Epizoological Tools for Acute Hepatopancreatic Necrosis ... version of... · at country level,...

EpizoologicalToolsfor

AcuteHepatopancreaticNecrosisDisease

(AHPND)inThaiShrimpFarming

THESISSUBMITTEDTOTHEUNIVERSITYOFSTIRLINGFORTHE

DEGREEOFDOCTOROFPHILOSOPHY

BY

NATTAKANSALEETID

JULY2017

INSTITUTEOFAQUACULTURE

2

Declaration

3

Declaration

Iherebydeclarethat:

(1)Thisthesisiscomposedbyme,

(2)Theworkdescribedismyownwork,exceptfortheliveshrimpmovementnetwork

data from Thailand, which I analyse, but which is obtained from the Thailand

DepartmentofFisheries,

(3)ThebibliographycontainsalltheliteraturethatIhaveusedinwritingthethesis,and

(4)Theworkcontainedinthesishasnotbeensubmittedforanyotherdegree.

Signature:____________________________________________

Signatureofsupervisor:_________________________________

Date:________________________________________________

Abstract

4

Abstract

Acutehepatopancreaticnecrosisdisease(AHPND)isanemergingbacterialinfectionin

shrimpthathasbeenwidespreadacrossthemajorworldshrimpproducingcountries

since2009.AHPNDepizooticshaveresultedinahugelossofglobalshrimpproduction,

similar to that caused by white spot disease in the 1990’s. The epizootiological

understandingofthespreadofAHPNDisstillinitsearlystages,however,andmostof

thecurrentlypublishedresearchfindingsarebasedonexperimentalstudiesthatmay

struggletocapturethepotentialfordiseasetransmissionatthecountryscale.Themain

aim of this research, therefore, is to develop epizootiological tools to study AHPND

transmission between shrimp farming sites. Some tools used in this research have

alreadybeenappliedtoshrimpepizoology,butothersareusedhereforthefirsttime

toevaluatethespreadofshrimpdiseases.

AccordingtoanepizootiologicalsurveyofAHPNDinThailand(Chapter3),thefirstcase

ofAHPNDinthecountrywasineasternshrimpfarmsinJanuary2012.Thediseasewas

thentransmittedtothesouthinDecember2012.Theresultsobtainedfrominterviews,

undertakenwith143samplefarmswerestratifiedbythreefarm-scales(large,medium

andsmall)andtwolocations(eastandsouth).Boththesouthern locationand large-

scalefarmingwereassociatedwithadelayinAHPNDonsetcomparedwiththeeastern

locationandsmall-andmedium-scalefarming.

The24riskfactors(mostlyrelatedtofarmingmanagementpractices)forAHPNDwere

investigatedinacross-sectionalstudy(Chapter3).Thisallowedthedevelopmentofan

AHPNDdecision tree for defining cases (diseased farms) and controls (non-diseased

farms)becauseat the timeof the studyAHPNDwasadiseaseofunknownetiology.

Resultsofunivariateandunconditional logistic regressionmodels indicated that two

farmingmanagementpractices related to theonsetofAHPND.First, theabsenceof

pondharrowingbeforeshrimpstockingincreasedtheriskofAHPNDoccurrencewithan

odds ratio (𝑂𝑅) of 3.9 (95 % CI1.3–12.6; P-value=0.01), whereas earthen ponds

decreasedtheriskofAHPNDwithan𝑂𝑅of0.25(95%CI0.06–0.8;P-value=0.02).These

findings imply that good farming management practices, such as pond-bottom

Abstract

5

harrowing, which are a common practice of shrimp farming in earthen ponds,may

contributetoovercomingAHPNDinfectionatfarmlevel.

Forthepurposesofdiseasesurveillanceandcontrol,thestructureoftheliveshrimp

movement networkwithin Thailand (LSMN)wasmodelled,which demonstrated the

high potential for site-to-site disease spread (Chapter 4). Real network data was

recordedover a 13-month period fromMarch 2013 toMarch 2014 by the Thailand

DepartmentofFisheries.Afterdatavalidation,c.74400repeatedconnectionsbetween

13801shrimpfarmingsiteswereretained.77%ofthetotalconnectionswereinter-

province movements; the remaining connections were intra-province movements

(23%).

TheresultsdemonstratedthattheLSMNhadpropertiesthatbothaidedandhindered

diseasespread(Chapter4).Forhinderingtransmission,thecorrelationbetween𝑖𝑛and

𝑜𝑢𝑡degreeswasweaklypositive,i.e.itsuggeststhatsiteswithahighriskofcatching

diseaseposedalowriskfortransmittingthedisease(assumingsolelynetworkspread),

and the LSMN showed 𝑖𝑛-𝑜𝑢𝑡 disassortative mixing, i.e. a low preference for

connections joining sites with high 𝑖𝑛 degree linked to connections with high 𝑜𝑢𝑡

degree.However,therewerelowvaluesformeanshortestpathlengthandclustering.

Thelattercharacteristicstendtobeassociatedwiththepotentialfordiseaseepidemics.

Moreover, theLSMNdisplayedthepower-law𝑃(𝑘)~𝑘/0 inboth 𝑖𝑛and𝑜𝑢𝑡degree

distributions with the exponents (𝛾) 2.87 and 2.17, respectively. The presence of

power-lawdistributionsindicatesthatmostsitesintheLSMNhaveasmallnumberof

connections,whileafewsiteshavelargenumbersofconnections.Thesefindingsnot

onlycontributetoabetterunderstandingofdiseasespreadbetweensites,therefore,

butalsorevealtheimportanceoftargeteddiseasesurveillanceandcontrol,duetothe

detectionofscale-freepropertiesintheLSMN.

Chapter5,therefore,examinedtheeffectivenessoftargeteddiseasesurveillanceand

control inrespecttoreducingthepotentialsizeofepizootics intheLSMN.Thestudy

untilisednetworkapproachestoidentifyhigh-riskconnections,whoseremovalfromthe

networkcouldreduceepizootics.Fivedisease-controlalgorithmsweredevelopedfor

the comparison: four of these algorithms were based on centrality measures to

Abstract

6

representtargetedapproaches,withanon-targetedapproachasacontrol.Withthe

targetedapproaches,technicallyadmissiblecentralitymeasureswereconsidered:the

betweenness(thenumberofshortestpathsthatgothroughconnectionsinanetwork),

connection weight (the frequency of repeated connections between a site pair),

eigenvector(consideringthedegreecentralitiesofallneighbouringsitesconnectedto

aspecifiedsite),andsubnet-crossing(prioritisingconnectionsthatlinkstwodifferent

subnetworks).Theresultsshowedthattheestimatedepizooticsizesweresmallerwhen

an optimal targeted approachwas applied, comparedwith the random targeting of

high-riskconnections.Thisoptimaltargetedapproachcanbeusedtoprioritisetargets

inthecontextofestablishingdiseasesurveillanceandcontrolprogrammes.

