ContentslistsavailableatScienceDirect
Epidemics
jo u rn al h om ep age : w w w . e l s e v i e r . c o m / l o c a t e / e p i d e m i c s
Epidemiological
modelling
for
the
assessment
of
bovine
tuberculosis
surveillance
in
the
dairy
farm
network
in
Emilia-Romagna
(Italy)
Gianluigi
Rossi
a,∗,
Giulio
A.
De
Leo
b,
Stefano
Pongolini
c,
Silvano
Natalini
d,
Simone
Vincenzi
e,f,
Luca
Bolzoni
c,gaDipartimentodiBioscienze,UniversitàdiParma,ParcoAreadelleScienze11/A,I-43124Parma,Italy bStanfordUniversity,HopkinsMarineStation,PacificGrove,CA93950,USA
cIstitutoZooprofilatticoSperimentaledellaLombardiaedell’Emilia-Romagna,ViadeiMercati,13/A,ParmaI-43122,Italy
dServizioVeterinarioeIgieneAlimentiAssessoratoPoliticheperlaSaluteRegioneEmilia-Romagna,VialeAldoMoro21,BolognaI-40127,Italy eCenterforStockAssessmentResearch,DepartmentofAppliedMathematicsandStatistics,UniversityofCalifornia,SantaCruz,CA95064,USA fDipartimentodiElettronica,InformazioneeBioingegneria,PolitecnicodiMilano,ViaPonzio34/5,I-20133Milan,Italy
gDepartmentofBiodiversityandMolecularEcology,ResearchandInnovationCentre–FondazioneEdmundMach,SanMicheleall’Adige,Trento,Italy
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received17March2014
Receivedinrevisedform11February2015 Accepted25February2015
Availableonline6March2015 Keywords: Epidemiologicalmodel Diseasesurveillance Networkmodel Bovinetuberculosis SEImodel
a
b
s
t
r
a
c
t
Assessingtheperformanceofasurveillancesystemforinfectiousdiseasesofdomesticanimalsisa chal-lengingtaskforhealthauthorities.Therefore,itisimportanttoassesswhatstrategyisthemosteffective inidentifyingtheonsetofanepidemicandinminimizingthenumberofinfectedfarms.
Theaimofthepresentworkwastoevaluatetheperformanceofthebovinetuberculosis(bTB) surveil-lancesysteminthenetworkofdairyfarmsintheEmilia-Romagna(ER)Region,Italy.AbTB-freeRegion since2007,ERimplementsanintegratedsurveillancestrategybasedonthreecomponents,namely rou-tineon-farmtuberculinskin-testingperformedevery3years,tuberculinskin-testingofcattleexchanged betweenfarms,andpost-morteminspectionatslaughterhouses.Weassessedtheeffectivenessof surveil-lancebymeansofastochasticnetworkmodelofbothwithin-farmandbetween-farmbTBdynamics calibratedondataavailableforERdairyfarms.Epidemicdynamicsweresimulatedforfivescenarios: thecurrentERsurveillancesystem,anosurveillancescenariothatweusedasthebenchmarkto charac-terizeepidemicdynamics,threeadditionalscenariosinwhichoneofthesurveillancecomponentswas removedatatimesoastooutlineitssignificanceindetectingtheinfection.Foreachscenarioweran MonteCarlosimulationsofbTBepidemicsfollowingtherandomintroductionofaninfectedindividual inthenetwork.Systemperformanceswereassessedthroughthecomparativeanalysisofanumberof statistics,includingthetimerequiredforepidemicdetectionandthetotalnumberofinfectedfarms duringtheepidemic.
Ouranalysisshowedthatslaughterhouseinspectionisthemosteffectivesurveillancecomponentin reducingthetimefordiseasedetection,whileroutinesurveillanceinreducingthenumberofmulti-farms epidemics.Ontheotherhand,testingexchangedcattleimprovedtheperformanceofthesurveillance systemonlymarginally.
©2015TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction
Bovine tuberculosis (bTB) caused by Mycobacterium bovis is amongthemajordiseasethreatstofarmanimalsworldwide(Skuce etal.,2012).bTBisachronicdiseasecharacterizedbyavariable andgenerallylongincubationperiod,anditsaetiologicalagenthas anexceptionallywiderangeofhosts,includinghumans,domestic
∗ Correspondingauthor.
E-mailaddress:gianluigi.rossi@nemo.unipr.it(G.Rossi).
andwildanimals,aswellashighpersistenceintheenvironment (Morrisetal.,1994).Besidesitsdirectimpacttothecattle indus-try,bTBisazoonoticdiseaseofgreatconcern.Forthesereasons, regulatoryrestrictionsareinplacetopreventthetradeofinfected animalsandtheirproductswithinandbetweencountriesandto reducetheriskofspilloverfromtheanimaltothehuman compart-ment.Mostcountriesandtransnationalareas(suchastheEuropean Union)withathrivingcattleindustryhavedevelopedsurveillance systemstopreventbTBoutbreaks(Cousins,2001).Despitethe con-trolefforts,bTBisstillendemicinmanycountries,bothdeveloped anddeveloping(WorldOrganizationforAnimalHealth,2008)and http://dx.doi.org/10.1016/j.epidem.2015.02.007
itseradicationhasprovedtobeverychallenging(Fitzgeraldand Kaneene,2013;IndependentScientificGroup,2007).Therefore,it isessentialtodevelopsurveillancestrategiesthatallowforarapid detectionofinfectedanimalsbothinendemicandbTB-freeareas. Theorganizationandthecomponentsneededforaneffective surveillancesysteminagiventerritorydependonbTB epidemi-ology,which islargely determinedbythecharacteristicsofthe farmingsysteminplace.Howthesefactorsmayinterplayto deter-minetheriskofoutbreaksinthespecificcaseofbTBhasbeenwell documentedintheUKandIreland,wheretherisingincidenceof theinfectioninthelast20yearshascausedsignificanteconomic losses(Abernethyetal.,2013;Reynolds,2006).Thedirectcosts ofbTBforUKtaxpayersin2009wereestimatedinaround
£
63M (about95MUSD)andover25,000cattlewereculled(Johnstonetal., 2011).Outdoorfarming,acommonbreedingsystemintheUKand Ireland,favourscontactsbetweenanimalsofdifferentherds,one ofthepathwaysforbTBtransmission.Inaddition,thisfarming sys-temexposescattletothepossiblyinfected wildlife,suchasthe Europeanbadger(Melesmeles)whosepopulationisendemically infectedbyM.bovisinlargeareasoftheBritishisles(Cheeseman etal.,1989;Griffinetal.,2005;Morrisetal.,1994).Anotherfactor associatedwithrecurrent bTBoutbreaksintheUK is between-farmmovementofliveanimals(Johnstonetal.,2011;Reillyand Courtenay,2007).Therefore,preventingcontactsbetweencattle of different herds and between cattle and wildlife are consid-eredeffectivemeasurestocontrolbTBriskintheUKandIreland (Johnstonetal.,2011;Phillipsetal.,2003;ReillyandCourtenay, 2007).Incountrieswhereindoorfarmingisthemostcommonbreeding practice,suchasItaly,bTBtransmissionrarelyoccursthrough con-tactswithinfectedwildlife.AstudyonbTBriskfactorsinNorthern ItalyshowedthatthemainriskfactorforbTBbreakdownwas cat-tlemovementbetweenfarms(Marangonetal.,1998).bTBisstill endemic,thoughatverylowprevalence,insomeItalianregions whileithasbeeneradicatedinothers.
