential prescribing of pioglitazone to such patients iscould lead to biasedestimatesofrisk. Tominimizechannellingbias,eachpioglitazoneRexposedpatient(exposuregroup)willbematchedwithupto 10pioglitazoneRunexposeddiabeticpatients(referencegroup)basedon: Propensity score (the probability to initiate pioglitazone therapy) will be evaluated with a logistic regressionmodelusingthefollowingproxymeasuresofseverityofdiabetesandvariablesaffecting the prescription of pioglitazone. The number and exact definitions of variables included in the propensityscoremayvarybetweenthedatabases.! Duration(years)oftreateddiabetesmellitusatcohortentry! Historyofdiabeticcomplicationsatcohortentry! Historyofmyocardialinfarctionorstrokeatcohortentry! Historyofcongestiveheartfailureatcohortentry! Yearofcohortentry! Duration(years)ofmembershipinmedicationdatabasepriortocohortentry! Numberofdifferentantidiabeticdrugclassespriortocohortentry Useofotherthiazolidinediones(otherthanpioglitazone)priortocohortentry Typeofantidiabetictreatmentimmediatelypriortocohortentryclassifiedas! notreatment! metforminonly! sulphonylureasonly! metforminandsulphonylureas! insulinwithorwithoutanyotherantidiabeticmedication Version2.0,20June2013 Page12of37

13 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy! other(includinguseofotherthiazolidinediones) Typeofmodificationinbaselineantidiabetictherapyatcohortentryclassifiedastreatmentswitchor introductionofanaddrontreatment Livingingeographicalareaforwhichcancerdataareavailable(usedintheNetherlandsonly) PatientsinthepioglitazoneRexposedgroupwillbematchedtopatientsinthereferencegrouptowithin±0.05 of the propensity score and exactly within the classes of the other matching variables. If any pioglitazoner exposedpatientsdonothaveanymatch,thematchingwillberepeatedforthesepatientsbylooseningthe matchingrangeupto±0.1forthepropensityscore.cautionisneedednottoloosenthematchingrangetoo muchbecausethiswouldreducethecomparabilityofthegroups. ChannellingbiasduetobaselineBMIandHbA1Clevelisalsolikely.BMIisamoderatelystrongindependent risk factor for bladder cancer (RR 5) and the EU SmPC specifically recommends pioglitazone use among overweightt2dmpatients.asbmiandhba1cdataareonlyavailableinsomedatasets,theseareincludedas covariatesinthedataanalysesratherthanascohortmatchingvariables. 9.2 Study)population) The target population consists of patients with type 2 DM whose antidiabetic treatment at cohort entry is modified to include pioglitazone or another antidiabetic medication. These patients are identified with the followinginclusion/exclusioncriteria. ) Inclusion)criteria:) Treatmentwithanyoralantidiabeticdrugsatanytimeintheavailablemedicationrecords. Baseline is modified (cohort entry point) to include pioglitazone (exposure group) or another antidiabeticmedication(referencegroup). Age 40yearsatcohortentry. Atleast12monthsofmedicationdatabasemembershipduringbaselineperiodpriortocohortentry. Exclusion)criteria: Diagnosisoftype1diabetes,gestationaldiabetes,orsecondaryandothertypesofdiabetesmellitus priortocohortentryusingavailabledatasources.fordetailedicdr10,icdr9,icpcandreadcodessee Appendix3. Patients who are entitled to special reimbursement for diabetes with ICDR10 diagnosis starting with E10, E12, E13, O24.4 indicativeoftype1diabetes,diabeteswithmalnutrition,otherdiabetesor gestationaldiabetes,respectively(specifictofindatasetonly). DiagnosisorhistoryofbladdercancerorinRsitubladdercancerpriortocohortentry. 9.3 VARIABLES) Outcome)variable) Bladdercancer Dateofdiagnosisofthefirstincidenceofbladdercanceraftertheentryintothestudycohortwillbeusedas theprimaryoutcomedate.thebladdercancerdefinitionwillinclude Version2.0,20June2013 Page13of37

14 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy malignantneoplasmofthebladderand carcinomainsituofthebladder. Sensitivityanalysiswillalsolookthefollowingadditionaloutcomes: neoplasmofuncertainorunknownbehaviourofthebladder. IntheFINandSWEdatasetsinformationfromcancerregistrieswillusedtodefineincidentbladdercancers. Other datasets will use hospital discharge and GP records to find incident bladder cancers. In the cancer registries, malignant bladder cancer is coded to ICDROR3 code C67 with behaviour code 3. In the Hospital dischargedata,malignantbladdercanceriscodedtoicdr9rcm188oricdr10codec67.innlgp,bladder canceriscodedtoicpccodeu76,andintheukcprdiscodedtoreadcodesb49. The specific codes malignant neoplasm of the bladder, carcinoma in situ of the bladder and neoplasm of uncertainorunspecifiedbehaviourofthebladderaregiveninthefollowingtable. ICDR10CM ICDR9CM ICDROR3 ICPC READ Malignant neoplasm of the bladder C C67 with behaviour code 3 Carcinomainsituofthebladder D C67 with behaviour code 2 Neoplasm of uncertain or unknown behaviour of the bladder D C67 with behaviour code 1 (PUNLMP with morphology code MR 8130/1 U76 B49 * B837 * B917, BB4111, BA04 * Text mining of the database contents used to identify in situ carcinomas and neoplasms of uncertain or unknownbehaviourofthebladder IntheFIN,SWEandUKCPRDGOLDRHESdatasets,thereis100%linkagetocancerregistries.Thedatacollection onallcancercasesiscompulsoryandtheinformantssubmittingdataoncancerpatientstotheregistryinclude all hospitals, physicians, pathological, cytological and haematological laboratories, and dentists. The data collectedbytheseregistersincludedemographicinformationonthepatient,medicaldatasuchastheprimary siteofthetumour(internationalclassificationofdiseasesforoncologycode),dateofdiagnosis,andtumour type and grade. In the FIN, SWE and UK CPRD GOLDRHES datasets, bladder cancer will be based on cancer registryrbasedinformation. TheNLPharmacyRHospitalandNLGPdatasetsonlyhavepartiallinkagetothecancerregistrydata(20R30%in NLPharmacyRHospitaland10%NLGPdatasets).ItisthusnotpossibletousecancerregistryRbaseddiagnosis forallsubjects.gprecordsdata(nlgpdataset)andhospitaldischargedata(nlpharmacyrhospitaldataset) will be used to identify bladder cancers. For patients living in geographical areas that can be linked to the Version2.0,20June2013 Page14of37