Withcomplexmodesofdiseasetransmission(i.e.long-distancetransmissionlikevialive

shrimpmovement,andlocaltransmission),an𝑆𝐸𝐼𝑅𝑆compartmental,individual-based

epizootic model was constructed for AHPND (Chapter 6). The 𝑆𝐸𝐼𝑅𝑆 modelling

uncoveredtheseasonalityofAHPNDepizooticsinThailand,whichwerefoundlikelyto

occurbetweenAprilandAugust(duringthehotandrainyseasonsofThailand).Based

on two movement types, intra-province movements were a small proportion of

connections,andtheyalonecouldcauseasmallAHPNDepizootic.Themainpathway

for AHPND spread is therefore long-distance transmission and regulators need to

increasetheefficacyoftestingfordiseasesinfarmedshrimpbeforemovementsand

improvetheconductofroutinemonitoringfordiseases.Theimplementationofthese

biosecurity practices was modelled by changing the values of the long-distance

transmissionrate(𝐵6789).Themodeldemonstratedthathighlevelsofbiosecurityon

liveshrimpmovements(𝐵6789 <1)ledtoadecreaseinthepotentialsizeofepizootics

inThaishrimpfarming.Moreover,thepotentialsizeofepizooticswasalsodecreased

whenAHPNDspreadwasmodelledwithadecreasedvalueforthelocaltransmission

rate 𝐵67;<6. Hence, not only did the model predict AHPND epizootic dynamics

stochastically, but it also assessed biosecurity enhancement, allowing the design of

effectivepreventionprogrammes.

Inbrief,thisthesisdevelopstoolsforthesystematicepizootiologicalstudyofAHPND

transmissioninThaishrimpfarminganddemonstratesthat:(1)atfarmlevel,current

Abstract

7

Thaishrimpfarmingshouldenhancebiosecuritysystemseveninlargerbusinesses,(2)

atcountry level, targeteddiseasecontrolstrategiesarerequiredtoestablishdisease

surveillanceandcontrolmeasures.Althoughtheepizootiologicaltoolsusedheremainly

evaluatethespreadofAHPNDinshrimpfarmingsites,theycouldbeadaptedtoother

infectiousdiseasesorotherfarmingsectors,suchasthecurrentspreadoftilapialake

virusinNiletilapiafarms.

Acknowledgments

8

Acknowledgments

Thisthesiswouldnothavebeenpossiblewithoutthesupportofmanypeople.Firstand

foremost, Iwould liketothankmysupervisorsDarrenGreenandFrancisMurray for

providingme with the opportunity to completemy PhD thesis at the University of

Stirling.

I especially want to thank my supervisor, Darren Green, who provided me with

direction, technical support and became more of a mentor and friend than a

supervisor. He read my numerous revisions and helped make some sense of the

confusion.Iamverygratefulforhispatience,motivationandimmenseknowledgein

epidemiology.

In addition, thanks aredue tomy sponsors, theAgricultural ResearchDevelopment

Agency (Public Organization) and the Thailand Department of Fisheries for their

financial support throughout my study. My sincere thanks also go to Dr. Jiraporn

KasornchandraandDr.PutthSongsangjindaforbelievinginmeandsupportingmein

everything.

Finally,Iwishtothanktomyparentsfortheirloveandsupportthroughoutmylife, and

tomybestfriends,Paranee,ThidaratandSooksri,whoalwayshelpmeandbelievethat

Icandoit.

Tableofcontents

9

Tableofcontents

Declaration……………........................................................................................................3

Abstract…………................................................................................................................4

Listoffigures……...........................................................................................................15

Listoftables………….......................................................................................................19

Chapter1- Generalintroduction...............................................................................21

1.1 Overviewofthethesis....................................................................................21

1.2 Theimportanceandcurrentstatusoftheshrimpfarmingsector.................23

1.3 Thesupplychainoffarmedshrimp,relatedregulationsforcontrolling

productionandsocialcommunitiesofshrimpproducersinThailand......................25

1.4 ThecharacteristicsofshrimpfarminginThailand..........................................27

1.5 Infectiousdiseasesinfarmedshrimpandtheirtransmission........................30

1.5.1 Theoccurrenceofdisease.......................................................................32

1.5.2 Riskfactorsforvibriosisoutbreaksinfarmedshrimp.............................35

1.5.3 Site-to-sitetransmissionofshrimpdiseases...........................................40

1.6 Researchoutline.............................................................................................42

1.7 References......................................................................................................46

Chapter2- Epidemiologicalandepizootiologicaltoolsfordesignofdiseaseprevention

andcontrolstrategies...................................................................................................61

2.1 Experimentalstudies......................................................................................62

Tableofcontents

10

2.2 Observationalstudies.....................................................................................63

2.3 Theoreticalstudies.........................................................................................66

2.3.1 Compartmentalepidemicmodelsformicroparasitesversus

macroparasites......................................................................................................66

2.3.2 Mass-actionversusnetworkmodels.......................................................69

2.3.2.1 Mass-actionmodels.................................................................................69

2.3.2.2 Networkmodels......................................................................................70

2.3.2.3 Networkmodelsfortargeteddiseasesurveillanceandcontrol..............72

2.3.3 Individual-basedsimulationmodels........................................................75

2.4 Applicationtoacutehepatopancreaticnecrosisdisease(AHPND)................76

2.5 References......................................................................................................78

Chapter3- Evaluatingriskfactorsfortransmissionofacutehepatopancreaticnecrosis

disease(AHPND)intheThaishrimpfarmingsector.....................................................90

3.1 Abstract..........................................................................................................91

3.2 Introduction....................................................................................................91

3.3 Methods.........................................................................................................93

3.3.1 DevelopmentofacasedefinitionanddecisiontreeforAHPND.............93

3.3.2 CandidateriskfactorsforAHPNDoccurrenceatfarmlevel....................94

3.3.3 Sampledesign..........................................................................................95

3.3.4 Surveydesign...........................................................................................96

3.4 DataAnalysis...................................................................................................97

Tableofcontents

11

3.4.1 DescriptiveanalysisoftheoccurrenceofAHPNDandotherdiseases....97

3.4.2 StatisticalanalysisoftheriskfactorsforAHPND....................................97

3.5 Results..........................................................................................................100

3.5.1 IdentificationofAHPNDcasesandcontrols..........................................100

3.5.2 DescriptiveepizoologyofAHPND..........................................................104

3.5.3 RiskfactorsforAHPNDtransmissionatfarmlevel................................106

3.6 Discussion.....................................................................................................109

3.7 References....................................................................................................113

Chapter4- Analysisofthenetworkstructureoftheliveshrimpmovementsrelevant

toAHPNDepizootic.....................................................................................................119

4.1 Abstract........................................................................................................120

4.2 Introduction..................................................................................................120

4.3 Methods.......................................................................................................122

4.3.1 Datasources..........................................................................................122

4.3.2 Identificationofliveshrimpmovementtypesbyprovincialscale........123

4.3.3 Provincialvisualisationfortheliveshrimpmovementnetwork

(LSMN)……………….................................................................................................124

4.3.4 LSMNadjacencymatrixfornetworkrepresentationandanalysisatsite

level……................................................................................................................124

4.3.5 Rewiringthenetwork............................................................................129

4.4 Results..........................................................................................................132

Tableofcontents

12

4.4.1 GeneralcharacteristicsoftheliveshrimpmovementnetworkofThailand

(LSMN)……….........................................................................................................132

4.4.2 VisualisingtheLSMNbasedonnationalprovincialcentres..................137

4.4.3 Descriptiveanalysisoftheliveshrimpmovementnetwork(LSMN)atsite

level…………...........................................................................................................139

