• No results found

Prostate cancer prevalence in New South Wales Australia: A population-based study

N/A
N/A
Protected

Academic year: 2021

Share "Prostate cancer prevalence in New South Wales Australia: A population-based study"

Copied!
8
0
0

Loading.... (view fulltext now)

Full text

(1)

Prostate

cancer

prevalence

in

New

South

Wales

Australia:

A

population-based

study

Xue

Qin

Yu

a,b,

*,

Qingwei

Luo

a

,

David

P.

Smith

a,c

,

Mark

S.

Clements

d,e,f

,

Dianne

L.

O’Connell

a,b,g,h

a

CancerResearchDivision,CancerCouncilNewSouthWales,Sydney,Australia

b

SydneySchoolofPublicHealth,theUniversityofSydney,Sydney,Australia

c

GriffithHealthInstitute,GriffithUniversity,GoldCoast,QLD,Australia

dDepartmentofMedicalEpidemiologyandBiostatistics,KarolinskaInstitutet,Stockholm,Sweden e

NordicInformationforActioneScienceCenter,Stockholm,Sweden

f

Swedishe-ScienceResearchCentre,KTH,DepartmentofMechanics,Stockholm,Sweden

g

SchoolofPublicHealthandCommunityMedicine,UniversityofNSW,Sydney,Australia

h

SchoolofMedicineandPublicHealth,UniversityofNewcastle,Newcastle,Australia

1. Introduction

Prostate cancer has become the most frequently diagnosed canceramongmenindevelopedcountriesaroundtheworld[1], andwilllikelyremainsointheforeseeablefuture[2].Itisamajor

burdenonhealthservicesinmosthighincomecountries,although itcanbedifficulttoaccuratelyassessthefullextentofthisburden. Whileitisfortunatethatmostprostatecancerpatientslivewith thediseaseformanyyearsafterdiagnosis,thisdoesmeanthatthe traditionalcancersurveillancemeasuresofincidenceand mortali-ty,whichcoveronlythetwoextremeendsofthediseasespectrum (diagnosis and death), are insufficient measures of the true magnitude ofthedisease burdenin a given population.In this regard,cancerprevalence–definedasthenumberorproportionof people alive in a population at a given date who have been diagnosedwiththedisease–providesinformationthatiscrucialto

ARTICLE INFO Articlehistory:

Received29October2014

Receivedinrevisedform24November2014 Accepted29November2014

Availableonline17December2014 Keywords: Cancerprevalence Cancerincidence Statisticalprojection Prostatecancer Epidemiology Australia ABSTRACT

Background: InformationonthecurrentandfuturenumbersofAustralianmenlivingwithprostate cancerislimited.Wedescribeamethodforestimatingcompleteprevalenceofprostatecancertoprovide ameasureoftheburdenofprostatecancerinAustralia.

Methods:ProstatecancerdatafromtheNewSouthWales(NSW)CentralCancerRegistrywereusedwith PIAMOD (Prevalence and Incidence Analysis MODel) software to estimatefuture prostate cancer prevalenceinNSW.Wefirstfittedparametricincidenceandsurvivalmodelsthenusedthemodelled incidence and survival estimates to calculate complete prevalence. The estimated and projected prevalenceincorporatepastobservedtrendsandtakeintoaccountdifferentassumptionsaboutfuture survivaltrends.Thesemodelswerevalidatedagainstobservedprevalencefromthecountingmethod. Results:Basedondatafor1996–2007,thenumberofmenlivingwithprostatecancerinNSWwas estimatedtoriseby59%to73%,from38,322in2007to60,910–66,160in2017.Theincreasingincidence ratesandtheageingpopulationwerethemajorcontributorstothisestimatedincrease. Validation suggestedthattheseprojectionswerereasonable,astheestimatedprevalencein1996–2007wasin goodagreementwiththecorrespondingprevalencecalculatedusingthedirectcountingmethod,and theincidencemodelsweresupportedbytherecentdataonprostate-specificantigentesting. Conclusions: Asthenumberofmenlivingwithprostatecancerisexpectedtoincreasedramaticallyinthe nextdecadeinAustralia,representingasignificantchallengetothehealthsystem,carefulplanningand developmentof ahealthcare systemableto respondto this increaseddemand is required.These projectionsareusefulforaddressingthechallengeinmeetingthecancercareneedsofmenwithprostate cancer.

ß2014TheAuthors.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/3.0/).

* Correspondingauthorat:CancerResearchDivision,CancerCouncilNewSouth Wales,POBox572,KingsCross,Sydney1340,NSW,Australia.Tel.:+61293341851; fax:+61283023550.

E-mailaddresses:xueqiny@nswcc.org.au,xue.yu@sydney.edu.au(X.Q.Yu).

ContentslistsavailableatScienceDirect

Cancer

Epidemiology

The

International

Journal

of

Cancer

Epidemiology,

Detection,

and

Prevention

j o urn a lhom e pa g e : ww w . ca nc e re pi d e mi ol o gy . ne t

http://dx.doi.org/10.1016/j.canep.2014.11.009

1877-7821/ß2014TheAuthors.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/ 3.0/).

(2)

theplanningandprovisionofhealthservicesrequiredbypatients intheperiodfollowingprimarytreatmentandbeforedeathfrom prostatecancer.

