<%BANNER%>






Evaluating the Combined Effectiveness of Influenza Control Strategies and Human Preventive Behavior
www.plosone.org ( Publisher's URL )
CITATION PDF VIEWER
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/IR00000630/00001
 Material Information
Title: Evaluating the Combined Effectiveness of Influenza Control Strategies and Human Preventive Behavior
Series Title: Mao L (2011) Evaluating the Combined Effectiveness of Influenza Control Strategies and Human Preventive Behavior. PLoS ONE 6(10): e24706. doi:10.1371/journal.pone.0024706
Physical Description: Journal Article
Creator: Mao, Liang
Publisher: PLoS ONE
Publication Date: October 17, 2011
 Notes
Abstract: Control strategies enforced by health agencies are a major type of practice to contain influenza outbreaks. Another type of practice is the voluntary preventive behavior of individuals, such as receiving vaccination, taking antiviral drugs, and wearing face masks. These two types of practices take effects concurrently in influenza containment, but little attention has been paid to their combined effectiveness. This article estimates this combined effectiveness using established simulation models in the urbanized area of Buffalo, NY, USA. Three control strategies are investigated, including: Targeted Antiviral Prophylaxis (TAP), workplace/school closure, community travel restriction, as well as the combination of the three. All control strategies are simulated with and without regard to individual preventive behavior, and the resulting effectiveness are compared. The simulation outcomes suggest that weaker control strategies could suffice to contain influenza epidemics, because individuals voluntarily adopt preventive behavior, rendering these weaker strategies more effective than would otherwise have been expected. The preventive behavior of individuals could save medical resources for control strategies and avoid unnecessary socio-economic interruptions. This research adds a human behavioral dimension into the simulation of control strategies and offers new insights into disease containment. Health policy makers are recommended to review current control strategies and comprehend preventive behavior patterns of local populations before making decisions on influenza containment.
Acquisition: Collected for University of Florida's Institutional Repository by the UFIR Self-Submittal tool. Submitted by Liang Mao.
Publication Status: Published
 Record Information
Source Institution: University of Florida Institutional Repository
Holding Location: University of Florida
Rights Management: All rights reserved by the submitter.
System ID: IR00000630:00001

Downloads
Full Text

PAGE 1

EvaluatingtheCombinedEffectivenessofInfluenza ControlStrategiesandHumanPreventiveBehaviorLiangMao *DepartmentofGeography,UniversityofFlorida,Gainesville,Florida,UnitedStatesofAmericaAbstractControlstrategiesenforcedbyhealthagenciesareamajortypeofpracticetocontaininfluenzaoutbreaks.Anothertypeof practiceisthevoluntarypreventivebehaviorofindividuals,suchasreceivingvaccination,takingantiviraldrugs,and wearingfacemasks.Thesetwotypesofpracticestakeeffectsconcurrentlyininfluenzacontainment,butlittleattentionhas beenpaidtotheircombinedeffectiveness.Thisarticleestimatesthiscombinedeffectivenessusingestablishedsimulation modelsintheurbanizedareaofBuffalo,NY,USA.Threecontrolstrategiesareinvestigated,including:TargetedAntiviral Prophylaxis(TAP),workplace/schoolclosure,communitytravelrestriction,aswellasthecombinationofthethree.All controlstrategiesaresimulatedwithandwithoutregardtoindividualpreventivebehavior,andtheresultingeffectiveness arecompared.Thesimulationoutcomessuggestthatweakercontrolstrategiescouldsufficetocontaininfluenzaepidemics, becauseindividualsvoluntarilyadoptpreventivebehavior,renderingtheseweakerstrategiesmoreeffectivethanwould otherwisehavebeenexpected.Thepreventivebehaviorofindividualscouldsavemedicalresourcesforcontrolstrategies