Estimation of Canine Leishmania Infection Prevalence in Six Cities of the Algerian Littoral Zone

17 Pages·2015·0.48 MB·English

RESEARCHARTICLE
Leishmania
Estimation of Canine Infection
Prevalence in Six Cities of the Algerian
Littoral Zone Using a Bayesian Approach
AmelAdel1,2,3,EmmanuelAbatih2,NikoSpeybroeck4,AbdelkrimSoukehal5,
RachidBouguedour6,KarimBoughalem7,AbdelmalekBouhbal7,MouloudDjerbal8,
ClaudeSaegerman3,DirkBerkvens2*
1InstituteofVeterinarySciences,UniversitySaadDahlab,Blida,Algeria,2InstituteofTropicalMedicine,
DepartmentofBiomedicalSciences,Nationalestraat155,Antwerpen,Belgium,3ResearchUnitof
a11111 EpidemiologyandRiskAnalysisappliedtoVeterinaryScience(UREAR-ULg),FundamentalandApplied
ResearchforAnimals&Health(FARAH),FacultyofVeterinaryMedicine,UniversityofLiège,Boulevardde
Colonster20B42,Sart-TilmanLiège,Belgium,4UniversitéCatholiquedeLouvain,IRSS-FSP,Clos
ChapelleauxChamps30,Bruxelles,Belgium,5UniversityHospitalofBeniMessous,Algiers,Algeria,6OIE
Sub-RegionalRepresentationforNorthAfrica,17Avenued’Afrique,ElMenzahV2091,Tunis,Tunesia,
7DirectiondesServicesVétérinaires,Ministèredel’AgricultureetduDéveloppementRural,12bdColonel
Amirouche,16000Algiers,Algeria,8RegionalVeterinaryLaboratoryofDraa-Ben-Kheda,Tizi-Ouzou,
Algeria
OPENACCESS
* [e-mail]
Citation:AdelA,AbatihE,SpeybroeckN,Soukehal
A,BouguedourR,BoughalemK,etal.(2015)
EstimationofCanineLeishmaniaInfection Abstract
PrevalenceinSixCitiesoftheAlgerianLittoralZone
UsingaBayesianApproach.PLoSONE10(3):
Alarge-scalestudyoncanineLeishmaniainfection(CanL)wasconductedinsixlocalities
e0117313.doi:10.1371/journal.pone.0117313
alongawest-easttransectintheAlgerianlittoralzone(Tlemcen,Mostaganem,Tipaza,
AcademicEditor:HenkD.F.H.Schallig,Royal
Boumerdes,Bejaia,Jijel)andcoveringtwosamplingperiods.Intotal2,184dogsweretest-
TropicalInstitute,NETHERLANDS
edwithanindirectfluorescentantibodytest(IFAT)andadirectagglutinationtest(DAT).
Received:August21,2014
Combinedmultiple-testingandseveralstatisticalmethodswerecomparedtoestimatethe
Accepted:December20,2014
CanLtrueprevalenceandtestscharacteristics(sensitivityandspecificity).TheBayesian
Published:March20,2015 fullmodelshowedthebestfitandyieldedprevalenceestimatesbetween11%(Mostaga-
Copyright:©2015Adeletal.Thisisanopenaccess nem,firstperiod)and38%(Bejaia,secondperiod).SensitivityofIFATvaried(infunctionof
articledistributedunderthetermsoftheCreative locality)between86%and88%whileitsspecificityvariedbetween65%and87%.DATwas
CommonsAttributionLicense,whichpermits
lesssensitivethanIFATbutshowedahigherspecificity(between80%and95%infunction
unrestricteduse,distribution,andreproductioninany
oflocalityor/andseason).AgeneralincreasingtrendoftheCanLprevalencewasnoted
medium,providedtheoriginalauthorandsourceare
credited. fromwesttoeast.Aconcordancebetweenthepresentresultsandtheincidenceofhuman
casesofvisceralleishmaniasiswasobserved,wherealsoamaximumwasrecordedforBe-
DataAvailabilityStatement:Fulldatasetis
includedinpublicationasTable2. jaia.TheresultsofthepresentstudyhighlightthedangerswhenusingIFATasa
goldstandard.
Funding:ThisworkwassupportedbyBelgian
Directorate-GeneralforDevelopmentCo-operation:
scholarshipAA.Thefundershadnoroleinstudy
design,datacollectionandanalysis,decisionto
publish,orpreparationofthemanuscript.
CompetingInterests:Prof.NikoSpeybroeckisa Introduction
PLOSONEEditorialBoardmember,butthisisno
LeishmaniasiscausedbyLeishmaniainfantumisendemicintheMediterraneanbasinwhere
wayaltershisadherencetoPLOSONEeditorial
policiesandcriteria. thedogisconsideredthemaindomesticreservoirforhumanvisceralleishmaniasis(VL)[1].
