Accuracy and usefulness of a clinical prediction rule and D-dimer testi

www.elsevier.com/locate/thromres

REGULARARTICLE

AccuracyandusefulnessofaclinicalpredictionruleandD-dimertestinginexcludingdeepveinthrombosisincancerpatients

MarcCarriera,AgnesY.Y.Leeb,ShannonM.Batesb,DavidR.Andersonc,PhilipS.Wellsa,d,⁎

ThrombosisProgram,DivisionofHematology,DepartmentofMedicine,UniversityofOttawa,Ottawa,Ontario,CanadabDepartmentofMedicine,MichaelG.DeGrooteSchoolofMedicine,McMasterUniversity,Hamilton,Ontario,CanadacDepartmentofMedicine,QueenElizabethIIHealthSciencesCentre,DalhousieUniversity,Halifax,NovaScotia,CanadadClinicalEpidemiologyProgram,TheOttawaHealthResearchInstitute,CanadaReceived29January2008;receivedinrevisedform2May2008;accepted8May2008Availableonline16June2008

aKEYWORDS

Deepveinthrombosis;Cancer;Diagnosis;

Clinicalpredictionrules

Abstract

Introduction:Deepveinthrombosis(DVT)canbesafelyandreliablyexcludedinpatientswithalowclinicalprobabilityandanegativeD-dimerresultbuttheaccuracyandutilityofsuchastrategyislesscertainincancerpatients.WesoughtocomparetheperformanceoftheWellspretestprobability(PTP)modelandD-dimertestingbetweenpatientswithandwithoutcancerandtoexaminetheutilityofthetwoPTPmodelclassificationschemes(low/moderate/highversusunlikely/likely)inexclud-ingDVTinpatientswithcancer.

Materialsandmethods:PooledanalysisofdatabasesfromthreeprospectivediagnosticstudiesevaluatingconsecutiveoutpatientswithsuspectedDVT.

Results:Atotalof2696patientswereevaluated.DVTwasdiagnosedin403(15%)patientsoverallandin83of200(41.5%)cancerpatients.ThePTPdistributionandtheprevalenceofDVTineachPTPcategoryweresignificantlydifferentbetweenpatientswithandwithoutcancer,regardlessoftheclassificationused(pb0.01).Inpatientswithcancer,thenegativepredictivevaluesofaloworunlikelyPTPscoreincombinationwithanegativeD-dimerresultwere100%(95%CI69.8%–100%)and100%

Abbreviations:DVT,Deepveinthrombosis;PTP,Pre-testprobability;LR+,positivelikelihoodratio;LR-,negativelikelihoodratio;NS,notstatisticallysignificant.

⁎Correspondingauthor.SuiteF6-49,1053CarlingAvenue,OttawaHospital,CivicCampus,Ottawa,OntarioK1Y4E9.Tel.:+16137985555x18769;fax:+16137615351.

E-mailaddress:pwells@ohri.ca(P.S.Wells).0049-3848/$-seefrontmatter©2008ElsevierLtd.Allrightsreserved.doi:10.1016/j.thromres.2008.05.002

178M.Carrieretal.

(95%CI82.8%–96.6%),respectively.However,thespecificitiesrangedfrom46.2%(95%CI27.1%–66.3%)to57.1%(95%CI41.1%–71.9%).Furthertestingwasrequiredin94%ofcancerpatientsusingthelow/moderate/highPTPclassificationandin88%usingtheunlikely/likelystratification.

Conclusions:Asinpatientswithoutcancer,thecombinationofaloworunlikelyPTPwithanegativeD-dimerresultcanexcludeDVTinpatientswithcancer.However,thisstrategyhaslimitedutilitybecauseveryfewcancerpatientspresentwiththiscombination.

©2008ElsevierLtd.Allrightsreserved.

Introduction

Patientswithmalignancyhaveahighriskofdevel-opingvenousthrombosis[1–3]andthetreatmentofthrombosiswithanticoagulantsiscomplicatedbyasubstantialriskofbleeding[3–6].Therefore,accu-ratediagnosisofdeepvenousthrombosis(DVT)isrequiredtoappropriatelytreatpatientswiththediseaseandavoidunnecessaryanticoagulationinthosewithoutvenousthrombosis.