Thereare,however,somesignificantchallengesindetermining thetrueprevalenceofprostatecancer.Population-widedataused toestimateprevalenceusuallylag3to5yearsbehindthecurrent yearduetothetimerequiredfordatacollection,compilation,and dissemination[3].Moreover,prevalenceestimatesusingobserved data from cancer registries can only provide limited duration prevalence(e.g.1-yearprevalence or5-yearprevalence),as the majorityofpopulation-basedcancerregistriesintheworldhave not been established long enough to capture all prior cancer diagnoses[4].Thus,anestimateofthecompleteprevalenceisoften derivedfromobserveddata usingstatistical models,whichcan alsobeextendedtoestimatefutureprevalence.

Estimating future prostate cancer prevalence is particularly complicated due to the changes in patterns of incidence and survivalthat have occurred since theintroduction of prostate-specificantigen(PSA)testinginthelate1980s[5].Thesemarked changes mean that historical data are a relatively unreliable foundationformodellingprevalence.Inarecentreportcomparing thefour mostwidely usedage–period–cohort (APC) models to projectcancer incidenceinCanada,noneoftheapproacheswas foundtoworkwellforprostatecancer[6].Asaresultofthis,some authorsavoidedincludingprostatecancer[7]whentheypredicted future cancer incidence for major cancer types due to the uncertaintyintheprojections.

In this study, we used a valid PIAMOD (Prevalence and IncidenceAnalysisMODel)method[8]anddatafromanAustralian population-basedcancerregistrytoestimatethefutureprevalence ofprostatecancerinthestateofNewSouthWales(NSW). 2. Methods

Thesoftwareweusedinthisstudy,PIAMOD[8],estimatesand projectscancerprevalenceasafunctionofmodelledincidenceand survivalestimates.Amoredetaileddescriptionofthemethodsfor usingPIAMODsoftwaretoestimatefuturecancerprevalencecan befoundinprevious publications[9,10].Inbrief,theprocessof usingthesoftwaretoestimateprevalenceinvolvesthreeprincipal steps: modelling incidence (by fitting APC models to obtain incidenceprojections), modelling survival (byfitting a mixture curemodel),andthentheestimationandprojectionofcomplete prevalence.ThesestepsareillustratedinFig.1,andthedataand

methodsinvolvedineachofthesestepswillbedescribedindetail below.

2.1. Data

Incidencedataforfirstprimaryprostatecancer(ICD-O3C61)

[11] diagnosed in 1972–2007 were extracted from the NSW CentralCancerRegistrydatabase.TheRegistrycoversapopulation of 7.2 million people, approximately one-third of the national population of Australia, and maintains a record of all cases of cancerdiagnosedinNSWresidentssince1972[12].TheRegistry generallyhashighstandardsofdatacompletenessandquality,and thedataareacceptedbytheInternationalAgencyforResearchon Cancer for publication in Cancer Incidence in Five Continents

[13,14]. We included cases aged 18–84 years at diagnosisand excludedcaseswhowerereportedtotheregistrythroughdeath certificateonly(DCO),orwhowerefirstidentifiedpost-mortem. TheproportionofDCOcasesintheRegistry,anindicatorofthe qualityofthecancerregistry,isgenerallylow(1.0%for1993–1997

[15]and0.9%for2004–2008[16]).Otherinputdatarequiredby thePIAMODsoftwarewereobtainedfromtheAustralianBureauof Statistics(ABS):all-causemortalitydataforNSWbysingleyearof ageandyear(1972–2007),andthecorrespondingmid-yearNSW residentialmalepopulationdatabysingleyearofageandcalendar year.Dataon PSAtests performed (1996–2012)fromMedicare Australiawereusedasacomplementaryvalidationforthefitted incidencemodels[17].

Caseswere followed up for survivalstatus to31 December 2007 throughrecord linkageof thecancer casesin the Cancer RegistrywiththedeathrecordsfromtheNSWRegisterofBirths, DeathsandMarriagesandtheNationalDeathIndex.2007wasthe mostrecentyear forwhich followupdata wereavailable. This significantlagwascausedbytheABSreviewingitsprocessesfor releaseofitsdataincludingcauseofdeath( http://www.cancer- institute.org.au/data-and-statistics/accessing-our-data/availabili-ty-of-nsw-central-cancer-registry-data#death-why-2008). 2.2. Ethicsstatement

Thisstudyinvolvesanalysisofroutinelycollecteddataandthe recordswerede-identified(name,address,dateofbirthhadbeen removed)beforebeingprovidedtotheresearchteam.Asalarge proportionof theindividuals wouldlikely havemoved or died since their diagnosis of cancer, which could have been up to 40yearsago,itwouldhavebeenimpracticabletoseekconsent,and thus the NSW Population and Health Service Research Ethics Committeewaived theconditionsforconsentandapprovedthe study(referencenumber:2009/03/139).

2.3. Modellingincidence

Incidenceratesareoneofthemainfactorsinpredictingfuture cancerprevalence.Prostatecancerincidencechangeddramatically inAustraliawiththeintroductionofPSAtestinginthelate1980s.A sharpinitialincreaseintheearly1990s(whichpeakedin1994at 187casesper100,000men)wasfollowedbyafallof10%peryear toaminimumof126per100,000menin1998,andthenanother increaseduring2001–2005[18].Thisincidencepatternposesa statistical challenge in terms of accurately projecting future incidencetrends.Whileincidencedataareavailablefor1972 on-wards we chose to follow the approach used by previous researchers [19] and modelled incidence using data from 1996to2007.Selectingthisperiodpotentiallyhelpsreducesome oftheimpactoftheintroductionofPSAtestingonourincidence models.

Fig.1. Flow chartshowing the useof PIAMODfor estimatingfuture cancer prevalence.