Amongthelatter,Emilia-Romagna(ER),locatedontheright sideoftheriverPovalley(NorthernItaly),hasbeendeclared offi-ciallyfreefrombovinetuberculosis(UEDecisionn◦2007/174/CE). ThisRegionhasanimportantfood-farmingindustrycharacterized byan intensivedairy production, and istheregionof originof Parmigiano-Reggianocheese.Accordingtothelatestofficial statis-tics(ItalianNationalStatisticsInstitute,2010),about550,000cattle arerearedin7343farmsinER.AsanepidemicofbTBinthisRegion wouldhave severeeconomic consequences, theregionalhealth authorities have implemented, in compliance with EU regula-tions(UEDecision2002/677/CE),anintegratedsurveillancesystem topreventbTB re-emergencein dairy cattle (Regional decision GPG/2010/1049).Thesystemisbasedonthreedetectionmethods, namely:(i)periodicroutineskin-testonallanimalsagedover24 months(RS);(ii)systematicskin-testonallexchangedcattle(ECT) atthedestinationfarm;and(iii)inspectionforbTBlesionsofall slaughteredanimals(SI).WhileERRegionhasmanagedtoremain bTB-freesofar,theeffectivenessoftheintegratedsurveillance sys-tem(anditsindividualcomponents)inthecaseofre-introduction oftheinfectionhasneverbeenassessed.
Thegoalofthisworkistoquantitativelyassesstheperformance ofthecurrentsurveillancesystemanditsindividualcomponentsin termsof(i)timeneededtodetectaprospectivenewlyintroduced bTBepidemicinthedairyfarmsofER,and(ii)themagnitudeofthe epidemic,intermsofnumberoffarmsinfectedbeforethedetection oftheepidemic.
SinceERhasbeenfreefrombTBformanyyears,noempirical dataontheepidemiologicaldynamicsoftheinfectionisavailable. Therefore,inordertotesttheperformanceofthethree surveil-lancemethodscurrentlyinplaceintheRegion,wedevelopedan individual-basedepidemicmodelcapableofsimulatingthespread
Dairy farms size (#individuals)
Frequency 0 500 1000 1500 0 250 500 750 1000 1250 1500 1750 2000
a)
Sample farms size (#individuals)
Frequency 05 0 150 250 0 500 1000 1500 2000
b)
Fig.1. DistributionofEmilia-Romagnadairyfarmssize.Everybarcorrespondstoa sizeintervalof50individuals(0–50,51–100,etc.).(a)Alldairyfarmssize distribu-tion(Min=1,Median=63,Mean=101.9,Max=1889);(b)sampleddairyfarmssize distribution(Min=1,Median=74,Mean=126.6,Max=1889).
ofbTBintheregionalnetworkofdairyfarmsfollowingtherandom introductionofinfectedanimalsinthesystem.
Considering that therearing systemin placein ERand that thecomponentsof bTBsurveillancedescribedabovearelargely widespreadin severaldeveloped countries, ourfindingscan be applicabletomanyfarmingandsurveillancesystemsworldwide.
2. Materialsandmethods
Toassesstheperformanceofthecurrentsurveillancesystem andeachofitscomponents,webuiltadata-driven,stochastic net-workmodelthatsimulatesbothwithin-farmandbetween-farm bTBdynamics.Weassumedthatthecontributionoftransmission routesotherthancattleexchanges–suchasairbornetransmission ortransmissionmediatedbywildlifeorfomites–wasnegligible withintheregionalsystemof industrialindoorfarms.Thus,we consideredcattleexchangesastheonlytransmissionrouteforbTB amongdairyfarms.
Since bTB is a chronic diseasecharacterized by slow trans-missiondynamics (see Agustoet al., 2011; Huang et al.,2013; Brooks-Pollocketal.,2014)theassumptionofendemicequilibrium withinagivenfarmisunrealistic.Then,weexplicitlyaccountedfor within-farmdiseasedynamicsdescribinghow epidemicsevolve insidefarmsfollowingtheintroductionofinfectedanimals.
WerepresentedthesystemofcattleexchangesofERasa con-tactnetworkwherenodesrepresentfarmsanddirectionaledges representbetween-farmanimalmovements.Wereproducedthe network of4353dairy farmsand 20 intermediarytrader farms (ITF)byusinginformationfromthecattlemovementdatasetas describedhereafter.
Allstatisticalanalysesandmodelsimulationswereperformed usingthesoftwareRwith“MASS”,“triangle”and“poilog”packages (http://www.r-project.com).
2.1. Cattletrademovementdata
Cattle movementdatawereprovidedbytheItalian National Bovinedatabase.Attheendof2010,4353dairyfarmsand20 inter-mediarytraderfarms(ITF)wereinoperationinER.Forallofthem wecollectedthefarmsize,i.e.numberofanimalsperfarm(Fig.1). ForalltheITFandasubsetof837dairyfarmswealsocollected individualrecordsofeverycattlemovement,incomingand outgo-ing,overa100-weektimespan(fromthebeginningofFebruary
Table1
Detaileddataonsampleddairyfarmsandintermediarytraderfarms(ITF).
Cattlemovementdairyfarmsdata(February-2009/December-2010) CattlemovementITFdata(February-2009/December-2010) TotaldairyfarmsinEmilia-Romagna 4353 TotalITFinEmilia-Romagna 20
Samplesdairyfarms 837 SampleITF 20
Sampleindividualrecords 106,499 Sampleindividualrecords 3571
Incomingindividuals Incomingindividuals
Total 9173 Total 3571
F1romERdairyfarms 7036 FromERdairyfarms 1774
Fromotherregions/countries 2137 Fromotherregions/countries 1797
Outgoingindividuals Outgoingindividuals
Total 28,954 Total 3476
Totaltofarms 6328 Totaltofarms 3211
TootherERdairyfarms 5597 TootherERdairyfarms 2136
Tootherregions/countries 731 Tootherregions/countries 1075
Totaltoslaughterhouse 21,851 Totaltoslaughterhouse 230
ToERslaughterhouse 10,195 ToERslaughterhouse 147
Tootherregionsslaughterhouse 11,656 Tootherregionsslaughterhouse 83
Death/otherdestinations 775 Death/otherdestinations 35
Samplefarmsmeanin-degree 1.91(SE±0.096) ITFmeanin-degree 24.50(SE±7.185) Samplefarmsmeanout-degree 1.67(SE±0.075) ITFmeanout-degree 19.25(SE±5.537)
2009totheendofDecember2010).Thefarmsweresampled strat-ifiedbyprovinceandtheobtainedsampleisrepresentativeofthe farmssizedistributionobservedintheregion(Fig.1and Supple-mentarymaterialsS1.1fordetails).Eachindividualrecordhada uniqueidentifiercodefortheanimal,thebirthdate,sexandrace oftheanimal,identifiercodesforthefarmsoforiginand desti-nation,codesforfarmsproductionsector(beef,dairyormixed), andthemovementdate.Weconsideredthemovementsbetween dairyfarmsonly,andweexcludedthemovementsofyoungcalves orend-of-lifecattlesenttobeeffarmsordirectlytothe slaughter-house.Thefinaldatasetwascomposedof15,501individualrecords reportinganimalmovement(Table1).
ITFactivityconsistsintradingcattle,notinrearingthem.Asa consequence,animalsstayinITFforjustafewdays(median2days, lower[upper]quartile1[8]days)beforebeingmovedtoa desti-nationdairyfarm,amuchshortertimecomparedtothetimethat individualsspendindairyfarms(1815days,onaverage).In addi-tion,thenumberofanimalssimultaneouslyheldinITFsisusually verylow(median0,lower[upper]quartile0[5]animals).Given thelimitednumberofanimalsandtheshorttimethatindividuals spendinsidethem,weassumedthatthebTBtransmissioninside ITFswasnegligiblecomparedtodairyfarms(asensitivityanalysis ontheeffectofthisassumptiononbTBdynamicsandthe effec-tivenessofsurveillanceisshownintheSupplementarymaterials, SectionS3.4).