PLOSONE|DOI:10.1371/journal.pone.0117313 March20,2015 1/17
CanineLeishmaniaInfectionintheAlgerianLittoralZone
Canineleishmaniasis(CanL)causedbyL.infantumisaseverezoonoticdiseasethataffects
millionsofdogs[2].Theparasitesaretransmittedbythebitesoffemalesandfliesofthegenus
Phlebotomus(Phlebotominae,Diptera)thatareinAlgeriaPhlebotomusperniciosusandP.longi-
cuspis[3,4].TheincubationperiodofCanLrangesfromafewmonthstoseveralyears[5].The
clinicalfeaturesofthediseasevaryfromsubclinicalself-limitinginfectiontofataldisease[5,6].
Bothdiseasedandsub-clinicallyinfecteddogsareinfectioustosandflyvectors,allowingtrans-
missionoftheparasitetootherdogsorhumans[7,8].Thatiswhypromptdiagnosisofinfected
dogsisessential.Microscopicexaminationofsmearsoflymphnodeandbonemarrowaspirates,
alongwithserologyandpolymerasechainreaction,arethemostfrequentlyuseddiagnostic
methodsforCanL[9,10].However,accordingtotheWorldOrganisationforAnimalHealthse-
rologyisthepreferredmethodfordiagnosisofCanLandVL,evenduringtheearlystagesofthe
disease:withapurportedsensitivityof96%andapurportedspecificityof98%,immunofluores-
cenceantibodytest(IFAT)isconsideredthemostsuitedtestforfielddiagnosis[11].
Afirstepidemiologicalsurveyinvolving462dogsofAlgiers,inwhichthreeserologicaltests
wereassessedindifferentdogcategories,showedthatIFATwasthemostsensitive.However,its
specificitywasconsiderablylowerinfarmdogs(65%)[12].Therefore,inthepresentstudy,IFAT
wascombinedwiththedirectagglutinationtest(DAT)tocarryoutacross-sectionalsurveyinsix
citiesoftheseacoastregionofAlgeria.Thisareawaschosenbecausethegeographicaldistribu-
tionofVLcoversallthehumidandsub-humidregionsinthenorthofthecountry[13–16].
Severalstatisticalmethodshavebeendevelopedtoassessthetrueprevalenceandthediag-
nosticaccuracyintheabsenceofatruegoldstandard[17].Thetrueprevalenceisthenumber
oftrulyinfectedindividualsofatestedpopulation[18].Whennoperfecttestisavailable,the
resultswillprovideanestimateofthetrueprevalenceknownastheapparentprevalence.This
isalsocalledtheseroprevalencewhenusingserologicaldiagnostictests.Latentclassmodelsfor
whichthetruediseasestatusisconsideredtobealatentvariablewereoftenusedforestimating
testaccuracyanddiseaseprevalenceintheabsenceofagoldstandard[19].Thefirstmodelsug-
gestedbyHuiandWalter[20]isusefulwhentwoormoretestsareappliedtothesameindivid-
ualsfromtwoormorepopulations.Thismethodusesthemaximumlikelihoodprocedure
undertheassumptionofconditionalindependencebetweentestresults,thatsensitivityand
specificityareunchangedinthetwopopulationsandthateachpopulationhasdistinctdisease
prevalence.Iftheseassumptionsarenotmet,weuseaBayesianapproachtodrawinferences
aboutthediseaseprevalenceandtestpropertieswhileadjustingforthepossibilityofcondition-
aldependencebetweentests[21].Bayesianstatisticsisatheoryforinterweavingnewandexist-
ingdatabytakingbothsourcesintoaccount[19].
TheobjectiveofthepresentstudywastoestimatethetrueprevalenceofcanLinalargedog
populationbeforeandafterthevectorseason,comparingtheresultsobtainedfromstandard
frequentistapproachesanddifferentestimationmodelsusedinaBayesianframework.Atthe
sametime,thediagnostictestcharacteristicsofIFATandDATwerealsoevaluated.
MaterialsandMethods
Ethicsstatement
AuthorisationtoconductthesurveywasobtainedfromtheDirectiondesServicesVétérinaires
(DSV,MinistryofAgriculture)andtheInstitutNationaldeMédecineVétérinaire(INMV,
MinistryofAgriculture).Theprotocol,includingbloodcollectionprocedures,wasapproved
bytheDoctoralCommitteeattheInstituteofTropicalMedicine,AntwerpandtheDoctoral
CommitteeoftheFacultyofVeterinaryMedicineattheUniversityofLiège.Privatelyowned
dogswereincludedinthestudy.Bloodcollectionwasdonebymeansofcephalicveinalpunc-
ture,afterminimalrestrainingandwithoutanaesthetic.AteachWilaya(locality,township),
PLOSONE|DOI:10.1371/journal.pone.0117313 March20,2015 2/17
CanineLeishmaniaInfectionintheAlgerianLittoralZone
Fig1.Locationoflocalitiessampled.1=Tlemcen,2=Mostaganem,3=Tipaza,4=Boumerdes,5=Bejaia,6=Jijel.