Clinicalpredictionmodelscombinecomponentsofthehistoryandphysicalexaminationtocategorizeapatient'sprobabilityofhavingadisease.TheWellspretestprobability(PTP)modelisvalidatedandwellestablishedfordiagnosingDVTinsymptomaticpatients.Thismodeloriginallystratifiedpatientsashavingalow,moderate,orhighlikelihoodofhavingDVTbutsubsequentanalysesshowedthatthemodelcouldbesimplifiedbystratifyingpatientsintoeitheralikelyorunlikelyriskgroup[7–10].WhenusedincombinationwithD-dimertesting[10],bothofthesestratificationscansafelymanagepatientswithsuspectedDVT.Thenegativepredictivevalue,orthelikelihoodofnothavingdisease,whenapatienthasaloworunlikelyPTPandanegativeD-dimerresult,rangesfrom99.1%to99.6%[8–10].Ithasalsobeendemonstratedthatitissafetowithholdanti-coagulanttherapyinpatientswhohaveanegativeD-dimerresultandeitheraloworunlikelyPTPatinitialpresentation[8–10].

ButtheclinicalutilityofsuchastrategyforrulingoutDVTinpatientswithcancerislesscertain.ThestudiesthatvalidatedPTPmodelsdidincludecancerpatientsbutthescoringdoesnottakeintoaccountthestrongriskfactorsforDVTthatareuniquetocancerpatients.Also,whilesomestudiesshowedthatD-dimertestinghadcomparablenegativepre-dictivevaluesinpatientswithcancercomparedtopatientswithoutcancer[11],othershavenot[12].Finally,thehighprevalenceofDVTincancerpatientsmaysufficientlyreducethenegativepredictivevalueofthesediagnostictoolsandrenderthemlessuseful[12].InaretrospectivestudythatevaluatedtheapplicationofaPTPmodelandD-dimertestingin

cancerpatientswithsuspectedDVT,acombinationofanegativeD-dimerresultandaloworlow-moderatePTPwasusefulinexcludingDVT[13].However,giventhatonly9%ofcancerpatientsscoredalowPTP,theauthorssuggestedthattheunlikely/likelyPTPstratificationmightbemoreusefulinclinicalprac-tice.Also,theconfidenceintervalofthenegativepredictivevalueofthiscombinationwasaslowas88%,suggestingthatupto12%ofpatientsmaybemisseddiagnosedasnothavingDVT.

TofurtherinvestigatetheutilityofPTPmodelingandD-dimertestingincancerpatients,weperformedacombinedanalysisof3largeprospectivecohortstudiestocomparetheperformanceoftheWellsPTPmodelandD-dimertestingbetweenpatientswithandwithoutcancerandtoexaminetheutilityofthetwoPTPmodelclassificationschemes(low/moder-ate/highversusunlikely/likely)inexcludingDVTinpatientswithcancer.

Materialsandmethods

WecombinedthedatabasesfromthreeprospectivediagnosticstudiesinconsecutiveoutpatientswithsuspectedDVTthatwereperformedat7Canadiantertiarycarecentersbetween1997and2002(seeAppendixAfornamesofcenters)[8–10].PatientswereincludedinthesestudiesiftheypresentedwithsymptomsofDVTateithertheemergencydepartmentsortheThromboembolismclinics.EachstudyevaluatedpatientsusingtheWellsPTPmodel(Table1)[7,10]andperformedD-dimertestingatinitialpresentation.OnestudyrandomizedpatientstoultrasoundimagingaloneorD-dimertestingfollowedbyultrasound[10].Patientswerethenmanagedaccordingtotheseresultsbyfollowingprespecifieddiagnosticstrategies.Inallthreestudies,patientswithaloworunlikelyPTPscoreandanegativeD-dimerresultandhadnofurthertesting,werenottreatedwithanti-coagulantsandwerefollowedfor3monthstodetectthromboticevents.Compressionultrasonographyand/orvenographyofthesymptomaticlegwereperformedinallpatientswhohadahighorlikelyPTPscoreorapositiveD-dimerresult.Inthesestudies,DVTwasconsideredpresentifanyofthefollowingoccurredatpresentationorduringthe3-monthfollow-upperiod:1)anysegmentofnon-compressibilityofthecommonfemoralveintothetrifurcationofthepoplitealveinoncompressionultrasonography[8–10];or2)intra-luminalfillingdefectofanylegveinoncontrastvenography[8–10].PatientswerealsoconsideredtohaveDVTiftheypresentedwithobjectively