Wedefinedthein-degreeandtheout-degreeforfarmiasthe numberoffarmsfromwhichfarmireceivesandthenumberof farmstowhichfarmisendscattle,respectively.Altogether,thejoint setsofin-andout-degreesarereferredtoasthein-andout-degree distributionsofthenetworkofdairyfarms.Sincemovementdata wereavailableforallITFsinoperationinER,wedirectlyderived thein-andout-degreeofeachITFinthenetworkaswellastheir distributions.
RegardingfarmsotherthanITFs,movementdatawere avail-ableonlyfor asubsetof837farmsoutofthetotal4353inthe ERnetwork.Thus,wederivedthein-andout-degreedistributions ofthissubset.Then,weestimatedtheparametersofasetof can-didateprobabilitydistributions(namely:power-law,log-normal, PoissonandYule)ontheavailabledataandselectedthe distribu-tionthatbestfittedthedatabyusinglog-likelihoodratio(LLR)test asshowninVuong(1989)andClausetetal.(2009).Finally,we usedtheselectedprobabilitydistributiontogenerateasynthetic networkof4353farmshavingthesametopologicalproperties(i.e. thein-andout-degreedistributions)observedinthesubsetof837 dairyfarms.Theprocedurefornetworkgenerationisdescribedin thefollowingsection.
2.2. Networkmodel
Togenerateanetworkof4353farmswiththesame topologi-calpropertiesobservedinthesubsetof837dairyfarms,wefirst assigned in- and out-degreesindependentof each othertothe 4353nodesinthenetwork(asin-andout-degreeswereweakly correlated,seeSupplementarymaterialsS1.3).Moreover,asboth in-degreesandout-degreeswereveryweaklycorrelatedwithfarm sizeintheobserveddataset,theywereassignedtonodes irrespec-tiveoffarm size(seeSupplementarymaterialsS1.3). According totheavailablemovementdata,weassumedthatneitherthe in-degreenortheout-degreecouldbelargerthanthenodesize,i.e. thenumberofanimalsinthefarm.
Onceweassignedtoeachnodeanin-degreeandanout-degree, weusedaheuristicalgorithmtoexactlymatchthenumberof out-goingconnectionsfromfarmsinthenetworktothatofingoing connectionstofarmsinthesamenetwork(Supplementary mate-rialsS2).Then,wecompletedthenetworkbyadding20additional nodesrepresentingthe20ITFandassignedtoeachofthemthe observedin-andout-degrees.
2.3. Within-farmmodel
Wedescribedthewithin-farmbTBdynamicsthrougha com-partmentalstochasticSusceptible–Exposed–Infected(SEI)model with homogeneous frequency dependent transmission (Agusto etal.,2011;Brooks-Pollocketal.,2014).Thecompartmentofthe susceptible individuals (S) representscattlethat have notbeen infectedyet.Followingsuccessfulinfection,eachindividualis clas-sifiedasexposed(E)butnotinfectiousyet.Theincubationperiod endswhentheinfectedindividualbecomesalsoinfectious(I).Then, theinfectiondynamicsforfarmicanberepresentedbythe follow-ingsystemofordinarydifferentialequations:
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
˙Ei=ˇiNSiIi i −(+)Ei− j/=i ijEi+ j/=i jiEj ˙Ii=Ei−Ii− j/=i ijIi+ j/=i jiIj (1)wherefarmsizeNi,i.e.thenumberofcattleinfarmi,wasassumed
constantintime,andthenumberofsusceptibleanimalsinfarm iwascomputedasSi=Ni−Ei−Ii;ˇirepresentsthewithin-farm
transmissionrate;therateatwhichexposedindividualsbecome infectious;ijthemovementrateofindividualsfromfarmitofarm
j,whichisproportionaltofarmiout-degree;andtherateatwhich individualsleavethefarmsystem,eitherbecausetheyaresentto
Table2
Thelistandratesofthepossibleeventsforthestochasticmodel.
Eventinfarmi Transition Rateatwhich eventoccurs Infection Ei→Ei+1 ˇSiIi/Ni Exposedindividualbecome
Infectious
Ei→Ei−1,Ii→Ii+1 Ei Exposedindividualissent
tofarmj
Ei→Ei−1,Ej→Ej+1 iEi Exposedindividualleaves
thesystem
Ei→Ei−1 Ei Infectiousindividualsent
tofarmj
Ii→Ii-1,Ij→Ij+1 iIi Infectiousindividualleaves
thesystem
Ii→Ii−1 Ii
slaughterhouse,tofarmsoutsidetheERsystem,orbecausethey die.
Toexplicitlyaccountforthestochasticnatureofthecontactand transmissionprocessesinalargenetworkoffarms,weran stochas-ticsimulationsthroughanevent-drivenapproachinwhichtime stepsbetweentwoconsecutiveeventsweredescribedthrougha Poissonprocesswithexponentiallydistributedmeans(Gillespie, 1977;Rohanietal.,2002).Alltheeventssimulatedbythestochastic infectionmodelarereportedinTable2.
2.4. Parameterestimation
The basic reproduction number represents the mean num-berofsecondaryinfectionscausedbyasingleinfectedindividual introducedintoacompletelysusceptible population(Diekmann etal.,1990).AsERisanofficiallybTB-freeregionsince7years,there arenoavailableoutbreaksdatatoestimateawithin-farmbasic reproductionnumber,R0.Thus,wesetR0valueinoursimulations
usingestimatesforbTBfoundintheliterature(Brooks-Pollockand Keeling,2009;Agustoetal.,2011;Conlanetal.,2012;O’Hareetal., 2014).Inparticular,wedescribedtheuncertaintyinR0by
assum-ingatriangulardistributionwithmode2.40(whichcorresponds tothemeanvalueofliteratureestimates),lowerlimit0.83,and upperlimit4.9(whichcorrespondtotheextremevaluesfound), seeSupplementarymaterialsS3.1for details.Infurther simula-tionswetestedtheeffectofR0variability,comparingtheobtained
resultsagainstthoseofsimulationsranusingafixedR0value(see
SupplementarymaterialsS3.2).
Moreover,inordertotesttherobustnessofourresultsagainst possiblevariationsinthewithin-farmbTBbasicreproduction num-ber,werunfurthersimulationsusingasfixedvaluethehighest (R0,i 4.9)andthelowest(R0,i 0.83)oftheliteratureestimates
(seeSupplementarymaterialsS3.3).
Theithfarmtransmissionrate,ˇi,wasindirectlyderivedfrom
theexpressionforthewithin-farmbasicreproductionnumberof model(1),i.e.assuminginfectioninithfarmonly,byusingNext GenerationMatrixtechniques(Diekmannetal.,1990):
ˇi=R0
(
j/=iij+)(j/=iij++)(2)
FollowingBarlowetal.(1997),wesetthemeantime(=1/) spentintheexposedclassEto202days.Likewise,parametersand mean[ij]wereestimatedfrommovementdataastheinverseofthe
meantimethatanimalsspentinsidethefarmbeforebeingmoved outofthesystemandtowardsotherERdairyfarms,respectively.
ITFsarecharacterizedbysubstantialdifferentbehaviourswith respecttodairyfarms.AsweassumedthatthereisnobTB trans-missioninsidethesefarms,wesetˇITF=0andweestimatedITF
specificmovement(ITF)andexit(ITF)ratesfromavailabledata.
2.5. Assessmentofthesurveillancesystem
WeusedthestochasticmodeldescribedinTable2toassessthe effectivenessofthethreesurveillancecomponentsofthecurrent surveillancesystemimplementedinERRegion.Specifically:(i)a systematiccontroloneveryfarmbasedonatuberculinskin-tests (RS).Thisisperformedoneveryindividualinthefarmolderthan24 months,whichrepresentsabout60%ofthetotalfarmpopulation. FollowingcurrentlegislationforbTB-freeareasinERallcattleherds aretestedwithaturnaroundperiodofthreeyears;(ii)askin-test performedatthedestinationfarmonmovedcattlewithin45days afterthearrival(ECT);(iii)avisualinspectionforbTBcharacteristic lesionsoncattleslaughtered(SI).