15 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy cancerregistry,thismethodofoutcomedetectionwillbevalidatedandsupplementedusingthecancerregistry data. TheCPRDGOLDdatasetisnotlinkedtothecancerregistrydata.Diagnosisofbladdercancerwillbebasedon GPrecordsdataonly. Allcausemortalityandbladdercancermortality Informationondateofalldeathsamongcohortmembers,anddeathswherebladdercancerisrecordedasthe cause of death will be obtained from the national death registries for FIN, SWE, and UK CPRD GOLDRHES datasets. A recent validation study in the UK CPRD Gold demonstrated 99% concordance between GP s mortalityrecordandnationaldeathregistryentries.causerspecificmortalityisnotavailableforthetwonl datasets or the UK CPRD GOLD dataset. Bladder cancer deaths will be identified as deaths with malignant bladdercancer(e.g.icdr10codec67)astheunderlyingcauseofdeath Drug)exposure) InFINandSWEdatasets,exposureisbasedonpurchasing.IntheNLGP,UKCPRDGOLDandUKCPRDGOLDR HESdatasets,exposureisbasedonGPprescriptions.IntheNLPharmacyRHospitaldataset,exposureisbased ondispensing. Diabetic medications are classified into the following antidiabetic drug exposure groups: pioglitazone, metformin, sulphonylureas, other thiazolidinediones, alpharglucosidase inhibitors, DPPR4 inhibitors, GLPR1 agonists,meglitinides,amylinanalogues,insulinsandotheroralantidiabeticmedications.basedontheactive substancescontainedincombinationproducts,theyareallocatedtoseveralexposuregroups.furtherdetails ofthediabeticmedicationsaregiveninappendix1. The amount of a specific drug is given as the number of defined daily doses (DDDs) or as actual amount dispensed or prescribed depending on the database. DDD describes how long the prescription would last if taken as one DDD per day and can converted into total amount contained using the Nordic article number whichidentifiesthemedicinepackage.basedontotalamountofdrugcontainedineachprescriptionwillbe modifiedintoadruguseperiodbydefiningthelengthoftheprescriptionplusanadditional50%treatmentgap period[nielsenetal.2008,2009].thepurposeofthetreatmentgapperiodistoavoidunnecessarybreaks betweentwoprescriptions.theresultingconsecutivedruguseperiodswithpossiblegapsbetweenthemare usedtodefineatimerdependentcurrentexposureperiodsindicatinginwhichtreatmentcategoriesthepatient isatanygiventimeduringthefollowrup. Based on the prescriptions/purchases/dispensing of pioglitazone and the subsequent consecutive drug use periodsthefollowingtimedependentvariablesforthepioglitazoneexposurewillbeconstructed: Pioglitazone)exposure) variable)(t)=)time) dependent)) Currentexposure(T) )Description) Indicatorofcurrentpioglitazoneuseeitherusedaloneorasacombination. Thisvariableisnotinthestatisticalanalysisbutonlyusedinthecalculationof durationofexposureandtimesincelastdose. Page15of37 Version2.0,20June2013

16 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Evervs.neveruse(T) Takingthevalue1(ever)assoonasoneprescriptionwithpioglitazonehas beenpurchased,and0(never)otherwise. Durationofexposure(T) Cumulativedose(T) Timesincelastdose(T) TimeRdependentcumulativesumofdurationsofpreviouspioglitazone exposureperiods.thedurationofexposurecategorieswillbeidenticaltothose usedinthekpncfinal(10ryearfollowrup)analysis. AsecondaryanalysiswillusethedurationofexposurecategoryfromtheKPNC 3 rd interim(5ryear)analysis:never,<12months,12r24months,andmorethan 24months. TimeRdependentcumulativesumofdrugconsumptionbasedonthedaily dosageofpioglitazonecontainingprescriptionsordispensingssinceentryinto thecohort.thecumulativedosecategorieswillbeidenticaltothoseusedin thekpncfinal(10ryearfollowrup)analysis. AsecondaryanalysiswillusetheexposurecategoryfromtheKPNC3 rd interim (5Ryear)analysis:Never,1R10,500mg,10,501R28,000mg,andmorethan 28,000mg. Currenttimeminusthetimeoftheendoflastcurrentexposuretopioglitazone containingprescriptionssinceentryintothestudycohort.categorisedinto Current,<1year,1R2years,2R3years,3R4years,4+years,Never Thecategorylevelsarepreliminary,andwillbebasedontheactualexposuredistribution FurtherdetailsaboutcalculationoftheexposurearegiveninAppendix Propensity)score)and)other)cohort)matching)variables) Variable)(F=)Fixed)at)cohort) )Description) entry,)t)=)time)dependent)) ) Propensity)score)variables) ) Durationoftreated diabetesmellitusatcohort entry(f) Historyofdiabetic complicationsatcohort entry(f) Historyofmyocardial Duration(years)oftreateddiabetesmellitusisapproximatedastheinterval betweenthefirstdiabetestherapyintheprescriptionrecords,anddateof cohortentry. Historyofanyofthefollowing: Diabeticretinopathyormaculopathy(N/Y) Peripheralneuropathy(N/Y) Chronickidneydisease(N/Y) Proteinuria(microormacro)(N/Y) Diabeticnephropathy(N/Y) Ketoacidosis(N/Y) Diabeticcoma(N/Y). DefinitionsoftheindividualvariablesaregiveninAppendix3. Classifiedas(Y/N),fordetaileddefinitionseeAppendix3. Version2.0,20June2013 Page16of37

17 infarctionorstrokeat cohortentry(f) Historyofcongestiveheart failureatcohortentry(f) Yearofcohortentry(F) Durationofdatabase membershipbeforecohort entry(f) Numberofdifferent antidiabeticdrugclasses priortocohortentry(f) Other)matching)variables) UseofotherTDZspriorto cohortentry(f) Typeofantidiabetic treatmentpriortocohort entry(f) Typeofmodificationin baselinetherapy(f) Geographicalarea(F) PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Classifiedas(Y/N),fordetaileddefinitionseeAppendix3. Calendaryearofcohortentry Duration(years)ofmembershipinmedicationdatabasepriortocohortentry. Scorefrom0to10withonepointfromeachofthefollowingclassesusedprior to cohort entry: metformin, sulphonylureas, other TDZs, alphaglucosidase inhibitors, DPPR4 inhibitors, GLPR1 agonists, meglitinides, amylin analogues, insulin,other.combinationproductscontributeseparatelytoeachdrugclass basedontheactivesubstancesincludedintheproduct. Useofotherthiazolidinediones(otherthanpioglitazone)priortocohortentry (N/Y) Typeofantidiabeticmedicationimmediatelypriortocohortentryclassifiedas: Nopharmacotherapy Metforminonly Sulphonylureaonly Metformin+sulphonylureasonly Insulinwithorwithoutotherantidiabeticmedications Anyotherantidiabeticmedicationsorcombinations Typeofmodificationinbaselineantidiabetictherapyatcohortentryclassified astreatmentswitchoradditionofnewtreatmenttoallpriortreatment(ifany exisited). SeeAppendix2. Livingingeographicalareaforwhichcancerdataareavailable(Y/N,usedinthe Netherlandsonly) Other)explanatory)variables) In the hospitalrbased morbidity datasets (FIN, SWE, NL PharmacyRHospital and UK CPRD GOLDRHES dataset) somecovariatesarenotavailable,whileothersmaybebasedonhospitalizations(moreseverecases)oronuse ofspecificmedications. In the two GPRbased morbidity datasets (NL GP dataset and UK CPRD GOLD) covariates are based on GP recordsoruseofspecificmedications. FurtherdetailsofthevariabledefinitionsforeachdatasetaregiveninAppendix3andAppendix4. Version2.0,20June2013 Page17of37

18 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Page18of37 Version2.0,20June2013 Covariates)(F)=)fixed)at) cohort)entry,) )T)=)time)dependent)) )Description) Gender(F) M/F Age(F) Ageatcohortentry Cigarettesmoking(F) Cigarettesmokingstatusatcohortentry.Ifnotavailablebeforecohortentry thefirstavailablerecordaftercohortentryisadopted.ifnocigarettesmoking dataavailable,codedas unknown. NosmokingdataisavailableinFIN,SWE&NLPharmacyRHospitaldatasets. Categorizedinto:Never,Ever(currentorformer),Unknown BodyMassIndexatcohort entry(f) Classifiedasmissing,<30,30R34.9and 35.Usedinthosedatasetswhere informationavailable,fordetaileddefinitionseeappendix3.ifnotavailable atbaseline,thefirstrecordwithin12monthsofcohortentryisadopted.ifno BMIdataisavailable,codedas missing. NoBMIdataisavailableinFIN,SWE&NLPharmacyRHospitaldatasets. HbA1C(F,T) Classifiedasmissing,<7.5%,7.5R8.9%, 9.0%.BaselineHbA1Cmeasurement willbemostrecentrecordwithin6monthspriortocohortentry.persons withnobaselinehba1cwillbecodedas missing. HbA1Cwillalsobemeasuredasatimevaryingcovariate,usingthemost recenthba1crecordatapointintime. NoHbA1CdataisavailableinFIN,SWE&NLPharmacyRHospitaldatasets PSAelevated(T) PSAelevatedatanygiventimeduringthefollowRup.ClassifiedasNevervs. everelevated;andneverelevatedvs.elevatedvs.notelevated. NoPSAdatainFIN,SWE&NLPharmacyRHospitaldatasets. NumberofPSAtests(T) CumulativenumberofPSAtestsduringthefollowRupperiod. Numberofurineproteintests (T) CumulativenumberofurineproteintestsduringthefollowRupperiod History)of)diabetic)complications(FordetailsseeAppendix3) Diabeticretinopathyor maculopathy(t) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Peripheralneuropathy(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Proteinuria(microormacro) (T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0