UsefulnessofaclinicalpredictionruleandD-dimertestinginexcludingDVTincancerpatients

Table1Pre-testprobability(PTP)modelforsuspecteddeepveinthrombosisWells’Criteria

Activecancer(patientreceivingtreatmentforcancerwithintheprevious6moorcurrentlyreceivingpalliativetreatment)

Paralysis,paresis,orrecentplasterimmobilizationofthelowerextremities

Recentlybedriddenfor3daysormore,ormajorsurgerywithintheprevious12wkrequiringgeneralorregionalanesthesia

LocalizedtendernessalongthedistributionofthedeepvenoussystemEntirelegswollen

Calfswellingatleast3cmlargerthanthatontheasymptomaticside(measured10cmbelowtibialtuberosity)

PittingedemaconfinedtothesymptomaticlegCollateralsuperficialveins(nonvaricose)Previouslydocumenteddeep-veinthrombosis

Alternativediagnosisatleastaslikelyasdeep-veinthrombosisPre-testProbabilityCategories:UnlikelyLikelyLow

ModerateHigh

179

Score

1

Table3ProportionofpatientsaccordingtoPTPcategoriesandD-dimerresultsinpatientswithandwithoutcancer

PatientsPatientsPvaluewithoutcancerwithcancer

NumberofPatientsClinical

ProbabilityLow

ModerateHighUnlikelyLikely

D-dimerResultPositiveNegative

2496n(%)1049(42.0)1046(41.9)401(16.1)1571(62.9)925(37.1)2070

828(40.0)1242(60.0)

200n(%)31(15.5)70(35.0)99(49.5)57(28.5)143(71.5)163

110(67.5)53(32.5)

b0.001b0.05b0.001b0.001b0.001b0.001b0.001

11

111

111-2

TotalScoreb2≥2≤01–2N2

theIL-Test(InstrumentationLaboratory,Lexington,USA)andtheMDAD-Dimerassay(TrinityBiotechUSA,St.Louis,USA).FortheSimpliRedtest,theresultwasconsiderednegativeifnoagglutinationwasseen.FortheIL-TestandMDAD-Dimer,theresultswereconsiderednegativeifthevaluewaslessthan200μg/L[14]and0.50μgFEU/ml,respectively.Thesensitivities,specificitiesandnegativepredictivevaluesoftheseassayshavebeenpreviouslydescribed[8,9].All3assayshavenegativepredictivevaluesof90%orhigherinexcludingsymptomaticDVTinoutpatientpopulations.TheoriginalresultsoftheD-dimertestingwereusedinthisanalysis.

confirmedvenousthromboembolismduringthe3-monthfollow-upperiod.PatientswithnormalimagingtestsandallpatientswithoutsymptomsofDVTorpulmonaryembolismduringfollow-upwereconsideredasnothavinghadDVT.

Cancerstatus

Cancerstatuswasdeterminedatthetimeofstudyenrollmentinall3studies.Patientswereclassifiedashavingcanceriftheyreportedashavingclinicalevidenceofcancer,werereceivingcancertreatment,hadreceivedcancertreatmentwithinthe6monthspriortostudyenrollment,orwerereceivingpalliativecare.Thepatient'scancerstatusrecordedatthetimeofstudyenrollmentwasusedinthisanalysis.