TosimulatetheRSweassumedthat,withinaturnaroundperiod, farmsaretestedinarandomorder(withafrequencyofabout4 farmsaday).Tosimulatetheothersurveillancecomponents,we assumedtotestallindividualsmovedbetweendairyfarmsand toallindividualssenttoslaughterhouses.Whenananimaltests positivetothetuberculinskin-test,bothforRSandECTcases,the procedurestoverifywhethertheanimalistrulyinfectedbyM.bovis lastusually2months.Theseproceduresconsistinacarcass inspec-tionforbTBtypicallesionsandinabacteriologicalexamination. Duringthisconfirmationperiod,cattlefromthesamefarmcannot betransferredtootherfarms.Thisprotocolwassimulatedinthe modelbyassumingthat,whenapositiveindividualisdetectedby theskin-test,thefarmcannothaveanycontactwithotherfarms, butthesimulationofdiseasedynamicslastsforother2months.
Theskin-testcurrentlyinuseinItalyisthecervicalSingle Intra-dermalTest(SIT).Intheirreview,delaRua-Domenechetal.(2006) showed that theSIT sensitivity estimates fallbetween 0.7 and 1.However,on-farmtestsensitivitycanbesubstantiallysmaller than in controlledlaboratoryexperiment trials asother factors than theimperfect natureof thetest may affecttest outcome, suchasthetraininglevel,careand experienceofthe veterinar-ianandtheactualprotocolused.Therefore,followingWelbyetal. (2012),wesimulatedtheuncertaintyintestsensitivitiesby ran-domlydrawingsensitivity values froma beta distributionwith mean0.60andcoefficientofvariation0.10.AnalogouslytoBarlow etal.(1997),weassumedareductionoftestsensitivityininfective butnotinfectiousanimals(meanvalue0.54).Sensitivityof post-morteminspectionattheslaughterhousescanalsobeveryvariable (Assegedetal.,2004;Fischeretal.,2005;Smithetal.,2013;van Asseldonketal.,2005;Welbyetal.,2012).Therefore,similarlyto Welbyetal.(2012),wesimulatedtheuncertaintyinpost-mortem inspectionasabetadistributionwithmean0.60andcoefficientof variation0.10(withareductionofsensitivityforexposedanimals asintuberculinskin-test).
In addition, with the intent to simulate different levels of accuracyandawarenessinthetestingprocedures,werana sen-sitivityanalysistoassesstheeffectofpost-morteminspectionand tuberculinskin-testsensitivityvaluesonsurveillancesystem per-formance (Hadorn and Stärk, 2008; Humblet et al., 2011).The resultsareshowninSupplementarymaterialsS4.
Diseasedynamicsfollowingtheintroductionofaninfected indi-vidualinthenetworkofdairyfarmsofERwereassessedunder fivescenarios.Firstly,wesimulatedabaselinescenarioofbTB epi-demicsspreading without surveillanceso astounderstand the potentialimpactsofuncontrolledbTBepidemicsonthesystem. Secondly,weransimulationsontheintegratedsurveillance sys-temcurrentlyinplace,whichincludesallthethreesurveillance components(Scenario1).Then,inordertoassesstheperformances ofeachcomponentofthesurveillancesystem,weanalyzedthree alternativescenariosinwhichweremovedoneofthethree com-ponentsatatimeandsimulatedepidemicdynamicswiththeother twosystemsinplace.InScenario2a,Scenario2bandScenario2c weremovedRS,ECTandSIcomponents,respectively(Fig.2).
Fig.2.Differentsurveillancesystemscenariossimulated.
Foreachscenarioweran10thousandstochasticMonteCarlo simulationsof diseasedynamicsuntil infectiondetectionor up to10 years at most.For each simulation we generated a net-workof4535dairyfarmsasdescribedaboveandassumed that aninfectedbutnotinfectiveindividualwasrandomlyintroduced inthenetwork.Tosimulatevariabletestsensitivity,thevaluesof skintestandpost-morteminspectionsensitivityweredrawnfrom therespectivebetadistributionseachtimeatestwasperformed. Then,the10thousandreplicateswereusedtoderiveanumberof statisticsondiseasedynamicsandontheperformanceofthe spe-cificsurveillancesystemunderassessment.Inparticular,foreach scenariowederived:(i)themeanand95thpercentileofthetime foranepidemictobedetected(whensurveillanceisinplace);(ii) themeanand95thpercentileofthenumberofinfectedfarms;(iii) thenumberofepidemicsthatinvolvedmorethanonefarm(i.e. multi-farmsepidemics);(iv)thenumberofundetectedepidemics after10years;and(v)thenumberofepidemicsthatwentnaturally extinctinlessthan10years.
3. Results
3.1. Farmsize,in-andout-degreedistributions,andparameters estimations
Thesizeof the4353ERcattlefarmswasquitevariable and rangedfromjustfewindividualstoalmosttwothousands:farm sizedistributionwashighlyskewedwithmeanequalto102and median63(Fig.1).
Theaveragenumberofoutgoing[ingoing]linksinthe837 sam-pledfarmswas1.67[1.91],themedian1[1]andthemaximum number23[29].Thein-degreeandout-degreedistributionswere not significantly different (Kolgomorov-Smirnov test, D=0.044, p=0.387)andwerebestfittedbyadiscretelog-normal distribu-tion(Fig.3andTable3).In-degreeandout-degreewereweakly correlatedwithfarmsize(Kendall’s=−0.028and0.20, respec-tively). The correlation between in- and out-degrees was also very weak(Kendall’s =−0.064). Therefore, for each of the 10 thousandrunswe generateda networkof 4353farmsby inde-pendently drawingfarms’ in-degree and out-degree fromtheir corresponding distributions, assuming no correlation between
1 2 5 10 20 12 5 1 0 2 0 5 0 100 200
log(sampled farms degree)
log(frequency)
Data in−degree Data out−degree Lognormal fit Power−law fit Poisson fit Yule fitFig.3.Pointsrepresentstheobservedin-(circles)andout-degree(triangles) dis-tributionsofthesampleddairyfarms.Linesrepresentthefittingfordifferent distributionstotheempiricaldata:discretelog-normal(solidline),Yule(dashed line),discretepower-law(dottedline)andPoisson(dashed-dottedline).
1 2 5 10 20 50 100 1 5 50 500 5 000
log(outbreaks size
)
log(counts)
Fig.4. Thedistribution(inlog–logscale)ofmaximumnumberofinfectedfarms (outbreaksize)ineachsimulatedoutbreak.
them.Theresultingnetworkswereweaklyconnected,withaverage densityof5.2×10−4(SD±8.7×10−6).Thevalueofparameters andmean[ij]estimatedfrommovementdatawas0.240y−1(S.E.
±0.044)and0.043y−1(S.E.±0.008),respectively,fordairyfarms and,19.981y−1(S.E.±1.499)and15.901y−1(S.E.±1.192)forITFs.
3.2. Thebaselinescenario:nosurveillance
In theabsenceof anysurveillancesystem,about halfofthe epidemicslastedfor10yearsormore(56.05%),whiletheothers (43.95%)naturallyfadedoutin amedian timeof14.92months (lower[upper]quartile5.95[32.89]months).Epidemicsizewas highlyskewed(Fig.4):themeannumberofinfectedfarmswas 2.91(S.E.±0.05),but62.66%oftheepidemicsinvolvedonlyasingle farm.Intheworst5%ofthecasesmorethan11farmsgotinfected, withamaximumof89.
Table3
TheresultoftheVuongtestonthedegreedatafordiscretelog-normaldistributionvs.discretepower-law,PoissonandYuledistributions.