doi:10.1371/journal.pone.0117313.g001
thestudywasauthorisedandsupervisedbytherespectiveInspectionVétérinairedeWilaya
(IVWdeTlemcen,Mostaganem,Tipaza,Boumerdes,JijelandBejaia),operatingundertheum-
brellaoftheDirectiondesServicesVétérinaires.TheofficialveterinaryofficeroftheInspection
VétérinairedeWilayainquestioncarriedoutthebloodcollectionafterobtainingoralpermis-
sionfromthedogowner.Thestudytookplaceontheterritoriesofthesixaforementioned
Wilayate(Tlemcen,Mostaganem,Tipaza,Boumerdes,JijelandBejaia).Thefieldstudiesdid
notinvolveendangeredorprotectedspecies.
Studyarea
Across-sectionalstudywasconductedinsixWilayate(singular:Wilaya=locality)oftheAlge-
rianlittoralzone(fromwesttoeast):Tlemcen,Mostaganem,Tipaza,Boumerdes,Bejaiaand
Jijel(Fig.1).TheoriginallistalsoincludedOranandAnnaba,butnocollaborationcouldbees-
tablishedtherebecauseofarabiesscare.Withintheselectedlocalities,surveyswerecarriedout
betweenFebruaryandApril2008(firstphase,beforethevectorseason,whichextendsfrom
MaytoOctober)andbetweenNovember2008andFebruary2009(secondphase,afterthevec-
torseason)[15].
Animals
Arandomsampleofthecaninepopulationofallmunicipalitiesateachlocationwasobtained
ineachcollectionphase.Nodistinctionbetweensexandagewasmade.Nostraydogswerein-
cludedastherewerenokennelsinthelocalitiesincludedinthesamplingframe.Withamaxi-
mumprevalenceof11%estimatedinAlgierswithinthegroupofstraydogs[12]andan
absoluteprecisionof5%,thesamplesizewas150dogs.Thus,werequestedtheveterinaryoffi-
cerstocollect200samplesineachcityineachseason.Allanimalswereinjudgedtobein
goodhealth.
Serology
Bloodfromthecephalicveinwascollectedindrylabelledtubes.Twoserologicaltestswere
used:anindirectimmunofluorescenceantibodytest(IFAT)andadirectagglutinationtest
PLOSONE|DOI:10.1371/journal.pone.0117313 March20,2015 3/17
CanineLeishmaniaInfectionintheAlgerianLittoralZone
(DAT).BothtestswereselectedafterevaluationindifferentgroupsofdogsinAlgiers[12]:
IFATshowedahighsensitivityinallgroups(≽90%),butalowspecificityinfarmdogs(65%),
whereasDAThadahighspecificityinallgroups(≽85%).
IFATwasperformedaccordingto[22]and[23].Theantigenconsistedofpromastigotesof
L.infantum.Anti-Leishmaniaantibodiesweredetectedbysecondaryantibodiesagainstrab-
bits,anti-dogimmunoglobulinsG(IgG)conjugatedwithfluoresceinisothiacyanate(FITC)
(Sigma-Aldrich,StLouis,MO,USA).A1:128dilutionwasusedascut-offvalue[24,25].
DATwasperformedusingacommercialkit,availablefromtheInstituteofTropicalMedi-
cine(Antwerp).Adilutionof1:320wasusedascut-offvalue[26,27].
Statisticalanalysis
Thedatageneratedbythesurveyproduce12sets(sixlocalitiestimestwosamplingperiods)of
fourdatapointseach(Table1).
Theapparentprevalence(laboratoryseroprevalence,proportionpositivetestresults)per
testandcombinationoftests:parallelinterpretation(atleastonepositivetestresultsignifiesa
positivedog)andserialinterpretation(bothtestsmustbepositivetobeacase)andtheirexact
95%confidenceintervalswerecalculated[l =qbeta(0.025,n ,n +1);l =
0.025 positive negative 0.975
qbeta(0.975,n +1,n )withqbetabeingthebetaquantilefunctionwithparameters
positive negative
(probability,shape1,shape2)andn...thenumberofpositiveornegativetestresults].Trueand
apparentprevalencebydiagnostictestandbysamplingperiodwereplottedonamapproduced
withQGIS2.0.1,usinggeoreferenceddataontheadministrativeregionsofAlgeria(obtained
fromhttp://www.diva-gis.org/gdata).
Concordancebetweentestresultswasexpressedintermsofindicesofpositiveandnegative
agreement[28]andtheir95%confidenceintervalswerecalculatedaccordingto[29].Compu-
tationsweredoneinR3.1.1(http://www.r-project.org).ThecodeisshowninS1Listing1.