Clinicalpretestprobability

Patientswereoriginallystratifiedintolow,moderateorhighriskgroupsofhavingDVTintwostudies[8,9]whilethethirdstudyhadstratifiedpatientsintounlikelyandlikelycategories[10].Inordertocomparetheutilityofthesetwoclassificationschemes,weclassifiedeachpatientashavingalow,moderateorhighriskofDVTaswellashavinganunlikelyorlikelyprobabilityofhavingDVTusingtheoriginalPTPscoresobtainedatstudyenrollment(Table1).

Statisticalanalysis

Wepooledtheresultsfromthe3studiesaccordingtothepatients'cancerstatus.Withineachpatientgroup(withorwithoutcancer),thesensitivity,specificity,negativepredictivevalue,andlikelihoodratio,alongwiththeir95%confidenceintervals(CI),werecalculatedseparatelyforthecombinationofalowPTPandanegativeD-dimerresultandthecombinationofanunlikelyPTPandanegativeD-dimerresult.Thedistribution

D-dimerassays

ThreedifferentD-dimerassayswereusedintheoriginalstudies:theSimpliREDassay(AGENBiomedical,Ltd.,Brisbane,Australia),

Table2Characteristicsoftheprospectivestudies

Batesetal.[8]Cancer

NoCancer

50461.961.339(7.7)

5263.765.3

17(32.7)

Characteristics

Number;nAge;y(mean)Femalesex;(%)

ConfirmedDVT;n(%)DVT:Deepveinthrombosis.

Andersonetal.[9]Cancer

5765.543.9

25(43.9)

Wellsetal.[10]Cancer

93.049.5

41(44.1)

NoCancer

101956.156.8

170(16.7)

NoCancer

97358.059.3

111(11.4)

180

Table4PrevalenceofDVTsaccordingtoPTPcategoriesandD-dimerresultinpatientswithandwithoutcancer

PatientswithoutPatientswithPvaluecancern(%)cancern(%)

ClinicalProbability

Low44(4.2)Moderate138(13.2)High138(34.4)Unlikely83(5.3)Likely237(25.6)D-dimerResultPositive219(26.5)Negative43(3.5)NS:notstatisticallysignificant.

3(9.7)

19(27.1)61(61.6)9(15.8)74(51.7)

NSb0.01b0.001b0.05b0.001

M.Carrieretal.

Inpatientswithoutcancer,42.0%wereconsideredtohavealowPTPwhile62.9%wereclassifiedashavinganunlikelyPTPforDVT.Incontrast,inpatientswithcancer,only15.5%and28.5%wereclassifiedashavingalowandunlikelyPTP,respectively.TheproportionofpatientsclassifiedintothelowandunlikelyPTPcategorieswassignificantlydifferentbetweenpatientswithandwithoutcancer(pb0.001).

TheprevalenceofDVTaccordingtoPTPcategoriesaresummarizedinTable4.BothPTPclassificationschemesappropriatelypredictedtheprevalenceofDVTinpatientswithandwithoutcancer.However,theprevalenceofDVTineachofthePTPgroupswasapproximately2-foldhigherinpatientwithcancerthaninpatientswithoutcancer.TheprevalenceofDVTincancerpatientswithalowandunlikelyPTPis9.7%and15.8%,respectively.D-dimerdistributionandprevalenceofDVT

TheD-dimerresultsweresignificantlydifferentbetweenpatientswithandwithoutcancer(pb0.01)(Table3).Inpatientswithoutcancer,60.0%hadanegativeD-dimerresultcomparedto32.5%inpatientswithcancer(pb0.001).

TheprevalenceofDVTaccordingtoD-dimerresultsaresummarizedinTable4.InbothD-dimerpositiveandnegativegroups,theprevalenceofDVTisapproximately2-foldhigherinpatientswithcancerthaninpatientswithoutcancer(pb0.05).Ofnote,7.8%ofcancerpatientswithanegativeD-dimerresulthadDVTconfirmed.AccuracyindicesforcombiningPTPandD-dimerTable5outlinestheaccuracyindicesforthecombina-tionsofaloworunlikelyPTPwithanegativeD-dimerresultinthetwopatientgroupsseparately.Thesensitiv-itiesofthecombinationsarehighinbothpatientgroups,resultinginnegativepredictivevaluesof99%orhigher.However,thespecificitiesrangefrom46.2%to70.4%.Inbothpatientgroups,thespecificitywasslightlyhigherusingtheunlikelyPTPclassificationthanthelowPTPclassification.