Discretelog-normaldistribution loglike.ratio mean.LLR sd.LLR Vuong p-Value
vs.Powerlaw 158.903 0.151 0.647 7.57 3.6e−14
vs.Poisson 314.182 0.299 1.735 5.58 2.4e−08
vs.Yule 107.584 0.102 0.583 5.69 1.3e−08
Table4
Themainindicatorforthesurveillancesystemperformances.Resultsfor5differentsurveillancescenarios:nosurveillance,1:currentsurveillance,2a:RSremoved,2b:ECT, 2c:SIremoved.
Scenario Mean(and95p)timeto detection(months)
Mean(and95p)numberof infectedfarms %ofmulti-farm epidemics %ofepidemicover10 years %ofepidemic extinct Nosurveillance – 2.91(11) 37.34% 56.05% 43.95% Scenario1 27.04(59.20) 1.12(2) 10.15% 0.10% 14.55% Scenario2a 33.77(76.91) 1.20(2) 16.80% 0.70% 16.23% Scenario2b 27.52(60.09) 1.11(2) 9.61% 0.14% 14.80% Scenario2c 35.45(71.34) 1.15(2) 12.01% 0.18% 33.57%
3.3. Performanceofthecurrentsurveillancesystem
Fig.5showsthecumulateddistributionofthetimeofepidemic detectioncomputedasthefractionofepidemicsmonthlydetected bythecurrentsurveillancesystem.Themeandetectiontimewas 27.04months(S.E.±0.19). Epidemicswereidentified more fre-quentlybyroutinesurveillance,RS,andbypost-morteminspection attheslaughterhouse,SI.In10years,RSidentified41.66%ofthe epidemicsand SIthe38.59%.Only5.10%oftheepidemicswere firstdetectedbyECT,i.e.intra-dermaltestingofexchanged cat-tle;14.55%oftheepidemicsfadedoutunreportedwhileonly0.1% oftheepidemicslasted10yearsormore.Theaveragenumberof infectedfarmsduringtheepidemicswas1.12(S.E.±0.003);the worst5%oftheepidemicslastedmorethan59.20monthsaffecting from2to5farmsatmost.Multi-farmepidemicsoccurredin10.15% ofthereplicates.Theperformancesofthecurrentsurveillanceand thealternativescenarios describedhereafteraresummarizedin Table4.
3.4. Performanceofalternativesurveillancesystems
Fig.6showsthecumulativefractionofmonthlydetected epi-demicsregisteredforasurveillancesystemimplementingonlytwo
0 20 40 60 80 100 120 0.0 0 .2 0.4 0 .6 0.8 1 .0 Time (months) Cumulativ e fraction of epidemics Total detected RS ECT SI 1 − extinct
Fig.5.Thecumulatedfractionofsimulatedepidemicsmonthlydetectedbythe cur-rentsurveillancesystem.Differentcolourscorrespondtothefractionofepidemics detectedbyeachsurveillancestrategy:inlightgreybyslaughterhouseinspection (SI),indarkgreybyroutinesurveillance(RS)andinblackbyexchangedcattletesting (ECT).Thesolidlinerepresentsthefractionoftotalofdetectedepidemicswhilethe dashedlinerepresentsthefractionofepidemicsnotnaturallyfadedout,orextinct, atanygiventime.
ofthethreesurveillancemethodscurrentlyinplace.Ouranalysis showed that removing RS testing from the integrated surveil-lance system (Scenario 2a) ledto an almost 7-monthdelay in thedetectiontimedrivenbyasignificantreductioninthe num-berofepidemicsdetectedstartingfromthesecondyear(Fig.6). Comparedwiththecurrent integrated surveillancesystem,this scenariowascharacterizedbyasmallincreaseinthemean num-berofinfectedfarms,inthetotalnumberofunreportedepidemics lasting10yearsormore(from0.10%to0.70%),andinthenumber ofunreportedepidemicsnaturallyfadingoutbeforetheendofthe simulationtime(from14.55%to16.23%).Intheworst5%ofthe cases,theepidemicsremainedundetectedforabout6andahalf years(76.91months)andinvolved2–5farms.
Removing the ECT (Scenario 2b) did not cause significant changesinanyoftheperformanceindicators.Intheworst5%ofthe casesthetimetodetectionwasabout60monthsandthenumber ofinfectedfarmsbetween2and7.
TheremovalofSI(Scenario2c)ledtoamorethan8months increaseinthemeandetectiontimemostlydrivenbyasignificant reductioninthedetectionratesincethefirstmonths.Thenumber ofundetectedepidemicsstillongoingafter10yearsdidnot signifi-cantlychange,whilethefractionofmulti-farmepidemicsincreased from10.15%inScenario1to12.01%.Thelargestchangein sce-nario2cwasrepresentedbythenumberofunreportedepidemics thatnaturallyfadedout,increasedfrom14.55%intheScenario1
0 20 40 60 80 100 120 0.0 0 .2 0.4 0 .6 0. 1 8 .0 Time (months) Cumulativ e
fraction of detected epidemics
Scenario 1 Scenario 2a Scenario 2b Scenario 2c
Fig.6.Thecumulatedfractionofsimulatedepidemicsmonthlydetectedbythe surveillancesysteminScenario1(solidline);Scenario2a(onlyECTandSI compo-nents,dottedline);Scenario2b(onlyRSandSIcomponents,dashed-dottedline); andScenario2c(onlyRSandECTcomponents,dashedline).
tothe33.57% inScenario2c,thatexplainsthelowernumberof totallydetectedepidemics(Fig.6).Intheworst5%ofthecases,the epidemicsremainedundetectedfor71monthsandinvolved2–7 farms.
4. Discussion
Thegoalofthis studywastoassesstheeffectiveness ofthe currentbTBsurveillancesysteminEmilia-Romagna(ER)andthe specificroleplayedbyitsthreecomponents:on-farmroutine test-ing,exchangedanimaltesting,andslaughterhouseinspection.To assesstheperformanceofthebTBsurveillancesystemwe com-putedtwomainindicators:thetimeneededtodetectaprospective newlyintroducedbTBepidemicinthedairyfarmsofER;andthe magnitudeoftheepidemic,intermsofnumberoffarmsinfected beforethedetectionoftheepidemic.Thetimetoepidemic detec-tionisaveryimportantmeasureofthesurveillanceperformance (HadornandStärk,2008);firstly,becauseanearlydetectionofthe epidemiclimitsthenumberofbothanimalsandfarmsinfected; secondly,becauseaccordingtoItalianandEuropeanlegislationsat least6monthswithoutbTBcasesarerequiredforfarmstoregain theirdisease-freestatus.Indeed,thetraderestrictionsfollowing anepidemicrepresentoneoftheworstconsequencesforthe cat-tleindustry(Gordon,2008).Themagnitudeoftheepidemicallows ustoassessthedamagescausedbyabTBincursionquantifying thenumber ofinfected farmsand, consequently,thetotal eco-nomicburden (i.e.lossofearnings,costsforanimal cullingand re-stocking).
Ouranalysisshowedthat theintegratedsurveillancesystem currentlyinplaceinERcaneffectivelyreducethenumberoffarms involvedina potentialepidemicofbTBwithrespecttoa base-linescenariowithnosurveillance(seeTable4).Themodelpredicts thataboutthe15%oftheepidemicscannotbedetectedbythe sys-tem.However,theundetectedepidemicsusuallyfadedoutduring thefirstninemonthsaftertheintroductionoftheinfectionand generallyinvolvedoneorfewindividualsinasinglefarm.
Numericalsimulationsalsoshowedthatpost-morteminspection attheslaughterhouse(SI)andon-farmroutinetesting(RS)werethe mosteffectivesurveillancemethods.Thesemethodswere signifi-cantlymoreeffectiveindetectingbTBintheERnetworkofdairy farmsthantestingcattleuponmovingtoanewfarm(ECT). Consid-eringthatthecattleexchangesweretheonlyhypothesizedroute ofbetween-farmtransmissioninthemodel,thisresultseems,at firstsight,unexpected.However,itcanbeexplainedbythelow bTBtransmissionrateandsomespecificproprietiesoftheERcattle system(suchasthelownetworkdensityandthelownumberof cattletraded).