Estimationofthetrueprevalenceandtestcharacteristicswasattemptedinseveralways.As
explainedindetailin[30],afulltwo-diagnostic-testmodel,assumingnoconditionalindepen-
dencebetweenthetwotests,consistsofthreeindependentequationswithsevenvariablesto
estimate(refertoTable1forT+/-;D+D-refertorespectivelytrulyinfectedandtruly
i
infection-freedogs):
pðTþTþÞ ¼W W W þð1(cid:2)W Þð1(cid:2)W Þð1(cid:2)W Þ
1 2 1 2 4 1 3 7
pðTþT(cid:2)Þ ¼W W ð1(cid:2)W Þþð1(cid:2)W Þð1(cid:2)W ÞW
1 2 1 2 4 1 3 7
pðT(cid:2)TþÞ ¼W ð1(cid:2)W ÞW þð1(cid:2)W ÞW ð1(cid:2)W Þ
1 2 1 2 5 1 3 6
pðT(cid:2)T(cid:2)Þ ¼W ð1(cid:2)W Þð1(cid:2)W Þþð1(cid:2)W ÞW W
1 2 1 2 5 1 3 6
¼1(cid:2)pðTþTþÞ(cid:2)pðTþT(cid:2)Þ(cid:2)pðT(cid:2)TþÞ
1 2 1 2 1 2
W ¼pðDþÞ ½pðD(cid:2)Þ¼1(cid:2)W (cid:3)
1 1
W ¼pðTþ jDþÞ
2 1
W ¼pðT(cid:2) jD(cid:2)Þ
3 1
W ¼pðTþ jDþ\TþÞ
4 2 1
W ¼pðTþ jDþ\T(cid:2)Þ
5 2 1
W ¼pðT(cid:2) jD(cid:2)\T(cid:2)Þ
6 2 1
W ¼pðT(cid:2) jD(cid:2)\TþÞ
7 2 1
PLOSONE|DOI:10.1371/journal.pone.0117313 March20,2015 4/17
CanineLeishmaniaInfectionintheAlgerianLittoralZone
Table1.Datastructureperlocalitypersamplinground.Tj:numberoftestresultswithi=test(1=IFAT,
i
2=DAT)andj=testresult(–=negative,+=positive).
DAT
-ve +ve
IFAT -ve T(cid:2)T(cid:2) T(cid:2)Tþ
1 2 1 2
+ve TþT(cid:2) TþTþ
1 2 1 2
doi:10.1371/journal.pone.0117313.t001
Thecurrentdatasethas36independentequations(sixlocalities×twoperiods×threeequa-
tions)and,assumingconditionaltestdependenceandspecificdiagnostictestcharacteristics
perlocalityperperiod,requires84variablestobeestimated(sixlocalities×twoperiods×seven
variables).Becausethenumberofvariablestoestimateexceedsthenumberofindependent
equations,priordeterministicorprobabilisticconstraintsneedtobeappliedtoallowestima-
tion[30].
Thestartingpointisthesimplifiedso-calledHui-WaltermodelAppendix[20],whereby
testsareassumedconditionallyindependentandtestsensitivityandtestspecificityareassumed
constantoverthetwosamplingperiodsperlocality.Perlocality,thismodelresultsinsixinde-
pendentequations(threeforthefirstsamplingperiodandthreeforthesecond)andsixvari-
ablestobeestimated(prevalencefirstperiod,prevalencesecondperiodandtwosensitivities
andspecificities–notethatthisresultsinsixvariables×sixlocalities=36variablesintotal).
