TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page1

PublishedinTheLancetOncology,http://dx.doi.org/10.1016/S1470-2045(15)00361-7

Prostatecancerscreeninginmenaged50–69years(STHLM3):aprospectivepopulation-baseddiagnosticstudyHenrikGrönberg,JanAdolfsson,MarkusAly,TobiasNordström,PeterWiklund,YvonneBrandberg,JamesThompson,FredrikWiklund,JohanLindberg,MarkClements,LarsEgevad,MartinEklund

Summary

BackgroundTheprostate-specificantigen(PSA)testisusedtoscreenforprostatecancerbuthasahighfalse-positiveratethattranslatesintounnecessaryprostatebiopsiesandoverdiagnosisoflow-riskprostatecancers.Weaimedtodevelopandvalidateamodeltoidentifyhigh-riskprostatecancer(withaGleasonscoreofatleast7)withbettertestcharacteristicsthanthatprovidedbyPSAscreeningalone.

MethodsTheStockholm3(STHLM3)studyisaprospective,population-based,paired,screen-positive,diagnosticstudyofmenwithoutprostatecanceraged50–69yearsrandomlyinvitedbydateofbirthfromtheSwedishPopulationRegisterkeptbytheSwedishTaxAgency.Menwithprostatecanceratenrolmentwereexcludedfromthestudy.ThepredefinedSTHLM3model(acombinationofplasmaproteinbiomarkers[PSA,freePSA,intactPSA,hK2,MSMB,MIC1],geneticpolymorphisms[232SNPs],andclinicalvariables[age,family,history,previousprostatebiopsy,prostateexam]),andPSAconcentrationwerebothtestedinallparticipantsenrolled.TheprimaryaimwastoincreasethespecificitycomparedwithPSAwithoutdecreasingthesensitivitytodiagnosehigh-riskprostatecancer.Theprimaryoutcomeswerenumberofdetectedhigh-riskcancers(sensitivity)andthenumberofperformedprostatebiopsies(specificity).TheSTHLM3trainingcohortwasusedtotraintheSTHLM3model,whichwasprospectivelytestedintheSTHLM3validationcohort.LogisticregressionwasusedtotestforassociationsbetweenbiomarkersandclinicalvariablesandprostatecancerwithaGleasonscoreofatleast7.ThisstudyisregisteredwithISCRTN.com,numberISRCTN84445406.

FindingsTheSTHLM3modelperformedsignificantlybetterthanPSAalonefordetectionofcancerswithaGleasonscoreofatleast7(p<0·0001),theareaunderthecurvewas0·56(95%CI0·55–0·60)withPSAaloneand0·74(95%CI0·72–0·75)withtheSTHLM3model.AllvariablesusedintheSTHLM3modelweresignificantlyassociatedwithprostatecancerswithaGleasonscoreofatleast7(p<0·05)inamultiplelogisticregressionmodel.AtthesamelevelofsensitivityasthePSAtestusingacutoffof≥3ng/mLtodiagnosehigh-riskprostatecancer,useoftheSTHLM3modelcouldreducethenumberofbiopsiesby32%(95%CI24–39)andcouldavoid44%(35–54)ofbenignbiopsies.

InterpretationTheSTHLM3modelcouldreduceunnecessarybiopsieswithoutcompromisingtheabilitytodiagnoseprostatecancerwithaGleasonscoreofatleast7,andcouldbeasteptowardspersonalisedrisk-basedprostatecancerdiagnosticprogrammes.

FundingStockholmCountyCouncil(StockholmsLänsLandsting).

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page2

Researchincontext

EvidencebeforethisstudyWesearchedPubMedbetweenJan1,2000,and31Dec,2009,usingsearchtermsrelatedtoprostatecancer,biomarkers,andscreeningforarticleswritteninEnglish.Prostate-specificantigen(PSA)iswidelyusedasaninitialscreeningtestforprostatecancerandislargelycreditedwiththereductioninprostatecancermortalityreportedduringthepasttwodecades.TheEuropeanRandomisedStudyofScreeningforProstateCancer(ERSPC)showeda21%reductioninprostatecancermortalitywithstructuredPSAscreeningafter13years.However,thepoorspecificityofPSAtranslatesintomanyunnecessaryprostatebiopsiesandoverdiagnosisoflow-riskprostatecancers.Subsequently,nogovernmentalbodyhasrecommendedstructuredPSAscreeningbecauseofthepotentialharmsofoverdiagnosis.PSAistheonlybiomarkerthathasbeenassessedprospectivelyandinrandomisedcontrolledtrials.AlternativeplasmaproteinbiomarkersotherthanPSAhavebeenproposedtoaddresstheseconcerns.InadditiontototalPSA,theotherplasmaproteinbiomarkersusedintheSTHLM3studywereselectedbyasystematicscientificliteraturesearchduring2010andtwosubsequentvalidationstudiesusingtheSTHLM2cohort(appendix).FiveadditionalbiomarkerswereselectedbasedontheirassociationswiththepresenceofprostatecancerwithaGleasonscoreofatleast7.Urine-basedmarkerswereexcludedbecausetheywouldrequireaprostatemassage,whichmakesthemunsuitabletouseinapopulationscreeningsetting.Onthebasisoftheliteratureandgeneticassessmentofpreviousstudies,254SNPswereselectedbasedontheirassociationwithprostatecancerrisk.