4.5 Discussion.....................................................................................................145

4.6 References....................................................................................................149

Chapter5- Target priority for targeted disease surveillance and control in the live

shrimpmovementnetworkofThailand......................................................................157

5.1 Abstract........................................................................................................158

5.2 Introduction..................................................................................................158

5.3 Materialsandmethods.................................................................................160

5.3.1 Datasourcefortheliveshrimpmovementnetwork(LSMN)................160

5.3.2 Disease-controlalgorithmsfortargeteddiseasesurveillanceand

control………….......................................................................................................161

5.3.3 Usingthedisease-controlalgorithms....................................................165

5.3.4 Characterisingthetargetedconnections..............................................166

5.4 Results..........................................................................................................166

5.4.1 Thenumberofsitesreachedinthenetwork........................................166

5.4.2 Reducingconnectedcomponentsinthenetwork.................................168

5.4.3 Thecharacteristicsoftargetedconnections.........................................171

Tableofcontents

13

5.5 Discussion.....................................................................................................172

5.6 References....................................................................................................176

Chapter6- Epizootic disease modelling in farmed shrimp using compartmental

epizooticnetwork-basedsimulations.........................................................................181

6.1 Abstract........................................................................................................182

6.2 Introduction..................................................................................................182

6.3 Materialsandmethod..................................................................................184

6.3.1 Theliveshrimpmovementnetwork(LSMN).........................................184

6.3.2 Localcontactsbetweenshrimpfarmingsites.......................................185

6.3.3 An𝑆𝐸𝐼𝑅𝑆compartmental,individual-basedepizooticmodelforacute

hepatopancreaticnecrosisdisease(AHPND).......................................................186

6.4 Results..........................................................................................................194

6.4.1 SeasonalityofAHPNDepizooticdynamics............................................194

6.4.2 Effectoflong-distanceandlocaltransmissiononAHPNDepizootic

dynamics..............................................................................................................195

6.4.3 GeographicdistributionsofAHPNDprevalenceatprovinciallevelin

Thailand……….......................................................................................................197

6.4.4 Predictiveperformanceofthe𝑆𝐸𝐼𝑅𝑆models......................................198

6.5 Discussion.....................................................................................................199

6.6 References....................................................................................................203

Chapter7- Generaldiscussion.................................................................................209

Tableofcontents

14

7.1 Summary.......................................................................................................209

7.2 Generaldiscussion........................................................................................210

7.2.1 Diseasecaseconfirmation.....................................................................210

7.2.2 Shrimpfarmingdatausedforepizoology.............................................211

7.2.3 Modellingdiseaseepizooticdynamics..................................................212

7.3 Futurework..................................................................................................213

7.3.1 Controlstrategiesforlocalnon-networkspread..................................213

7.3.2 Coinfectionepizooticmodels................................................................213

7.3.3 Geographicalinformationsystems(GIS)forshrimpfarmingsites........214

7.4 Conclusions...................................................................................................214

7.5 References....................................................................................................217

Appendices

AppendixA: Shrimpdiseasepictures 221

AppendixB: Questionsusedinbrieftelephonesurvey 223

AppendixC: Nationalprovincialcentresandabbreviation 224

Listoffigures

15

Listoffigures

Figure1.1Volumeandpercentageof shrimpproducts (rawshrimpandvalue-added

shrimp)importstotheUSAmarketbythemajorproducingcountriesin2014…............24

Figure1.2Volumeandpercentageof shrimpproducts (rawshrimpandvalue-added

shrimp)importstoEUmarketsbythemajorproducingcountriesin2014…………………..24

Figure1.3TheshrimpproductionchaininThailand…………………………………………………….25

Figure1.4OverviewofshrimpfarminginThailandfortwomajorshrimpspecies:tiger

shrimpandwhitelegshrimp,1999–2013……………………………………………………………………28

Figure1.5DistributionofThaishrimpfarmingsitesbyprovince………………………………….29

Figure1.6Theepidemiologicaltriad…………………………………………………………………………..32

Figure1.7Majorroutesinsite-to-sitetransmissionofshrimpdiseases……………………….40

Figure 1.8 Outline of the “Epizootiological tools for AHPND in Thai shrimp farming”

research…………………………………………………………………………………………………………………….45

Figure 2.1 Potential pathway for disease transmission via live shrimpmovements in

Thailand…………………………………………………………………………………………………………………….70

Figure 2.2 Epidemic network models often are often characterised by these five

simulatednetworks……………………………………………………………………………………………………71

Figure 3.1 A flow chart of the methodology used in evaluating risk factors for

transmission of acute hepatopancreatic necrosis disease (AHPND) in Thai shrimp

farming…………………………………………………………………………………………………………………….100

Figure 3.2 TheAHPNDdecision tree for determinationof higherAHPNDprobability,

lowerAHPNDprobability,andnoAHPND…………………………………………………………………102

Listoffigures

16

Figure 3.3 Report of disease status stratified according to geographic location and

farm-scalebetweenJanuary2012andMay2013……………………………………………………..105

Figure3.4ThecumulativeincidenceofAHPNDbetweenJanuary2012andMay2013,

accountingtotworegions………………………………………………………………………………………..106

Figure3.5ROCcurvesforAHPNDmodels…………………………………………………………………109

Figure 4.1 A small weighted directed network and its matrix of the shortest paths

𝐿>?……………………………………………………………………………………………………………………………127

Figure 4.2 An example of a rewiring process which generates a new network by

swappingtheendpointsoftwo-pairconnectionsinanetwork………………………………….130

Figure4.3Stronglyconnectedcomponentofadirectednetworkwitheightsites……..131

Figure 4.4 Weakly connected component of a bidirectional network with eight

sites…………………………………………………………………………………………………………………………132

Figure4.5Circa13800shrimpfarmingsiteslocatedinfiveregionsand37provincesof

Thailand…………………………………………………………………………………………………………………..133

Figure 4.6 Diagrammatic representation of LSMN demonstrating the Thai shrimp

farmingindustrystructure……………………………………………………………………………………….134

Figure4.7Distributionofthenumberofrepeatedconnectionsoverthe13-monthstudy

period(March2013–March2014)ofliveshrimpmovementsinThailand………………….135

Figure4.8Distributionofthenumberofshrimpmovedoverthe13-monthstudyperiod

(March2013–March2014)ofliveshrimpmovementsinThailand…………………………….136

Figure4.9Theprovincialstructureofthe liveshrimpmovementnetworkofThailand

(LSMN)overa13-monthperiod(March2013–March2014)……………………………………..138

Figure 4.10 The weighted degree distributions for the LSMN plotted on a log–log

scale…………………………………………………………………………………………………………………………141

Listoffigures

17

Figure 4.11 The distribution of weighted path lengths in the live shrimpmovement

networkofThailand(LSMN)isshownasafractionoftotalconnections……………………143

Figure5.1Schematicexplainingdisease-controlalgorithmswithandwithouttargeted

approachesfortargeteddiseasesurveillanceandcontrolfortheliveshrimpmovement

networkofThailand(LSMN)……………………………………………………………………………………..162

Figure 5.2 Evaluating the disease-control algorithms against the network

reachability…………………………………………………………………………………………………………….167

Figure5.3Resultsofdifferentstepsizesofthebetweennessalgorithmcomparedtothe

randomalgorithmat250removals…………………………………………………………………………..168

Figure 5.4 Evaluating the disease-control algorithms against the weakly connected