Slaughterhouseinspectionturnedouttobeparticularly effec-tiveasanearlydetectiontooloftheepidemics.Thisseemedtobe duetothefairlyregularflowofcattlesentouttoslaughterhousesat theendoftheirproductivecycle,onaverageat5yearsofage. There-fore,whenanoutbreakoccurredinafarm,SIwaslikelytheearliest diagnosticinterventionabletodetectitcomparedtoECTandRS. However,thelargenumberofepidemicsthatnaturallyfadedout whenSIwasabsent(Scenario2c)pointsoutthatmostofthe out-breaksthatremainedundetectedattheendofthefirstyearwere likelytogetextinctafterwardswithoutdevelopingintosignificant epidemics.
On-farmroutine surveillancewas themosteffective in pre-ventingmajor,ormulti-farm,outbreaks.Thiswasprobablydue tothelow valueofwithin-farmbTBbasicreproductionnumber (R0).Indeed,alowwithin-farmR0 isfollowedbyalow number
ofinfectedindividualsinsidefarms,thatmeansalownumberof infectedindividualsexchangedorsenttoslaughterhouses.Inthese epidemiologicalconditions,RSwasthemosteffectivemethodto
detectoutbreaksbeforebTBspreadstootherfarms.Moreover,this methodwasalsothemosteffectiveindetectingthesmallportion ofminorepidemicsthatcanpersistunnoticedformanyyears.
TheperformancesofSIandRS,bothwithregardtotimetobTB detectionandtofractionofmulti-farmepidemics,were compara-bleandwhichoneperformedbetterdependedonthevalueofthe basicreproductionnumber(R0),seeSupplementaryMaterialsS3.3,
andonthevaluesofskin-testandSIsensitivity,seeSupplementary materialsS4.2.
AstheR0 valueincreased,thenumberofdetectedepidemics
bytheSItendedtoincreasecomparedtothenumberofepidemics detectedbyRS(seeSupplementarymaterialsS3.3).Thisresultcan beexplainedconsideringthatalargerwithin-farmdisease preva-lence,asaconsequenceofalargerwithin-farmR0,impliesahigher
probabilityofsendinginfectedindividualstotheslaughterhouse intheearlystagesofepidemics,speedingupbTBdetection,while thetimingofRSisnotaffectedbyfarmprevalence.
Thevaluesofskin-testandcarcassinspectionsensitivityplayed animportantroleaswell.Thiswasparticularlysignificantinthe caseofslaughterhouseinspection,sinceaverylargerangeofvalues ofinspectionsensitivityhasbeenestimatedindifferentcontexts, from0.1to0.65(Assegedetal.,2004;Fischeretal.,2005;Smith etal.,2013;vanAsseldonketal.,2005;Welbyetal.,2012).
TheRSturnaroundperiod,i.e.thetimeframeinwhichallfarms aretestedonceforbTB,wassetto3years,ascurrentlyisinER region.Theregionaladministrationdecidedtomovethisperiod from2to3yearsin2010.Ourresultssuggestedthatthischange hada negligible impactontheoverall surveillancesystem per-formanceand, moreover,thatmoving itto4 yearswould have marginalimpactaswell(SupplementarymaterialsS5).
ThelimitedperformanceimprovementprovidedbyECTraised questions about its usefulness and whether it could be safely droppedoutfromERintegratedsurveillancesystemwithout sig-nificantlyweakeningit.Ofcourse,thiskindofdecisionshouldalso bebasedoncost-benefitanalysis.Ontheotherhand,theroleof ECTonanimalsenteringabTB-freeregionisnotnegligible, espe-ciallyforcattleoriginatingfromregionswherebTBisendemicand ifpre-movementtesting-incontrasttopost-movementtesting– isadoptedtopreventtheriskofdiseaseintroduction(Cleggetal., 2008;Schilleretal.,2011).
Inagreementwithourresults,Shittuetal.(2013)showedthat SIplaysanimportantroleinbTBdetectionintheUK,bothinlow incidenceandendemicareas,butishighlyinfluencedby inspec-tionsensitivity.AnotherstudyconductedinBelgiumbyWelbyetal. (2012)showedthatSIisoneofthemosteffectivesurveillance strat-egyforbTB.However,furtheranalysesperformedwithourmodel showedthattheuseoftheSIcomponentonlydetermineda signif-icantreductioninthesurveillancesystemperformancecompared toasurveillancebasedonSIcoupledwithroutineskin-testing(see SupplementarymaterialsS6).Thisresultisinagreementwith find-ingsbyFischeretal.(2005)andvanAsseldonketal.(2005),who showedthatSIsurveillancealoneisineffectiveindetectingbTBin TheNetherlands.ThiswasconfirmedbySchöningetal.(2013)on UScattleherds,whoshowedhowSIneededtogoalongwiththe skin-testcomponentinordertoprovideaneffectivesurveillance system.
Inthisstudy,cattlemovementdatawereavailableforasubsetof thenetworkofdairyfarmsinER.Thesedatawereusedtoinferthe fundamentaltopologicalpropertiesofthewholesysteminterms ofconnectivity,soastogeneratesyntheticnetworksofthesame sizeoftheERone.Whileweareconfidentthatourapproach pro-videsasolidpreliminaryunderstandingofbTBdynamicsonthe networkofdairyfarmsinER,wedidnothavethepossibilityto derivemoresophisticatedmetrics,suchastheclustering coeffi-cientandtheDiseaseFlowCentrality,thatwouldhelptohavemore detailedpredictiononhowdiseasesspreadthroughthenetwork,
asshownbyBajardietal.(2011)andNataleetal.(2011).We simu-lateddiseasedynamicsonastaticnetwork,ignoringseasonalityin cattlemovementandthedynamicalnatureofcattletrade. Dynam-icalnetworkscaneffectivelyrepresentthetime-varyingstructure ofthenetwork–whichiscrucialtosimulatethedynamicsofacute andhighlycontagiousdiseases,suchasFood-and-MouthDisease (Bajardietal.,2011,2012;VernonandKeeling,2009).However, sincebTBischaracterizedbyslowinfectiondynamicsanditisable toremainunnoticedforalongperiodoftime,webelievethatitis notessentialtotrackthesmallstructuralchangesoccurringona shorttimescaleforsuitablydescribebTBdynamics.
Otherstudieswerepreviouslyconductedoncattlemovements inItalybyNataleetal.(2009,2011).Inparticular,theyfoundthat thedegreedistributionofcattlemovementwasbestapproximated byafat-tailedpower-lawprobabilitydistribution.Ourdegree dis-tributionsbestfittingwasobtainedthroughadiscretelog-normal distribution,characterizedbylowerright-tailsthanthepower-law (Fig.3).Thiscouldbeaconsequenceofisolatingthedairysystem, i.e.bynotconsideringthemovementstowardsfatteningfarmsand slaughterhouses.
Thisstudywasspecificallyperformedonthedairycattle sys-tem,ignoringthebeefsector.ThemainreasonwasthatthebTB surveillancesysteminERisinplaceonthedairysectoronly. More-over,furtherreasonssupportedthischoice:(i)thehighereconomic valueofthedairysectorintheRegion;(ii)thefactthatgenerally beeffarmsdonotexchangeanimalswithotherfarms,theyonly receivemalecalveswithin100daysofagefromdairyfarmsand sendanimalstoslaughterhousesattheendofthefattening pro-cess;(iii)thetwosystemsarealmostfullyseparated,sothereis nopossibilityofspill-overfromthebeeftothedairysystem;and (iv)dairycattlelivesubstantiallylongerthanbeefcattle,i.e.upto 5yearsvs.2respectively,thusepidemicshavemorepossibilityto developinthedairysystembecauseofbTBlongincubationperiod. Our analysesshowedthat thefarms in-degree and the out-degreewerenotcoupled.Inotherwords,farmshavingahigher in-degree,i.e.thatimportrelativelymorecattlefromotherfarms intheRegion,donotnecessarilyhaveahighout-degree,andvice versa.AsimilarresultisoutlinedinVolkovaetal.(2010)onScottish farms,suggestingthatthiscouldbearecurrentpatternforcattle herdsindifferentmanagementcontexts.