AddedvalueofthisstudyWeshowedthattheSTHLM3model,acombinationofplasmaproteinbiomarkers,geneticpolymorphisms,andclinicalvariables,cansignificantlyimproveprostatecancerscreeningspecificitywiththesamesensitivitycomparedwiththePSAtestingusingacutoffofatleast3ng/mL.TheSTHLM3modelcanidentifycancerswithaGleasonscoreofatleast7inmenaged50–69yearsandidentifyclinicallysignificantprostatecancersinthePSAconcentrationrangeof1–3ng/mL.TheSTHLM3modelalsoincludestwonovelbiomarkersandgeneticsmarkersthathavenotbeenpreviouslyincludedinaprospectivediagnosticstudy.Additionally,STHLM3isthefirstlarge-scalediagnosticstudyinprostatecancerwherebiopsydecisionisprospectivelybasedontheresultsfromthepredefinedSTHLM3model.Finally,STHLM3ispopulationbased,thusminimisingselectionbiasandincreasingthegeneralisability.

ImplicationsofalltheavailableevidenceTogetherwiththeresultsfromtheEuropeanRandomizedStudyofProstateCancer,thefindingsfromtheSTHLM3studyindicatethatprostatecancermortalitycanbereducedbutwithsubstantiallyfewerbiopsiesandreducedoverdiagnosis.

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page3

IntroductionLevelsofprostate-specificantigen(PSA)arewidelyusedasaninitialscreeningtestforprostatecancerandislargelycreditedwiththereductioninprostatecancermortalityreportedduringthepasttwodecades.1,2EvenwithevidencethatPSA-basedscreeninghasreducedprostatecancermortality,nogovernmentalbodyhasrecommendedstructuredPSAtestingbecauseofthepotentialharmsofoverdiagnosis.3,4Nonetheless,opportunisticscreeningisfrequent,resultingin1000000prostatebiopsiesannuallyintheUSA.5Thereisgrowingconcernabouttheincreasingincidenceofseriousinfectionscausedbymultidrug-resistantbacteriaandrelatedinfectiouscomplications(eg,2%ofpatientsdevelopsepticaemiainStockholm)afterprostatebiopsy.6,7Alternativeplasmaproteinbiomarkershavebeenproposedtoaddresstheseconcerns;however,nonehasbeenprospectivelyassessedinscreeningstudies.8,9Atleast100singlenucleotidepolymorphisms(SNPs)havebeenidentified,accountingforabout30%oftheinheritedriskforprostatecancer.10CombinationofPSAwithageneticscorebasedontheseSNPshasbeensuggestedtoincreasethespecificityofprostatecancertesting.11,12Wereportarisk-basedmodelforprostatecancerscreeningthatcombinesPSA,SNPs,clinicalvariables,andestablishedandnovelplasmaproteinbiomarkers(theSTHLM3model).WeaimedtoassesswhethertheSTHLM3modelcouldincreasethespecificityofdetectingmenwithhighrisk(Gleasonscoreofatleast7)prostatecancerandthussubstantiallyreducetheproportionofmenundergoingprostatebiopsy,whilemaintainingthesamesensitivitytodetecthighriskprostatecancersasthePSAtestalone(whichusesathresholdofatleast3ng/mL).

Methods

StudydesignandparticipantsSTHLM3wasaprospective,population-based,diagnosticstudyfollowingapaired,screen-positivedesign,inwhichwecomparedtheSTHLM3modelwithPSAinmenaged50–69yearsfromStockholm,Sweden.Men,irrespectiveofanycomorbidityexceptprostatecancer,wererandomlyselectedbydateofbirthfromtheSwedishPopulationRegisterkeptbytheSwedishTaxAgencyandinvitationswerepostedtothem.Thetwoscreeningmethods,PSAandtheSTHLM3model,werebothtestedineachstudyparticipant.PSAistheonlybiomarkerprospectivelyassessedinpopulation-basedscreeningtrialswithapositiveeffectonprostatecancermortality2andtheclinicalusefulnessofotherbiomarkersinthiscontextislimited.Subsequently,wechosetouseaPSAconcentrationofatleast3ng/mLasthereferencetoinferthesamemortalityeffectasseeninthesetrials.TheSTHLM3modelisatestconsistingofacombinationofplasmaproteinbiomarkers(PSA,freePSA,intactPSA,hK2,MSMB,andMIC1),geneticmarkers,clinicalvariables(age,familyhistory,previousprostatebiopsy),andaprostateexam(digitalrectalexamandprostatevolume).PlasmaproteinbiomarkersusedinSTHLM3wereselectedfromascientificliteraturesearchandtwosubsequentvalidationstudies(appendix).Forthegeneticmarkers,wetestedthe254SNPsshowntobeassociatedwithprostatecancerinourpreviousstudies.11,13TheseSNPswerecombinedinageneticscoreusingoddsratiosestimatedfromcohortsinthesepreviousstudies(appendix).11,13WesubsequentlyrankedtheSNPsaccordingtotheirpvalueandincludedSNPsinthegeneticscoreintheorderoftherankedlist.SNPsthatcouldnotbegenotypedreliablywereexcludedfromthescore,leaving232SNPsintheSTHLM3model.STHLM3wasdoneintwoseparatephases.TheSTHLM3trainingcohortrecruitedmenfromMay,2012,toMay,2013,followedbytheSTHLM3validationcohort,whichrecruitedmenfromAugust,2013,toDecember,2014.Thetrainingcohort(appendix)wasusedtotrainandpredefinetheSTHLM3modelalgorithm.ThevalidationcohortwasusedtoprospectivelytesttheSTHLM3algorithm.Localandgovernmentalethicscommitteesreviewedandapprovedthestudyprotocol.Allparticipantsprovidedwritteninformedconsent.