components(WCC)………………………………………………………………………………………………….170

Figure5.5Resultsofdifferentstepsizeswhencomparingthesubnet-crossingalgorithm

andrandomalgorithmat250removals…………………………………………………………………….171

Figure6.1Frequencyofnumberofsitememberspersub-district……………………………..186

Figure6.2Densityplotsof fitteddistributionsof thedata for incubationperiodsand

fallowperiods………………………………………………………………………………………………………….188

Figure6.3Designand implementationofanalgorithmforan𝑆𝐸𝐼𝑅𝑆compartmental,

individual-basedepizooticmodelforshrimpdiseaseinThailand………………………………192

Figure6.4Meannumberofinfectedsitesperseedforone-monthepizootics…………..195

Figure6.5Expectedoutcomesoftheapplicationofbiosecuritymeasuresonliveshrimp

movementsinThailand…………………………………………………………………………………………….196

Figure6.6EffectsoflargerlocalspreadinThaishrimpfarmingsectors……………………..197

Figure6.7GeographicdistributionsofAHPND-infectedprovincesinThailand…………..198

Listoffigures

18

Figure6.8ROCcurvesofthreetestmodelsidentifyingthepresenceofAHPNDinThai

shrimpfarmingsites…………………………………………………………………………………………………199

Listoftables

19

Listoftables

Table1.1Reviewsofriskfactorsforvibriosisinshrimpfarming…………………………………..38

Table2.1Threewell-knowncompartmentalmodelsandtheirapplicationto

microparasiteinfections…………………………………………………………………………………………….67

Table2.2Centralitymeasuresstudied in fivenetworksof farmedanimalmovements

resultinginoptimalstrategiesfortargeteddiseasesurveillanceandcontrol……………….74

Table3.1CandidateriskfactorsforAHPNDoccurrenceatfarmlevel…………………………..94

Table3.2CriteriausedforclassifyingThaishrimpfarmsintothreescales:small,medium

andlarge…………………………………………………………………………………………………………………...95

Table 3.3 Theoutcome from the telephone survey (Phase1) followedby face-to-face

interviews(Phase2)………………………………………………………………………………………………….103

Table3.4Cross-tabulationofoutcomesforcaseandcontrolsamples……………………….104

Table3.5ThestatisticallysignificantriskfactorsforAHPNDwithoddsratios(𝑂𝑅s)and

95%confidenceintervals…………………………………………………………………………………………107

Table3.6UnconditionallogisticregressionanalysisofriskfactorsforAHPND……………….108

Table3.7Cross-validationresultsontheAHPNDmodelsobtainedfromunconditional

logisticregression…………………………………………………………………………………………………….108

Table 4.1 Degree properties of the live shrimp movement network of Thailand

(LSMN)………………………………………………………………………………………………………………….…140

Table4.2Descriptionofthenumberofconnectionsbetweenseed-producingsitesand

ongrowingsitesbasedontheweighteddegreeoftheLSMN…………………………………….142

Table4.3Descriptionofthenumberofconnectionsbetweenseed-producingsitesand

ongrowingsitesbasedonthenon-weighteddegreeoftheLSMN……………………………..142

Listoftables

20

Table 4.4 Estimated maximum and mean reach, size of giant strongly connected

components(GSCCs),andsizeofgiantweaklyconnectedcomponent(GWCCs)forboth

theLSMNandtherewiredLSMNs…………………………………………………………………………….145

Table 5.1 Source and destination site types of the top 1 000 removals from the

betweenness-based algorithm shown by probabilities (in percentages) in the total

numberofremovals,andinthewholeconnections………………………………………………….172

Table5.2Sourceanddestinationsitetypesofthetop1000removalsfromthesubnet-

crossingbasedalgorithmshownbyprobabilities(inpercentages)inthetotalnumberof

removals,andinthewholeconnections…………………………………………………………………..172

Table6.1Akaike'sInformationCriterion(AIC)valuesoffitteddistributions………………188

Table 6.2 Real pattern of AHPND epizooticswithin shrimp farming sites of Thailand

reportedinJuly2013………………………………………………………………………………………………..193

Chapter1Generalintroduction

21

Chapter 1 - General introduction

1.1 Overviewofthethesis

Farmedpenaeidshrimparethehighestvaluespeciesinworldaquacultureproduction

atapproximatelyUSD22000millionin2014(FAO,2016b).Shrimpisthemostimportant

internationally tradedfisherycommodity inboththeUnitedStatesofAmerica (USA)

and the EuropeanUnion (EU)markets (FAO, 2016a). Further, shrimp farming drives

economic growth for many countries, provides a source of income and better

livelihoodsforproducers,anddevelopsmanyrelatedbusinessesinthewholeshrimp

industry.

Thailand isoneof the top shrimp-producing countries (FAO,2016b),withanannual

productionofaround500000–600000 tonnesbasedon2014 figures (Undercurrent

News,2014),andwith85%of this total soldoutside thecountry (Alam,2015).The

shrimpsupplychain inThailandcomprisesofhatcheries,ongrowing farmers, traders

and brokers, shrimp auction markets and processing plants (Alam, 2015). Farmed

shrimprepresentsoneofthemajoragriculturalproductsdrivingthegrowthinannual

Thai gross domestic product (GDP) from 2.9 % in 2015 to 3.2 % in 2016 (National

Economic and Social Development Board, 2017). Moreover, about 30 % of Thai

labourers are in the agriculture sector (TheWorld Bank, 2015). The shrimp farming

sector is therefore a key element in allowing exporting countries like Thailand to

improvetheirsocialandeconomiccircumstances.

Nevertheless, the growthand sustainabilityof shrimp farming is affectedbydisease

outbreaks, mainly caused by microparasites such as viruses, bacteria, fungi and

protozoans.Thitamadeeetal.(2016)describerecentdiseasesthreateningAsianshrimp

farming,thebiggestsourceofshrimpworldwide.They indicatethatanewemerging

disease,acutehepatopancreaticnecrosisdisease(AHPND)isofmostconcern,together

withthereoccurrenceofviraldiseasessuchaswhitespotsyndromevirusandyellow

headvirus(Thitamadeeetal.,2016).

Importantly, thewidespreadpresenceofdiseases leads to theuseof chemicals and

antibioticsinfarming(Chenetal.,2015;Holmströmetal.,2003;Ricoetal.,2012;Uchida


22

etal.,2016).Theresiduesofthesechemicalsandantibioticsnotonlyhaveapotentially

adverse effect on human health such as causing the development of antibiotic

resistance(Rochaetal.,2016)andsomeareactuallytoxic(Somjetlerdcharoen,2002),

buttheyalsoleadtointernationaltradedisputes(FDA,2016)anddramaticpollutionof

theenvironment(LeandMunekage,2004).Thesenegativeimpactsofchemicalsand

antibioticsonhumanhealth,tradeandtheenvironmentinferthatdiseaseprevention

and controls (e.g. biosecurity, good farm management and disease surveillance

measures)arethebestmanagementinterventionsforshrimpfarming(Brugereetal.,

2017;ChinabutandPuttinaowarat,2005).