19 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Page19of37 Version2.0,20June2013 otherwise. Diabeticnephropathy(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Chronickidneydisease(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Ketoacidosis(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Diabeticcoma(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. History)of)relevant)comorbidities)(FordetailsseeAppendix3)) Othercancers(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Myocardialinfarctionor stroke(t) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Peripheralvasculardisease (T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Othervasculardisease(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Congestiveheartfailure(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Chronicpulmonary obstructivedisease(t) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. History)of)relevant)medications(fordetailsseeAppendix4) Statinuse(T) Prioruseofstatinsorstatincombinations Classifiedasnever(0)orever(1)useevaluatedatanygiventimeduringthe followrup;startingat1ifprioruseatcohortentry,0otherwise. ARBuse(T) Prioruseofangiotensinreceptorblockers(ARB)

20 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Page20of37 Version2.0,20June2013 Classifiedasnever(0)orever(1)useevaluatedatanygiventimeduringthe followrup;startingat1ifprioruseatcohortentry,0otherwise. ACEuse(T) Prioruseofangiotensinconvertingenzyme(ACE)inhibitors Classifiedasnever(0)orever(1)useevaluatedatanygiventimeduringthe followrup;startingat1ifprioruseatcohortentry,0otherwise. BPHdruguse(T) Prioruseofdrugforbenignprostatichypertrophy(BPH) Classifiedasnever(0)orever(1)useevaluatedatanygiventimeduringthe followrup;startingat1ifprioruseatcohortentry,0otherwise. Useofotherdiabetic medications(t) Aseparatevariableforeachofthefollowingclasses:metformin, sulphonylureas,otherthiazolidinediones,alpharglucosidaseinhibitors,dppr4 inhibitors,glpr1agonists,meglitinides,amylinanalogues,insulinsandother oralantidiabeticmedications. Classifiedasnever(0)orever(1)useevaluatedatanygiventimeduringthe followrup;startingat1ifprioruseatcohortentry,0otherwise. History)of)bladder)comorbidities(FordetailsseeAppendix3) Urinaryincontinence(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Urinarytractinfection(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Pyelonephritis(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Urolithiasis(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Hematuria(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. UrinaryRetention(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Neurogenicbladder(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. Catheterization(T) Classifiedasnever(0)orever(1)withtheconditionevaluatedatanygiven

21 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy timeduringthefollowrup;startingat1ifconditionexistsatcohortentry,0 otherwise. 9.4 FOLLOW6UP)PERIOD)) ) Periodcoveredbythedatasets Dataset) Start)of)coverage) End)of)Follow6up) FIN 01January June2011 NLPharmacyRHospital 01January June2011 NLGP 01January June2011 SWE 01July June2011 UKCPRDGOLD 01January June2011 UKCPRDGOLDRHESsubset 01January December2010 Cohortentry/StartofFollowRup CohortentryandstartoffollowRupisdefinedasthedatewhenthebaselinetherapy(includingno pharmacotherapy)wasmodifiedtoincludepioglitazoneoranotherantidiabeticmedicine. ) Baselineobservationperiodbeforecohortentry Eachpatientmusthaveatleastoneyearofmembershipintherelevantstudydatabasepriortocohortentry. Thisperiodisusedtocharacterisebaselinetherapy,baselinemedicalhistory,andbaselinecovariates.This baselineobservationperiodwillstartfrom onwardsinfinland,from onwardsinsweden, andfromstartofdatabasemembershipforthenlandukdatasets.inthefinnishandswedishdatasetsthe coverageforeachpatientisassuredbyverifyingtheplaceofdomicilenotbeingabroad12monthspriortothe indexdate(fin,swe). EndofFollowRup Inthebladdercanceranalyses,eachpatientwillbefollowedRupfromcohortentryuntilthedateofdiagnosisof thefirstincidentbladdercancer,startofotherthiazolidinediones(otherthanpioglitazone)atoraftercohort entry,endofmembershipofthedatabase,endofdatabasecoverage,deathor30june2011,whicheveroccurs first. Inthemortalityanalyses,eachpatientwillbefollowedRupfromcohortentryuntildeath,startofother thiazolidinediones(otherthanpioglitazone)atoraftercohortentry,endofmembershipofthedatabase,end ofdatabasecoverageor30june2011,whicheveroccursfirst. 9.5 CENSORING)OF)FOLLOW6UP) TheNLGP,NLPharmacyRHospital,UKCPRDGOLDandUKCPRDGOLDRHESdatasetshavedefinableentryand exitdatesforeachperson,somortalitydataisnotrequiredforcensoring.exitdatewillbeusedasafollowrup censoringdate. The FIN and SWE death registers provide data on dates and causes of death with personal identification number,sex,age,placeofresidenceandcausesofdeathreportedbyicdr10diagnosiscodes,andwillbeused tocensorfollowrup.dateofdeathwillbeusedasacensoringeventinthebladdercancerincidenceanalyses, Version2.0,20June2013 Page21of37

22 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy andasanendpointforthemortalityanalyses.thefinpopulationinformationsystemregisterandtheswe Total Population Register (maintained by Statistics Sweden) contain immigration and emigration date informationandwillbeusedtocensorfollowrup.thenlpharmodatabasenetworkcontainsdatesofdeath fromhospital,pharmacyandgprecords,whicharesupplementedthroughlinkagetonationaldeathregistries, whichprovidedateofdeath,butnotcauseofdeath. 9.6 DATA)SOURCES) FinlandandSwedenbothhavewellRdevelopedpopulationRwideregistersystemswithlongitudinalfollowRup data.thepersonsareidentifiedintheregisterswithauniquepersonalidentificationnumber(pid)andthus therecordscanbelinkedforresearchpurposesonsubjectlevelbetweentheregisters.permissionstousethe data may be received upon providing a research plan to the authorities responsible for the registers.the protocolisalsosubjectedtorelevantethicalcommittee sreviewandapproval.thedatapermitprocessesand timelines for data permit granting by the authorities responsible for the registers vary between different registersandcountries,aswellastherequirementsforthelocalresearchcollaborationandthepublicationof results.registersthatareusedinthestudyaredescribedbelowindetail.thestartingyearsoftherelevant datasourcesfortheproposedstudyareshowninthetablebelow. In Finland, the nationwide prescription registers contain information on outpatient medication purchases in pharmacies[furuet#al.2010].theregistercontainsinformationonallpurchasedprescribedmedicineswithin the reimbursement scheme excluding, for example, some very cheap medicines and contraceptives. In addition, patients who are entitled for special refunds (72% or 100% reimbursement) of medicine expenses basedontheirchronicconditioncanbeidentifiedusingtheregistryforreimbursedmedications.forexample, thespecialreimbursementcategory103isfordiabetesandthecategory203for Chronicasthmaandsimilar chronicobstructivepulmonarydisease includescopdpatientswiththeinternationalclassificationofdiseases (ICDR10,10 th revision)diagnosecode J44.Thisdatabaseisusedforidentificationofcomorbidities. InSweden,thenationwideprescriptionregisterscontaininformationonoutpatientmedicationpurchasesin pharmacies [Furu et# al. 2010]. All prescribed medicines purchased in community pharmacies are included irrespectiveofreimbursementstatus. Thestudydatawillbeextractedfromtherelevantregisters.Theextractionprocess,subjecttocountryspecific adaptations, is as follows. First, patients in the study cohort with the unique personal identification (PID) number are identified from the prescription register of each country. A unique dummy study identification number(sid)iscreatedforeachpidbytheprescriptionregisterholder.thelistwiththepidrsidpairswillbe providedfromtheprescriptionregisterholderstotheotherregisterholders.eachregisterholderextractsthe relevant data according to the study protocol and links the data to the PID R SID list. Each register holder decodesthedatabydestroyingthekeybetweenthepidandsidpermanently.finally,eachregisterholder providesthederidentifiedstudydatatoepidresearch.thus,onlyderidentifieddatawillbeprovidedtoepid Researchforperformingtheanalysisofthestudydata. Table)Starting)years)of)the)relevant)registers)in)Finland)and)Sweden) Data Finland) Sweden Drug)exposure) (antidiabetic) drugs) and) TheFinnishPrescription Register/TheRegistryfor TheSwedish PrescribedDrugRegister Version2.0,20June2013 Page22of37