(3)

APCmodelscanbefittedusingdifferentapproaches,including thegeneralisedlinearmodel(GLM),thegeneralisedadditivemodel (GAM) and the Bayesian model.We used the GLM polynomial methodin thisstudy, implemented usingthe PIAMODsoftware

[8].Thisapproachconsidersage,yearofdiagnosisandbirthcohortas continuousvariables,andusesasmoothingmethod,apolynomial function,tofitthedata.Inthepolynomialmethod,age,periodand cohorttrendsaremodelledbylog-linearregressionforthemean parameterofthePoissondistribution.Forouranalysisthelinear termforperiodwasexcludedwhenestimatingtheparametersin ordertoavoidthecollinearityduetothelinearrelationshipbetween age,periodandcohort[8].Theregressioncoefficientswereobtained usingthemaximumlikelihoodmethod.

The parameters of the APC models were estimated using observedincidencefor1996–2007andthenthismodelwasused forforward(after2007)andbackward(before1996)projections. Theresultingfittedincidenceestimateswereusedasinputsfor estimatingfuture prevalence(for2008–2017). Themost appro-priatemodelwasselectedbasedon thelikelihoodratiostatistic (LRS)combinedwithknowledgeoftheepidemiologyofprostate cancerinAustraliabecause‘...forecastingisnotpossiblewithout sufficientknowledgeoftheepidemiologyofagivencancer...’[20]. 2.4. Modellingsurvival

Atwo-stepprocedurewasusedtomodelsurvival.First,relative survivalwas estimated and tabulated,and thena mixture cure model was fitted to the tabulated relative survival estimates. Survivalrelativetothegeneralpopulationwascalculatedinthis studybecauseweusedall-causemortalitydatafroma population-basedcancerregistry.We derivedtabulatedrelativesurvivalfor prostatecancerfromincidenceandfollow-updatausingthePohar– Permemethod[21].Inlinewiththemodellingofincidencewealso useddataforthe period1996–2007 formodellingsurvival.This periodwaschosenbecausesurvivalpatternsfromtheearlierperiod of1972–1995weremarkedlydifferentfromthesurvivalpatterns thatoccurredaftertheintroductionofPSAtesting,andthepre-PSA testingpatternswouldbeunlikelytooccuragaininthefuture.

Wemodelledthesetabulatedrelativesurvivalestimateswitha mixture cure model [22], which assumes that patients can be dividedintotwodistinctgroupsbasedontheirprognosis:thosewho arecuredandthosewhoarenot.Thedefinitionof‘cure’,asusedin thiscontext,isthat‘cured’patientshaveamortalityrateequalto thatofpeopleofthesameageandsexinthegeneralpopulation

[22].Themodelparametersarethentheproportioncuredandthe scaleandshapeparametersoftheWeibulldistribution,whichwas assumed to describe the survival distribution for the uncured patients.Themodelparameterswereestimatedbymeansofa non-linearregressionprocedureusingPROCNLIN(SAScodeprovidedby RobertaDeAngelis)andtakingtheinverseofthevariancesofthe observedsurvivalasweights[22].Inthisstudy,theWeibullmixture modelwasstratifiedbyagegroup(18–64,65–74and75–84years) soparameterswereestimatedforeachagestratumatthereference yearofdiagnosis(centralpointoftheperiod1996–2007),andthe periodeffectvariedbyage.

Themodelledsurvivalestimateswereextrapolatedbackward to 1972 and forward to 2017, using the assumption that the survivaltrendswouldbedynamicandwouldhavethesame‘slope’ as that for the 1996–2007 observed data. The model-based estimates of survivalfrom this mixture cure model wereused asinputsintoPIAMODforthenextstepoftheanalysis.

2.5. Prevalenceprojections

ProstatecancerprevalencewasthenestimatedusingPIAMOD softwarewiththemodelledprostatecancerincidence,model-based

relative survival, and all-cause mortality as inputs. Because PIAMOD can only provide results for closed age groups and populations, ourprevalenceestimates include casesup toage 84yearsonly.FurtherdetailsonthePIAMODmethodhavebeen previously described by Mariotto et al. [23]. The PIAMOD prevalenceestimatesfor1996–2007werevalidatedby compar-ing theestimatedprevalencewiththelimitedduration preva-lence estimates derived by counting the number of cases diagnosedbetween1972and2007[24],whichisconsideredto bethemostreliableestimateforpopulationscoveredbyaregistry forasufficientlengthoftime[25].

2.6. Sensitivityanalyses

Furtheranalyseswerealsoundertakentoinvestigatetheeffects of different assumptions regarding future incidence rates and survivalontheestimatedfutureprevalence.First,weestimatedthe futureprevalencebychoosingtwodifferentAPCincidencemodels. Second, we estimated prostate cancer prevalence based on the assumptionthatthefuturesurvivalratewillremainconstantatthe levelobservedin2007ratherthancontinuingtofollowthetrendsin theobserveddata.Third,werepeatedtheprimaryanalysesusing data with longer survival follow-up (1990–2007) to assess the sensitivityofthecuremodeltothelengthoffollow-upanditsimpact ontheprojectedprevalence.Finally,weassumedthattherewould benochangesinincidenceratesinfutureyears,sothe2007 age-specificprevalencerateswereappliedtotheprojected2017 age-specific male populationin NSW. Thisgave an estimate of the prevalencein2017duetopopulationgrowthandageingonly. 3. Results

Atotalof49,866casesoffirst primaryprostatecancerwere diagnosedin1996–2007inNSW.Afterexcluding2685casesaged 85 years or over, 47,181 cases (median age: 69 years) were includedintheincidenceandprevalenceanalyses.