Overall,22%ofpatientswithoutcancerhadalowPTPandanegativeD-dimerresultand36%hadanunlikelyPTPandanegativeD-dimerresult.Inpatientswithcancer,

57(51.8)4(7.8)b0.001b0.05

ofthePTPscores,theproportionofpatientswithanegativeD-dimerresultandtheproportionofpatientswithDVTineachPTPgroupwerecomparedbetweenthe2patientgroupsusingChi-squaretests.P-valueslessthan0.05wereconsideredstatisticallysignificant.

Results

Atotalof2696patientswereevaluated.Thechar-acteristicsofthestudyparticipantsaredisplayedinTable2.Deepveinthrombosiswasdiagnosedin403(15%)patients;13ofthesewereconfirmedduringthe3-monthfollow-up.Atotalof200cancerpatientswereincluded.Ofthesepatients,83(41.5%)werediagnosedwithDVT.Atotalof463patientsdidnothaveD-dimerperformed.Thirtypatientsfromthe3originalstudiesdidnothavesufficientinformationinthedatabasesandwereexcludedfromthisanalysis.

PTPdistributionandprevalenceofDVT

ThePTPdistributionwassignificantlydifferentbetweenpatientswithandwithoutcancer(pb0.01)(Table3),regardlessoftheclassificationschemeused.ThemajorityofpatientswithoutcancerwereclassifiedashavinganunlikelyPTP(62.9%)whereasthemajorityofpatientswithcancerwerestratifiedashavingalikelyPTP(71.5%).

Table5AccuracyindicesofaloworunlikelyPTPcategoryincombinationwithanegativeD-dimerresulttoexcludeDVTinpatientswithandwithoutcancerPTP&D-minerresult

Sensitivity(%,95%CI)Specificity(%,95%CI)NPV(%,95%CI)LR+(%,95%CI)LR-(%,95%CI)

Patientswithoutcancerlowandnegative(n=557)

95.0(81.8–99.1)62.9(59.6–66.1)99.6(98.6–99.92.56(2.29–2.87)0.079(0.021–0.31)

Pateintswithcancerlowandnegative(n=12)

100(31.0–100)46.2(27.1–66.3)100(69.8–100)1.85(1.30–2.65)0(0–0.000

unlikelyandnegative(n=4)

91.7(82.1–96.6)70.4(67.8–72.9)99.3(98.5–99.7)3.1(2.78–3.46)0.12(0.055–0.25)

unlikelyandnegative(n=24)

100(59.8–100)57.1(41.1–71.9)100(82.8–99.6)2.33(1.–3.31)0(0–0.00019)

PTP:Pre-testprobability;LR+:positivelikelihoodratio;LR-:negativelikelihoodratio.

UsefulnessofaclinicalpredictionruleandD-dimertestinginexcludingDVTincancerpatients

181

only6%and12%ofpatientshadaloworunlikelyPTPandanegativeD-dimerresult,respectively.

Discussion

WeinvestigatedthesafetyandutilityofexcludingDVTinpatientswithcancerusingdifferentPTPcategories(loworunlikely)incombinationwithD-dimertesting.Asinpatientswithoutcancer,loworunlikelyPTPcategoriesincombinationwithanegativeD-dimerresulthaveveryhighnegativepredictivevaluesandshouldbeabletosafelyruleoutDVTinpatientswithcancer.However,becauseD-dimerlevelsareoftenelevatedincancerpatientsintheabsenceofDVT,theD-dimertesthaslowspecificityinbothlowandunlikelyPTPcategories.Consequently,theclinicalutilityofD-dimertestingislimitedincancerpatientsbecauseveryfewpatientshaveloworunlikelyPTPandanegativeD-dimerresultatinitialpresentationandsothevastmajorityofpatientswillrequireadditionaltestingtoconfirmorexcludeadiagnosisofDVT.Ofthetwocategoriza-tionschemes,theunlikely/likelyPTPstratificationmaybemoreclinicallyusefulbecauseithashigherspecificityandnegativepredictivevaluethanthelowPTPcategorywhencombinedwithanegativeD-dimerresult.