andavoidunnecessarysocio-economicinterruptions.Thisresearchaddsahumanbehavioraldimensionintothesimulation ofcontrolstrategiesandoffersnewinsightsintodiseasecontainment.Healthpolicymakersarerecommendedtoreview currentcontrolstrategiesandcomprehendpreventivebehaviorpatternsoflocalpopulationsbeforemakingdecisionson influenzacontainment.Citation: MaoL(2011)EvaluatingtheCombinedEffectivenessofInfluenzaControlStrategiesandHumanPreventiveBehavior.PLoSONE6(10):e24706. doi:10.1371/journal.pone.0024706 Editor: PetterHolme,Umea University,Sweden Received April13,2011; Accepted August16,2011; Published October17,2011 Copyright: 2011LiangMao.Thisisanopen-accessarticledistributedunderthetermsoftheCreativeCommonsAttributionLicense,whichpermits unrestricteduse,distribution,andreproductioninanymedium,providedtheoriginalauthorandsourcearecredited. Funding: ThisresearchwassupportedinpartbytheMarkDiamondResearchFundandtheHughCalkinsAppliedGISAward,bothfromtheUniversityatBuffalo, StateUniversityofNewYork.PublicationofthisarticlewasfundedbytheUniversityofFloridaOpen-Accesspublishingfund.Thefundershadnorole instudy design,datacollectionandanalysis,decisiontopublish,orpreparationofthemanuscript. CompetingInterests: Theauthorhasdeclaredthatnocompetinginterestsexist. *E-mail:liangmao@ufl.eduIntroductionDuringthepastdecade,influenzahasobtainedunprecedented attentionduetowidespreadoccurrenceofnovelviruses,suchas thebirdfluin2003andtheswinefluin2009[1,2].Recent estimatesbytheCenterofDiseaseControlandPrevention(CDC) indicatedthatthe2009swinefluisresponsiblefor274,000 hospitalizationsand12,470deathsintheUnitedStates[3].These staggeringhealthburdenscallforeffectivemeasurestocontroland preventfutureoutbreaks.Thecontrolofinfluenzaprimarily involvesapplyinghealthresourcestoaffectedpeople,knownas controlstrategies,forexample,medicaltreatmentforinfected individuals,closureofaffectedworkplaces/schools,andtravel restrictiontoaffectedcommunities[4].Thepreventionof influenzaemphasizeshealthypeopleanddependsontheir voluntarybehavioragainstthedisease,referredtoasthe preventivebehavior.AsrecommendedbyCDC,thepreventive behavioragainstinfluenzaincludereceivingvaccination,wearing facemasks,washinghandsfrequently,takingantiviraldrugs,and others[5]. Whiledevisingvariouscontrolstrategiesandevaluatingtheir effectiveness,fewstudieshaveincorporatedthepreventive behaviorofindividuals[6,7].Inmostcases,individualsareoften assumedtopassivelycomplywithcontrolstrategies,buttheir activepreventionagainstthediseasehasbeenoverlooked.In reality,thepreventivebehaviorofindividualsalsoreduces infectionsandtakeseffectconcurrentlywithtypicalcontrol strategies.Forinstance,individualsmayvoluntarilyprotect themselvesfrominfection,oncetheyrealizesomecontrol strategiesbeingappliedtotheirfamilymembers,colleagues,or communities[8,9,10].Byfar,thecombinedeffectivenessofcontrol strategiesandindividuals’preventivebehaviorremainsunclear, andlittleattentionhasbeenpaidtothisissue.Lackofsuch knowledgemaybiasestimationofhealthresourcesneededto suppressanoutbreak,andmisleadtherealpracticeofinfluenza containment. Thepurposeofthisarticleistoevaluatethecombined effectivenessofcontrolstrategiesandindividualpreventive behavior.Agent-basedstochasticsimulationsareusedtoinvestigatethreecontrolstrategies,includingtheTargetedAntiviral Prophylaxis(TAP),workplace/schoolclosure,andtravelrestriction,aswellascombinationsofallthree.Theurbanizedareaof Buffalo,NewYork,USA,istakenasastudyarea.Thecontrol effectivenesswithandwithoutconsideringindividualpreventive behavioriscomparedtoindicateifthereexistsasignificant difference.Cost-effectivestrategiesaresuggestedbasedonthe comparisonanalysis.Theremainderofthisarticleisorganizedas follows.Themethodsectionthatfollowsreviewstwoestablished influenzamodelsforsimulationanddescribesthedesignofcontrol strategiesbeingsimulated.Theresultsectionpresentsand comparesthesimulationresults.Thediscussionsectionconcludes thisarticlewithimplications. PLoSONE|www.plosone.org1October2011|Volume6|Issue10|e24706