Intermsofdiseasepreventionandcontrols,in1998theThaiauthoritieslaunchedthe

NationalDiseaseSurveillanceandMonitoringProgrammeforShrimpFarming(NACA,

2017).Furthermore,themovementsofshrimpbetweensourcesitesanddestination

sites,themostcommonpathwayforsite-to-sitediseasetransmission,arecontrolledby

the aquatic animal trade regulation of Thailand, B.E.2553 (2010). Authorised users

recordreal-lifeliveshrimpmovementsinacomputersystem;datathatisreferredtoin

thisresearchastheLiveShrimpMovementNetworkorLSMN.Althoughtheresultsof

networkmodellingprovideagooddescriptionofdiseasespreadandareutilisedinmost

control programmes (Green et al., 2012; Keeling and Eames, 2005;Werkman et al.,

2011),thedatafromtheLSMNhasneverbeenappliedinnetworkmodellingtoexamine

diseasespread.

In order to protect the Thai shrimp farming from AHPND and other diseases,

epizootiological studies are needed. Four epizootiological questions are therefore

analysed in this research using a variety of tools. The investigated epizootiological

questionsconsistof:

(1)WhataretheriskfactorsforthespreadofAHPNDatfarmlevel?(Chapter3);

(2)Howdoes thenetwork structureof live shrimpmovements influence site-to-site

diseasetransmission?(Chapter4);

(3) How can we identify those live shrimp movements at high risk of disease

transmissionfromsitetosite?(Chapter5);and


23

(4)WhatistheoverallAHPNDprevalencewhenthediseaseiswidespreadthroughboth

long-distanceandlocaltransmission?(Chapter6)

These research outcomes can be used to improve existing disease prevention and

controlmeasuresandregulations, i.e. inrespecttobiosecurity,certificationschemes

forshrimpfarminganddiseasesurveillanceandcontrolprogrammes.Thebackground

ofshrimpfarmingisdescribedinmoredetailinthenextsectioninordertojustifythe

importanceofthisresearch.

1.2 Theimportanceandcurrentstatusoftheshrimpfarmingsector

Giventhattheworldpopulationisexpectedtoriseto12.3billionin2100(Gerlandet

al., 2014), the aquaculture sector has a high potential to produce large amounts of

humanfoodcomparedwiththefisherysector.Ascanbeseeninthestatisticalreportof

theFoodandAgricultureOrganizationoftheUnitedNations(FAO,2016b),thereisa

clearcontrastbetweenaquacultureandcapturetrendsoverthe29yearsfrom1985to

2014: aquaculture productionhas increased gradually from10 to 70million tonnes,

whileproductionfromfishcapturehasremainedstableatc.80–90milliontonnes.

Aquacultureoffersvariousfoodcommodities: fish,crustaceansandmolluscs.Among

the product varieties, penaeid shrimp (tiger shrimpPenaeusmonodon andwhiteleg

shrimpLitopenaeusvannamei)aretwoofthedominantspeciesforinternationaltrade.

They play an important role in food consumption, and drive economic growth and

enhancepeople’slivelihoodinmanyagriculturalcountries(FAO,2016b).Thecountries

shown in Figures 1.1 and 1.2, respectively,were themajor shrimp exporters to the

UnitedStatesofAmerica(USA)andtheEuropeanUnion(EU)in2014(FAO,2015).

Moreover,farmedshrimpisahigh-valueproduct.InJanuary2017,shrimpprices(per

kg)weretwotimesmoreexpensivethanPangasiussp.,acommercialfreshwaterfish

(FAO,2017).Thehighlevelsofincomethatitispossibletomakefromshrimpmeans

that shrimp farming has become widespread (Filose, 1995). The net income, for

example,ofsmall-scaleintensivefarminginIndiawasaround2000USDperhectare

and 9 000 USD per hectare for medium-scale intensive farming (Bhattacharya and

Ninan,2011).


24

These show the importance of shrimp farming for global food supply and socio-

economicstatus.Whendiseaseiswidespread,therefore,thehugeeconomiclossisthe

obviousoutcomethroughoutthesupplychainoffarmedshrimp.

Figure 1.1 Volume and percentage of shrimp products (raw shrimp and value-added shrimp) imports to the USA market by the major producing countries in 2014 (unit: thousand tonnes). The total shrimp imported into USA was around 570 thousand tonnes. Thailand was the fifth-largest supplier (FAO, 2015).

Figure 1.2 Volume and percentage of shrimp products (raw shrimp and value-added shrimp) imports to EU markets by the major producing countries in 2014 (unit: thousand tonnes). The total shrimp imported into the EU was around 790 thousand tonnes. Thailand was the 14th-largest supplier (FAO, 2015).

6.7

8

11.8

17.9

20.2

32.5

64.6

73.8

92.5

103.4

108.8

0 20 40 60 80 100 120

Guyana

Honduras

Peru

Malaysia

Mexico

China

Thailand

Vietnam

Ecuador

Indonesia

India

Thousandtonnes

(5)Thailand

Coun

tries

(19.1%)

(18.2%)

(16.3%)

(13%)

(11.3%)

(5.7%)

(3.5%)

(3.1%)

(2.1%)

(1.4%)

(1.2%)

15.618.218.8

22.725

28.835.535.8

40.744

49.755.1

66.283.2

93.1

0 20 40 60 80 100 120

IndonesiaThailandGermanyBelgium

SpainChina

NetherlandsCanada

BangladeshDenmarkVietnam

GreenlandArgentina

IndiaEcuador

Thousandtonnes

Coun

tries

(14)Thailand

(11.8%)(10.5%)

(8.4%)(7%)

(6.3%)(5.6%)(5.1%)

(4.5%)(4.5%)

(3.6%)(3.2%)(2.9%)

(2.4%)

(2%)(2.3%)


25

1.3 Thesupplychainoffarmedshrimp,relatedregulationsforcontrollingproductionandsocialcommunitiesofshrimpproducersinThailand

ThissectionoutlinesthesupplychainofThaishrimpproduction.ItalsodescribesThai

regulations(i.e.movementcontrols,sitecertificationandfarmingregistration),andthe

socialcommunitiesofshrimpproducersinThailandthathavebeenincorporatedinto

ourresearch.

The supply chain for farmed shrimp in Thailand is simple (Figure 1.3). For hatchery

productionofshrimpseed,thewildbroodstockofPenaeusmonodoniseithercaptured

from the sea or cultured in a breeding programme, whereas using broodstock of

Litopenaeusvannameifromabreedingprogrammeisacommonpractice(Lebeletal.,

2010).Alam(2015)andUddin(2008)demonstratethattheshrimpindustryinThailand

entailsthreestepsforpassingtheproductbetweenahatcherysiteanddomesticand

globalmarkets.Inthisresearch,however(Chapters4,5and6),thenetworkmodelling

ofdiseaseepizooticshasbeen focusedonthe transmissionbetweenseed-producing

sites (hatcheries and nurseries) and ongrowing sites, a process that denotes a large

numberofliveshrimpmovementsforfarming.Theremainingsteps(i.e.movementsof

chilledorfrozenshrimpfromongrowingsitestotraders,brokers,processingplantsor

auctionmarkets)shouldnotposeariskofspreadingtoshrimpfarmingsites.

Figure 1.3 The shrimp production chain in Thailand (modified from Alam, 2015 page 103). The live shrimp movement data used in the research demonstrate the movements of live shrimp (shrimp seed) between hatchery, nursery and ongrowing sites, as shown in the box.