Thismodelisthusentirelydrivenbydeterministicconstraints.Thestartingmodelis
(p,i2{1,2}forrespectivelyperiod1andperiod2):
i
p ðTþTþÞ ¼W W W þð1(cid:2)W Þð1(cid:2)W Þð1(cid:2)W Þ
1 1 2 1 3 5 1 4 8
p ðTþT(cid:2)Þ ¼W W ð1(cid:2)W Þþð1(cid:2)W Þð1(cid:2)W ÞW
1 1 2 1 3 5 1 4 8
p ðT(cid:2)TþÞ ¼W ð1(cid:2)W ÞW þð1(cid:2)W ÞW ð1(cid:2)W Þ
1 1 2 1 3 7 1 4 6
p ðT(cid:2)T(cid:2)Þ ¼W ð1(cid:2)W Þð1(cid:2)W Þþð1(cid:2)W ÞW W
1 1 2 1 3 7 1 4 6
¼1(cid:2)p ðTþTþÞ(cid:2)p ðTþT(cid:2)Þ(cid:2)p ðT(cid:2)TþÞ
1 1 2 1 1 2 1 1 2
p ðTþTþÞ ¼W W W þð1(cid:2)W Þð1(cid:2)W Þð1(cid:2)W Þ
2 1 2 2 9 11 2 10 14
p ðTþT(cid:2)Þ ¼W W ð1(cid:2)W Þþð1(cid:2)W Þð1(cid:2)W ÞW
2 1 2 2 9 11 2 10 14
p ðT(cid:2)TþÞ ¼W ð1(cid:2)W ÞW þð1(cid:2)W ÞW ð1(cid:2)W Þ
2 1 2 2 9 13 2 10 12
p ðT(cid:2)T(cid:2)Þ ¼W ð1(cid:2)W Þð1(cid:2)W Þþð1(cid:2)W ÞW W
2 1 2 2 9 13 2 10 12
¼1(cid:2)p ðTþTþÞ(cid:2)p ðTþT(cid:2)Þ(cid:2)p ðT(cid:2)TþÞ
2 1 2 2 1 2 2 1 2
W ¼p ðDþÞ ½p ðD(cid:2)Þ¼1(cid:2)W (cid:3)
1 1 1 1
W ¼p ðDþÞ ½p ðD(cid:2)Þ¼1(cid:2)W (cid:3)
2 2 2 2
W ¼p ðTþ jDþÞ;W ¼p ðTþ jDþÞ
3 1 1 9 2 1
W ¼p ðT(cid:2) jD(cid:2)Þ;W ¼p ðT(cid:2) jD(cid:2)Þ
4 1 1 10 2 1
W ¼p ðTþ jDþ\TþÞ;W ¼p ðTþ jDþ\TþÞ
5 1 2 1 11 2 2 1
W ¼p ðT(cid:2) jD(cid:2)\T(cid:2)Þ;W ¼p ðT(cid:2) jD(cid:2)\T(cid:2)Þ
6 1 2 1 12 2 2 1
W ¼p ðT(cid:2) jDþ\T(cid:2)Þ;W ¼p ðTþ jDþ\T(cid:2)Þ
7 1 2 1 13 2 2 1
W ¼p ðT(cid:2) jD(cid:2)\TþÞ;W ¼p ðT(cid:2) jD(cid:2)\TþÞ
8 1 2 1 14 1 2 1
PLOSONE|DOI:10.1371/journal.pone.0117313 March20,2015 5/17
CanineLeishmaniaInfectionintheAlgerianLittoralZone
TheHui-Walterdeterministicconstraintsareϑ =ϑ ;ϑ =ϑ ;ϑ =ϑ =ϑ =ϑ ;ϑ =ϑ =
3 9 4 10 5 7 11 13 6 8
ϑ =ϑ andthemodelthusbecomes:
12 14
p ðTþTþÞ ¼W W W þð1(cid:2)W Þð1(cid:2)W Þð1(cid:2)W Þ
1 1 2 1 3 5 1 4 6
p ðTþT(cid:2)Þ ¼W W ð1(cid:2)W Þþð1(cid:2)W Þð1(cid:2)W ÞW
1 1 2 1 3 5 1 4 6
p ðT(cid:2)TþÞ ¼W ð1(cid:2)W ÞW þð1(cid:2)W ÞW ð1(cid:2)W Þ
1 1 2 1 3 5 1 4 6
p ðT(cid:2)T(cid:2)Þ ¼W ð1(cid:2)W Þð1(cid:2)W Þþð1(cid:2)W ÞW W
1 1 2 1 3 5 1 4 6
p ðTþTþÞ ¼W W W þð1(cid:2)W Þð1(cid:2)W Þð1(cid:2)W Þ
2 1 2 2 3 5 2 4 6
p ðTþT(cid:2)Þ ¼W W ð1(cid:2)W Þþð1(cid:2)W Þð1(cid:2)W ÞW
2 1 2 2 3 5 2 4 6
p ðT(cid:2)TþÞ ¼W ð1(cid:2)W ÞW þð1(cid:2)W ÞW ð1(cid:2)W Þ
2 1 2 2 3 5 2 4 6
p ðT(cid:2)T(cid:2)Þ ¼W ð1(cid:2)W Þð1(cid:2)W Þþð1(cid:2)W ÞW W
2 1 2 1 3 5 1 4 6
W ¼p ðDþÞ ½p ðD(cid:2)Þ¼1(cid:2)W (cid:3)
1 1 1 1
W ¼p ðDþÞ ½p ðD(cid:2)Þ¼1(cid:2)W (cid:3)
2 2 2 2
W ¼p ðTþ jDþÞ
3 1 1
W ¼p ðT(cid:2) jD(cid:2)Þ
4 1 1
W ¼p ðTþ jDþÞ
5 1 2
W ¼p ðT(cid:2) jD(cid:2)Þ
6 1 2
ThismodelisruninWinBUGS[31].ThecodeisshowninS2Listing2.Modelfitandcon-
cordancebetweenpriorconstraintsanddataareassessedbymeansofBayes−pvalue,DICand
p (see[30]foradetaileddescriptionoftheuseofthesestatistics).Thepriorconstraintsap-
D
pliedtothetwoprevalencesarenecessarybecauseoftheexistenceoftwomirror-symmetricso-
lutionswhensolvingtheRogan-Gladenequation[32]:
p0 ¼pSeþð1(cid:2)pÞð1(cid:2)SpÞ
with:
p0 ¼apparent prevalence
p¼true prevalence
Se¼test sensitivity
Sp¼test specificity
Solutionsare:{p =p;Se =Se;Sp =Sp}and{p =1−p;Se =1−Sp;Sp =1−Se}.Applyinga
1 1 1 2 2 2
(realistic)prioruniform{0,0.5}topforcesthesolutiontowards{p ,Se ,Sp }.