PAGE 2

MaterialsandMethodsEpidemicmodels,includingmathematicalandcomputer models,havebeenextensivelyusedtoinvestigatediseasecontrol strategies,becauseoftheireaseandflexibilitytodealwithdifferent scenarios.Theclassicmathematicalmodel,theSIRmodel,andits variantsemploydifferentialequationstodescribecontinuous variationsbetweenthreesubpopulations,i.e.,thesusceptible, infectiousandrecovered[11,12].Variouscontrolstrategiesare oftenexpressedasdifferentinitialconditions(e.g.,thesizeof susceptiblepopulation)orparametersettings(e.g.,theinfection rate)ofdifferentialequations.Thecomputer-basedsimulation modelshaverecentlygainedtheirimpetusinepidemiology [13,14,15,16].Thesemodelsstudypopulation-levelhealthoutcomesthroughthesimulationofindividualsandtheirmicrointeractions.Controlstrategiescanberepresentedbyaltering individuals’healthstatusandtheirbehavior,suchasendowing themimmunityagainstinfectionandprohibitingtheirout-ofhomeactivities.Alloftheseepidemicmodelsprovidesolid platformstoevaluateandcomparealternativestrategies,thus informinghealthpolicymaking[17,18].Epidemicmodelswithandwithoutindividualpreventive behaviorEpidemicmodelswithoutconsideringindividualpreventive behaviorarewidelyseenintheliteratureandhereinafterreferred toas‘influenza-only’models,becausetheyprimarilyfocuson influenzatransmission.Inthisresearch,aninfluenza-onlymodel isimplementedinthestudyarea,whichincludesatotalnumber of985,001individuals.Theseindividualslivein967censusblock groupsand400,870householdsaccordingtoUScensus2000 [19],andcarryoutdailyactivitiesin36,839businesslocations [20].Themodelinvolvesanagent-basedstochasticsimulation, discretetimesteps,andspatiallyexplicitrepresentationof individuals.Eachindividualisamodelingunitwithasetof characteristics(e.g.,age,occupation,infectionstatus,locationand timeofdailyactivities)andbehaviors(e.g.,travelingbetween locationsforactivitiesandhavingcontactwithotherindividuals) [21,22].Individualsandhouseholdsaresimulatedunderthe constraintsofcensusdatasothatthemodeledpopulation matchestheageandhouseholdstructureoftherealstudyarea. Individualsarealsoassignedtobusinesslocationstorepresent theirdailyactivities,suchasworking,shopping,eatingout,etc. (FigureS1).Thecontactsbetweenindividualstakeplacewhen individualsmeetatthesametimeandlocation,suchashomes, workplaces,shops,andrestaurants.Becauseindividualstravel overtimeandlocation,theirmobilityweavesaspatio-temporally varyingcontactnetwork(SeeTextS1Section1.1).Throughsuch anetwork,influenzavirusesdiffusefromoneindividualto another.Eachindividualisallowedtotakeoneoffourinfection statusduringatimeperiod,i.e.susceptible,latent,infectious,and recovered.Theprogressofinfectionstatusfollowsthenatural historyofinfluenza,includingthelatent,incubation,and infectiousperiods(TableS1).Duringtheinfectiousperiod, individualsmaymanifestsymptomsandbecomesymptomatic. Toinitiatethediseasetransmission,fiveinfectiousindividualsare randomlyseededintothestudyareaatthefirstdayofsimulation, whichthenlastsfor150days.Ineachday,themodeltraces susceptiblecontactsofinfectiousindividuals,andstochastically identifiesthenextgenerationofinfectionsusingtheMonte-Carlo method(SeeTextS1Section1.2). Inordertofurtherconsiderindividualpreventivebehavior,this researchemploysanagent-based‘dual-diffusion’stochasticmodel thatsimulatestheconcurrentdiffusionofbothinfluenzaand individualpreventivebehavior[23].Thepreventivebehavioris consideredasapracticeorinformationthatalsodiffusesover contactnetworksthroughinter-personalinfluence.Thesetwo diffusionprocessesinteractwithoneanother,i.e.,thediffusionof influenzamotivatesthepropagationofpreventivebehavior,which inturnlimitstheinfluenzadiffusion[24,25,26].Inthemodel,the diffusionofinfluenzaissimulatedsimilarlytotheinfluenza-only modelaforementioned.Thediffusionofindividualpreventive behaviorispropelledbytwotypesofinter-personalinfluence throughthecontactnetwork:oneistheperceivedinfectionrisk andtheotheristheperceivedsocialstandard.Theformeris representedastheproportionofinfluenzacasesamongan individual’scontacts,whilethelatterisexpressedastheproportion ofbehavioraladoptersamongthecontacts[23].Individualsare simulatedtoevaluatethesetwoproportionseverydaythroughthe contactnetwork.Onceeitherproportionexceedsacorresponding threshold,anindividualwillbeconvincedtoadoptandpractice preventivebehavior[27,28].Theestimationofindividualized thresholdstowardadoptionisbasedonahealthbehavioralsurvey approvedbytheSocialandBehavioralSciencesInstitutional ReviewBoard,UniversityatBuffalo,StateUniversityofNew York.Thewaiverofinformedconsentwasobtainedfromthe universityreviewboardforthisresearch(SeeTextS1Section2 andFiguresS2–S3).Comparedtotheinfluenza-onlymodel, individualsinthedual-diffusionmodelhaveadditionalcharacteristics,suchastheiradoptionstatusofpreventivebehaviorand thresholdstowardadoption.Individualsalsohavemorebehaviors, forexample,evaluatinginfectionrisksandsocialstandardsfrom theircontacts,makingdecisiontoadopt,andcarryingout preventivebehavioragainstinfluenza.Forillustrativepurposes, theuseoffluantiviraldrugs(e.g.,Tamiflu)istakenasanexample ofpreventivebehaviorinthesimulation,becauseitsclinical efficacyismoreconclusivethanotherbehaviors,forinstance, washinghandandwearingfacemasks.Specifically,ifanindividual usesantiviraldrugs,thechanceofbeinginfectedandinfecting otherscanbereducedby70%and40%,respectively[14,29]. Implementationdetailsofthesetwomodelsarenotthefocusof thisarticle,andreaderscouldrefertoTextS1Section1.3and TableS2.InfluenzacontrolstrategiesInfluenzacontrolstrategiesaremostlyappliedatthreelevels: theindividuallevel,grouplevel,andcommunitylevel.Foreach level,onestrategyisselectedforsubsequentinvestigation,namely, aTargetedAntiviralProphylaxis(TAP)strategyattheindividual level,aworkplaceclosurestrategyatthegrouplevel,andatravel restrictionstrategyatthecommunitylevel,asshowninTable1. Detaileddescriptionsofthethreestrategiesareprovidedbelow. First,theTAPstrategyidentifiessymptomaticindividuals (influenzacases),searchestheirhouseholdmembers,andthen targetsantiviraldrugstoalltheseindividuals[13,30].Thisstrategy hasbeenrecommendedtobequiteeffectiveifstockpilesof antiviraldrugsaresufficientandinfectionscanbequicklydetected [4].Toaccountforlimitedhealthpersonnel,thisresearchassumes thatonlyaproportionofinfluenzacases,60%(60%TAP)and 80%(80%TAP),canbeidentifiedduringaday,followingthe designbyGermannetal.[13].Second,theworkplaceclosure strategyshutsdownaproportionofworkplaces/schoolswhere influenzacasesareidentified[31].Thisstrategyhasbeen suggestedtobeusefultosociallydistanceindividuals,delaythe diseasespread,andwintimefordevelopingvaccinesandantiviral drugs[32].FollowingtheworkbyFergusonetal.[33],alow-level scenario(10%WC)closes10%affectedworkplacesand100% affectedschoolsduringaday,whileahigh-levelscenarioCombinedControlEffectsforInfluenza PLoSONE|www.plosone.org2October2011|Volume6|Issue10|e24706