IntheUKand inother countrieswhere bTBisendemic,the highcostofthesurveillanceanderadicationprogramme,namely
£
74–99million ayear forUK (Smithand Clifton-Hadley, 2008; TorgersonandTorgerson,2008)hasbeenputunderscrutiny.Our resultsindicatethataproposalforsimplificationorstrengthening ofanintegratedsurveillancesystemshouldbealwaysbasedona rigorous,quantitativeunderstandingoftheroleplayedbyeachof itscomponents.Infactunderstandingtheactualeffectivenessof aspecificcontrolstrategy,isnotalwaysstraightforward,andcan beaffectedbyfactorslikenetworktopology,thetradingsystem, rearingmethods,testfrequencyandtestsensitivity.Moreover,the increasingdensityofungulatesandotherpotentialwildlife reser-voirspeciesoccurringin severalEuropeancountriesposesnew challengesfortheexistingsurveillancesystemseveninareasthat arecurrentlybTB-free(Schöningetal.,2013).A priori information on the network structure of cattle exchangesduringnon-emergencyperiodscanhelporienting con-trolstrategiestopreventepidemicsinareascharacterizedbyahigh numberofindustrialfarms(Bajardietal.,2012;Gilbertetal.,2005). Ourframework,basedonamathematicalmodellingapproach, pro-videddecision-makerswithapowerfulcost-effectivetooltoassess theeffectivenessofthecurrentbTBsurveillancesysteminER,by highlightingstrengthsandweaknessesitsdifferentcomponents.
Moreover, the methodological approach proposed here will allowthehealth authoritiestoassesstheeffectivenessoffuture alternative strategies for bTB detection that are not currently
implementedinER,suchasantigendetectingtests(ELISA)onmilk andbloodsample(vanAsseldonketal.,2005).Furthermore,by tak-ingintoconsiderationthecostofdifferenttestingstrategies,itwill bepossibletouseourmodellingapproachtoruncost-effectiveness analysesofdifferentoptionsofsurveillance(Andersonetal.,2013; Cameron,2012).
Acknowledgements
We would like to thank Marco Tamba for the constructive discussionsaboutbovinetuberculosissurveillanceandtheherd systemofEmilia-Romagna.
Thisstudywasfunded byIZSLERwithintheresearchproject AUTIFIN-TBC-EMILIA-2011 ‘Development of an Epidemiological ModelforBovineTuberculosisinEmilia-Romagna’.
AppendixA. Supplementarydata
Supplementarymaterialrelatedtothisarticlecanbefound,in theonlineversion,atdoi:10.1016/j.epidem.2015.02.007.
References
Abernethy,D.A.,Upton,P.,Higgins,I.M.,McGrath,G.,Goodchild,A.V.,Rolfe,S.J., Broughan, J.M., Downs, S.H.,Clifton-Hadley, R.,Menzies, F.D., de la Rua-Domenech,R.,Blissitt,M.J.,Duignan,A.,More,S.J.,2013.Bovinetuberculosis trendsintheUKandtheRepublicofIreland,1995–2010.Vet.Rec.172,312.
Agusto,F.B.,Lenhart,S.,Gumel,A.B.,Odoi,A.,2011.Mathematicalanalysisofamodel forthetransmissiondynamicsofbovinetuberculosis.Math.MethodsAppl.Sci. 34,1873–1887.
Anderson,D.P.,Ramsey,D.S.L.,Nugent,G.,Bosson,M.,Livingstone,P.,Martin,P.A.J., Sergeant,E.,Gormley,A.M.,Warburton,B.,2013.Anovelapproachtoassessthe probabilityofdiseaseeradicationfromawild-animalreservoirhost.Epidemiol. Infect.141,1509–1521.
Anonymous,2008.Bovinetuberculosis.In:ManualofDiagnosticTestsand Vac-cinesforTerrestrialAnimals.WorldOrganizationforAnimalHealth(OIE),Paris, France,pp.683–697.
Asseged,B.,Woldesenbet,Z.,Yimer,E.,Lemma,E.,2004.Evaluationofabattoir inspectionforthediagnosisofMycobacteriumbovisinfectionincattleatAddis Ababaabattoir.Trop.Anim.HealthProd.36,537–546.
Bajardi,P.,Barrat,A.,Natale,F.,Savini,L.,Colizza,V.,2011.Dynamicalpatternsof cattletrademovements.PLoSONE6,e19869.
Bajardi,P.,Barrat,A.,Savini,L.,Colizza,V.,2012.Optimizingsurveillanceforlivestock diseasespreadingthroughanimalmovements.J.R.Soc.Interface9,2814–2825.
Barlow,N.D.,Kean,J.M.,Hickling,G.,Livingstone,P.G.,Robson,A.B.,1997.A simu-lationmodelforthespreadofbovinetuberculosiswithinNewZealandcattle herds.Prev.Vet.Med.32,57–75.
Brooks-Pollock,E.,Keeling,M.J.,2009.Herdsizeandbovinetuberculosispersistence incattlefarmsinGreatBritain.Prev.Vet.Med.92,360–365.
Brooks-Pollock,E.,Roberts,G.O.,Keeling,M.J.,2014.Adynamicmodelofbovine tuberculosisspreadandcontrolinGreatBritain.Nature511,228–231.
Cameron,A.R.,2012. Theconsequencesofrisk-basedsurveillance:developing output-basedstandardsforsurveillancetodemonstratefreedomfromdisease. Prev.Vet.Med.105,280–286.
Cheeseman,C.L.,Wilesmith,J.W.,Stuart,F.A.,1989.Tuberculosis:thediseaseandits epidemiologyinthebadger,areview.Epidemiol.Infect.103,113–125.
Clauset,A.,Shalizi,C.,Newman,M.,2009.Power-lawdistributionsinempiricaldata. SIAMRev.51,661–703.
Clegg,T.A.,More,S.J.,Higgins,I.M.,Good,M.,Blake,M.,Williams,D.H.,2008.Potential infection-controlbenefitforIrelandfrompre-movementtestingofcattlefor tuberculosis.Prev.Vet.Med.84,94–111.
Conlan,A.J.K.,McKinley,T.J.,Karolemeas,K.,Pollock,E.B.,Goodchild,A.V.,Mitchell, A.P.,Birch,C.P.D.,Clifton-Hadley,R.S.,Wood,J.L.N.,2012.Estimatingthehidden burdenofbovinetuberculosisinGreatBritain.PLoSComput.Biol.8,e1002730.
Cousins,D.V.,2001.Mycobacteriumbovisinfectionandcontrolindomesticlivestock. Rev.Sci.Tech.20,71–85.
delaRua-Domenech,R.,Goodchild,A.T.,Vordermeier,H.M.,Hewinson,R.G., Chris-tiansen,K.H.,Clifton-Hadley,R.S.,2006.Antemortemdiagnosisoftuberculosis incattle:areviewofthetuberculintests,gamma-interferonassayandother ancillarydiagnostictechniques.Res.Vet.Sci.81,190–210.
Diekmann,O.,Heesterbeek,J.,Metz,J.,1990.Onthedefinitionandthe compu-tationofthebasicreproductionratioR0inmodelsforinfectiousdiseasesin
heterogeneouspopulations.J.Math.Biol.28,365–382.
Fischer,E.A.J.,vanRoermund,H.J.W.,Hemerik,L.,vanAsseldonk,M.A.P.M.,de Jong,M.C.M.,2005.Evaluationofsurveillancestrategiesforbovinetuberculosis (Mycobacteriumbovis)usinganindividualbasedepidemiologicalmodel.Prev. Vet.Med.67,283–301.