3.1. Incidencemodels

SixrelativelysimpleAPCmodels(APC101,102,201,202,103, and 301) wereplotted against the observed incidence and are showninFig.2.Toselectwhichmodelsweremostappropriatefor predictingfutureincidencetrendsbeyondtheavailabledatafor NSW(to2007),theprojectedincidencetrendsfromthesemodels were compared to more recent observed incidence data for prostate cancer from similar jurisdictions: data from another Australianstateupto2012[26],andfromtheUnitedStates[27]

and NewZealand[28].Allthesedatasuggest thatfromaround 2008–2010prostatecancerincidencetendedtostabiliseoreven fall.Inaddition,NSWdataonPSAtestinghavealsoshownatrend towardsareductioninthenumberofPSAtestsbeingperformed (Fig.3), whichinturnislikely toresultina slowerincrease in prostatecancerincidence[29].ThustheAPCmodels102and103, which showedslowerincreasingincidencetrends,were consid-ered to bethemostappropriate modelswithwhich to project incidencefor2008–2017(Fig.2).Themodel-fit-statistics( Appen-dix)alsosuggestthatthesetwomodelsareareasonablygoodfitto theobservedincidencedata.

3.2. Survivalmodels

Five-year prostate cancer relative survival trends (assuming dynamicsurvivalbeyondtheobserveddatawindow)byyearof diagnosisareshowninFig.4.Itcanbeseenthatthefittedvaluesfor five-year relative survival were in close agreement with the observeddatafortheperiod1996–2007.

(4)

3.3. Projectedprevalence

Basedonthesedata,thenumberofmenwithprostatecancerin NSWisprojectedtorisebyapproximately59%to73%in10years: from38,322in2007to60,910–66,160patientsin2017(Table1).

Fig.5showsthePIAMODprevalenceestimatesbasedondatafrom 1996to2007(withAPC102and103incidencemodels)aswellas theobservedprevalence(to2007).Weobserved(Fig.5)thatthe PIAMODestimates(1996–2007)wereingoodagreementwiththe prevalence (1996–2007) calculated from the direct counting methodusingobserveddatafrom1972to2007.

For thesake of brevity,we only presentsensitivity analysis resultsfromincidencemodelAPC102here(Table2).Itsuggested that the impact of variations in survival (either assuming a constant trend or using a longer time series) on the projected prevalencewaslimited(2%).Thisisduetothehighsurvivalfor prostatecancer,whichmeansthereisverylittleroomforfurther improvement.Iftheeffectofthegrowthintheageingpopulationis

solely considered, prostate cancer prevalence is estimated to increaseby26%from2007to2017.

4. Discussion

Thisstudyusedpopulation-basedcancerregistrydataandthe PIAMODsoftware to model current and future prostate cancer prevalenceinNSW.Ourresultsindicatedthatinthe10yearsfrom 2007to2017thenumberofmenlivingwithadiagnosisofprostate cancerinNSWislikelytorisebybetween59%(lowbound)and73% (highbound).Wefoundthatfortheperiod1996–2007therewas closeagreementbetweenthefittedandobservedprevalence,so webelievethatourpredictionsarereasonableandthatthetrue futureprevalenceislikelytoliebetweenthesetwopredictions.We would however, like to emphasise that there is considerable uncertaintyintheseprojections,andthatourmodelsarelikelyto becomelessaccurateastimepasses,especiallyconsideringhow

Fig.2.Comparisonofage–period–cohortincidencemodelsandobservedincidenceratesforprostatecancerinNSW,Australia.

Fig.3.Estimatedage-standardisedprostatecancerincidencerates(1972–2007),andprojectedincidencerates(1996–2017)andage-standardisedratesofPSAtesting(1996– 2012)inNSW,Australia.

(5)

difficultitistopredictthefuturetrendsinprostatecancertesting behaviours.

Despitethedifficultiesinvolved,severalpreviouslypublished international studies have estimated future prostate cancer prevalence.Gattaetal.[30]usedthecompletenessindexmethod withdatafrom76Europeancancerregistries,andreporteda17.6% increaseinprostatecancerprevalencefrom2003to2010dueto demographic changes (which is a very similar result to our estimates of increase in prevalence due to population changes only:2.5%vs2.6%annualincrease).AnAmericanstudybyMariotto etal. [31]usedthePIAMODmethod[8] anddata from9SEER

registries, and estimated a 41.3% increase in prostate cancer prevalencefrom2010to2020.Giventhedifferentmethods,time periodsanddatausedinthesestudiesitisverydifficulttomake anydirectcomparisonsbetweenthesestudiesandours,butitdoes seematleastthatthepredictionofanincreasingtrendinprostate cancer prevalence is common amongst studies in developed countries.

Several Australian studies of cancer prevalence [32–34]

provided estimates of limited duration prevalence. This means thattheyreportedthenumberofpeoplealiveatthedateofinterest withadiagnosisofcancerwithinapastspecificnumberofyears (e.g. 2, 10, or 25 years). Our 36 year prevalence estimate at 2007wascomparabletothosereportedby theCancerInstitute NSW [33] and the Australian Institute of Health and Welfare (AIHW)[34],althoughdifferenttimeperiodswereusedinthese studies.Thiscloseagreementisnotsurprising,astheresultswere basedonthesame(CancerInstituteNSWreport)orsimilar(AIHW report)datasources.However,ourstudyextendedtheseresultsby providingestimatesofthefutureprevalencethataremoreuseful forhealthserviceplanningforprostatecancerpatients. Extrapo-latingtheseprojectionstothenationalpopulationwouldequateto about 185,700 to 201,700 men living with prostate cancer in Australia in 2017, representing a significant challenge to the

Fig.4.Comparisonoffittedfive-yearrelativesurvivalwithobservedsurvivaldataforprostatecancerinNSW,Australia.