AlthoughthePTPWellsmodelwasnotdesignedtoassignDVTriskspecificallyinpatientswithcancer,themodelwasabletoseparatecancerpatientsintodifferentriskgroups.However,becausecancerisanindependentriskfactorforvenousthromboembolism[2],significantlyfewercancerpatientswereintheloworunlikelyPTPcategories:only16%ofpatientswithcancerhadalowPTPcomparedto42%inpatientswithoutcancer(differenceof26%;95%CI18.1%–35.0%);and29%ofpatientwithcancerwereclassifiedintheunlikelycategoryincomparisonto63%inpatientswithoutcancer(difference34%;95%CI24.1%–44.7%).More-over,themodelrevealedthatcancerpatientshavea2-foldhigherprevalenceofDVTcomparedwithpatientswithoutcancerinallPTPgroups(Table4).Inotherwords,theprevalenceofDVTisnotcomparablebetweenpatientswithandwithoutcancerdespitebeingcategorizedintothesameriskgroup.Thisisanimportantreminderthatitisprudenttoapplyclinicalpredictionrulesforriskassessmentonlytopatientpopulationsinwhomthemodelwasvalidatedspecifically.

TheaccuracyoftheD-dimertoexcludeDVTinpatientswithloworunlikelyPTPwassimilarbetweenpatientswithorwithoutcancer,withnegativepredictivevaluesrangingfrom99%to100%.Thiswassomewhatunexpectedbecausethepredictive

valuesofadiagnostictestareinfluencedbytheprevalenceofthedisease,suchthatahighpre-valencewilllowerthenegativepredictivevalueofatest.SincetheoverallprevalenceofDVTinpatientswithcancerinourstudywas42%,andtheprevalenceofDVTinthelowandunlikelyPTPcategorieswas10%and16%respectively,itwasanticipatedthatthenegativepredictivevaluewouldbelower.However,becausethesensitivityoftheD-dimerassaysaresufficientlyhigh(99%to100%)incancerpatients,thenegativepredictivevalueswerepreserveddespiteahighprevalenceofdisease.Incontrast,thespecifi-cityoftheD-dimertestwasquitelowincancerpatients.EveninthelowPTPcategory,thespecificityofD-dimertestwasonly46%,indicatingthat54%ofpatientswithalowPTPwillhaveapositiveD-dimertestdespitenothavingDVT.

Inthethreestudieswecombined,only6%ofcancerpatientsscoredalowPTPandhadanegativeD-dimerresult.Thus94%ofpatientswithcancerrequiredfurthertestingtoruleoutorconfirmadiagnosisofDVTafterperformingPTPmodelingandD-dimertesting.Iftheunlikely/likelycategoriza-tionwasused,88%ofpatientswouldhaveneededfurtherinvestigations.TheseresultssuggestthatastrategyusingPTPmodelingandD-dimerdoesnoteffectivelyorefficientlyexcludeDVTinmostpatientswithcancer,despitethehighnegativepredictivevaluedemonstrated.Indeed,itmaybemorecost-effectivetoavoidD-dimertestingincancerpatientsandproceedwithotherdiagnosticstudies,suchascompressionultrasonography.

Thecost-effectivenessofdiagnosticstrategiescombiningPTPmodelingtoD-dimerisunknownincancerpatientswithsuspectedDVT.Arecentcosteffectivenessanalysis[15]showedthatdiagnosticstrategiesusingD-dimersarecost-effectiveinelderlypatientswithsuspectedPEuntil80yearsoldsuggestingthatthecost-sparingeffectofD-dimerismaintaineduntiltheproportionofpatientwithD-dimernegativeandlowandintermediatePTPisabove12.5%.Inourstudy,6%and12%ofcancerpatientshadaloworunlikelyPTPandanegativeD-dimerresult,respectivelysuggestingthatadiag-nosticstrategyusingD-dimermightnotbecosteffective.However,prospectivestudiesarerequiredtoconfirmthesefindings.