23 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy other)relevant)drugs)) Cancer) Mortality) Co6morbidities) ) Immigration)and) emigration)dates) ReimbursedMedications 1994 FinnishCancerRegister 1953) TheCauseofDeathRegister 1969) TheFinnishPrescription Register/TheRegistryfor ReimbursedMedications 1994 TheFinnishHospitalCare Register 1967) TheFinnishPopulation InformationSystemRegister 1950)) July)1,)2005) TheSwedishCancerRegistry 1958) TheCausesofDeathRegister 1952) TheSwedish PrescribedDrugRegister July)1,)2005) TheNationalPatientRegister 1987) TheSwedishtotalpopulation register Institutionalization)) ) InstitutionalCareRegister (SocialHILMO) 2000) The PHARMO database network is a populationrbased patientrcentric data tracking system which started in 1986 that includes high quality information of patient demographics, drug dispensing, hospital morbidity, clinical laboratory, pathology and general practitioner and other information. The PHARMO databases are linked to form a database network on a patient level and contain information required for the study. The PHARMOPharmacyRHospitaldatasetispartofthePHARMOdatabasenetworkandcontainsinformationfor3.2 million communityrdwelling inhabitants of 65 municipal areas in the Netherlands. Some databases (Cancer registry,gp,clinicallaboratory,inhospitalpharmacydata)onlyoverlappartiallywiththe3.2millionpatients includedinthepharmopharmacyrhospitaldataset.togatherthemostinformationpossiblethepharmogp datasetisanalyzedseparatelyfromthepharmopharmacyrhospitaldataset. ThePHARMOCommunityPharmacydatabaseisanationallyrepresentativedatabasethatwasstartedin1986 andincludesdispensinginformationfromcommunitypharmaciesinthenetherlands.thedatabaseincludesall GP or specialist prescribed and pharmacy dispensed healthcare products on the Dutch market. Of each dispensed drug, the Anatomical Therapeutic Chemical (ATC) code, dispensing date, prescriber, prescribed dosageregimen,dispensedquantity,costsandtheestimateddurationofuseareavailable.thecostsofdrugs are estimated from a third payer perspective and represent reimbursement costs per dispensing record includingvat.alldrugdispensingdatainthepharmacydatabaseareencodedaccordingtostandardsbased Version2.0,20June2013 Page23of37

24 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy upon the ZRIndex drug database ( It is possible to identify and classify drug use followedovertime,bothonthebasisofnationalandinternationalclassificationschemesandonthebasisofa drug sindividualactiveingredientsandadministrationforms.thedatabasecoversapproximately12%ofthe totaldutchpopulationeachcalendaryear,isavailableinrhouse,anddoesnotrequireseparatepermissionfor useineachstudy. TheHospitalizationdatabase(DutchNationalMedicalRegister(LMR)( databasecomprisingallhospitaladmissionsinthenetherlands,i.e.,admissionsformorethan24hoursand admissions for less than 24 hours for which a bed is required. These records include detailed information concerningtheprimary(mandatory)andsecondary(optional)dischargediagnoses,proceduresanddatesof hospital admission and discharge. All diagnoses are coded according to the International Classification of Diseases, 9th edition (ICDR9RCM). Procedures are coded according to the Dutch classification of procedures ( CvVR ClassificatievanVerrichtingen ). ThepartofthedatabasethatislinkedtothePHARMOcommunity pharmacydatabaseisavailableinrhouseatpharmoanddoesnotrequireseparatepermissionforuseineach study.allhospitalizationsofpatientsrepresentedinthecommunitypharmacydatabasearecoveredbythe linkage.thelinkeddatasetisreferredtoasnlpharmacyrhospitaldatasetthroughoutthisdocument. TheEindhovenCancerRegistry(ECR)isapopulationRbasedregistry(coveringademographicregionwith2.4 millioninhabitants)whichismaintainedbythecomprehensivecancercentresouthandcollatesrecordsonall newly diagnosed cancer patients in the southreastern part of the Netherlands. The ECR is notified for new casesofcancerbysixpathologydepartments,tengeneralhospitalsandtworadiotherapyinstitutes.trained registry personnel subsequently actively collect on site data on patient characteristics, diagnosis, tumour staging, cormorbidity at diagnosis, socioeconomic status at diagnosis and treatment received directly after diagnosis (e.g. chemotherapy (yes/no), radiation therapy and surgery). The ECR is linked with the PHARMO PharmacyRHospitaldataset.Theoverlapisapproximately30%(1millionpatientlives).Thisdatabaseisnota PHARMOinRhousedatabase,butanexternal(partnership)databaseandpermissionofascientificreviewboard isneededtoaccessthedata. The PHARMO GP database is part of the PHARMO database network. The database is a longitudinal observationaldatabasethatcontainsdatafromcomputerrbasedpatientrecordsofgeneralpractitioners(gps) throughoutthenetherlands,whovoluntarilychosetosupplydatatothedatabase.gpshavecompletecontrol usageoftheirdata,throughthesteeringcommitteeandarepermittedtowithdrawdataforspecificstudies. CollaboratingpracticesarelocatedthroughouttheNetherlandsandthecollaboratingGPsarecomparableto othergpsinthecountrybasedonageandgender. The PHARMO GP database currently includes about 1.5 million patients. This is the cumulative number of patientswhohaveeverbeenpartofthedynamiccohortofpatientswhowereregistered.turnoveroccursas patientsmoveandtransfertonewpractices.therecordsof transferredout patientsremaininthedatabase andareavailableforretrospectivestudieswiththeappropriatetimeperiods. Detailedclinicalinformationandprescriptiondrugsdataarecapturedinthisdatabase,includingamongother things demographic data (age, sex, patient identification, GP registration information),diagnoses,physicianr linkedindicationsfortherapy,comorbidities,drugprescriptions,laboratoryvalues(e.g.,potassium,creatinine), doctorinattendance,referralstospecialists,anda medicalchart containingfreertextasnotedbythegeneral practitioner.diagnosesandsymptomsarerecordedbasedontheinternationalclassificationofprimarycare (ICPC),whichcanbemappedtoICDR9codes,butdiagnosesandcomplaintscanalsobeenteredasfreetext, prescription data such as product name, quantity dispensed, dosage regimens, strength, and indication are Version2.0,20June2013 Page24of37