Table1

Estimated(2007)andprojected(2017)prevalenceofprostatecancerinNSW, Australia.

2007(Baseline)estimate Projectedprevalenceestimatesin2017and percentincreasefrombaseline

Lower bounda % Increase Upper boundb % Increase 38,322 60,910 59% 66,160 73% a

BasedonincidencemodelAPC102.

b

BasedonincidencemodelAPC103.

(6)

Australianhealthsystem.Moreover,theseprojectionsarelikelyto beunderestimatedbecausewedidnotincludecasesaged85years oroverintheestimationandthosewithsecondprimaryprostate cancer. There would be approximately 6922–7518 men aged 85yearsorolderlivingwithprostatecancerinNSWin2017ifwe applytheproportionof cases in this oldestgroupin 2007(for whichwe havedata) totheprojected 2017totalprevalence in NSW.Thecorrespondingnationalfigurewouldbebetween21,103 and22,922prevalentcasesagedover85years;thishighlightsthe importanceofincludingtheelderlypopulationinstudiesofcancer survivors.

Theadvantageofthemethodweusedisthatthemodelsfor incidence and survival were fitted separately, similar to the methodadoptedbyCrouchetal.[35],withtheestimatesoffuture prevalencebeingobtainedasaproductofboth.Thismethodallows ustoconstructandevaluatedifferentscenariosregardingfuture incidenceandsurvival,aswellaspopulationageing,andtomodel theireffectsonfutureprevalenceestimates.

Our modelsarebasedonbothstatisticalandclinical founda-tions.First,mostprojectionmodelsforprostatecancerarelikelyto be poor due to the fluctuation in incidence caused by the introductionofPSAtestinginthelate 1980s[6]andthiswould inevitably lead to biased estimates of both incidence and prevalence. Hence, from a statistical point of view it is well justifiedtoexcludetheperiodwherethedatafluctuate consider-ably,ashasalsobeendonebyotherresearchers[19].Asageneral rule,itismoreappropriatetoseparatethepatientpopulationinto asetofhomogeneousgroupsandexcludethosegroupsthatadd littlevaluetothemodellingoffutureincidenceandprevalence. Therefore,themorerecentdatahavegreaterpredictivepowerthan much older data, and in the case of prostate cancer this is particularly important, as the introduction of PSA testing has significantlyalteredthewayinwhichdiseasefirstpresentsandits subsequentmanagement.Modellingtherecentdataallowsfora moreclinicallyrelevantpredictionoffutureprevalence.

Projectionsofcancerprevalencedependontheestimatesand assumptionsoffuturetrendsinincidenceandsurvival.Wefound thattheprojectedincidencetrendwasthemostimportantfactor inestimatingfutureprostatecancerprevalence,aswasshownin theresults,withtheestimatedprevalencebeingverysensitiveto theincidencemodelsselected.Forexample,ifwechoseincidence modelAPC202,whichhadthelowestLRS(Appendix),thiswould resultina 164%increaseinprostatecancerprevalenceby2017 (ratherthan59–73%),whichwebelievetobehighlyunlikely.This demonstratesthatitmayberiskytoselecttheincidencemodel basedon statisticalassessmentcriteriaonly,andshowswhyan understandingoftheepidemiologyofthecancerunderstudyis extremely important for the selection of themost appropriate modelsforpredictingfuturetrends.

Basedonthemodelsweselected,therewouldbemuchslower growthin the incidence rates of prostate cancer in theperiod 2008–2017thanwasseenfrom2001onward,and ittended to flattenoutneartheendoftheperiod.Basedonourunderstanding ofthedrivingforcesbehindchangesinprostatecancerincidence and independent data from similar jurisdictions [26–28], we

believetheseassumptionsarereasonable.Manystudiesshowan associationbetweenthelevelofPSAtestingandprostatecancer incidence[18,29,36],sothetrendinPSAtestingshouldprovide someinsightintothefuturetrendsinincidence.Recentdatafrom theMedicareBenefits Scheme(Australia’s universalhealth care scheme)suggestedthatthepeakofPSAtestinginNSWmayhave beenreached,althoughitstillremainsataveryhighlevel(Fig.3). However,itisasyetunclearhowpatternsoffutureprostatecancer testinginAustralia willbeaffectedbyrecentrecommendations suchasthoseoftheU.S.PreventiveServices TaskForceagainst routinePSAscreeningformenofanyage[37],andtheAmerican Urological Association’smoreconservative recommendationsto onlytestmenaged55to69[38].EvidencefromtheUSAindicates thattheimpactislikelytobeminimal[39,40].

Ageisanotherimportantfactorinpredictingprostatecancer incidencepatterns,astheriskofdiagnosisincreasesexponentially afterage50.Thenumberofmenagedover65yearsinAustraliais projectedtoincreasefrom1.23millionin2006(representing12.1% ofthetotalmalepopulation)to1.78million(15.7%ofthetotal male population) in the year 2016 [41]. This ageing of the population will inevitably lead toa significant increase in the numberofmendiagnosedwithprostatecancer.Inregardstothe extentofthisincrease,incidenceprojectionsfromCanadaandthe USAmayprovidesomeinsightintofutureAustraliantrends,asthe populationagestructuresofthesecountriesarebroadlysimilarto thatofAustralia[41].TheincidenceinCanadawasprojectedto increaseby39%from2009to2021duetopopulationageingonly

[2],whilethecorrespondingincreaseintheUSAisestimatedtobe 30%overtheperiod2010to2020[42].Ourprojectedincreasedue topopulationchangesonly(26%from2007to2017)issimilarto theestimatesfromCanadaandtheUSA.