PAGE 3

(33%WC)closes33%affectedworkplacesand100%affected schools.Third,thetravelrestrictionstrategyaimstoreducethe tripsintoandoutofaffectedcommunities[33,34].Eachofthe967 censusblockgroupsinthestudyareaistreatedasacommunity. FollowingtheprojectbyGermannetal.[13],alow-levelscenario (10%TR)restricts10%tripsintoandoutofallaffected communities,whileahigh-levelscenarioprohibits50%trips (50%TR).Inadditiontotestingthethreecontrolstrategies individually,thecombinationsofallthreearealsoevaluated.A low-levelcombinationscenario(referredtoastheCombined-Low) includesallthreestrategiesattheirrespectivelowlevels.Likewise, ahigh-levelcombination(referredtoasCombined-High)contains allthreesinglestrategiesathighlevels. ThethreecontrolstrategiesandtheircombinationsinTable1 aresimulatedbytheinfluenza-onlymodelandthedual-diffusion model,respectively.Resultsfromtheinfluenza-onlymodel indicatetheeffectivenessofcontrolstrategieswithoutindividual preventivebehavior.Meanwhile,outcomesfromthedual-diffusion modelshowthecombinedeffectivenessofbothcontrolstrategies andindividualpreventivebehavior.Thesetwomodeledeffectivenessarecomparedtoabaselineepidemicscenario,which representsaworstsituationofnocontrolstrategiesandno preventivebehavior.Allstrategiesareassumedtobeimplemented atthetimewhenthecumulativenumberofinfluenzacases exceeds1,000(1 % ofthepopulation),andlastuntiltheendofthe epidemic.Individualshavingorhavingnotadoptedpreventive behavioraretreatedthesamebyallcontrolstrategies,sothatthe controleffectivenessfromtwomodelsarecomparable. ForeachmodelandeachstrategyscenarioinTable1,the simulationisperformed50realizationstoreducerandomness, resultinginatotalof1,000realizations(5strategies 6 2 scenarios 6 2models 6 50realizations).Eachsimulationrecords thetimeandlocationofeveryinfectioneventduringa150-day period.Foreachstrategyscenario,thecontroleffectivenessis measuredbyanepidemiccurvethatdepictsthenumberofdaily newinfluenzacasesfromDay1toDay150.Thenumberofdaily newcasesisaveragedfrom50modelrealizations,andthenplotted againsttimetoformanaveragedepidemiccurve(Figure1). Associatedcharacteristicsofthisepidemiccurvearealsoderived, includinganoverallattackrate(thepercentageofinfluenzacases inthepopulation)andepidemicpeaktime(Table2).Fortheease ofcomparison,arelativeeffectivenessofacontrolstrategyisalso calculatedasanindexrangingfrom0to1.Therelative effectivenessisdefinedasaratiooftheattackratereducedbya strategyfromthebaselinetothebaselineattackrate,i.e.,(Baseline attackrate 2 Attackrateunderastrategy)/Baselineattackrate.A zerovaluerepresentsthebaselinescenariowithoutanycontrol strategy(theattackrateundernon-strategy=thebaselinerate), whileahighervaluecloseto1indicatesthatacontrolstrategy producesasmallerattackrate.Aneffectivestrategyisexpectedto producealowepidemiccurve,smallattackrate,andhighrelative effectiveness.Inthisresearch,anepidemicisassumedtobe successfullycontained,iftheoverallattackrateisbelow5%.This isbecausereportedinfluenzaepidemicsoftenhavea5%orhigher attackrate[35,36]. Thespatialeffectivenessofcontrolstrategiesisalsoofinterest, andthusaseriesofinfectionintensitymapsaredisplayedin Figure2.Theinfectionintensityrepresentsthedensityoftotal infectionsaspointsoccurringwithineverygeographicunit (50m 6 50m)duringtheentire150-dayepidemic.Theintensity valueateachcelllocationisalsotheaveragefrom50model realizationsandisconvertedtoaunitofinfectionspersqkm2for theeaseofcomparison.Aneffectivestrategyisexpectedtoreduce infectionintensityateverylocation,andmeanwhileconfinethe spatialextentofaffectedareas.Results TargetedAntiviralProphylaxis(TAP)strategyatthe individuallevelOnaverage,thebaselineepidemicscenario(redcurvesin Figure1)causesan18.6%ofthepopulationdevelopinginfluenza symptoms(Table2).TheepidemicpeaksatDay77with approximately6,000newcasesoccurringatthepeaktime.The applicationof60%TAPand80%TAPscenario(bluecurvesin Figure1A–B)significantlyreducestheoverallattackrateto6.87% and4.74%,respectively.ThesetwoTAPscenariosalsopostpone thepeaktimeby5–13days.Withoutconsideringpreventive behavior,the80%TAPscenarioseemseffectivetocontainthe epidemic,becauseitmanagestolessentheoverallattackrate underthe5%epidemiccriterion. Byfurtheraddingthepreventivebehavior(hereinafterabbreviatedasPB),both60%TAP + PBand80%TAP + PBscenarios (greencurvesinFigure1A–B)resultinevenlowerattackrates around4.3%(Table2).Theepidemicpeakscanbelimitedaround 1,000dailynewcases,whilethepeaktimeremainssimilartothe baselinescenario.Thisisbecausethediffusionofpreventive behaviorquicklyexhauststhepoolofsusceptibleindividuals,and Table1. Designandsimulationofcontrolscenarios.EpidemicModels Strategies Influenza-onlymodel (withoutpreventivebehavior,PB) Dual-diffusionmodel (withpreventivebehavior) Baselinescenario Nocontrolstrategies Nopreventivebehavior N/A # 1:TAP Low:60%casesLow:60%cases + PB High:80%casesHigh:80%cases + PB # 2:School/workplaceclosure(WC) Low:100%schools + 10%workplacesLow:100%schools + 10%workplaces + PB High:100%schools + 33%workplacesHigh:100%schools + 33%workplaces + PB # 3:Travelrestriction(TR) Low:10%tripsLow:10%trips + PB High:50%tripsHigh:50%trips + PB # 1 + # 2 + # 3 Low:combinedbyalllows''aboveLow:combinedbyalllows''above + PB High:combinedbyallhighs''aboveHigh:combinedbyallhighs''above + PB doi:10.1371/journal.pone.0024706.t001 CombinedControlEffectsforInfluenza PLoSONE|www.plosone.org3October2011|Volume6|Issue10|e24706