ProceduresThestudyparticipantswereenrolledandbloodwasdrawnat67clinicallaboratoriesinStockholmcollaboratingwithSTHLM3.Previousprostatebiopsysamplerecords(within10yearsbeforeinclusioninSTHLM3)weretakenfromhighlyaccuratehealth-careregisters,14togetherwithself-reportedfamilyhistoryofprostatecancer(first-degreerelatives).PSAlevelswereanalysedinallpatientsandinthosewithaPSAconcentrationofatleast1ng/mL,additionalbiomarkerswereanalysed(geneticandplasmaproteinmarkers).MenwithPSAconcentrationsoflessthan1ng/mLhaveverylowprevalenceofhigh-riskcancers.ThePCPTtrial15showedthatalmosthalfthestudypopulationhadaPSAconcentrationlessthan1ng/mLandonly7·2%ofhigh-riskcancerswereidentifiedinthese.Moreover,menwithaPSAconcentrationoflessthan1ng/mLhaveaverylowriskofdyingwithprostatecancerwithin25yearsoftheirdiagnosis.16Thislow-riskgroupwasthereforeexcludedfrombeingtestedwiththeadditionalbiomarkers.MenwithaPSAconcentrationofatleast3ng/mLoranSTHLM3modelindicatinghighriskwereconsideredtobeatanincreasedriskofhigh-riskprostatecancerandwerereferredtoaurologist.Theurologistperformeddigitalrectalexams,prostatevolumemeasurements,andtransrectalprostatebiopsy.Accordingtoastandardisedbiopsyprotocol,10corebiopsiesweretakeniftheprostatevolumewaslessthan35cm.and12corebiopsiesweretakenifthevolumewasgreaterorequalto35cm..Asinglepathologistassessedallbiopsiestoreduceinterobservervariance(appendix).ParticipatingurologistsandthepathologistwereblindedtobiomarkerresultsandPSAconcentration.PlasmaproteinbiomarkerswereanalysedusingThermoFisherScientific’sISACmultiplexplatform.GenotypingwasdoneusingtheQuantStudio12KFlexReal-TimePCRSystem(appendix).

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page4

OutcomesTheprimaryaimofSTHLM3wastoincreasethespecificityofacombinedprostatecancertestcomparedwiththePSAtestwithoutdecreasingthesensitivityofhigh-riskprostatecancer.Theprimaryendpointswerethenumberofdetectedhigh-riskprostatecancers(sensitivity)andthenumberofperformedprostatebiopsies(specificity).

StatisticalanalysisLogisticregressionwasusedtotestforassociationsbetweenpredictors(biomarkersandclinicalvariablesmeasuredintheSTHLM3trainingcohort)andcancerswithaGleasonscoreofatleast7(cancerswithGleasonscoresof6andbenignbiopsiesweretreatedascontrols).Non-linearmodelsandsemi-supervisedmodelswerealsoexaminedbutdidnotimprovepredictiveperformance.WeletτdenotetheSTHLM3modelcutoffyieldingthesamesensitivityasaPSAconcentrationofatleast3ng/mLtodetectcancerswithaGleasonscoreofatleast7.OnthebasisoftheSTHLM3trainingcohortdata,acutoffofτ*waschosenusingfive-foldcross-validation,suchthatthesensitivityoftheSTHLM3modelwithcutoffτ*wasestimatedtobe7%higherthanwithaPSAconcentrationofmorethan3ng/mL.Thisoversamplingwasdonetoensurethatτwasestimable(ie,thatasmanyormorehigh-riskcancerswouldbediagnosedwithτ*asanSTHLM3modelcutoffforbiopsycomparedwithaPSAconcentrationofatleast3ng/mLasacutoffforbiopsy).Assumingtherelativefalse-positivefractionoftheSTHLM3modelcomparedwithPSAtobe0·81(forarelativetruepositivefractionof1·07),wecalculatedthatasamplesizeof48000menwouldbeneededtoshow,with92%power,anon-inferiorsensitivityandsuperiorspecificityoftheSTHLM3model(appendix).ThetrainedmodelwasusedprospectivelyintheSTHLM3validationcohort.StudyparticipantswerereferredtohaveabiopsysampletakeniftheyhadapredictedSTHLM3-modelriskofhavingcancerwithaGleasonscoreofatleast7exceedingτ*oraPSAconcentrationofatleast3ng/mL.Thepaired,screen-positivedesignisanefficientdesigntoestimaterelativechangesintestcharacteristics.17,18Apaired,screen-positivedesignwasusedintheSTHLM3studytoestimateratiosbetweentheSTHLM3modelandPSAtestsforthesensitivityforhighriskprostatecancers,forthefalse-positivefractionofmenwithbenignbiopsiesandcancerswithaGleasonscoreof6,andforthenumberoftotalbiopsies.AftertheSTHLM3validationcohortclosed,wecomputedanestimateofτ,τ^,bysettingtherelativesensitivityforhigh-riskcancersequalto1andsolvingforτ(thatis,thesamenumberofcancerswithaGleasonscoreofatleast7foraPSAconcentrationofatleast3ng/mLandfortheSTHLM3modelcutoffbeingatleastτ).CIswerecomputedwiththenon-parametricbootstrapmethodwith1000bootstrappeddatasets.Briefly,foreachbootstrappeddataset,wesolvedfortheSTHLM3modelcutoffyieldingarelativetruepositivefractionequalto1andthencalculatedtheproportionofpatientsthathadabiopsytaken,thepercentageofpatientswithatumourofGleasonscore6thatweresparedbiopsy,thepercentagereductioninbenignbiopsies,andthenumbersofdetectedcancersstratifiedbyGleasonscoreandtotalcancerlengthaboveorbelow10mm.CIswerethencomputedwithmethodsdescribedbyEfronandcolleagues.19Age-stratified(5-yearstrata)resultsandresultsusingotherendpoints20(numberofallprostatecancers,Gleasonscoreofatleast4+3,andCAPRAscore20ofatleast3asdependentvariables)werecomputedasadditionalanalyses.Formodelcomparisons,wecalculatedtheareaunderthecurve(AUC)with95%CIcalculatedusingthebootstrapmethod.Allpvaluesaretwo-sidedandapvalueoflessthan0·05wasconsideredsignificant.WeusedRstatisticalsoftwareversion3.1forallanalyses(appendix).ThisstudyisregisteredwithISRCTN.com,numberISRCTN84445406Subsequenttoregistration,butbeforeenrolment,thestudydesignwaschangedfromarandomiseddesigntothatreportedhere.Continuousqualitycontrolswereimplementedforallprocessesanddataintegritywasmonitoredregularlythroughoutthestudy(appendix).