Incontrasttotheselectionoftheincidencemodels,different assumptionsoftrendsinfuturesurvivalhadlessofanimpacton theestimatedfutureprevalencethanwasseenwithchangesin incidence rates and population ageing (Table 2). This may be becausesurvivalrates forprostatecancerare alreadyveryhigh (fiveyearrelativesurvivalbeingover90%).Therefore,alessrobust assumption(ofcure)wouldnothavetoogreatanimpactonthe predicted prevalence. Sensitivity analysis indicates that the predictedprevalence estimateswerenotsensitive tothelength ofthesurvivalfollow-up.

Cancer prevalence is a function of incidence, survival and population ageing. Itis challenging to completely separatethe contributionofeachofthesefactors,asitisnotsimplyanadditive relationship.Forexample,anincreasedincidenceofPSAdetected caseswilllikely leadtoprolongedsurvivalevenifno improve-mentsintreatmentoccur.Nevertheless,weconductedsensitivity analysesdesignedtoinvestigatethesensitivityoftheestimated future prevalence to different factors. Compared to 2007, our resultsindicate that populationageing and growthonlywould resultinanincreaseofaround26%inthenumberofmenliving withprostate cancer in 2017(assuming incidenceand survival remainstaticfortheperiod2007–2017).Thisalsoimpliesthatthe remaining predicted increase in prevalence for 2017 would be causedbytheincreasedincidenceratesandimprovedsurvival.As

Table2

Impactofvariousfactorsontheprojectedprevalenceofprostatecancerin2017,NSWAustralia.

2007(Baseline)estimate Projectednumberofprevalentcases(2017)usingdifferentassumptionsa

Assumingconstant survivaltrend

%Change Basedondata for1990–2007

%Change Basedonpopulation changebonly

%Change

38,322 60,313 57% 60,163 57% 48,269 26%

a

IncidencemodelAPC102wasusedintheseanalyses.

b

(7)

discussed previously, changes in survival patterns would only contributeabout2%tothechangeintheestimatedprevalencein 2017,and thuspopulationageingandincreasedincidencerates should be the major contributors to the future increase in prevalence.Althoughtheseresultsareconsistentwithprevious international studies [2,42], we would emphasise that it is extremelydifficulttoaccurately isolatetheeffectsofindividual factors,andtheresultsofthissensitivityanalysisonlyprovidea roughindicationofthecontributionofeachofthesevariables.

The methodused here hasseveral limitations that must be consideredifitistobeappliedtootherdata.Astheseprevalence projections are based on assumptions about incidence and survival, anyfuture changesin themethods usedtoundertake prostatebiopsyorchangestoPSAtestingthresholdscouldhave markedshort-termeffects,leadingtounreliableprojections.Itis thereforeimportanttoemphasisethattheseprojectionsarevery sensitive to any future changes in prostate cancer diagnostic practicesand theaccuracy oftheprojections tendstodecrease with time. Thus, we suggest that these projections should be updatedasnewdatabecomeavailableandassumption underpin-ningthe projectionsbe periodicallyreviewed. Also, integrating dataonPSAtestingtrendsandthemethodandnumberofbiopsy corestakenwithobservedincidencedatawouldhaveincreased theaccuracyofourmodel,butasregistrydataarenotroutinely linkedwithdataonscreeninghistory,andthesoftwareweused doesnot allowfortheintegrationof suchfactors,this wasnot possible. This may warrant further investigation in future modellingexercises,asthemodelaccuracycouldbesignificantly increasedbyincorporatingdataregardingimportantpredictorsof incidence[43]. The finallimitation of ourmethods is that the assumptionofcureisnotalwaysreasonableforprostatecancer becauseofitshighsurvivalrate[44].Whencuredoesnotoccur,the mixturecure modelused here provides anapproximationonly

[22].However,aspreviouslydescribed,wefoundthatalessrobust assumption of cure would only have a small impact on the predictedprevalence.

Despite these limitations, this study does provide a more complete measure of the future burden of prostate cancer in Australiathanhaspreviouslybeenavailable.Withanincreasing numberofmenbeingdiagnosedwithprostatecancerandlivingfor manyyearsafterdiagnosis,this isanareaofresearchofcritical importancefortheplanningofhealthcareservices.Despiteitshigh prevalence(about 20%of thetotalcancer survivor population), researchonprostatecancersurvivorsisrelativelyrare,withonly 5%ofallstudiesofcancersurvivorshavingaspecificfocusonthese men[45].Bycomparison,breastcancersurvivorshavebeenthe focusof40%onstudiesofcancersurvivors,despiterepresentinga similarproportionofthesurvivorpopulationasprostatecancer survivors(22%)[45].Asprostatecancerprevalenceisexpectedto increasedramaticallyinthenextdecade,itisimportantthatmore isknownaboutthispopulationandthatreliableprojectionsofthe futureburdenareavailable.Itishopedthatthisstudywillbeginto addresssomeoftheseissuesand willhelpenablean evidence-basedapproachtohealth-careplanningandservicedelivery. Conflictofintereststatement

Theauthorsdeclarethattheyhavenoconflictofinterests. Authorshipcontribution

XQYandDOCconceivedtheproject;DPSandMSCassistedwith furtherrefinementoftheproject.XQYobtainedthecancerregistry data,QLperformedthedataanalysis,andXQYprovidedoversight ofthedata analysiswithinputs fromDOC, DPS andMSC. XQY draftedthemanuscriptwithhelpfromQLforSection2and3.DOC,

DPS, MSCand QL revisedthe manuscript.Allauthors read and approvedthefinalversionofthemanuscript.