25 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy enteredintothecomputer.prescriptionsincludemedicationsprescribedbythegp(pleasenotethatspecialist prescriptionsaregenerallynotincludedinthegpdata).thenationaldatabaseofdrugs,maintainedbythe RoyalDutchAssociationfortheAdvancementofPharmacy,enablesthecodingofprescriptions,accordingto theanatomicaltherapeuticchemicalclassificationschemerecommendedbytheworldhealthorganization. ThesystemcomplieswithEuropeanUnionguidelinesontheuseofmedicaldataformedicalresearchandhas been validated for pharmacorepidemiological research. For confidentiality reasons, the database is strictly anonymized.thedatabaseisavailableinrhouseandnoseparatepermissionisrequiredforuseineachstudy. ThedatasetforthisstudyisreferredtoasNLGPdatasetthroughoutthisdocument. The NL GP dataset is linked with the Eindhoven Cancer Registry through the pharmacy dataset. Overlap betweengpdataset,pharmacy datasetandtheeindhovencancerregistryisapproximately patient lives. UKCPRDGOLDRHEScontainstheanonymisedlongitudinalmedicalrecordsmanagedbyGPsworkingtheNHS primarycaresetting.approximately8%ofallgpscontributetothecprddatabase,whichhascollecteddata prospectively from The data contains demographic information, diagnoses, prescriptions, tests and referrals.approximately56%ofcprdgoldpracticestakepartintheenglishlinkageprogram(cprdgoldr HESsubset),whichincludespatientlevellinkagetotheinpatientHospitalEpisodeStatistics(HES),allowingfor hospitalization diagnoses and procedures to be captured. Records in both settings include the unique NHS number, which is used for linkage purposes. In a similar way, CPRD GOLD data is linked to Cancer Registry (includinghistology,stageandgrading)anddeathcertificatedata(includingdateandcause).protocolsusing CPRDGOLDandlinkeddataaresubjecttoapprovalbyanIndependentScientificAdvisoryCommittee. 9.7 SAMPLE)SIZE) Theestimatednumberofpatientswhowereeverexposedandwhowereneverexposedtopioglitazoneand estimatesnumberofbladdercancersexpectedduringthefollowruparepresentedinthetablebelowforeach dataset. Powercalculationsfor i) eachofthesixdatasetsseparately, ii) poolingallmorbiditydatasetsformetaranalysis arepresentedforthecomparisonofevervs.neverexposedtopioglitazone.inpoolingofalldatasetstheuk CPRDRHESdatasethasbeendropped,asitisasubsetoftheUKCPRDGOLDdataset.Inadditionoverlapofthe NL datasets must be removed (~10%). Power calculations for the cohort study (1 to 10 matching) are presented.forindividualdatasetstheassumeddataspecificbackgroundratesandfollowruptimesgiveninthe firsttable,andforpooleddatasetstheaveragevaluesprovidedarepresentedinthesecondtable. Forthecohortstudythepoweriscalculatedfortheeffectsizes ontherelativeriskscale.AtwoRsided type1errorrateof5%isused.calculationswerecarriedoutusingthecpowerfunctionofthehmiscpackagein therrprogram[rdevelopmentcoreteam2008;petersonetal.1993;lachinandfoulkes1986;schoenfeld 1983]. ) Version2.0,20June2013 Page25of37

26 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Table)Estimated)patients)exposed)and)unexposed)to)pioglitazone)and)power)calculations)for)the)cohort)design) for)individual)datasets)and)pooled)datasets.) ) Finland) Sweden) NL) Pharmacy6 Hospital) UK)CPRD) GOLD6HES) subset) NL)GP) UK)CPRD) GOLD) Everexposedto pioglitazone Neverexposedto pioglitazone (10:1matched) MeanfollowRup time(years) Totalpersonyears AnnualIncidence rate (Incidence/ ) Cancercases (total) Power(%) IndividualCohort Studies RR=1.2 RR=1.3 RR=1.4 RR=1.5 RR= Table)Power)calculations)for)the)cohort)design)(1)to)10)matching))for))pooled)dataset.) Pooleddataset FollowRuptimet(years) 5.5 Annualincidencerateλ) (Incidence/100000) 33.7 Cumulativeprobabilityof developingbladder 0.19 cancer(%) Power(%) Page26of37 Version2.0,20June2013

27 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Cohort)studies) RR=1.2 RR=1.3 RR=1.4 RR=1.5 RR= ) 58.0) 76.3) 87.8) Data)Management) FINandSWE: TheFINcomponentofthestudywillbeundertakenbyEPIDResearchandtheSwedishpartbyCPE.Fullaudit trail starting from raw data obtained from register holders, and ending with statistical tables and graphs in reports will be maintained. In Finland data management, tabulations, graphics, and statistical modelling in carriedoutwithrdataranalyseslanguage( report"r:regulatorycomplianceandvalidationissues:aguidancedocumentfortheuseofrinregulated Clinical Trial Environments" ( In Sweden all analyses and data managementarecarriedoutusingsassoftwareversion9.3(sasinstituteinc.cary,nc,usa).sourcecodeof datamanagementanddataanalysesiskeptforinspectionforfiveyearsafterpublicationofresults.thestudy may be inspected by the sponsor s study independent representative(s) or inspected by the competent authorities.finlandandswedenwillshareacommonstatisticalanalysisplan(sap). NL: The PHARMO Institute for Drug Outcomes Research will coordinate and conduct the study with the NL databases.pharmohasdirectaccesstotheinrhousederidentifieddata,butnotthecancerregistrydata.the study is funded by Sponsor who has only access to the study results but not to the original data. All data analysisandreportingiscarriedoutaccordingtothestudyprotocol.ateamofresearcherswillconductthe study,whichiscoordinatedbythelocalprojectleader.allprogrammingwillbeindependentlyreviewedbya trainedresearcherandallresultswillbequalityreviewedbyaseniorresearcher. AlldatawillbeanalysedusingSASEnterpriseGuideversion4.3(SASInstituteInc.,Cary,NC,USA).Allanalyses willbeconductedunderwindowsusingsasversion9.2.sourcecodeofdatamanagementanddataanalysesis keptforinspection. UK: The CPRD GOLD and CPRD GOLDRHES components of the study will be undertaken by CPRD, and results providedtothesponsor.alldataanalysisandreportingwillbecarriedoutaccordingtothestudyprotocoland subsequentdataspecificationdocument.thestudywillbeoverseenbytheheadofresearch,managedand conductedbyaseniorresearcher,withappropriatequalityreview.alldatawillbeanalysedusingstatav11.2 andcodingkeptandarchivedforinspection. PooleddatasetformetaRanalysis ThepooleddatasetwillbeheldbytheprincipalinvestigatorinFinland,anddatamanagementwillbehandled asdescribedaboveforfinland.thepooleddatasetwillcontainthecoresetofvariablesthatareavailableinall datasets.smoking,bmi,hba1candotherlaboratoryrbasedvariablearenotavailableinalldatasetsandwill Version2.0,20June2013 Page27of37