1 1 1
Asecondapproachalsolimitsthenumberofvariablestobeestimatedto36.Contraryto
theHui-Waltermodel,itallowsforconditionaltestdependenceanddifferenttestcharacteris-
ticsindifferentlocality-periodcombinations.Reductionof84variablestobeestimatedto36is
achievedbycombiningthelocalitiesintotwogroups,basedonthelevelofapparentprevalence:
afirstgroupcontainsthetwowesternmostlocalities(TlemcenandMostaganem),thesecond
groupthefourotherlocalities(Tipaza,Boumerdes,BejaiaandJijel).The36variablestobeesti-
matedarethus:12prevalencesand4groups×6variablespergroup(ϑ ...ϑ ).TheWinBUGS
2 7
listingisshowninS3Listing3.
Asstatedbefore,thefullmodelforthecurrentdataset,assumingconditionaldependence
anddifferentvariablevaluesforthesixlocalities×twosamplingperiodscombinationsthushas
36independentequationsand84variables.Itisobviousthatthismodelisover-specifiedand
thatexternalinformationisrequiredinordertoobtainestimatesforthevariables.Theexternal
PLOSONE|DOI:10.1371/journal.pone.0117313 March20,2015 6/17
CanineLeishmaniaInfectionintheAlgerianLittoralZone
(prior)informationwasextractedfrom[12]:thisarticledemonstratedthatthediagnostictest
characteristics(bothforIFATandDAT)dependonthetypeofdogonwhichthetestisap-
plied.Thetypeofdogwasnotunequivocallyestablishedinthecurrentsurvey,althoughmost
oftheanimalscouldbeclassifiedaseitherfarmdogorguarddog.Thelimitsusedfortheprior
informationwerethereforetheminimumvalueofthelowerlimitsofthe95%confidenceinter-
valsobtainedforfarmdogsandguarddogsandthemaximumvalueofthecorresponding
upperlimits.Bayes-p,DICandp valueswereusedtoguidemodeladjustment,whichinthis
D
caseconsistedmainlyofrefiningtherangesoftheprioruniformdistributions.Sinceonlyval-
uesforsensitivityandspecificitywereavailablefrom[12],amodelbasedoncovarianceswas
usedtomodelconditionaltestdependence[21].TheWinBUGSlistingofthebasicmodelis
showninS4Listing4.
Results
Atotalof2,184dogsweresampledduringthetwophases(1,180duringthefirstsamplingperi-
odand1,004duringthesecond).Thedetailedresultsperlocality,periodandtestaregivenin
Table2.
Table3andFigs.2and3showtheapparentprevalences,theproportionpositiveresultsper
localityandsamplinground.IFAT-basedapparentprevalenceisnearlyalwayshigherthanthat
basedonDAT(theonlyexceptionisTipazaduringthesecondsamplinground).Withineach
locality,IFAT-basedapparentprevalenceisverysimilarforthetwosamplingphases.Thesame
istrueforDAT-basedseroprevalence,withtheexceptionofTlemcen,BoumerdesandBejaia
wherehighervaluesareobservedduringthesecondsamplinground.Parallelandserialinter-
pretationoftheresultsaremuchmorevariable:thereisnodifferencebetweenthetwosam-
plingroundsfortheserialinterpretationbasedapparentprevalenceforthetwomostwestern
Table2.Contingencytablesperlocalitypersamplingperiodwithpositive(+ve)andnegative(-ve)resultsinIFATandDAT.
Phase1 Phase2
IFAT DAT IFAT DAT
-ve +ve Total -ve +ve Total
Tlemcen -ve 107 12 119 -ve 94 13 107
+ve 47 16 63 +ve 29 16 45
Total 154 28 182 Total 123 29 152
Mostaganem -ve 81 0 81 -ve 120 3 123
+ve 50 12 62 +ve 27 19 46
Total 131 12 143 Total 147 22 169
Tipaza -ve 111 49 160 -ve 123 19 142
+ve 56 26 82 +ve 17 50 67
Total 167 75 242 Total 140 69 209
Boumerdes -ve 114 21 135 -ve 59 7 66
+ve 54 17 71 +ve 17 24 41
Total 168 38 206 Total 76 31 107
Bejaia -ve 76 28 104 -ve 97 11 108
+ve 64 30 94 +ve 32 63 95
Total 140 58 198 Total 129 74 203
Jijel -ve 117 30 147 -ve 96 7 103
+ve 32 30 62 +ve 21 40 61
Total 149 60 209 Total 117 47 164
doi:10.1371/journal.pone.0117313.t002
PLOSONE|DOI:10.1371/journal.pone.0117313 March20,2015 7/17
CanineLeishmaniaInfectionintheAlgerianLittoralZone
Table3.Apparentprevalenceswith95%confidenceintervalperlocality,samplingphaseandtest:
IFAT=indirectimmunofluorescencetest;DAT=directagglutinationtest;PAR=parallelinterpreta-
tionofthetwotestresults;SER=serialinterpretationofthetwotestresults.