Ourstudyisconsistentwiththeliterature.Inaprospectivestudyof1739patientsevaluatingtheroleoftheSimpli-REDD-dimerassayandcompressionultrasoundinpatientswithsuspectedDVT,compar-ablyhighnegativepredictivevalueswerefoundincancerpatients(97%;95%CI%–100%)andpatientswithoutcancer(97%;95%CI96%–98%)[11].Inanothercohortstudyof327consecutivepatientsreferredforasuspicionofDVT,thenegativepredictivevalueand

182

specificityoftheSimpliREDD-dimerinpatientswithcancerclassifiedinthelowPTPcategorywere100%(95%CI85%–100%)and58%(95%CI42%–72%),respectively[13].Incontrast,anotherretrospectiveanalysisofprospectivestudiesinvolvingconsecutiveoutpatientsfoundasignificantlylowernegativepredictivevalue(79%vs.97%;p=0.008)andspecifi-city(48%vs.82%;pb0.001)fortheSimpliREDD-Dimerinpatientswithcancer[12].Inthislaststudy,thesensitivityoftheD-dimerassaywaslowat86%.

Weacknowledgesomelimitationstoourstudy.First,threedifferentD-dimerassayswithdifferentoperatingcharacteristicswereused.FiftypatientswithcancerhadaMDAD-Dimerassayandwerepreviouslyshowntohavesimilarresultscomparedtothegroupingofthiscurrentstudy[8].Therewasinsufficientnumberofcancerpatientstoexamineandcomparetheperformanceofthe3D-dimerassays.Second,wedidnothaveanyobjectiveconfirmationofthepatients'activecancerdiagnosis.Diagnosisofcancerwasbasedonselfreportingandthatcouldhaveleadtomisclassification,especiallyinpatientswithoccultmalignancythatwasnotyetdiagnosedatthetimeofstudyenrolment.Finally,thetotalnumberofcancerpatientsfromthe3studieswassmall,leadingtowideconfidenceintervalsforouraccuracyindicesestimates.Forexample,wecannotexcludewith95%confidencethatthenegativepredictivevalueofalowPTPwithanegativeD-dimerresultisaslowas69.8%;i.e.,30.2%ofcancerpatientswiththiscombinationoftestresultscouldinfacthaveDVT.Inconclusion,asinpatientswithoutcancer,thecombinationofaloworunlikelyPTPwithanegativeD-dimerresultcanexcludeDVTinpatientswithcancer.However,incontrasttopatientswithoutcancer,thecombinationofaloworunlikelyPTPandanegativeD-dimerresultoccursrelativelyuncom-monlyand,therefore,excludesveryfewcancerpatientsandlimitstheclinicalutilityofthisstrategy.OurresultssuggestthatD-dimertestingisnotusefulincancerpatientsandthesepatientsshoulddirectlyproceedtohavediagnosticimaging.

Acknowledgements

P.WellsisarecipientofaCanadaResearchChair.M.CarrierisarecipientofaCanadianInstituteforHealthResearchFellowship.

AppendixA

QEIIHealthSciencesCenter,Halifax,NovaScotia,Canada.

OttawaHospital–Civiccampus,Ottawa,Ontario,Canada.

M.Carrieretal.

LondonHealthSciencesCenter,London,Ontario,Canada.

St.Paul'sHospital,Vancouver,BritishColumbia,Canada.

McMasterUniversityMedicalCentre,Hamilton,Ontario,Canada.

HendersonHospital,Hamilton,Ontario,Canada.St.Joseph'sHospital,Hamilton,Ontario,Canada.

References

[1]BlomJW,DoggenCJM,OsantoS,RosendaalFR.Malignan-cies,prothromboticmutations,andtheriskofvenousthrombosis.JAMA2005;293(6):715—22.