28 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy thusnotbeincludedinthemetaranalysis.asthereisoverlapofpatientswithinthe2nlpharmodatasetsand withinthe2ukcprddatasets,individualpatientswillonlybeincludedonceinthepooleddataset. Data)retention)and)archiving) Allstudydataandsupportingdocumentswillberetainedforaminimumoffive(5)yearsafterthestudyends. Securearchiveswillbemaintainedfortheorderlystorageandretrievalofallstudyrelatedmaterial.Anindex shallbepreparedtoidentifythelocationofthearchivedcontents.accesstothearchiveswillbecontrolledand limitedtoauthorisedpersonnelonly. 9.9 DATA)ANALYSES) The principles of the statistical analysis are outlined below. More detailed statistical analysis plans will be writtenseparatelyforeachdatasetandforthemetaranalysis. Missing)data) Ifavariableistotallymissingfromadatabase(forexamplesmokingandBMI),itisexcludedfromtheanalysis inthatdatabase.ifavariableismissingforonlysomeofthepatientswithinadatabaseamissingdatacategory isaddedandusedintheanalysis. Population)summary) Thecharacteristicsofthestudypopulationwillbedescribedwithdescriptivestatisticsattheentryintothe studycohort.thefollowingvariableswillbeincludedinthepopulationsummary:age,gender,yearofentry, duration of diabetes mellitus, history of relevant diabetic medications, history of relevant concomitant medications,historyofdiabeticcomplications,historyofbladdercomorbiditiesandhistoryofotherrelevant comorbidities.) Thepopulationsummarywillbestratifiedaccordingtopioglitazoneexposureusingneverusevs.everuseas theexposurevariable.summariesforeachcountrywillbeprovidedseparately.summariespoolingthecountry specificresultswillbeprovidedforthosevariablesthatareavailablefromeachcountry. Descriptive)analysis) Comparisonsforthevariouspioglitazoneexposuredefinitionswillbeperformed. Crudebladdercancerincidence andmortality rates with 95% confidence intervals (CI) will be estimated for eachpioglitazoneexposuredefinitionseparatelywithinthestrataof gender,age,yearofentry,durationof disease, history of relevant diabetic medications, history of relevant concomitant medications, history of diabeticcomplications,historyofbladdercomorbiditiesandhistoryofotherrelevantcomorbidities. Crude incidence rates for each country will be provided separately. Summaries pooling the country specific resultswillbeprovidedforthosevariablesthatareavailablefromeachcountry. The incident bladder cancer cases occurring during the followrup period are described in detail using the followinginformationwhereavailable: patient sageatdiagnosis Version2.0,20June2013 Page28of37

29 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy tumourstaging(behaviour,morphologycode5 th digit), histologygradeanddifferentiation(morphologycode6 th digit), treatmentsusedandtheirtemporalassociationwiththetimeofdiagnosis. Emphasiswillbemadetopresentthetemporalassociationbetweenthetimeofdiagnosisandexposureto pioglitazone(evervs.never;cumulativeexposure;timesincestart,timesincelastdose). Formal)analysis)of)individual)cohorts) Thehazardratio(HR)estimateswith95%CIsforeachpioglitazoneexposuredefinitionwillbeestimatedusing theconventionalcox sproportionalhazardsmodelwithacountingprocessapproachwhichenablesthefollowr uptimeofeachpatienttobesplitintoseveralperiodsandthusallowsadjustmentsforrelevantbaselineand timerdependentcovariatesinthemodelspecification.separateanalysesofbladdercancerincidence,bladder cancermortalityandallrcausemortalitywillbeperformedforeachdataset. Apropensityscorematchedcohortanalysis(Lietal2012,Spreeuwenberg2005)willbeperformed. Inthecohortanalysistwomodelswillbeused:i)thecrudemodelwithpioglitazoneexposureasevervsnever exposed included only and ii) the base model with age at cohort entry, sex and use of metformin, sulphonylureas,insulinsorotherantibiabeticdrugseachclassifiedasevervs.neverexposedincludedinthe crudemodel.inadditionanadjustedmodelwillbecreatedusingthefollowingforwardandbackwardselection procedure: 1. Identify candidate covariates as those covariates with at least 5% prevalence in pioglitazone or referencegroupandwiththeprvalueoftheunivariateassociationbetweenthecovariateandbladder cancerincidence< Startwiththebasemodel 3. Foreachcandidatecovariateproduceanewadjustedmodelwiththecandidatecovariateoneatatime addedinthebasemodel. 4. Consider each candidate covariate as a potential confounder if the relative change in the HR of pioglitazoneexposureisatleasta10%relativewhencomparingtothebasemodel.thisprocesswill generateasetofpotentialconfounders. 5. Addallpotentialconfounderssimultaneouslyinthebasemodel. 6. Remove each potential confounder one at a time from the model and see if this results in a 10% relativechangeinthehrofpioglitazoneexposure.ifanypotentialconfounderdoesnotfulfilthe10% threshold,droptheonewiththesmallestrelativechangeinhr.repeattheprocessuntilnofurther changesareneeded.thiswillbetheadjustedmodel. Thecovariatesincludedinadjustedmodelwillbeusedintheanalysisoftheeffectofdifferentpioglitazone exposure definitions (never/ever exposed; duration of exposure; cumulative dose; time since last dose) on bladdercancerincidence,allcausemortalityandbladdercancermortalityaswellasinthesensitivityanalyses. Stratified)analyses) The risk models will be presented stratified, if there are sufficient event, by baseline variable used in the matchingorpropensityscore:) Durationoftreateddiabetes UseofotherTZDs(otherthanpioglitazone)priortocohortentry HistoryofChronicKidneydiseaseorrenalimpairment Version2.0,20June2013 Page29of37

30 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Historyofproteinuria(microormacro) Historyofdiabeticnephropathy Sensitivity)analysis) Thefollowingsensitivityanalyseswillbeundertaken: 1. An analysis to assess the impact of exclusion of unmatched pioglitazonerexposed patients in the propensityscorematchedcohort.thecharacteristicsoftheunmatchedpioglitazonerexposedpatients arecomparedtothoseofthematchedpioglitazonepatients.thebladdercancerincidenceratesare compared between these two patient groups. If the incidence rates do not differ then no further analysisisrequired.iftheincidenceratesaredifferent,thenanincidencedensitysamplingfromthe referencegroupwillbeperformedtoformanewreferencegroupwithsimilardistributionoftheyear ofcohortentryasinthepioglitazoneexposedpatients.theresultingcohortsareanalysedusingacox modeltoassessthedifferencesbetweenthematchedandunmatchedanalyses.thisanalysiswillonly beundertakenifmorethan2.5%ofthepioglitazonepatientsarenotmatched. 2. Ananalysisexcluding allbladdercancersoccurringwithin12 monthsaftercohortentrytoallowfor bladdercancerlatency. 3. Ananalysisexcludingadenocarcinomasandsquamouscellcarcinomaswillbeperformedtolookatrisk withintransitionalcellurothelialtumours. 4. AnanalysiswherefollowRuptimeiscensoredwhenthefirstgapofatleast4monthsoccursbetween two prescriptions will be performed to assess the impact of removing patients who have gone to nursinghomes(finandswedatasetsonly). 5. Analysis to assess the impact of adjusting/not adjusting for smoking status, BMI and HbA1C informationinnlandukdatasets(nlandukdatasetsonly). 6. AnalysisoftheimpactofusingfirstsmokingandBMIrecordaftercohortentryforpersonswithno baselinesmokingorbmirecordontheadjustedriskestimates(nlandukdatasetsonly). 7. Analysiscomparingriskestimatesfromfullcohortvs.anincidentsubRcohort.TheincidentsubRcohort will only include patients with at least 12 months of database membership before first diabetes treatment(nlandukdatasetsonly).note:thesweandfincohortsareincidentcohorts. 8. Analysisoftheassociationofevervs.neveruseofpioglitazonewithbladdercancerincidencewhen insulinuseisincludedasacumulativedurationintheadjustedmodel. 9. Analysis in which the pioglitazone exposure group is changed from at least one prescription or dispensingofpioglitazonetoatleast2prescriptionsordispensingwithina6monthperiod.toensure immortal time bias is not introduced, the cohort entry date will be amended for both exposed and unexposedpatientstobetheoriginalcohortentrydateplussixmonths. 10. Analysis in which the definition of incident bladder cancer is broadened to include neoplasms of uncertainandunknownbehaviour.thebladdercancerriskmodelswillbereanalysed,ifthereareat least5%increaseinthebladdercancereventcountafterinclusionofneoplasmofuncertainbehaviour orunknownbehaviourofthebladderinthedefinitionofanincidentbladdercancerevent.thiswill assessthespecificityofanyobservedassociation.) Sensitivityanalyseswillalsobeundertakenonanysignificantunplanneddatahandlingdecisions. Thefollowingvalidationanalysiswillbeundertaken: In the NL PharmacyRHospital, NL GP and UK CPRD GOLDRHES datasets, the cancer registry data, GP recorddataandhospitalrecordsdatawillbecomparedtoassessthevalidityofusinggporhospital dischargedataforbladdercancerdiagnosis. Version2.0,20June2013 Page30of37