RoleofthefundingsourceThefunderofthestudyhadnoroleinstudydesign,datacollection,dataanalysis,datainterpretation,orwritingofthereport.Thecorrespondingauthorhadfullaccesstoallthedatainthestudyandtheyhadfinalresponsibilityforthedecisiontosubmitforpublication.

ResultsOf260000menaged50–69yearswitharesidentialaddressinStockholm,145905wererandomlyselectedandinvitedtothestudy.1633menwithprevalentprostatecanceratrecruitmentwereexcluded(figure1).BetweenMay28,2012,andMay31,2013,11130menwererecruitedtotheSTHLM3trainingcohort.Theresultsfromthetrainingcohortaresummarisedintheappendix.113082menaged50–69yearswereinvitedtoparticipateintheSTHLM3validationcohortand1263menwereexcludedbecauseofpreviousprostatecancerdiagnosis.BetweenAug5,2013,andDec30,2014,47688(42%)menchosetoenrolintheSTHLM3validationcohort(figure1).OnthebasisoftheresultsofthePSAtestandtheSTHLM3model,orboth,7606meninthevalidationcohortwererecommendedtourologicalconsultationand5426(71%)hadaprostatebiopsysampletaken.Baselinecharacteristicsofthevalidationcohortareshown(table1).AllvariablesusedintheSTHLM3modelweresignificantlyassociatedwithhigh-riskprostatecancers(p<0·05)inamultiplelogisticregressionmodel.STHLM3didsignificantlybetter(p<0·0001)thanPSAfordetectionofhigh-riskprostatecancers;theAUCwas0·56(95%CI0·55–0·60)withPSAaloneand0·74(95%CI0·72–0·75;table2).InastepwiseAUCanalysis,theestablishedriskfactors(age,familyhistory,andbiopsyhistory),thecombinedbiomarker

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page5

score(geneticscoreandplasmaproteinbiomarkers),andtheprostateexam(digitalrectalexamandprostatevolume)allindependentlyimprovedthetestcharacteristicsoftheSTHLM3model(table2).UsingthesamesensitivityfortheSTHLM3modelasforthePSAtesttodetecthigh-riskprostatecancers(correspondingtoτ^=0·10;95%CI0·09–0·12;ie,10%predictedrisk),theSTHLM3modelwouldreducethenumberofbiopsiesby32%(95%CI24–39).Moreover,usingtheSTHLM3modelwouldhavereducedthenumberofbenignbiopsiesby44%(95%CI35–54)(figure2).Ofthe603high-riskcancersidentifiedbytheSTHLM3model,124(21%)wereidentifiedinthePSArange1–3ng/mL.Furthermore,therewasnosignificantdifferenceinthetotalcancerlengthofthehigh-riskprostatecancersdetectedbytheSTHLM3comparedwiththosedetectedwithaPSAofatleast3ng/mL(p=0·82;figure2).ThenumberofcancerswithaGleasonscoreof6identifiedbytheSTHLM3modelwas722,whereas867wereidentifiedusingPSAalone;areductionof17%(95%CI7–26).The104cancerswithaGleasonscoreof6notidentifiedbytheSTHLM3modelallhadatotalcancerlengthoflessthan10mminthebiopsysamples(figure3).Subanalysisof5-yearagestratawithafixedsensitivity(10%risk)forhighriskprostatecancersineachageclassshowedsimilarperformanceoftheSTHLM3modelinallageclasses(appendix).SensitivityanalysisshowedthattheSTHLM3modelperformanceisrobusttoendpointsotherthanaGleasonscoreofatleast7(eg,aGleasonscoreofatleast4+3prostatecancersorCAPRAscore0–2;table3).ComparingtheSTHLM3modeltoPSAaloneexcludingtheprostateexam(digitalrectalexamandprostatevolume)maintainedtherobustnessoftheSTHLM3model(appendix).