Hatcherysite

Nurserysite

Ongrowing site

Trader/broker

Processingplant

Auctionmarket

Domesticmarket

Globalmarket

Broodstock

Seed

Sea

Seed

Seed


26

Uddin (2008) indicates that thesupplychain for farmedshrimpstarts fromhatchery

sites,whichproduceshrimpseedforongrowingsites. Insteadofdirectsellingtothe

ongrowing sites, hatchery sites pass some of their production to nursery sites at

nauplius stage or initial postlarval stage. Then, nursery sites rear the seed from the

naupliusuntilthepostlarval(PL)stage(mostlyPL10;rearedfor20days)beforeselling

theproductiontotheongrowingsites(FAO,2014).Commonly,theproductionperiod

for whiteleg shrimp (L. vannamei) is 105–120 days at pond level with a density at

400000–500000shrimpperhectare,obtainingaharvestsizeof21–25gforprocessing

plants(Wyban,2007).Someproducerspracticeahigherdensitystockingof900000–

1200000shrimpperhectareforatargetedsizeof12–18gincaseofpartialharvest,

and24gforfinalharvest.Currently,shrimpfarmingcangenerateaproductioncapacity

of twoor threecyclesperyear (Limsuwan,2009).Shrimpfarmingwithaproduction

capacityofthreecyclesperyeartendstohaveahighriskofdiseasesbecauseafallow

periodtotreatanddisinfectpathogensisshorterthanthatoffarmingoneortwocycles

peryear(Cocketal.,2009;Muniesaetal.,2015).

The movements of live shrimp in each of the steps mentioned above are closely

recordedintheliveshrimpmovementrecord(KongkeoandDavy,2010;Yamprayoon

andSukhumparnich,2010).Thisrecordfollowstheaquaticanimaltraderegulationof

Thailand,B.E.2553 (2010).All producersmust inform theproperauthorities (i.e. the

ThailandDepartmentofFisheriesstaffandtheirrepresentatives)aboutthemovements

of shrimp. Insteadof apaper-based system to collect the shrimpmovementdata, a

computer-basedsystemhasbeenusedbyThaiauthoritiessinceMarch2013andsuch

electronic records subsequently are printed on a paper for checking (Songsanjinda,

2013).Moreover,allshrimpfarmingsitesmustberegisteredlegallyandtheirfarming

management practices should be inspected under the governmental certification

schemes(KongkeoandDavy,2010).Farmingstandardsandcertificationaredeveloped

toenhance foodsafetyonaquacultureandsustainability includingenvironmentand

livelihoodreasons(Corsinetal.,2007;Piumsombunetal.,2005;Pongthanapanichand

Roth,2006;YamprayoonandSukhumparnich,2010).Thecriteriaforfarmingstandards

includegoodhealthmanagementof farmedshrimp,diseasepreventionandcontrol,


27

andtheapplicationofmovementdocumentsfortraceabilityintheshrimpproduction

chain(NationalBureauofAgriculturalCommodityandFoodStandards,2014).

As a regulatory requirement for the control of shrimp production, the live shrimp

movementrecordinThailandprovidesnewandusefuldataforepizootiologicalstudies.

Thisofficialrecordof liveshrimpmovementscanhelptoindicatepotentialroutesof

infectiousdiseasetransmissionfromsitetosite,andsince2013thesedataaremore

readilyobtainedandanalysed.TheresearchpresentedinChapters4,5and6isthefirst

studytousethisdatasourceforshrimpepizoology,however.Moreover,priortothis

thesis,thereisnoevidencethattheliveshrimpmovementdatahasbeenutilisedasa

partofdiseasepreventionandcontrolintheThaifarmingcertificationscheme(i.e.for

farmmonitoringprogramme).

Inaddition,socialcommunitiesofshrimpproducers,i.e.shrimpfarmerclubs,havean

influence on disease prevention and controls. Shrimp farmer clubs support better

farmingpractices(Kassametal.,2011),contributingtoadecreaseofdiseaseoutbreaks

(KongkeoandDavy,2010).Importantly,theannualconferencesarrangedbytheseclubs

generateanexchangeoftheideasbetweenproducersandhelptoimproveknowledge

aboutfarmpracticesandshrimphealthmanagement.WhenthecausalagentofAHPND

remainedunknown, the Thai shrimp farmer clubsparticipated in settingup suitable

broodstock feedingpractices inhatcheries. Shrimp fry frombroodstock treatedwith

non-livefeedsbecameakeyagreementbetweenshrimpsellersandbuyers(Suratthani

Shrimp Farmers Club, 2014). Consequently, the role of polychaete worms, bivalve

molluscs and other live feeds in disease transmission to farmed shrimp could be

decreased.Hence,socialfarmingcommunitieshaveparticipatedintheeffectivenessof

diseasecontrolstrategiesinThailand.

1.4 ThecharacteristicsofshrimpfarminginThailand

Figure 1.4 shows the approximately 20 000 shrimp farming sites in Thailand that

together generate up to 600 000 tonnes of shrimp production annually (2011).

Productiondecreasedsubstantiallyin2013,however,mainlyduetodiseaseproblems

(FAO,2013a).ThefigurealsoillustratesaneweraintheThaishrimpindustryin2003,


28

whenwhitelegshrimpwereintroducedtotheThailandin2003.Theintroductionofnew

shrimpspecies,togetherwiththedevelopmentofnewtechnologiesandinnovations,

ledtoalargeincreaseintotalproductionwithasmallernumberofshrimpfarmingsites.

Thesesitesadoptmoreintensivesystems,whichoftendeveloppoorwaterqualityand

stressfulconditioninfarmedshrimp(Kautskyetal.,2000).

Figure 1.4 Overview of shrimp farming in Thailand for two major shrimp species: tiger shrimp and whiteleg shrimp, 1999–2013. The left axis of the graph presents the number of ongrowing sites. The right axis shows the yield (in tonnes) of farmed shrimp production (modified from Thailand DoF, 2016).

Regarding the geographic location of shrimp farming sites in Thailand (Figure 1.5),

Thailandhasanapproximately2600km-longshorelinealongtheGulfofThailandand

theAndamanSea(Tookwinasetal.,2005),withalargenumberofshrimpfarmingsites

are intensively established along these coastal areas. The remainder, called inland

farmingsites,aresituatedawayfromtheshoreline,drawingwaterfromrivers,canals

orlakes.Notonlyarethesewaterbodiesthesourceofwaterforfarming,buttheyare

alsousedfordischargeofnutrientsand,concomitantly,pathogensduringwater-pond

discharge (Barraza-Guardado etal., 2013;Marchand etal., 2014). Thismeans thata

groupofneighbouringsitesthatsharenaturalresourceshaveasharedriskofdisease

transferthroughhydrologicalconnectivity.

0

200000

400000

600000

800000

0

10000

20000

30000

40000

1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013Years

40 000

30 000

20 000

10 000

800 000

600 000

400 000

200 000

00

Numbe

rofo

ngrowings

ites

Farm

edsh

rimpyield(to

nne)

Whitelegshrimp(2003)

AHPND(late2011)

Totalshrimpyield(tonne)Tigershrimpyield(tonne)Whiteleg shrimpyield(tonne)


29

Figure 1.5 Distribution of Thai shrimp farming sites by province. Data summarised from the live shrimp movement data between March 2013 and March 2014 and figure illustrated using the tmap package, in the 𝑹 Programme Environment (R foundation for statistical computing, 2015). The data were also used to construct network models in Chapters 4, 5 and 6.