31 Meta6analysis)) Version2.0,20June2013 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy Thepooleddatasetwillbeanalysedusingthesamemethodsandapproachastheindividualcohortanalyses. TheCoxmodelswillincludeacohortdatasetidentifierasacategoricalcovariate.Themodellingapproachwill besimilartothestepwisemodellingdescribedabove. Two separate metaranalyses will be performed. The first metaranalysis will use a pooled dataset containing common variables from all datasets. If smoking, BMI, HbA1C and urinary/renal marker information can be obtainedfortheswedataset,thenthesecondmetaranalysiswilluseasubsetofthepooleddatasetincluding only datasets where information on these variables is also available (i.e. excluding FIN dataset and NL PharmacyRHospital dataset). If these variables cannot be obtained for the SWE dataset, then only the first metaranalysiswillbeperformed Quality Control The study will be conducted as specified in this protocol. All revisions to the protocol shall be properly documented as protocol amendments and when necessary such protocol amendments are delivered to registerholder(s)wheneveramendment(s)tothedatapermissionsarerequired. ThestudyprotocolhasbeenwrittenbyfollowingtheCodeofConductbytheEuropeanNetworkofCentresfor PharmacoepidemiologyandPharmacovigilance[ENCePP2011]thatprovidesasetofrulesandprinciplesfor postrauthorization studies with regard to the best practices and transparency, thereby promoting scientific independence of such studies. The ENCePP is a project led by the European Medicines Agency to further strengthen the postrauthorization monitoring of medicinal products in Europe by facilitating the conduct of multircentre, independent, postrauthorization studies focusing on safety and on benefit/risk. The study has beenregisteredtotheencepp serregister( willalsobepublishedonthesamesite. ThestudyprotocolalsofollowsthekeyelementsoftheGuidelineforGoodPharmacoepidemiologyPractices byinternationalsocietyforpharmacoepidemiology[ispe2007],andtherecentdraftguidanceforindustryand FDAStaff BestPracticesforConductingandReportingPharmacoepidemiologicSafetyStudiesUsingElectronic HealthcareDataSets [FDA2011] Strengths)and)limitations)of)the)research)methods) StatisticalPower Severalpreviousstudieshaveobservedthateveruseofpioglitazoneisweaklyassociatedwithbladdercancer (pooledrelativerisk=1.2).thesixindividualdatasetsinthisprotocolareeachtoosmalltodetectarr<1.5. However,themetaRanalysisofthepooleddatasethasapredictedpowerof85%todetectaRR=1.5,anda powerof73%fordetection ofarr=1.4.themetaranalysisisnotsufficientlypowered(32%)todetectvery weakassociationsofrr=1.2.thisstudyhasincludedthemaineuropeandatasourcesthatcontaininformation toenablethecontrollingofpotentialchannellingbias,detectionbiasandconfounding.itispossiblethatthere aresomeothersmalleuropeandatasets,buttheirinclusionwouldonlyhaveminimalimpactonimprovingthe study spowertodetectveryweakassociations. Page31of37

32 Studydesign PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy ApotentiallimitationofusingmatchingisthatanypioglitazoneRexposedpatientwithoutamatchedcontrol would be excluded from the analyses. A risk minimisation strategy has been included, whereby a wider matching criterion would be applied propensity score of to any unmatched cases after the first round of matching. In addition, if the unmatched pioglitazone users change the incidence rates of the outcomes, a sensitivity analysis will be undertaken to compare the matched analysis to an unmatched analysis with the propensityscoreincludedasacovariate. Confoundingandbias While a weak association may be real and causal in nature, they are equally as likely to simply reflect differencesbetweentheexposedandreferencegroup.potentialsourcesofdifferencebetweenpioglitazoner exposed and pioglitazonerunexposed groups that could lead to misleading results include channelling bias, detectionbiasanduncontrolledconfounding. PioglitazoneRtreated patients are likely to have more severe diabetes than patients treated with other antidiabetic medications, and severity of diabetes is believed to be an independent risk factor for bladder cancer.pioglitazoneisalsorecommendedforuseinobesepatients,andobesityisanotherindependentrisk factorforbladdercancer.channellingbiasisprobablythemostimportantissuesinpharmacoepidemiologyand observationalresearchstudies.previousobservationalstudieshavebeenlimitedintheirabilitytocontrolfor channellingbias.inthisprotocol,particularlyattentionhasbeengiventochannellingbiasanditscontrolby matchingonpropensityscoresbasedonseveralproxymeasuresofdiabeticseverity. Fewofthepreviousstudieshaveconsidereddetectionbias.Microalbuminuriaandproteinuriamaybemore common in patients treated with pioglitazone. Diabetics with microalbuminuria or proteinuria are likely to undergothoroughrenalandbladderexaminationandbesubjecttofurthersurveillancethanpatientswithout microalbuminuriaorproteinuria.thiswouldleadtoabiaswithpioglitazonerexposedpatientsbeingmorelikely tohaveabladdercancerdetectedbecauseofurinaryinvestigations.thisstudywillexaminetheoccurrenceof microalbuminuria, proteinuria and other urinary bladder/renal signs and symptoms as possible sources of detectionbias.theukandnldatasetswillhavedataonthesevariables,butthefinandswehospitalbased dischargerecordsdonotcontainlaboratorydatabutincludeonlydiagnosesforproteinuria. WeattempttocontrolfortheconfoundingbiasbyusingbaselineandtimeRdependentvariablesofpotential confounders in the statistical analyses. Although HbA1C, smoking and BMI are not available in the hospital discharge datasets, sensitivity analyses in the UK and NL datasets will investigate the role they play on pioglitazonerelatedrisk,afteradjustingforthechannellingbias. Exposure Prescriptionsdonotcontaininformationonthediagnosis.Itisassumedthatpatientsusingoralantidiabetic medications(atccodea10b)areoftype2diabeticpatients.afurthercriterionofage>40yearshasbeen addedtoavoidinclusionofpatientswithtype1diabetes.type1diabetes,diabeteswithmalnutrition,other diabetes,orgestationaldiabetesareconsideredasexclusioncriteria. Prescriptiondatabasesdonotcontaininformationonmedicationuseinhospitalsornursinghomes.Thismay causes missrclassification of the current exposure and underestimation of the cumulative exposure and cumulativedoseinsomepatients.theeffectofsuchbiasisevaluatedinasensitivityanalysiswherefollowrup Version2.0,20June2013 Page32of37