DiscussionWehaveshownthatacombinationofplasmaproteinbiomarkers,geneticpolymorphisms,andclinicalvariablescanimprovethespecificityofprostatecancerscreeningsignificantlycomparedwithPSAinmenaged50–69years.UseoftheSTHLM3modelinstructuredscreeningcouldreducethenumberofprostatebiopsysamplestakenbyaboutathirdcomparedwiththeuseofPSAscreening.Importantly,thiscanbeachievedwithoutcompromisingthenumberofhigh-riskcancersdiagnosed.WeidentifiedanequalGleasonscoredistributionforcancersdetectedbyPSAversustheSTHLM3model,apartfromsmallcancerswithaGleasonscoreof6(figure3).TheclinicalusefulnessoftheSTHLM3modelisfurtherdisplayedintheappendixwheretheSTHLM3modeliscomparedwithPSAinahypotheticalexampleof10000testedmen.WeaimedtodevelopamodelforprostatecancerscreeningwithbettertestcharacteristicsthanPSA.PSAwaschosenasacomparisonbecauseitisthemostwidelyusedscreeningbiomarkerandtheonlybiomarkerthathasbeenassessedprospectivelyinarandomisedcontrolledtrialwithprostatecancermortalityasanendpoint.TheEuropeanRandomizedStudyofProstateCancer(ERSPC)2showeda21%decreaseinprostatecancermortalityusingPSAscreeningafter13yearsoffollow-up.ToinferthesamemortalityreductionasreportedinERSPC,wefixedthesensitivityoftheSTHLM3modelfordetectionofhigh-riskcancerstobeequaltothatofaPSAconcentrationofatleast3ng/mL.UseoftheSTHLM3modelcouldresultina17%reductioninthenumberofGleasonscore6cancersthatwerebiopsied.AlthoughsomedebateremainsastowhetherGleasonscore6cancerscanprogresstohigher-gradediseaseandalthoughpostoperativeupgradingofGleasonscore6cancersisnotuncommon,longitudinalcohortstudies21,22suggestthatmostoftheselesionsareindolentandthereforeneednotreatment.TheSTHLM3modelwouldpermitasubstantialreductioninthenumberofmenneedlesslybeingdiagnosedandtreatedforsmallandinsignificantprostatecancers.Gleasonscore6cancersnotdiagnosedbytheSTHLM3modelwerealllessthan10mmintotalcancerlength,indicatingthatmostofthesecancerswereclinicallyinsignificant.Informationfromtheprostateexam(prostatevolumeanddigitalrectalexam)hasbeenshowntobeimportantforpredictingcancerswithaGleasonscoreofatleast7andimprovestheSTHLM3model.12Fromapracticalstandpoint,toavoidhavingtodoadigitalrectalexamandprostatevolumemeasurementsonallmen,weforeseeasituationwhereanindividualprostatevolumethresholdforabiopsysamplerecommendationiswrittenexplicitlyintheurologyreferralbasedontheresultfromtheSTHLM3model(eg,onthebasisoftheSTHLM3model,abiopsysampleisrecommendediftheprostatevolumeis<50cm.orifthedigitalrectalexamispositive).ThegeneticscoredidnotimprovetheoverallpredictiveperformanceoftheSTHLM3modelasmuchasshowninthetrainingcohort(appendix);however,itdidimprovepredictiveperformance(table2),itisinexpensivetomeasure(andthepriceisconstantlydropping),itonlyneedstobemeasuredonceinaman’slifetime,anditisimportantformenwithaveryhighgeneticrisk(meninthetopdecileofthegeneticscorehavea25%riskofcancerwithaGleasonscoreofatleast7).Toputthesenumbersinperspective,a30%reductioninprostatebiopsiestranslatesto300000fewerproceduresannuallyintheUSA.5Havingaprostatebiopsysampletakencancausepainandrectalbleedingandincreasestheriskofaseriousinfectionwithmultiresistantbacteria.6Thus,theimprovedspecificityoftheSTHLM3modelcouldresultinsavingsintermsofreducedtreatmentmorbidity,coststotheindividualpatient,andtothehealth-caresystem.Afullhealtheconomicassessmentwillbepublishedinaseparatereport.FuturereportswillalsoincludeanassessmentofincreasingthePSAconcentrationlevelcutoffsfortheSTHLM3modeltesting(setto1ng/mLintheSTHLM3model)becausethegroupofmenwithaPSAconcentrationbetween1ng/mLand2ng/mLshowedalowprevalenceofcancerswithaGleasonscoreofatleast7.Twocommercialtestsbasedonthekallikreinsareavailable,theProstateHealthIndex(PHI)andthe4KScore.UsedasareflextesttoPSA,bothhavebeenfoundtoincreasespecificitycomparedwithPSA,butattheriskofmissing5–15%ofhigh-riskcancersthatwouldnormallybedetectedinaclinicalsetting.9,23–25STHLM3avoidsthisproblemwiththe