The different geographic locations of shrimp farming sitesmean that variouswater

salinitylevelsareusedforrearingshrimp.Ahigh-salinitysystem,withasalinityrange

between10–30ppt,isappliedtoshrimpfarminglocatednearthecoast,whereasthe

inlandfarmingoperatesunderalow-salinitysystemof3–4ppt(Flahertyetal.,2000;

Tookwinas et al., 2005). Importantly, low salinity farming causes adecreased innate

immuneabilityinshrimpandalsoreducestheirresistancetobacterialdiseases(Wang

andChen,2006;WangandChen,2005).Thesalinityparameterisonlyoneofanumber

ofenvironmentalfactorsthataffectsusceptibilitytoinfectioninshrimpfarmingsites.

The occurrence of diseases, however, is dependent on the nature of the disease

(microparasiticormacroparasitic)anditstransmission,whichcanbeexplainedinterms

oftheepidemiologicaltriad(Section1.5.1).

GulfofThailand

AndamanSea ShrimpfarmingindexNon-shrimp farmingprovinces

0to500sites500to1000sites1000to1500sites1500to2000sites


30

1.5 Infectiousdiseasesinfarmedshrimpandtheirtransmission

MicroparasiteshavebeenamajorcauseofseverediseaseoutbreaksinAsianshrimp

farmingoverthepast30years(Flegel,2012).Sixtypercentof lossesareduetoviral

diseases,20%duetobacterialdiseases(particularlyvibriosisasdescribedinDisease

box1),andtheremaininglossesduetootherparasites(Flegel,2012).Vaccinationand

immunostimulants are a challenge for inhibiting disease spread in shrimp farming

(Johnson et al., 2008; Namikoshi et al., 2004). An example of successful vaccines is

provided on vaccination trials with P. monodon to induce resistance to white spot

syndromevirus(Vaseeharanetal.,2006). Nevertheless,thereisevidencethattheiruse

rarely succeeds in general farming for two reasons: the lackof anadaptive immune

systeminshrimp,andthepresenceofmultiplepathogenswithinsites(Camposetal.,

2014;Cocketal.,2009;Gräslundetal.,2003;GräslundandBengtsson,2001;Supungul

etal.,2015).Epizootiologicalstudiestodescribetheoccurrenceofdiseasesandtheir

transmissionarethereforecurrentlyrequiredintheshrimpfarmingindustry.

Toincreaseunderstandingofthedynamicsofdiseaseepizooticsinshrimpfarming,the

natureofmajorshrimpdiseases,i.e.vibriosis,whitespotdisease,yellowheaddisease,

infectioushypodermalandhaematopoieticnecrosis,acutehepatopancreaticnecrosis

disease and taura syndrome, together with their routes of transmission, are briefly

outlinedinthissection(Diseaseboxes1–6andAppendixA).Importantly,inChapter3

“Evaluating risk factors for transmission of acute hepatopancreatic necrosis disease

(AHPND) in the Thai shrimp farming sector”, the visible clinical signsof thesemajor

shrimpdiseaseshavebeenusedinthedifferentiationoftheAHPND-affectedsitesfrom

sitesaffectedbyotherdiseases,andthusitisimportanttoexplainthoseclinicalsigns

here.


31

Diseasebox1:Vibriosis

Vibrio species that commonly cause vibriosis in farmed shrimp are, for example,

V.alginolyticus,V.harveyi,V.parahaemolyticusandV.penaeicida(Saulnieretal.,2000).

ClinicalsignsofinfectedshrimpvarywiththetypeofVibriospp.Infectedshrimpcanhave

paleandopaquemusclesandblackstripesonthelateralcephalothoraxpresentininfected

shrimp(Longyantetal.,2008).Abright-redsyndromeiscausedbyV.harveyi,wherethe

infected shrimp shows red discoloration spots on the abdomen (Soto-Rodriguez et al.,

2010).ShrimpinfectedwithV.choleraeshowanexteriorvisualappearanceoflegyellowing

(Caoetal.,2015).Vibriosisleadstolowsurvivalratesinhatcheriesandongrowingsites.In

many cases, outbreaks of vibriosis have caused mass mortality among small shrimp in

hatcherysites,suchasinTaiwanin1994(Liuetal.,1996),andChinain1995(Vandenberghe

et al., 1998). Vibriosis also largely occurred in farmed shrimp in Sri Lanka in 2010

(HeenatigalaandFernando,2016).

Diseasebox2:Whitespotdisease

Whitespotdisease(WSD)isaninfectionofshrimpbywhitespotsyndromevirusorWSSV

(OIE,2013b).WSDwasfirstreportedinTaiwanin1992(Chouetal.,1995).WSDspreadhas

amongmanyoftheshrimpproducingcountries inAsia (e.g.China, Japan,Korea,Thailand,

IndiaandBangladesh)withinoneortwoyearsofitsfirstdetection(Escobedo-Bonillaetal.,

2008),andinNorthAmericain1995(Lightner,1999).MostshrimpinfectedwithWSSVexhibit

white spots on their exoskeleton and lesions on the cephalotholax (Cheng et al., 2013;

Rodríguez et al., 2003), however poor water quality such as high alkalinity or bacterial

diseasemayalsocausethesewhitespots(OIE,2013b).WSDinfectionresultsinhighmortality

infarmedshrimp,upto100%withinoneortwoweeks(Rajendranetal.,1999;Wuetal.,

2005).


32

Diseasebox3:Yellowheaddisease

Yellowheaddisease(YHD)isaninfectionofshrimpbyyellowheadvirusorYHV(OIE,2013b).

The first YHD epizootic was identified in Asia in 1990 (Walker and Winton, 2010). The

economic losses from YHD outbreaks were reported at an estimated USD 3 million in

Thailand(2008),forexample(Senapinetal.,2010).Grosssignsofdiseasedshrimpinclude

yellowishcolourationandswollencephalothorax(Lio-Poetal.,2001).Cumulativemortality

of 60–70%has been reported amongP.monodon and L. vannamei cultured in earthen

ponds(Senapinetal.,2010).

1.5.1 Theoccurrenceofdisease

Anunderstandingofdiseaseoccurrenceisimportanttoshrimpepizoology.Alldiseases

aremultifactorial,asmanifestedbytheepidemiologicaltriadmodel(Figure1.6).The

diseasesexistwhenthereisinteractionbetweenpathogen,hostandtheenvironment.

Figure 1.6 The epidemiological triad (from Rockett, 1999 page 10).