PAGE 4

thusfewerindividualscanbeinfected.Theresultsindicatethatif thepreventivebehaviorofindividualsisconsidered,boththe60% TAP + PBand80%TAP + PBachieveasimilarcontroleffectiveness (=0.77),leadingtomildattackratesthatmaynotqualifyasan epidemic.The60%TAP,ratherthanthe80%TAP,wouldbe sufficientenoughtocontaintheepidemic.Healthagenciesonly Figure1.Simulatedepidemiccurvesresultingfromcontrolscenarioswith/withoutconsideringpreventivebehavior(PB). Thecurve depictsthenumberofdailynewinfluenzacasesduringthecourseofanepidemic.(A)60%TAP;(B)80%TAP;(C)10%WC;(D)33%WC;(E)10%TR;(F) 50%TR;(G)Combined-Low;(H)Combined-High. doi:10.1371/journal.pone.0024706.g001 CombinedControlEffectsforInfluenza PLoSONE|www.plosone.org4October2011|Volume6|Issue10|e24706

PAGE 5

needtoprepareasmallerstockpileofantiviraldrugsthanwould otherwisebeingexpected.Workplaceclosure(WC)strategyatthegrouplevelTurningtotheworkplaceclosurestrategies(bluecurvein Figure1C),the10%WCscenarioslightlyreducestheoverall attackratesto11.87%,anddelaysthepeaktimeonlyalittle (Table2).Incontrast,the33%WCscenario((bluecurvein Figure1D)lessenstheattackratetoamuchlowerlevelof4.86%, andadvancesthepeaktimebyapproximately1week.Forthe purposeofcontainingtheepidemic,the33%WCscenario,i.e., theclosureof33%affectedworkplaces,isneededtoachievean attackrateunder5%. Byfurtherincludingtheindividualpreventivebehavior(green curvesinFigure1C–D),the10%WC + PBand33%WC + PB scenariosproduceamuchsmallerattackrateof3.99%and1.83%, respectively(Table2).Thetimetoreachepidemicpeaksisshortened to61–66days,roughly2weeksearlierthanthebaselinescenario. Therelativeeffectivenessof10%WCscenarioisdoubledby consideringpreventivebehavior.Aprimaryreasonisthatanumber ofsusceptibleindividualsvoluntaryprotectthemselvesfrominfection. Theseindividuals,therefore,cannotbeinfectedorinfectothersat workplacesandschools,largelylimitingthediseasetransmission.The comparisonsuggeststhatgiventhepreventivebehaviorofindividuals iscounted,a10%workplaceclosurestrategy,insteadofthe33%one, wouldbeadequatetocontainaninfluenzaepidemic.Travelrestriction(TR)atthecommunitylevelSurprisingly,the10%TRscenario(Figure1E)alonecausesan evenworsesituationthanthebaselinescenario.Theoverallattack ratereaches20%,andis1.4%higherthanthebaselinerate, leadingtoanegativeeffectiveness(= 2 0.07inTable2).Apossible reasonisthatthetravelrestrictionstrategyextendsthetimeof individualsspentathome,therebyintensifyingthewithin-home transmission.Sinceonly10%oftripsintoandoutofaffected communitiesarerestricted,thediseasecanstillbeeasily transportedfromoneaffectedcommunitytoanotherthrough the90%unrestrictedtrips.Theepidemicthusdevelopsfasterand affectsmoreindividuals.Asthetravelrestrictionlevelelevatesto 50%(the50%TRinFigure1F),muchmoretripsintoandoutof affectedcommunitiesarerestricted.Althoughtheinfectionsat homesareintensified,mostinfectionscanonlytakeplacewithin communities,insteadofbetweencommunities.Asaresult,the overallattackratedropsto5.91%andtheepidemicpeakisgreatly mitigated.Nevertheless,the50%TRdoesnotsufficetocontain theepidemic,becausetheattackrateremainsabove5%. Thesimulationresultsaredistinctlydifferentifadding individualpreventivebehavior(greencurvesinFigure1E–F). The10%TR + PBscenarioproducesamuchbetteroutcomethan thatfromthe10%TRalone,becausetherelativeeffectiveness jumpsfrom 2 0.07to0.62.Theoverallattackrateandepidemic peaksizeareremarkablyreduced,althoughtheattackrate remainsabove5%(Table2).The50%TR + PBscenarioturnsout tobeeffectiveforinfluenzacontainment,becausetheoverall attackratecanbeloweredto1.65%,muchlessthanthe5% epidemiccriterion.CombinedcontrolstrategiesThecombinedstrategies(thebluecurvesinFigure1G–H) outperformeachofthethreesinglestrategies.Thetotalinfections canbecontainedfarbelow5%ofthepopulation,withasmall peaksizeunder1,000cases.Particularly,theCombined-High scenarioiscapableofpreventingtheepidemic,givenonly0.68% ofthepopulationbeinginfected(Table2).Amongthethreesingle strategies,theTAPstrategyreducesinfectionswithinhouseholds, theworkplaceclosurestrategytendstopreventinfectionsat Table2. Controleffectivenessofscenarioswith/withoutpreventivebehavior(PB).ScenariosOverallattackrate(%)Epidemicpeaktime(Days)RelativeEffectivenessbBaseline18.60[18.52,18.74]a77[64,92]0.00 60%TAP6.87[0.00,8.89]90[3,136]0.63 80%TAP4.74[0.00,7.49]82[3,145]0.75 60%TAP + PB4.31[0.00,5.20]71[3,104]0.77 80%TAP + PB4.30[0.00,4.96]76[5,102]0.77 10%WC11.87[11.36,11.90]80[64,71]0.36 33%WC4.86[0.00,5.42]69[4,94]0.74 10%WC + PB3.95[0.00,4.99]66[3,98]0.79 33%WC + PB1.83[0.00,2.46]61[3,103]0.90 10%TR20.00[19.91,20.11]74[64,85] 2 0.07 50%TR5.91[0.00,6.61]65[6,89]0.68 10%TR + PB7.10[0.00,8.70]67[3,97]0.62 50%TR + PB1.65[0.00,2.11]60[3,89]0.91 CombinedLow5.00[4.33,5.73]86[72,103]0.73 CombinedHigh0.72[0.00,0.94]60[5,108]0.96 CombinedLow + PB1.95[0.32,2.34]75[7,108]0.90 CombinedHigh + PB0.68[0.00,0.91]52[4,102]0.96aAllmeasuresaretheaveragesof50modelruns,and95%confidenceintervalsareshowninbrackets.bRelativeeffectiveness=(Baselineattackrate-Attackrateunderastrategy)/Baselineattackrate. T-testshowsthattherelativeeffectivenesswithandwithoutPBissignificantlydifferent( p -value=0.043). doi:10.1371/journal.pone.0024706.t002 CombinedControlEffectsforInfluenza PLoSONE|www.plosone.org5October2011|Volume6|Issue10|e24706