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page6

design—ie,prospectivelybiopsyingbasedontheSTHLM3modelresults,andthefixedsensitivitybetweentheSTHLM3modelandPSA.TheSTHLM3modelincludesallfourkallikreinsincludedin4KScore.Acomparativestudybetween4KScoreandPHIhasshownthattheyhavesimilarperformance.26The4KScorewasrecentlyassessedinthelargeUKProtecTstudy.27Althoughitisdifficulttocomparetheresultsbetweendifferentstudies,webelievethattheresultsfromProtecTandtheresultspresentedherebothshowthevalueofusingastructuredpredictionalgorithmforbiopsyrecommendations.27Severalotherprostatecancerriskcalculatorsarealsoavailable,combiningPSA,freePSA,andclinicalvariables.28Toourknowledge,theSTHLM3modelistheonlymodelthathasbeenprospectivelyassessedinalargediagnosticstudy.Population-baseddatafromtheStockholmPSAandBiopsyRegistry14formorethan420000menbeingtestedforPSAduringthepast10yearssuggeststhattheuseofclinicalriskcalculatorsisverylimited,atleastinStockholm.InStockholm,thefollowingvariablesareavailableinroutineclinicalcare:age,totalPSA,repeatedPSAtesting(PSAvelocity),freePSA,familyhistory,prostatevolume,PSAdensity,digitalrectalexam,andpreviousbiopsyhistory.However,thedetectionrateofhigh-riskcancersinclinicalpracticeandusingPSAaloneinthisstudyaresimilar.TheconclusionisthatPSAalone,asusedinSTHLM3,isatleastasgoodasdiagnosticpracticefordetectinghigh-riskcancers.Theseresultsshowthedifficultyofinterpretingseveralbiomarkersandclinicalvariablesatthesametime.Somecliniciansandhealth-caresystemsuseage-adjustedPSAcutoffsforbiopsyrecommendation.InSTHLM3,theAUCimprovementofusingageasapredictortogetherwithPSAwasverysmall(table2).Thisindicatesthatage-specificPSAcutoffsdonotimprovemuchonusingPSAaloneinthisstudy.TheSTHLM3modelincludestwonovelbiomarkersforprostatecancer.β-microseminoprotein(MSMB)isoneofthemostcommonproteinsinhumansemenandishighlyexpressedinnormalprostate.Severalstudies29,30haveindicatedthatMSMBisdownregulatedinprostatecancer,particularlyinhigh-gradetumours,andtissueexpressionMSMBisassociatedwithbiochemicalrecurrencesafterradicalprostatectomy.Macrophageinhibitorycytokine1(MIC1),alsoknownasGDF15,isinvolvedininflammationregulationandapoptoticpathwaysininjuredtissues.In-vitrostudies31suggestthatitplaysanimportantpartintheprogressionofprostatecancer.HighserumconcentrationsofMIC1werehighlyassociatedwithmetastaticprostatecancerinaSwedishstudyof1442menwithprostatecancer.32ComparingAUCstatisticsfromotherpublishedbiomarkerstudiesarenotinformativeasthereceiveroperatingcharacteristic(ROC)curvesandAUCvaluesinthisstudyarerelative.Absolutetrueandfalse-positiveratescannotbeassessedinascreen-positivedesignbecausethetruediseaserateisnotknown.Subsequently,comparisonsofrelativeROCcurvesareonlyvalidwithintheSTHLM3datasetandcannotbecomparedwithotherstudies.Forexample,becauseofthemarkedage-relatedprevalenceofprostatecancer,adifferenceof5–10yearsinmeanageofthestudypopulationwillsubstantiallychangetheROCcurve.Similarly,thePSArangeofthestudypopulationwillgreatlyaffecttheROCcurve.InsteadofusingAUCstatisticsasamainoutcome,weusedaclinicallyinterpretableendpoint,thenumberofprostatebiopsies.Ourstudyhassomelimitations.WecannotassesstheperformanceoftheSTHLM3modelinaretestingsituationbecausethetestingwasdoneonlyonceinthisstudy.Furthermore,STHLM3isadiagnostictrial,whichisnotdesignedtoaddresslong-termoutcomes—eg,prostatecancermortality(althoughwebelievetheresultsshowninfigure2indicatethatanyeffectonmortalitybyusingtheSTHLM3modelinsteadofPSAwillbesmall).Weareplanningfollow-upstudiestoaddresstheseresearchquestions.Atthetimeofstartofthisstudyin2011,wechosetouseaGleasonscoreofatleast7asthemainoutcomeinSTHLM3.Theviewofwhatisaclinicallysignificantprostatecanceriscontroversialandvarieswithintheprofessionalcommunity.TheSTHLM3modelisflexibleandwhenusingotheroutcomestheperformanceissimilar(table3).STHLM3wasdoneinStockholm,Sweden,andmostparticipantswereofnorthernEuropeandescent.However,robustevidenceexists9thatsuggeststhatproteinbiomarkersinprostatecancerpredictoutcomesimilarlyinotherwhitepopulationsindifferentpartsoftheworld.Thus,wepredictthattheSTHLM3modelwouldbegeneralisableamongthesepopulations.AlthoughmostoftheSNPsusedinthisstudyarealsosignificantlyassociatedwithprostatecancerinotherethnicities,33,34differencesexistbetweenSNPprofilesfromgroupsofdifferentethnicorigin.TheuseoftheSTHLM3modelintheseethnicgroupsneedstobevalidated.ThestrengthsoftheSTHLM3studyarethatitispopulationbasedwithaprospectivedesignandusesapredefinedalgorithmtestedintheindependentSTHLM3validationcohort.Moreover,theSTHLM3studyconsistsofalargerandomsubsampleofthemalepopulationaged50–69yearsinStockholm,whereopportunisticPSAtestingiscommon.14NosignificantdifferencesinageandpreviousPSAtestingwerereportedbetweenparticipantsinSTHLM3andmenundergoingPSAtestinginStockholm14andtheeducationlevelofparticipantsinSTHLM3wassimilartothatinthegeneralmalepopulationinStockholm.35Importantly,asubanalysisincluding35%ofPSA-naïveparticipantsshowedthattheSTHLM3modelworkedequallywellinthesemenasinmenwithaprevioushistoryofPSAtesting(33%reducedbiopsiesforPSnaivemenand31%formenwithpreviousPSA).Modernimagingtechniques,suchasMRIcombinedwithtargetedbiopsies,willmostlikelyfurtherreduceoverdiagnosisandunnecessaryprostatebiopsies.36WithuseoftheSTHLM3model,MRIcouldbeusedmorecost-effectivelybyidentifyingmenatincreasedriskofclinicallysignificantprostatecancerforsubsequentMRIreferral.Additionally,theidentificationofadditional,newbiomarkerscanbeincorporatedwithintheSTHLM3modelastheyarise,ultimatelyimprovingitsperformance.Wehaveestablishedasystematicassessmentprogrammeofpotentialnewmarkersusingthe58000samplesobtainedinSTHLM3and26000samplesfromearlierstudies.26WearguethattheSTHLM3studyhasseveralimportantnovelaspects,whichincreasesitspotentialtochangeclinicalpractice.TheSTHLM3modelidentifiedclinicallysignificantcancers(withaGleasonscoreofatleast7)inmenwithlowPSAconcentrationranges(1–3ng/mL),representing40%ofthepopulationaged50–69years.STHLM3isalarge-