Firstround Secondround
Tlemcen IFAT 0.35(0.28–0.42) 0.37(0.28–0.46)
DAT 0.15(0.10–0.21) 0.24(0.16–0.32)
PAR 0.41(0.34–0.49) 0.47(0.38–0.56)
SER 0.09(0.05–0.14) 0.13(0.08–0.20)
Mostaganem IFAT 0.43(0.35–0.52) 0.27(0.21–0.35)
DAT 0.08(0.04–0.14) 0.13(0.08–0.19)
PAR 0.43(0.35–0.52) 0.29(0.22–0.36)
SER 0.08(0.04–0.14) 0.11(0.07–0.17)
Tipaza IFAT 0.34(0.28–0.40) 0.32(0.26–0.39)
DAT 0.31(0.25–0.37) 0.33(0.27–0.40)
PAR 0.54(0.47–0.61) 0.41(0.34–0.48)
SER 0.11(0.07–0.15) 0.24(0.18–0.30)
Boumerdes IFAT 0.34(0.28–0.41) 0.38(0.29–0.48)
DAT 0.18(0.13–0.24) 0.29(0.21–0.39)
PAR 0.45(0.38–0.52) 0.45(0.35–0.55)
SER 0.08(0.05–0.13) 0.22(0.15–0.32)
Bejaia IFAT 0.47(0.40–0.55) 0.47(0.40–0.54)
DAT 0.29(0.23–0.36) 0.36(0.30–0.43)
PAR 0.62(0.54–0.68) 0.52(0.45–0.59)
SER 0.15(0.10–0.21) 0.31(0.25–0.38)
Jijel IFAT 0.30(0.24–0.36) 0.37(0.30–0.45)
DAT 0.29(0.23–0.35) 0.29(0.22–0.36)
PAR 0.44(0.37–0.51) 0.41(0.34–0.49)
SER 0.14(0.10–0.20) 0.24(0.18–0.31)
doi:10.1371/journal.pone.0117313.t003
localities(TlemcenandMostaganem),butishigherduringthesecondsamplingroundforthe
fourmoreeasternlocalities.
Indicesofpositiveandnegativeagreementandtheirconfidenceintervalsareshownin
Table4:theoverallindexofpositiveagreementwas0.52(95%CI:0.48–0.55)andtheoverall
indexofnegativeagreementwas0.79(95%CI:0.77–0.80).
Threetrendsareobservedintheindicesofagreement:(i)indicesofnegativeagreementare
invariablyhigherthanthoseofpositiveagreement,(ii)indicesofagreementarehigherduring
thesecondsamplingperiodthanduringthefirstperiod,(iii)thereisanincreaseinindicesof
positiveagreementalongthewest-easttransect.
TheresultsoftheHui-WaltermodelareshowninTable5andFig.4.Withtheexceptionof
Tlemcen,Bayes-pvaluesforalllocalitieslieclosetounity,indicatingpoormodelfit.AtTlem-
cen,thep valueforthefirstperiod,calculatedfromtheposteriormeansofthevariablenodes
D
ismarkedlysmallerthanthatcalculatedfromthemeanposteriorprobabilities,againindicating
lackofconcordancebetweenpriorinformationanddata[30].Thesamelackoffitisobserved
forthemodel,basedonthegroupedlocalities:theoverallp equals-14.1andtheBayes-pval-
D
uesvarybetween0.41and0.91.
Thefullmodeldidresultinagoodfit(p =27.92,DIC=234.52andBayes-pvaluesbetween
D
0.47and0.72)andyieldedsignificantcovariancesbetweenspecificities,whoseposteriorproba-
bilitydistributionsdidnotincludezero,indicatingconditionaldependence.Theestimatesob-
tainedfromthismodelareshowninTable6andFig.5.Theresultsindicateanincreasein
PLOSONE|DOI:10.1371/journal.pone.0117313 March20,2015 8/17
CanineLeishmaniaInfectionintheAlgerianLittoralZone
Fig2.Apparentprevalencepertestpersamplingperiodperlocality.Fromtoptobottom:IFATfirstperiod;IFATsecondperiod;DATfirstperiod;DAT
secondperiod.
doi:10.1371/journal.pone.0117313.g002
Fig3.Apparentprevalenceperlocality.Fromtoptobottom:onlyIFAT,onlyDAT,serialinterpretationofthetwotestresultsandparallelinterpretationof
thetwotestresults.
doi:10.1371/journal.pone.0117313.g003
PLOSONE|DOI:10.1371/journal.pone.0117313 March20,2015 9/17
CanineLeishmaniaInfectionintheAlgerianLittoralZone
Table4.ConcordancebetweenIFATandDAT:indicesofpositiveagreement(IPA)andnegativeagree-
ment(INA)and95%confidenceintervals.