Acknowledgements

ThisworkissupportedbytheProstateCancerFoundationof Australia(PCFA–YI0410).BothXueQinYuandDavidSmithare supported by NHMRC Early Career Fellowships (550002 and 1016598).WewouldliketothanktheNSWCentralCancerRegistry forprovidingthedataforthisstudyandClareKahnforeditorial assistance.

AppendixA. Supplementarydata

Supplementarydataassociatedwiththisarticlecanbefound, in the online version, at http://dx.doi.org/10.1016/j.canep. 2014.11.009.

References

[1]JemalA,BrayF,CenterMM,FerlayJ,WardE,FormanD.Globalcancerstatistics. CACancerJClin2011;61:69–90.

[2]QuonH,LoblawA,NamR.Dramaticincreaseinprostatecancercasesby2021. BJUInt2011;108:1734–8.

[3]SiegelR,NaishadhamD,JemalA.Cancerstatistics,2013.CACancerJClin 2013;63:11–30.

[4]CapocacciaR,ColonnaM,CorazziariI,DeAngelisR,FrancisciS,MicheliA,etal. MeasuringcancerprevalenceinEurope:theEUROPREVALproject.AnnOncol 2002;13:831–9.

[5]Neppl-HuberC,ZappaM,CoeberghJW,RapitiE,RachtanJ,HolleczekB,etal. Changesinincidence,survivalandmortalityofprostatecancerinEuropeand theUnitedStatesinthePSAera:additionaldiagnosesandavoideddeaths.Ann Oncol2012;23:1325–34.

[6]Cancer ProjectionNetwork(C-projection).Long-termprojectionmethods: comparison of age–period–cohort model-based approaches. Edmonton, Canada:AlbertaHealthServices,2010.

[7]Dyba T, Hakulinen T. Do cancer predictions work. Eur J Cancer 2008;44:448–53.

[8]VerdecchiaA,DeAngelisG,CapocacciaR.Estimationandprojectionsofcancer prevalencefromcancerregistrydata.StatMed2002;21:3511–26.

[9]YuXQ,ClementsM,O’ConnellD.Projectionsofcancerprevalencebyphaseof care:apotentialtoolforplanningfuturehealthserviceneeds.JCancerSurviv 2013;7:641–51.

[10]YuXQ,SmithDP,ClementsMS,PatelMI,McHughB,O’ConnellDL.Projecting prevalencebystageofcareforprostatecancerandestimatingfuturehealth serviceneeds:protocolforamodellingstudy.BMJOpen2011;1:e000104.

[11]FritzA,PercyC,JackA,ShanmugaratnamK,SobinL,ParkinD,etal. Interna-tionalclassificationofdiseasesforoncology.3rded.Geneva,Switzerland: WorldHealthOrganisation,2000.

[12]YuXQ,O’ConnellDL,GibberdRW,CoatesAS,ArmstrongBK.Trendsinsurvival andexcessriskofdeathafterdiagnosisofcancerin1980–1996inNewSouth Wales,Australia.IntJCancer2006;119:894–900.

[13]ParkinDM,MuirCS.Cancerincidenceinfivecontinents.Comparabilityand qualityofdata.IARCSciPubl1992;45–173.

[14]BrayF,RenJS,MasuyerE,FerlayJ.Globalestimatesofcancerprevalencefor 27sitesintheadultpopulationin2008.IntJCancer2013;132:1133–45.

[15]TraceyE,RoderD,BishopJ,ChenS,ChenW.CancerinNewSouthWales incidenceandmortality2003.Sydney:TheCancerInstituteNSW,2005.

[16]CurrowD,ThomsonW.CancerinNSW:incidencereport2009.Sydney:Cancer InstituteNSW,2014.

[17]AustralianGovernment.MedicareAustraliaMedicareBenefitsSchedule(MBS) itemstatisticsreports.AustralianGovernment;2014.

[18]SmithDP,SupramaniamR,MarshallVR,ArmstrongBK.Prostatecancerand prostate-specific antigen testing in New South Wales. Med J Aust 2008;189:315–8.

[19]MollerH,FairleyL,CouplandV,OkelloC,GreenM,FormanD,etal.Thefuture burdenofcancerinEngland:incidenceandnumbersofnewpatientsin2020. BrJCancer2007;96:1484–8.

[20]ClaytonD,SchifflersE.Modelsfortemporalvariationincancerrates.II:Age– period–cohortmodels.StatMed1987;6:469–81.

[21]PermeMP,StareJ,EsteveJ.Onestimationinrelativesurvival.Biometrics 2012;68:113–20.

[22]VerdecchiaA,DeAngelisR,CapocacciaR,SantM,MicheliA,GattaG,etal.The cure for colon cancer: results from the EUROCARE study. Int J Cancer 1998;77:322–9.

[23]MariottoAB,YabroffKR,FeuerEJ,DeAngelisR,BrownM.Projectingthe numberofpatientswithcolorectalcarcinomabyphasesofcareintheUS: 2000–2020.CancerCausesControl2006;17:1215–26.

(8)

[24]Krogh V, Micheli A.Measure ofcancer prevalencewith a computerized program:anexampleonlarynxcancer.Tumori1996;82:287–90.

[25]GailMH,KesslerL,MidthuneD,ScoppaS.Twoapproachesforestimating diseaseprevalencefrompopulation-basedregistriesofincidenceandtotal mortality.Biometrics1999;55:1137–44.