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page7

scalepopulation-basedstudy,thusminimisingselectionbiasandincreasinggeneralisability.TheSTHLM3modelalsoincludestwonovelplasmabiomarkersandgenetics,whichhaveneverbeforebeenincludedinaprospectivediagnosticstudy.Overall,theSTHLM3modelcanbeusedasanaidtoidentifyhighriskprostatecancersinmenaged50–69years,withaPSAconcentrationofatleast1ng/mL,reducingthenumberofprostatebiopsiesandthedetectionofclinicallyinsignificantdisease,whilemaintainingthesensitivitytoclinicallysignificantprostatecancer.

AcknowledgmentsWethankallstudyparticipants,theSTHLM3coremanagementgroupfortakingcareofallcontactwithparticipantsandorganisingthedatabases,KIBiobankatKarolinskaInstitutetfortakingcareofallbloodsamples,aliquotingplasma,extractingDNA,anddoingallgenotyping,KarolinskaUniversityHospitalLaboratoryfororganisingsamplehandlinginmorethan60outpatientlaboratoriesthroughoutStockholm,theSTHLM3outpatienturologiststakingcareofanddoingprostatebiopsiesinmorethan7000STHLM3participants,UnilabsABinStockholmandHistocenterABinGöteborgfortheirveryhighqualityhandlingofmorethan80000prostatebiopsycoresforfurtherpathologyassessment,LaboratoriemedicininFalunforgreathelpwiththetrainingcohort,theThermoFisherScientificteaminUppsalaandtheQuantStudioteaminEuropeandCaliforniaforbeingtruepartnersintheSTHLM3project,andtheSTHLM3ScientificAdvisoryBoardwhohavereviewedandadvisedoneverythingfromthestudyprotocoltothefinalresults.ThemainfunderoftheSTHLM3studyistheStockholmCountyCouncil(StockholmsLänsLandsting)whoisthemainproviderofhealthcareinStockholm.AuxiliaryfundingforpilotsandinfrastructurewasprovidedbytheSwedishCancerSociety(Cancerfonden),RestaurantsAgainstCancer(RAC),andtheSwedishResearchCouncil(Vetenskapsrådet),OddFellowinVästerås,SwedishResearchCouncilforHealthWorkingLifeandWelfare(FORTE),andSwedishe-ScienceResearchCenter(SeRC).TheSTHLM3studyisapartoftheLinnaeusCenterCRISP“Predicationandpreventionofbreastandprostatecancer”fundedbytheSwedishResearchCouncil.

References1 EtzioniR,GulatiR,TsodikovA,etal.Theprostatecancerconundrumrevisited:treatmentchangesandprostate

cancermortalitydeclines.Cancer2012;118:5955–63.2 SchroderFH,HugossonJ,RoobolMJ,etal.Screeningandprostatecancermortality:resultsoftheEuropean

RandomisedStudyofScreeningforProstateCancer(ERSPC)at13yearsoffollow-up.Lancet2014;384:2027–35.3 LoebS,BjurlinMA,NicholsonJ,etal.Overdiagnosisandovertreatmentofprostatecancer.EurUrol2014;65:

1046–55.4 MoyerVA,ForceUSPST.Screeningforprostatecancer:USPreventiveServicesTaskForcerecommendation

statement.AnnalsInternMed2012;157:120–34.5 LoebS,VellekoopA,AhmedHU,etal.Systematicreviewofcomplicationsofprostatebiopsy.EurUrol2013;64:

876–92.6 AlyM,DyrdakR,NordstromT,etal.Rapidincreaseinmultidrugresistantentericbacillibloodstreaminfection

afterprostatebiopsy—a10-yearpopulation-basedcohortstudy.Prostate2015;75:947–56.7 CarignanA,RoussyJF,LapointeV,ValiquetteL,SabbaghR,PepinJ.Increasingriskofinfectiouscomplications

aftertransrectalultrasound-guidedprostatebiopsies:timetoreassessantimicrobialprophylaxis?EurUrol2012;62:453–59.

8 SchroderFH,RoobolMJ.Definingtheoptimalprostate-specificantigenthresholdforthediagnosisofprostatecancer.CurrOpinUrol2009;19:227–31.

9 BruzzeseD,MazzarellaC,FerroM,etal.Prostatehealthindexvspercentfreeprostate-specificantigenforprostatecancerdetectioninmenwith“gray”prostate-specificantigenlevelsatfirstbiopsy:systematicreviewandmeta-analysis.TranslRes2014;164:444–51.

10 AlOlamaAA,Kote-JaraiZ,BerndtSI,etal.Ameta-analysisof87040individualsidentifies23newsusceptibilitylociforprostatecancer.NatGenet2014;46:1103–09.

11 AlyM,WiklundF,XuJ,etal.Polygenicriskscoreimprovesprostatecancerriskprediction:resultsfromtheStockholm-1cohortstudy.EurUrol2011;60:21–28.

12 KaderAK,SunJ,ReckBH,etal.Potentialimpactofaddinggeneticmarkerstoclinicalparametersinpredictingprostatebiopsyoutcomesinmenfollowinganinitialnegativebiopsy:findingsfromtheREDUCEtrial.EurUrol2012;62:953–61.