Firstround Secondround
Tlemcen IPA 0.35(0.23–0.48) 0.43(0.29–0.57)
INA 0.78(0.73–0.84) 0.81(0.76–0.87)
Mostaganem IPA 0.32(0.20–0.45) 0.56(0.42–0.69)
INA 0.76(0.69–0.82) 0.88(0.84–0.92)
Tipaza IPA 0.33(0.24–0.43) 0.74(0.63–0.80)
INA 0.68(0.61–0.73) 0.87(0.83–0.90)
Boumerdes IPA 0.31(0.20–0.43) 0.67(0.54–0.77)
INA 0.75(0.70–0.80) 0.83(0.76–0.89)
Bejaia IPA 0.39(0.30–0.49) 0.75(0.67–0.81)
INA 0.62(0.55–0.70) 0.81(0.76–0.87)
Jijel IPA 0.49(0.39–0.59) 0.74(0.64–0.82)
INA 0.79(0.73–0.83) 0.87(0.82–0.91)
Overall IPA 0.37(0.33–0.42) 0.68(0.64–0.72)
INA 0.73(0.71–0.75) 0.85(0.83–0.87)
doi:10.1371/journal.pone.0117313.t004
prevalencefromwesttoeastandanincreaseinprevalencefromthefirstsurveyperiodtothe
second,althoughcredibilityintervalsoverlap.AlsonoteworthyistheincreaseinIFATspecific-
ityfromthefirstsurveyperiodtothesecond,althoughthesameremarkaboutcredibility
intervalsapplies.
Discussion
TheapparentprevalenceobtainedwithIFATvariedbetween30%(Jijel)and47%(Bejaia)dur-
ingthefirstphaseandfrom27%(Mostaganem)to47%(Bejaia)duringthepost-vectorial
phase,whiletheestimatewithDATwaslower(8%atMostaganemto31%atTipazaforthe
firstphaseand13%atMostaganemto36%atBejaiaduringthesecond).Theseestimateslieto-
wardsthehigherendoftherangeofresultsreportedforothercountriesinthemediterranean
basinusingIFAT:Tunisia18-53%[33,34];Morocco9-19%[35,36];Malaga(Spain)35%[37];
westernLiguria(Italy)30%[38];Corfu(Greece)50%[39].Results,obtainedwhenusingDAT,
rangedfrom6%inPortugal[40]to27%inCorsica[41]to28%inTurkeywitha100%agree-
mentwithIFATresults[42].
Table5.EstimatesfromtheHui-Waltermodel.
Locality Pr1 Pr2 SeFAT SpIFAT SeDAT SpDAT Bayes-p pD DIC
Tlemcen 0.26 0.26 0.72 0.81 0.52 0.92 0.51,0.56 0.49,2.62(2.13,2.08) 17.6,19.7(19.3,19.3)
Mostaganem 0.39 0.33 0.63 0.72 0.24 0.92 0.95,0.89 -9.9,-28.7(1.74,2.45) -4.5,6.4(20.9,22.2)
Tipaza 0.27 0.38 0.32 0.57 0.33 0.59 0.98,0.99 4.06,-56.12(3.31,3.09) 34.2,-16.1(33.52,40.73)
Boumerdes 0.11 0.27 0.83 0.73 0.80 0.88 0.58,0.85 1.66,2.48(2.52,1.99) 20.3,22.5(21.1,22.0)
Bejaia 0.15 0.30 0.93 0.66 0.90 0.83 0.95,0.99 1.58,2.52(2.48,2.01) 28.3,33.2(29.1,32.7)
Jijel 0.31 0.36 0.59 0.70 0.53 0.74 0.75,0.97 -14.26,-44.69(2.85,2.04) 6.68,-18.03(23.8,28.7)
Pr =prevalencefirstperiod;Pr =prevalencesecondperiod;Se =sensitivityIFAT;Sp =specificityIFAT;Se =sensitivityDAT;Sp =
1 2 IFAT IFAT DAT DAT
specificityDAT;p ,DICtopline=valuescalculatedfromposteriorvaluesofvariablenodesresp.forfirstperiodandsecondperiod,(bottomline)=values
D
calculatedfromposteriorprobabilities
doi:10.1371/journal.pone.0117313.t005
PLOSONE|DOI:10.1371/journal.pone.0117313 March20,2015 10/17