Fitzgerald,S.D.,Kaneene,J.B.,2013.Wildlifereservoirsofbovinetuberculosis world-wide:hosts,pathology,surveillance,andcontrol.Vet.Pathol.50,488–499.
Gilbert,M.,Mitchell, A.,Bourn, D.,Mawdsley,J., Clifton-Hadley,R.,Wint, W., 2005.CattlemovementsandbovinetuberculosisinGreatBritain.Nature435, 491–496.
Gillespie,D.,1977.Exactstochasticsimulationofcoupledchemicalreactions.J.Phys. Chem.93555,2340–2361.
Gordon,S.,2008.BovineTB:stoppingdiseasecontrolwouldblockallliveexports. Nature456,700.
Griffin,J.M.,Williams,D.H.,Kelly,G.E.,Clegg,T.A.,O’Boyle,I.,Collins,J.D.,More,S.J., 2005.Theimpactofbadgerremovalonthecontroloftuberculosisincattleherds inIreland.Prev.Vet.Med.67,237–266.
Hadorn,D.,Stärk,K.,2008.Evaluationandoptimizationofsurveillancesystemsfor rareandemerginginfectiousdiseases.Vet.Res.39,57–68.
Huang,Z.Y.X.,deBoer,W.F.,vanLangevelde,F.,Xu,C.,Jebara,K.B.,Berlingieri,F., Prins,H.H.T.,2013.Dilutioneffectinbovinetuberculosis:riskfactorsforregional diseaseoccurrenceinAfrica.Proc.R.Soc.B280,20130624.
Humblet,M.-F.,Moyen,J.-L.,Bardoux,P.,Boschiroli,M.L.,Saegerman,C.,2011.The importanceofawarenessforveterinariansinvolvedincattletuberculosisskin testing.Transbound.Emerg.Dis.58,531–536.
Italian National Statistics Institute (ISTAT), 2010. 6◦ Censimento Generale
dell’Agricoltura.http://censimentoagricoltura.istat.it/
Johnston,W.T.,Vial,F.,Gettinby,G.,Bourne,F.J.,Clifton-Hadley,R.S.,Cox,D.R.,Crea, P.,Donnelly,C.A.,McInerney,J.P.,Mitchell,A.P.,Morrison,W.I.,Woodroffe,R., 2011.Herd-levelriskfactorsofbovinetuberculosisinEnglandandWalesafter the2001foot-and-mouthdiseaseepidemic.Int.J.Infect.Dis.15,e833–e840.
IndependentScientificGroup,2007.BovineTuberculosis:TheScientificEvidence. DEFRAPublications,London.
Marangon,S.,Martini,M.,DallaPozza,M.,Neto,F.,1998.Acase-controlstudyon bovinetuberculosisintheVenetoRegion(Italy).Prev.Vet.Med.34,87–95.
Morris,R.S.,Pfeiffer,D.U.,Jackson,R.,1994.TheepidemiologyofMycobacteriumbovis infections.Vet.Microbiol.40,153–177.
Natale,F.,Giovannini,A.,Savini,L.,Palma,D.,Possenti,L.,Fiore,G.,Calistri,P.,2009.
NetworkanalysisofItaliancattletradepatternsandevaluationofrisksfor potentialdiseasespread.Prev.Vet.Med.92,341–350.
Natale,F.,Savini,L.,Giovannini,A.,Calistri,P.,Candeloro,L.,Fiore,G.,2011. Evalua-tionofriskandvulnerabilityusingaDiseaseFlowCentralitymeasureindynamic cattletradenetworks.Prev.Vet.Med.98,111–118.
O’Hare,A.,Orton,R.,Bessell,P.R.,Kao,R.R.,2014.Estimatingepidemiological param-etersforbovinetuberculosisinBritishcattleusingaBayesianpartial-likelihood approach.Proc.R.Soc.B281,1783.
Phillips,C.J.C.,Foster,C.R.W.,Morris,P.A.,Teverson,R.,2003.Thetransmissionof Mycobacteriumbovisinfectiontocattle.Res.Vet.Sci.74,1–15.
Reilly,L.A.,Courtenay,O.,2007.Husbandrypractices,badgersettdensityandhabitat compositionasriskfactorsfortransientandpersistentbovinetuberculosison UKcattlefarms.Prev.Vet.Med.80,129–142.
Reynolds,D.,2006.AreviewoftuberculosisscienceandpolicyinGreatBritain.Vet. Microbiol.112,119–126.
Rohani,P.,Keeling,M.J.,Grenfell,B.T.,2002.Theinterplaybetweendeterminismand stochasticityinchildhooddiseases.Am.Nat.159,469–481.
Schiller,I.,Waters,W.R.,Vordermeier,H.M.,Jemmi,T.,Welsh,M.,Keck,N.,Whelan, A.,Gormley,E.,Boschiroli,M.L.,Moyen,J.L.,Vela,C.,Cagiola,M.,Buddle,B.M., Palmer,M.,Thacker,T.,Oesch,B.,2011.BovinetuberculosisinEuropefromthe perspectiveofanofficiallytuberculosisfreecountry:trade,surveillanceand diagnostics.Vet.Microbiol.151,153–159.
Schöning,J.M.,Cerny,N.,Prohaska,S.,Wittenbrink,M.M.,Smith,N.H.,Bloemberg,G., Pewsner,M.,Schiller,I.,Origgi,F.C.,Ryser-Degiorgis,M.-P.,2013.Surveillance ofbovinetuberculosisandriskestimationofafuturereservoirformationin wildlifeinSwitzerlandandLiechtenstein.PLoSONE8,e54253.
Shittu,A.,Clifton-Hadley,R.S.,Ely,E.R.,Upton,P.U.,Downs,S.H.,2013.Factors asso-ciatedwithbovinetuberculosisconfirmationratesinsuspectlesionsfoundin cattleatroutineslaughterinGreatBritain,2003–2008.Prev.Vet.Med.110, 395–404.
Smith,N.H.,Clifton-Hadley,R.,2008.BovineTB:don’tgetridofthecatbecausethe micehavegone.Nature456,700.
Smith,R.L.,Schukken,Y.H.,Lu,Z.,Mitchell,R.M.,Grohn,Y.T.,2013.Developmentof amodeltosimulateinfectiondynamicsofMycobacteriumbovisincattleherds intheUnitedStates.J.Am.Vet.Med.Assoc.243,411–423.
Skuce,R.A.,Allen,A.R.,McDowell,S.W.J.,2012.Herd-levelriskfactorsforbovine tuberculosis:aliteraturereview.Vet.Med.Int.2012,621210.
Torgerson,P.,Torgerson,D.,2008.Doesrisktohumansjustifyhighcostoffighting bovineTB?Nature455,1029.
vanAsseldonk,M.A.P.M.,vanRoermund,H.J.W.,Fischer,E.A.J.,deJong,M.C.M., Huirne,R.B.M.,2005. Stochasticefficiency analysisofbovine tuberculosis-surveillanceprogramsintheNetherlands.Prev.Vet.Med.69,39–52.
Vernon,M.,Keeling,M.J.,2009.RepresentingtheUK’scattleherdasstaticand dynamicnetworks.Proc.R.Soc.B:Biol.Sci.276,469–476.
Volkova,V.V.,Howey,R.,Savill,N.J.,Woolhouse,M.E.J.,2010.Potentialfor transmis-sionofinfectionsinnetworksofcattlefarms.Epidemics2,116–122.
Vuong,Q.,1989.Likelihoodratiotestsformodelselectionandnon-nested hypothe-ses.Econom.J.Econom.Soc.57,307–333.
Welby,S.,Govaerts,M.,Vanholme,L.,Hooyberghs,J.,Mennens,K.,Maes,L.,VanDer Stede,Y.,2012.BovinetuberculosissurveillancealternativesinBelgium.Prev. Vet.Med.106,152–161.