[2]HeitJA,SilversteinMD,MohrDN,PettersonTM,O'FallonWM,

MeltonLJ,etal.Riskfactorsfordeepveinthrombosisandpulmonaryembolism:apopulation-basedcase-controlstudy.ArchInternMed2000;160(6):809—15.

[3]CogoA,BernardiE,PrandoniP,GirolamiB,NoventaF,

SimioniP,etal.Acquiredriskfactorsfordeep-veinthrombosisinsymptomaticoutpatients.ArchInternMed1994;154(2):1—8.

[4]HuttenBA,PrinsMH,GentM,GinsbergJS,TijssenJGP,

BullerHR.Incidenceofrecurrentthromboembolicandbleedingcomplicationsamongpatientswithvenousthrom-boembolisminrelationtobothmalignancyandachievedinternationalnormalizedratio:aretrospectiveanalysis.JClinOncol2000;18(17):3078—83.

[5]PrandoniP,LensingAW,PiccioliA,BernardiE,SimioniP,

GirolamiB,etal.Recurrentvenousthromboembolismandbleedingcomplicationsduringanticoagulanttreatmentinpatientswithcancerandvenousthrombosis.Blood2002;100(10):3484—8.

[6]EltingLS,EscalanteCP,CooksleyC,AvritscherEB,KurtinD,

HamblinL,etal.Outcomesandcostsofdeepvenousthrombosisamongpatientswithcancer.ArchInternMed2004;1(15):1653—61.

[7]WellsPS,HirshJ,AndersonDR,LensingAWA,FosterG,Kearon

C,etal.Accuracyofclinicalassessmentofdeep-veinthrombosis.Lancet1995;345(61):1326—30.

[8]BatesSM,KearonC,CrowtherM,LinkinsL,O'DonnellM,

DouketisJD,etal.Adiagnosticstrategyinvolvingaquantitativelatexd-dimerassayreliablyexcludesdeepveinthrombosis.AnnInternMed2003;138(10):787—94.

[9]AndersonDR,KovacsMJ,KovacsG,ForgieMA,KearonC,

DreyerJF,etal.Combineduseofclinicalassessmentandd-dimertoimprovethemanagementofpatientspresentingtotheemergencydepartmentwithsuspecteddeepveinthrombosis(theEDITEDStudy).JThrombHaemost2003;1(4):5—51.

[10]WellsPS,AndersonDR,RodgerMR,ForgieM,KearonC,Dreyer

J,etal.Evaluationofd-dimerinthediagnosisofsuspecteddeep-veinthrombosis.NEJM2003;349(13):1227—35.

[11]tenWoldeM,KraaijenhagenRA,PrinsMH,BullerHR.The

clinicalusefulnessofD-dimertestingincancerpatientswithsuspecteddeepvenousthrombosis.ArchInternMed2002;162(16):1880—4.

[12]LeeAYY,JulianJA,LevineMN,WeitzJI,KearonC,WellsPS,

etal.Clinicalutilityofarapidwhole-bloodD-dimerassayinpatientswithcancerwhopresentwithsuspectedacutedeepvenousthrombosis.AnnInternMed1999;131(6):417—23.

[13]DiNisioM,RutjesNWS,BullerHH.Combineduseofclinical

pretestprobabilityandD-dimertestincancerpatientswith

UsefulnessofaclinicalpredictionruleandD-dimertestinginexcludingDVTincancerpatients

clinicallysuspecteddeepvenousthrombosis.JThrombHaemost2006;4(1):52—7.

[14]KovacsMJ,MacKinnonKM,AndersonD,O'RourkeK,Keeney

M,KearonC,etal.AcomparisonofthreerapidD-dimermethodsforthediagnosisofvenousthromboembolism.BrJHaematol2001;115(1):140—4.

183

[15]RighiniM,NendazM,LeGalG,BounameauxH,PerrierA.

Influenceofageonthecost-effectivenessofdiagnosticstrategiesforsuspectedpulmonaryembolism.JThrombHaemost2007;5(9):1869—77.