[26]ThursfieldV.In:ThursfieldV,StainesC,GilesG,FarrugiaH,eds.Cancerin Victoria:statistics&trends2012.Melbourne:CancerCouncilVictoria,2013.

[27]HowladerN,NooneAM,KrapchoM,GarshellJ,MillerD,AltekruseSF,etal. SEERcancerstatisticsreview,1975–2011.BasedonNovember2013SEERdata submission, postedto theSEER web site.April 2014 ed. Bethesda,MD: NationalCancerInstitute,2014.

[28]MinistryofHealth.Cancer:newregistrationsanddeaths2010.Wellington: MinistryofHealth,2013,2013.

[29]GregorioDI,KulldorffM,SheehanTJ,SamociukH.Geographicdistributionof prostatecancerincidenceintheeraofPSAtesting,Connecticut,1984to1998. Urology2004;63:78–82.

[30]GattaG,MalloneS,vanderZwanJM,TramaA,SieslingS,CapocacciaR,etal. CancerprevalenceestimatesinEuropeatthebeginningof2000.AnnOncol 2013;24:1660–6.

[31]MariottoAB,YabroffKR,ShaoY,FeuerEJ,BrownML.Projectionsofthecostof cancercareintheUnitedStates:2010–2020.JNatlCancerInst2011;103:117–28.

[32]BrameldKJ,HolmanCD,ThrelfallTJ,LawrenceDM,DeKierkNH.Increasing ‘activeprevalence’ofcancerinWesternAustraliaanditsimplicationsfor healthservices.AustNZJPublicHealth2002;26:164–9.

[33]TraceyE,BakerD,ChenW,StavrouE,BishopJ.CancerinNewSouthWales: incidence,mortalityandprevalence,2005.Sydney:CancerInstituteNSW, 2007.

[34]AustralianInstituteofHealthandWelfare.Cancersurvivalandprevalencein Australia: period estimates from 1982 to 2010. Asia Pac J Clin Oncol 2013;9:29–39.

[35]CrouchS,SmithA,PainterD,LiJ,RomanE.Determiningdiseaseprevalence fromincidenceandsurvivalusingsimulationtechniques.CancerEpidemiol 2014;38:193–9.

[36]SmithDP,ArmstrongBK.Prostate-specificantigentestinginAustraliaand associationwithprostatecancerincidenceinNewSouthWales.MedJAust 1998;169:17–20.

[37]MoyerVA.Screeningforprostatecancer:U.S.PreventiveServicesTaskForce recommendationstatement.AnnInternMed2012;157:120–34.

[38]CarterHB,AlbertsenPC,BarryMJ,EtzioniR,FreedlandSJ,GreeneKL,etal.Early detectionofprostatecancer:AUAguideline.JUrol2013;190:419–26.

[39]DrazerMW,HuoD,SchonbergMA,RazmariaA,EggenerSE.Population-based patternsandpredictorsofprostate-specificantigenscreeningamongolder menintheUnitedStates.JClinOncol2011;29:1736–43.

[40]PollackCE,PlatzEA,BhavsarNA,NoronhaG,GreenGE,ChenS,etal.Primary careproviders’perspectivesondiscontinuingprostatecancerscreening. Can-cer2012;118:5518–24.

[41]AustralianInstituteofHealthandWelfare.OlderAustraliaataglance4th edition.4thed.Canberra:AIHW,DOHA,2007.

[42]SmithBD,SmithGL,HurriaA,HortobagyiGN,BuchholzTA.Futureofcancer incidenceintheUnitedStates:burdensuponanaging,changingnation.JClin Oncol2009;27:2758–65.

[43]PickleLW,HaoY,JemalA,ZouZ,TiwariRC,WardE,etal.Anewmethodof estimatingUnitedStates and state-levelcancerincidence countsfor the currentcalendaryear.CACancerJClin2007;57:30–42.

[44]YuXQ,DeAngelisR,AnderssonTM,LambertPC,O’ConnellDL,DickmanPW. Estimatingtheproportioncuredofcancer:somepracticaladviceforusers. CancerEpidemiol2013;37:836–42.

[45]HarropJP,DeanJA,PaskettED.Cancersurvivorshipresearch:areviewofthe literatureandsummaryofcurrentNCI-designatedcancercenterprojects. CancerEpidemiolBiomarkersPrev2011;20:2042–7.

References

Related documents

Because these practices are ideological, and because the purpose of ‘proof-of-status’ technologies is to verify identification in relation to the state, the debate over, and

• The results of this study revealed nurses in hospitals that are on the Journey to Magnet perceive that decision making is shared between nursing management/adminis- tration and

Physicians who participate in what the legislation calls “Alternative Payment Models” (APMs) will be eligible to receive a variety of extra payments or incentives based on the

organizations (namely hospitals and health systems, medical group practice, ambulatory care, long-term care, etc.); however, graduates are also prepared for administrative

a) House prices have been on a rapid increase since 2011, in particular the double digital growth during the years between 2013 and 2014. b) The results for price-to-rent ratios

Put the following categories into order of importance: Close Sámi relatives, The Sámi language, Reindeer herding and related activities, Keeping contacts with the Sámi

In male genitalia it is distinguished by: the long ventrad fibula (as in L. laisalis ); the elongate, straight or hook-shaped, sometimes branching distal sacculus process

Investment trusts reported net disinvestment of £0.4 billion in the fourth quarter of 2008 compared with a net disinvestment of £0.2 billion in the third quarter of 2008, while