33 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy time is censored when the first gap of at least 4 months occurs between two prescriptions. In Finland the informationoninstitutionalization(otherthanhospitalization)isusedtoavoidsuchgaps. TheFinnishprescriptiondatabasecontainsinformationonallmedicationspurchasedprovidedthattheircost exceedsthethresholdforbasicreimbursementthatappliestoallprescribedmedications.inthisstudybiasdue topossiblymissingsomeofthepurchasedmedicationsisnotexpected,becausethepatientsareentitledto specialreimbursementandthereforetheyareexpectedtohavealltheirdiabetesmedicationsregistered. Thisstudywillusethesameexposurecategoriesasusedinthefinal(10Ryear)followRupoftheKPNCcohort study. This will enable comparisons to be made between this study and the KPNC cohort study s 10Ryear followrupanalyses.theprimaryanalyseswillalsoberepeatedusingthesameexposurecategoriesashekpnc 3 rd interim (5Ryear) analysis. These same exposure categories were also used in the French cohort study (CNAMTS2011). Cancerdata The cancer registers have nationrwide coverage, except in NL where linkage is to a regional registry that contains more detailed information than the national registry. The data collection on all cancer cases is compulsory and the informants submitting data on cancer patients to the registry include all hospitals, physicians, pathological, cytological and haematological laboratories, and dentists. The completeness and accuracyoftheregisterdataareconsideredgoodforscientificresearch[teppoetal.1994;korhonenetal. 2002;Barlowetal.2009;vanHerkRSukeletal.2011].Therefore,reportingbiasfortheoutcomesofinterestin thecurrentstudyisnotconsideredtohavemajorimpactontheresults. Since2007allthreemorphologytypesofcancershavebeenregisteredintotheFinnishCancerRegistry.Before thatonlymalignantcancersandin#situcancerswereincludedlackingpunlmps. LosstofollowRupfromnationwideregisters(FINandSWE)meansemigrationtoothercountries.IntheUKand NLdatasetsattritionmaybeduetopersonsmovingtoageographicareanotcoveredbythePHARMO/CPRD Golddatabases.Itisunlikelythatsuchattritionwouldbedifferentbetweenthetreatmentgroups.Therefore attritionbiasisnotconsideredtohavemajorimpactontheresults. 10. PROTECTION)OF)HUMAN)RIGHTS) Toensurethefulldataprotectionofpatients,alltheresearchdataineachcountryisanonymized.Approval fromrelevantethical/researchreviewboardswillberequiredbeforeconductingthestudy. 11. MANAGEMENT)AND)REPORTING)OF)ADVERSE)EVENTS/REACTIONS) Thestudydesignisaretrospectivecohortbasedonsecondaryuseofexistinghealthcarerecords.Therefore, noadverseeventorseriousadversereportingwillbeperformedbeyondthedefinedstudydeliverables. Version2.0,20June2013 Page33of37

34 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy 12. COMMUNICATION)OF)STUDY)RESULTS) TheprincipalandcoRinvestigatorswillwritethestudyreports.ThereportisdeliveredtotheSponsor.Basedon these results the principal and corinvestigators will corauthor scientific manuscript(s) of the results to be published. The publication strategy has been defined in the research agreement between the principal investigatorsandthesponsor. The study has been registered to the ENCePP s ERregister ( The results will also be published on the same site. TheprincipalinvestigatorwillseekENCePPSealofApproval.Thecriteriaandprocessforsharingtheanalytical countryspecificdatasetsandmetaranalysisdatasetforthirdpartiesaredescribedinappendix5.accordingto theenceppcodeofconduct,theprincipalinvestigatorisresponsibleofpublicationoftheresults. A summary of the main results of the study, whether positive or negative and including results from prematurelyterminatedstudies,willalwaysbemadeavailabletothepublic.anabstractofthestudyfindings willbeprovidedthroughtheencepperregisterofstudieswithinthreemonthsfollowingthefinalstudyreport. TheprincipalinvestigatormayasktheENCePPSecretariattodelaythepublicationofthisabstractforalimited periodpendingresponsetopeerrreviewcomments.theoutcomeofastudywillalwaysbepresentedinan objectiveandtruthfulmannerprovidingacomprehensiveandaccuratedescriptionofthefindings.innoway shalltheinterpretationandpresentationoftheresultsbeaimedtowardsanycommercial,financialorpersonal interests. The study Sponsor is entitled to view the final results and interpretations thereof prior to submission for publication and to comment in advance of submission as agreed in the research contract and without unjustifiablydelayingthepublication. 13. APPENDICES) APPENDIX)1.)ANTIDIABETIC)MEDICATIONS) APPENDIX)2.)CALCULATION)OF)EXPOSURE) APPENDIX)3.)VARIABLE)DEFINITIONS) APPENDIX)4.)OTHER)MEDICATIONS) APPENDIX)5.)CRITERIA)AND)PROCESS)FOR)SHARING)THE)ANALYTICAL)COUNTRY)SPECIFIC) DATASETS)AND)META6ANALYSIS)DATASET)FOR)THIRD)PARTIES) Version2.0,20June2013 Page34of37

35 14. REFERENCES) PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy BarlowL,WestergrenK,HolmbergL,TalbäckM.ThecompletenessoftheSwedishCancerRegister asample surveyforyear1998.acta#oncologica2009,48:27r33. CNAMTS,Caissenationaledel assurancemaladie.riskofblabbercancerinpeoplewithdiabetestreatedwith pioglitazoneinfrance:agroupstudyonsniiramandpmsidata.finalreport7/06/ CancerIncidenceinSweden2010.OFFICIALSTATISTICSOFSWEDEN,Statistics HealthandMedicalCare. ENCePP.TheENCePPCodeofConductforScientificIndependenceandTransparencyintheConductof PharmacoepidemiologicalandPharmacovigilanceStudies.21November2011,EMA/929209/ FinnishCancerRegistry.[Online]CancerStatisticsatwww.cancerregistry.fi,updatedon FDA,FoodandDrugAdministration.GuidanceforIndustryandFDAStaff.BestPracticesforConductingand ReportingPharmacoepidemiologicSafetyStudiesUsingElectronicHealthcareDataSets.February Furu K, Wettermark B, Andersen M et al. The Nordic countries as a cohort for pharmacoepidemiological research.basic#clin#pharmacol#toxicol.2010,106:86r94. ISPE,InternationalSocietyforPharmacoepidemiology.GuidelinesforGoodPharmacoepidemiological Practices.Revision2:April KorhonenP,MalilaN,PukkalaEetal.TheFinnishCancerRegistryasFollowRUpSourceofaLargeTrialCohort. Acta#Oncologica.2000,41:381R388. LachinJM,FoulkesMA.EvaluationofsamplesizeandpowerforanalysesofsurvivalwithallowancesfornonR uniformpatiententryandlossestofollowrup,nonrcomplianceandstratification.biometrics.1986,42: LewisJDet#al.Riskofbladdercanceramongdiabeticpatientstreatedwithpioglitazone.Interimreportofa longitudinalcohortstudy.diabetes#care.2011,34:916r922. LewisJDet#al.Riskofbladdercanceramongdiabeticpatientstreatedwithpioglitazone.4 th Interimanalysis(8R year) report with data from January 1, 1997 to December 31, Unpublished data, 30 May fa486.rlmOahaInljyqA4IpR9BcxaNb3i?field=documents.otherDocument%5B0%5D&id=2903 Version2.0,20June2013 Page35of37

36 PanEuropeanMultiDatabaseBladderCancerRiskCharacterisationStudy LiW,MacDonaldTM,MackenzieIS.Pioglitazoneandbladdercancer:apropensityscorematchedcohortstudy. BrJClinPharmacol Nielsen, L H et al. Using prescription registries to define continuous drug use: how to fill gaps between prescriptions.pharmacoepidemiology#and#drug#safety.2008,17:384r388.) Nielsen, L H et al. Estimating the effect of current, previous and never use of drugs in studies based on prescriptionregistries.pharmacoepidemiology#and#drug#safety.2009,18:147r153. PetersonB,GeorgeSL.Controlled#Clinical#Trials.1993,14: R Development Core Team. R: A Language and Environment for Statistical Computing. Vienna, SchoenfeldD.SampleRSizeFormulafortheProportionalRHazardsRegressionModel.Biometrics.1983,39: SpreeuwenbergMAetal.Themultiplepropensityscoreascontrolforbiasinthecomparisonofmorethantwo treatmentarms:anintroductionfromacasestudyinmentalhealth.medical#care.#2005,48:166r174. TeppoL,PukkalaE,LehtonenM.DataqualityandqualitycontrolofapopulationRbasedcancerregistry.Acta# Oncologica.1994,33:365R369. van HerkRSukel MP, Lemmens VE, PollRFranse LV, Herings RM, Coebergh JW. Record linkage for pharmacoepidemiological studies in cancer patients. Pharmacoepidemiol# Drug# Saf. 2012;21(1):94R103. Epub 2011/08/04.# Version2.0,20June2013 Page36of37

37