PAGE 6

workplaces,andthetravelrestrictionlimitsthediseasetransmissionbetweencommunities.Thesethreesinglestrategieswork togetherascomplements,leadingtoasignificantimprovementin controleffectiveness(relativeeffectiveness 0.9).Withoutconsideringpreventivebehavior,theCombined-Highscenarioseems necessarytocontaintheepidemic,whiletheCombined-Low scenarioisinsufficient.Thisargument,however,maybechanged byincorporatingindividualpreventivebehavior(greencurvesin Figure1G–H).TheCombined-Low + PBscenarionowisadequate toreducetheoverallattackratebelow5%andthuscontainthe epidemic,whilethehigh-levelscenarioisnolongeranecessity.Spatialeffectivenessofcontrolstrategiesandpreventive behaviorBasedonthecomparisonanalysisabove,theTAP60% + PB, 10%WC + PB,50%TR + PB,andtheCombined-Low + PB scenariosaresuggestedtobecost-effectiveincontrollinginfluenza epidemics.Therefore,theirspatialeffectivenessisfurtherexaminedandcomparedthroughinfectionintensitymaps(Figure2). Fordescriptionpurposes,themappedinfectionintensityisfurther categorizedinto6levels,i.e.,verylow(0–50infections/km2),low (50–100),moderate(100–200),high(200–500),veryhigh(500– 1,000),andextremelyhigh( 1,000). Figure2.Intensitymapsofcumulativeinfectionsfortheentireepidemic. (A)Baselinescenario,(B)60%TAP + PB,(C)10%WC + PB,(D)50% TR + PB,and(E)CombinedLow + PB.Thecolorramprepresentsthe150-daycumulativenumberofinfectionspersqkm2ata50m 6 50mcelllocation. Theinfectionintensityisfurthercategorizedinto6levels,i.e.,verylow(050infections/km2),low(50100),moderate(100200),high(200500),very high(5001,000),andextremelyhigh( 1,000). doi:10.1371/journal.pone.0024706.g002 CombinedControlEffectsforInfluenza PLoSONE|www.plosone.org6October2011|Volume6|Issue10|e24706