13 LindmarkF,ZhengSL,WiklundF,etal.H6Dpolymorphisminmacrophage-inhibitorycytokine-1geneassociatedwithprostatecancer.JNatlCancerInst2004;96:1248–54.

14 NordstromT,AlyM,ClementsMS,WeibullCE,AdolfssonJ,GronbergH.Prostate-specificantigen(PSA)testingisprevalentandincreasinginStockholmCounty,Sweden,despitenorecommendationsforPSAscreening:resultsfromapopulation-basedstudy,2003–2011.EurUrol2013;63:419–25.

15 ThompsonIM,GoodmanPJ,TangenCM,etal.Theinfluenceoffinasterideonthedevelopmentofprostatecancer.NEnglJMed2003;349:215–24.

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page8

16 VickersAJ,CroninAM,BjorkT,etal.Prostatespecificantigenconcentrationatage60anddeathormetastasisfromprostatecancer:case-controlstudy.BMJ2010;341:c4521.

17 SchatzkinA,ConnorRJ,TaylorPR,BunnagB.Comparingnewandoldscreeningtestswhenareferenceprocedurecannotbeperformedonallscreenees.Exampleofautomatedcytometryforearlydetectionofcervicalcancer.AmJEpidemiol1987;125:672–78.

18 PepeMS,FengZ,JanesH,BossuytPM,PotterJD.Pivotalevaluationoftheaccuracyofabiomarkerusedforclassificationorprediction:standardsforstudydesign.JNatlCancerInst2008;100:1432–38.

19 EfronB.Betterbootstrapconfidenceintervals.JAmStatAssoc1987;82:171–185.20 PunnenS,FreedlandSJ,PrestiJCJr,etal.Multi-institutionalvalidationoftheCAPRA-Sscoretopredictdisease

recurrenceandmortalityafterradicalprostatectomy.EurUrol2014;65:1171–77.21 RiderJR,SandinF,AndrenO,WiklundP,HugossonJ,StattinP.Long-termoutcomesamongnoncurativelytreated

menaccordingtoprostatecancerriskcategoryinanationwide,population-basedstudy.EurUrol2013;63:88–96.

22 EggenerSE,BadaniK,BarocasDA,etal.Gleason6prostatecancer:translatingbiologyintopopulationhealth.JUrol2015;194:626–34.

23 LoebS,SandaMG,BroylesDL,etal.Theprostatehealthindexselectivelyidentifiesclinicallysignificantprostatecancer.JUrol2015;193:1163–69.

24 VedderMM,deBekker-GrobEW,LiljaHG,etal.Theaddedvalueofpercentageoffreetototalprostate-specificantigen,PCA3,andakallikreinpaneltotheERSPCriskcalculatorforprostatecancerinprescreenedmen.EurUrol2014;66:1109–15.

25 VickersAJ,GuptaA,SavageCJ,etal.Apanelofkallikreinmarkerpredictsprostatecancerinalarge,population-basedcohortfollowedfor15yearswithoutscreening.CancerEpidemiolBiomarkersPrev2011;20:255–61.

26 NordstromT,VickersA,AsselM,LiljaH,GronbergH,EklundM.ComparisonBetweenthefour-kallikreinpanelandprostatehealthindexforpredictingprostatecancer.EurUrol2015;68:139–46.

27 BryantRJ,SjobergDD,VickersAJ,etal.Predictinghigh-gradecanceratten-coreprostatebiopsyusingfourkallikreinmarkersmeasuredinbloodintheProtecTstudy.JNatlCancerInst2015;publishedonlineApr11.DOI:10.1093/jnci/djv095.

28 vanVugtHA,KranseR,SteyerbergEW,etal.Prospectivevalidationofariskcalculatorwhichcalculatestheprobabilityofapositiveprostatebiopsyinacontemporaryclinicalcohort.EurJCancer2012;48:1809–15.

29 WhitakerHC,Kote-JaraiZ,Ross-AdamsH,etal.Thers10993994riskalleleforprostatecancerresultsinclinicallyrelevantchangesinmicroseminoprotein-betaexpressionintissueandurine.PLoSOne2010;5:e13363.

30 DahlmanA,RexhepajE,BrennanDJ,etal.EvaluationoftheprognosticsignificanceofMSMBandCRISP3inprostatecancerusingautomatedimageanalysis.ModPathol2011;24:708–19.

31 BruzzeseF,HagglofC,LeoneA,etal.Localandsystemicprotumorigeniceffectsofcancer-associatedfibroblast-derivedGDF15.CancerRes2014;74:3408–17.

32 BrownDA,LindmarkF,StattinP,etal.Macrophageinhibitorycytokine1:anewprognosticmarkerinprostatecancer.ClinCancerRes2009;15:6658–64.

33 XuJ,MoZ,YeD,etal.Genome-wideassociationstudyinChinesemenidentifiestwonewprostatecancerrisklociat9q31.2and19q13.4.NatGenet2012;44:1231–35.

34 HanY,SignorelloLB,StromSS,etal.GeneralizabilityofestablishedprostatecancerriskvariantsinmenofAfricanancestry.IntJCancer2015;136:1210–17.

35 StatisticsSweden(SCB).Statistikdatabasen.http://www.statistikdatabasen.scb.se(accessedMarch1,2015).36 FuttererJJ,BrigantiA,DeVisschereP,etal.Canclinicallysignificantprostatecancerbedetectedwith

multiparametricmagneticresonanceimaging?Asystematicreviewoftheliterature.EurUrol2015;68:738.

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page9

FiguresandTables

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page10

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page11

TheLancetOncology:http://dx.doi.org/10.1016/S1470-2045(15)00361-7 Page12