PAGE 7

Thebaselinescenario(Figure2A)inducesanextremelyhigh intensityofinfectionsinthecentralbusinessdistrictofthestudyarea. Theinfectionintensitydecreases inanoutwarddirectiontosuburbs. Thisisbecausethecentralbusinessdistricthasthedensestresidential populationandhighlyconcentrate dbusinesslocations.Comparedto thebaselinemap,the60%TAP + PBscenario(Figure2B)greatly reducestheinfectionintensitiesallaroundthestudyarea,althoughthe centralbusinessdistrictretainsahigh-levelintensity.Thespatial effectivenessof10%WC + PB(Figure2C)issimilartothe60% TAP + PB,butmoderateinfectionsaremorescatteredinthesuburban areas.The50%TR + PBiscapableofconfiningthewidespreadof influenzaoverthestudyarea(Figure2D),leavingonlyasmallnumber ofseparatedareaswithhighinfectionintensity.Thesehotspotsare mostlocatedwithinCBD,universit ycampuses,andlargeindustrial plants,wherealargenumberofpeopleworkandlive.Thisisprobably becausethecity-widetravelsofindividualsarepartiallyprohibited, andhencediseasecanonlydeveloplocally.Finally,theCombined low + PBscenarionotonlyreducestheintensityoftheinfectionsatall locations,butalsoconfinesspatialextentofdiseasespread(Figure2E). Theinfectionsinthecentralbusinessdistrictarereducedtoa moderatelevel,whileavastproportionofthestudyareahasonlya smallnumberofinfections.DiscussionInsummary,previousstudiesoninfluenzacontainmenthaveonly consideredtheeffectivenessofapplyingcontrolstrategies,while overlookingtheeffectivenessfromindividuals’preventivebehavior. Thisresearchestimatesthecombinedeffectivenessofbothcontrol strategiesandindividualpreventivebehavior.Theresultsimplythat previousstudiesoncontrolstrategiesareincomplete,andthecontrol effectivenessmightbeunder-estimated.Thecomparisonbetween twomodelresultsindicatesthatpreventivebehaviorofindividuals hasanextraeffectiveness,inadditiontotheeffectivenessfromtypical controlstrategiesalone.Thisextraeffectivenessproducesaneven smallerattackrateofinfluenza,lowerepidemicpeak,andearlier peaktime.Byconsideringthecombinedeffectiveness,thecontrolof influenzaepidemicsmaynotrequireasmuchhealthresourcesas estimatedinpreviousstudies.Forexample,the80%TAPstrategy couldbereplacedbythe60%one,reducingtheburdenoflocal agenciestopreparehealthresources.Likewise,the10%workplace closurestrategy,ratherthanthe33%strategy,wouldbesufficientto controltheseasonalinfluenzaepidemicinthestudyarea.Enormous socio-economicdisruptionscouldbepossiblyavoided.Alow-level combinationofthethreestrategiesisrecommendedtosuppress influenzaepidemicinthestudyarea,whileahigh-levelcombination isnolongeramust.Particularly,withthehelpofindividual preventivebehavior,the50%travelrestrictionstrategyandthelowlevelcombinedstrategycansuccessfullyconfinesthespatial dispersionofinfluenzainthestudyarea. Similartoanymodelinganalysis,thisresearchhasanumberof limitations.First,thesimulationmodelsfocusononeUS metropolitanarea,oneinfluenzavirusstrain,andonepreventive behavior.Itispossiblethatthemodeloutcomesvaryindifferent citiesanddifferentdiseaseparameters,suchasapandemicinfluenza virus.Theinterpretationofmodeloutcomesshouldbelimitedto seasonalinfluenzaandinthestudyarea.Althoughtheuseof antiviraldrugsistakenasanexampleinthisresearch,the methodologycanbeeasilyextendedtootherpreventivebehavior, suchaswashinghandsandwearingfacemasks,oncetheir preventiveefficacyisconclusivelyquantified.Second,themass mediaalsoinfluencespeople’sdecisiontoadoptpreventive behavior,especiallyfordiseasesthatarehighlyinfectiousorpose severehealthrisks,suchasthesevereacuterespiratorysyndrome (SARS).Thisresearchhasnotmodeledthemassmediabecauseits effectsonflu-relatedpreventivebehaviorremaininconclusive.In addition,theseasonalinfluenzasimulatedinthisresearchhasa relativelymildinfectivityandlimitedrisks,thusisusuallynotafocus ofmassmediaattention.Third,themodelassumesthatindividuals adoptpreventivebehaviorimmediatelyafterthethresholdeffects happen.Inreality,individuals’adoptionofabehaviormaytakea relativelylongerperiodasitmayinvolveanumberofpsychological steps[37].Amoresophisticatedbehavioralapproachmayimprove themodelingreality,butalsoincreasethecomplexityofmodel structure.Atrade-offbetweenmodelperformanceanddetaillevels isalwaysachallengeformodelers[38].Ongoingresearchis intendedtoaddresstheselimitationsandchallenges. Controlstrategiesenforcedbyhealthagenciesandpreventive behaviorvoluntarilypracticedbythepublicaretwointertwined componentsofdiseasecontainment.Ignoringeithercomponentmay preventusfromeffectivelymitigatingburdensofinfluenzaonpublic health.Itishardtoresistcitingandrephrasingtheargumentby Funketal.[26]that‘‘individualself-initiatedbehaviorcanchange thefateofanoutbreak,anditscombinedeffectivenesswithcontrol strategiesrequiresproperunderstandingifwearetofully comprehendhowthesecontrolmeasureswork’’.Thisresearch attemptstofusethehumanbehavioraldimensionintothestudyof controlstrategies,andthusoffersmorecomprehensiveunderstandingsondiseasecontainment.Healthagenciesarerecommendedto gainpriorknowledgeaboutbehavioralpatternsoflocalpeople beforechoosinginfluenzacontrolstrategies.Thefindingsofthis researchcallforareviewofcurrentcontrolstrategiesandre-estimate thehealthresourcesthatarenecessarytocontainepidemics.Itis believedthatsuchareviewwouldshednewinsightsonimproving controleffectivenessforloominginfluenzapandemics.SupportingInformationFigureS1Thesimulationofcontactnetwork. The assignmentofindividualstohouseholds,workplaces,serviceplaces andneighborhouseholdsbasedontheattributeandspatial informationofindividuals. (TIF)FigureS2Estimateddistributionofthethresholdof infectionrisksbygender. The X axisindicatestheproportion ofinfluenzacasesinthecontactsofaparticipantthatisneededto convincetheparticipanttoadopt.The Y axisshowsthefrequency ofsuchproportionoccurringinthesurveyresults. (TIF)FigureS3Estimatedthresholddistributionofadoption pressurebygender. The X axisindicatestheproportionof adoptersinthecontactsofaparticipantthatisneededtoconvince theparticipanttoadopt.The Y axisshowsthefrequencyofsuch proportionoccurringinthesurveyresults. (TIF)TableS1Modelparametersforsimulatinginfluenza. (DOCX)TableS2Modelparametersforsimulatingpreventive behavior. (DOCX)TextS1(DOC)AuthorContributionsConceivedanddesignedtheexperiments:LM.Performedtheexperiments: LM.Analyzedthedata:LM.Wrotethepaper:LM. CombinedControlEffectsforInfluenza PLoSONE|www.plosone.org7October2011|Volume6|Issue10|e24706

PAGE 8

References1.NeumannG,NodaT,KawaokaY(2009)Emergenceandpandemicpotentialof swine-originH1N1influenzavirus.Nature459:931–939. 2.FergusonNM,FraserC,DonnellyCA,GhaniAC,AndersonRM(2004)Public HealthRiskfromtheAvianH5N1InfluenzaEpidemic.Science304:968–969. 3.CDC(2010)UpdatedCDCEstimatesof2009H1N1InfluenzaCases, HospitalizationsandDeathsintheUnitedStates,April2009–April10,2010. Atlanta:CentersforDiseaseControlandPrevention. 4.FergusonNM,KeelingMJ,EdmundsWJ,GaniR,GrenfellBT,etal.(2003) Planningforsmallpoxoutbreaks.Nature425:681–685. 5.CDC(2008)PreventingSeasonalFlu.Atlanta:CentersforDiseaseControland Prevention. 6.FergusonN(2007)Capturinghumanbehaviour.Nature446:733. 7.VardavasR,BrebanR,BlowerS(2007)Caninfluenzaepidemicsbeprevented byvoluntaryvaccination.PLoSComputationalBiology3:e85. 8.FunkS,GiladE,WatkinsC,JansenVAA(2009)Thespreadofawarenessandits impactonepidemicoutbreaks.ProceedingsoftheNationalAcademyofSciences 106:6872–6877. 9.BeckerMH(1976)Thehealthbeliefmodelandpersonalhealthbehavior. Thorofare:SLACKpress. 10.FunkS,GiladE,JansenV(2010)Endemicdisease,awareness,andlocal behaviouralresponse.JournalofTheoreticalBiology264:501–509. 11.KermackWO,McKendrickAG(1927)Acontributiontothemathematical theoryofepidemics.ProceedingsRoyalStatisticsSociety115:700–721. 12.SattenspielL,DietzK(1995)Astructuredepidemicmodelincorporating geographicmobilityamongregions.MathematicalBiosciences128:71–91. 13.GermannTC,KadauK,LonginiJr.IM,MackenCA(2006)Mitigation strategiesforpandemicinfluenzaintheUnitedStates.Proceedingsofthe NationalAcademyofSciences103:5935–5940. 14.LonginiIM,HalloranME,NizamA,YangY(2004)Containingpandemic influenzawithantiviralagents.AmericanJournalofEpidemiology159: 623–633. 15.HalloranME,FergusonNM,EubankS,IraM,LonginiJ,CummingsDAT, etal.(2008)Modelingtargetedlayeredcontainmentofaninfluenzapandemicin theUnitedStates.ProceedingsoftheNationalAcademyofSciences105: 4639–4644. 16.ChaoDL,HalloranME,ObenchainVJ,LonginiJr.IM(2010)FluTE,a publiclyavailablestochasticinfluenzaepidemicsimulationmodel.PLoS ComputionalBiology6:e1000656. 17.CoburnBJ,WagnerBG,BlowerS(2009)Modelinginfluenzaepidemicsand pandemics:insightsintothefutureofswineflu(H1N1).BMCMedicine7:1–8. 18.DuttaA(2008)TheEffectivenessofPoliciestoControlaHumanInfluenza Pandemic:ALiteratureReview.WorldBankPolicyResearchWorkingPaper No4524:SocialScienceResearchNetwork. 19.USCensus(2000)U.S.Census2000DataReleases.Washington,DC. 20.ReferenceUSA(2009)U.S.BusinessDatabases.Omaha,NE:ReferenceUSA Inc.,Omaha,NE. 21.MaoL,BianL(2010)Spatial-temporaltransmissionofinfluenzaanditshealth risksinanurbanizedarea.Computers,EnvironmentandUrbanSystems34: 204–215. 22.BianL(2004)Aconceptualframeworkforanindividual-basedspatiallyexplicit epidemiologicalmodel.EnvironmentandPlanningB31:381–395. 23.MaoL,BianL(2011)Agent-basedSimulationforaDual-DiffusionProcessof InfluenzaandHumanPreventiveBehavior.InternationalJournalofGeographicalInformationScience25(9):1371-1388. 24.KissIZ,CassellJ,ReckerM,SimonPL(2010)Theimpactofinformation transmissiononepidemicoutbreaks.Mathematicalbiosciences225:1–10. 25.HatzopoulosV,TaylorM,SimonPL,KissIZ(2011)Multiplesourcesand routesofinformationtransmission:implicationsforepidemicdynamics. Mathematicalbiosciences. 26.FunkS,Salathe M,JansenVAA(2010)Modellingtheinfluenceofhuman behaviouronthespreadofinfectiousdiseases:areview.JournalofTheRoyal SocietyInterface7:1247–1256. 27.GranovetterM(1978)Thresholdmodelsofcollectivebehavior.American journalofsociology83:1420–1443. 28.ValenteTW(1996)Socialnetworkthresholdsinthediffusionofinnovations. SocialNetworks18:69–89. 29.HaydenFG(2001)Perspectivesonantiviraluseduringpandemicinfluenza. PhilosophicalTransactionsoftheRoyalSocietyofLondonSeriesB356: 1877–1884. 30.LonginiIM,NizamA,XuS,UngchusakK,HanshaoworakulW,etal.(2005) Containingpandemicinfluenzaatthesource.Science309:1083–1087. 31.BurkeDS,EpsteinJM,CummingsDAT,ParkerJI,ClineKC,etal.(2006) Individual-basedcomputationalmod elingofsmallpoxepidemiccontrol strategies.AcademicEmergencyMedicine13:1142–1149. 32.KelsoJK,MilneGJ,KellyH(2009)Simulationsuggeststhatrapidactivationof socialdistancingcanarrestepidemicdevelopmentduetoanovelstrainof influenza.BMCPublicHealth9:117. 33.FergusonNM,CummingsDA,FraserC,CajkaJC,CooleyPC,etal.(2006) Strategiesformitigatinganinfluenzapandemic.Nature442:448–452. 34.CamitzM,LiljerosF(2006)Theeffectoftravelrestrictionsonthespreadofa moderatelycontagiousdisease.BMCmedicine4:32. 35.CDC(2008)KeyFactsAboutSeasonalInfluenza(Flu).Atlanta:Centersfor DiseaseControlandPrevention. 36.CoxNJ,SubbaraoK(2000)GlobalEpidemiologyofInfluenza:Pastand Present.AnnualReviewsinMedicine51:407–421. 37.WeinsteinND,SandmanPM(1992)Amodeloftheprecautionadoption process:Evidencefromhomeradontesting.HealthPsychology11:170–180. 38.BrooksR,TobiasA(1996)Choosingthebestmodel:Levelofdetail,complexity, andmodelperformance.Mathematicalandcomputermodelling24:1–14.CombinedControlEffectsforInfluenza PLoSONE|www.plosone.org8October2011|Volume6|Issue10|e24706