Pathogens, in termsofmicroparasites suchasvirusesandbacteria, causenumerous

infectionsinshrimpfarming,despitethefactthatmanycontrolmeasurestoprevent

microparasitic infectionshavebeendevelopedand implemented.Microparasitesare

distinguished from macroparasites by their small size, the short time required to

completetheirgeneration,andtheirhighability fordirect reproductionwithinhosts

(Anderson and May, 1981). Thus, the incidence and prevalence of disease due to

microparasites often tends to be high, particularly in farming conditions like shrimp

rearing.Oneoftheimportantmeasurestopreventmicroparasiticinfectionsinshrimp

EnvironmentHost

Pathogen


33

farming is the use of shrimp seed produced under specific pathogen free (SPF)

conditions(Lightneretal.,2009;Mossetal.,2012).Thismeasure,however,maynotbe

sufficientduetothecomplexityinthelifecyclesofpathogens.Forexample,ifWSSVis

latent,WSD-infectedshrimpmaynotbedetectedwithacommercialiseddiagnostictest

(HeandKwang,2008;Khadijahetal.,2003).Inaddition,inSPFconditions,shrimpare

only free from specifically targeted pathogens, and thus remain at risk from non-

targetedorunknownpathogen(Barmanetal.,2012).Hence,stockingSPFshrimpisjust

onemeasureforpreventingandcontrollingdiseaseoccurrenceinshrimpfarming.The

pathogenmaynotbedetectedinallcasesofSPFshrimp,andsuchshrimpfarmingsites

canbeinfectedviaotherpathways.

Shrimparehostsofdiseasesinthisthesis.Shrimparesusceptibletoawidevarietyof

pathogens,andespeciallytoviruses(Lightner,2011;Thitamadeeetal.,2016).Shrimp

susceptibility to a particular disease is affected by the species concerned (Bell and

Lightner, 1984; Lightner et al., 1998;Overstreet et al., 1997), tolerance to infection

(Hameedetal.,2000;Witteveldtetal.,2006),andlifehistorystage(Aguirre-Guzmánet

al., 2001; Sudha et al., 1998). Many severe viral outbreaks in shrimp are due to

persistent infections at a low level (Walker andWinton, 2010). Shrimp can also be

infectedbymultiplepathogenssuchasbyhepatopancreaticparvovirusandmonodon

baculovirus(Flegeletal.,2004).Thisevidencedemonstrateskeyconditionsofdisease

occurrence,although,inreality,avarietyofenvironmentalfactorsinfluencethehealth

offarmedshrimp.

Themajorenvironmentalfactorsaffectingshrimpfarminghavebeenevaluatedfortheir

association with diseases. Environmental factors that affect shrimp health include

climate (Piamsomboon et al., 2016), seasonality (Boonyawiwat, 2009) and storms

(Zhangetal.,2016). TheexperimentsinRahmanetal.(2006)andRahmanetal.(2007)

showedthatahighwatertemperatureof33°Cwasrelateddirectlytoareductionin

shrimpmortalityfromWSD(Diseasebox2),comparedwithatemperatureof27°C.The

lowest(0.5%)andhighestsalinitylevels(5.4%)wereassociatedwithhighWSD-related

mortalityinshrimp(Ramos-Carreñoetal.,2014).Environmentalfactorsmayalsotrigger

diseasereoccurrence,suchas inthecaseof infectionwith IHHN(Diseasebox4)and

AHPND(Diseasebox5).AshrimppondinfectedwithIHHNhasaprobabilityofrepeated


34

IHHNonsetwhenthetemperaturewasbelow24°C(Montgomery-Brocketal.,2007).In

addition,ahighpHlevel,at8.5–8.8,wasidentifiedasassociatedwithrepeatedAHPND

onsetinaffectedshrimppondsinMalaysia(AkazawaandEguchi,2013).Thisevidence

demonstrates that the environment affects the susceptibility to disease of farmed

shrimp.

Diseasebox4:Infectioushypodermalandhaematopoieticnecrosis

Infectious hypodermal and haematopoietic necrosis (IHHN) is infection of shrimp by

infectioushypodermalandhaematopoieticnecrosisvirus(OIE,2013b).Thefirstoccurrence

of IHHNinfarmedshrimpwasobservedinHawaii inthe1980’sduetothemovementof

IHHN-infectedpostlarvaefromCentralorSouthAmerica(Lightneretal.,1983a).Thecuticle

of the shrimp is found to be whitish in diseased shrimp and, generally, therewas high

mortality (> 80%)after shrimpmoulting (Lightner et al., 1983b). IHHN-infected shrimp

often exhibit runt-deformity syndrome, resulting in a reduced growth rate, and a high

presenceofrostrum,antennaeorcuticledeformity(Chayaburakuletal.,2005;Kalagayanet

al.,1991).

Diseasebox5:Acutehepatopancreaticnecrosisdisease

Acute hepatopancreatic necrosis disease (AHPND) occurs when shrimp are infected by

specificstrainsofVibrioparahaemolyticusbacteria(Tranetal.,2013).Further,thebacteria

produceatoxinthatdamagesthehepatopancreasofshrimp(Laietal.,2015).AHPNDhas

beeninvadingmajorshrimpproducingareas:China(2009),Vietnam(2010),Malaysia(2011),

Thailand (2011/2012),Mexico (2013),and,most recently, thePhilippines (2015). Itcauses

high mortalities, up to 100 % within 35 days post stocking (Eduardo and Mohan, 2012;

Kasornchandra, 2014). Gross signs can be seen in ongrowing ponds, i.e. empty gut and

stomach,andpaleandatrophiedhepatopancreasofaffectedshrimp(NACA,2014).


35

1.5.2 Riskfactorsforvibriosisoutbreaksinfarmedshrimp

The link between evidence for vibriosis in varied conditions and the actual field

conditionsandfarmingpracticesthat leadstoAHPNDarereviewedhere(Table1.1).

The geographic location of the farm has been suggested as an important factor in

increasingtheproductivityofshrimpfarmingbutalsointhesusceptibilitytoinfection

of shrimp farming sites (Zhu and Dong, 2013). Some examples of the relationship

betweenthelocationoffarmingsitesandthevibriosisaregiveninTable1.1(part1).

Gopaletal.(2005)foundthatthenumbersofVibrioinshrimpfarmingofinIndiawere

higheralongthewestcoastthanthesouthcoast,ataround102cfupermlwater.The

establishmentoffarmsnearhumancommunitiesalsoincreasedtheriskofvibriosisdue

to largeamountsofheavyorganicmaterial fromthehumancommunityflowing into

naturalsources(Mohneyetal.,1994;ReillyandTwiddy,1992).Farmslocatedcloseto

estuaries commonly facedwidely fluctuating salinity levels,withhigh fluctuationsof

salinity from 35 % to 5 and 15 % being associated with an increased risk of

V. alginolyticus infection in farmed whiteleg shrimp (Wang and Chen, 2005). Farms

established in, or close to, agriculture areas risked contamination from methyl

parathion (pesticides) that led to increased susceptibility to V. parahaemolyticus

infectioninshrimp(Roqueetal.,2005).

Climateappearstoinfluenceoutbreaksofvibriosisamongshrimpfarms.Thewetseason

hasbeenrelatedtothegrowthofV.cholerae(ReillyandTwiddy,1992),whilecooler

temperatures,at20°C,havebeenrelatedtoanincreasedriskofV.penaeicidaamong

cultured blue shrimp P. stylirostris (Goarant et al., 2000); in turn, warmer water

(changedfrom27to32°C)hasbeenrelatedtoanincreasedriskofV.alginolyticusfor

culturedL.vannamei(Chengetal.,2005).

Themostimportantfactorsthatappeartoinfluencevibriosisoutbreaks,however,are

inappropriate farming management practices. Shrimp fed untreated Artemia risked

V. parahaemolyticus and V. harveyi infections (Quiroz-Guzmán et al., 2013). High

phytoplankton dynamics incurred an abundance of Vibrio spp. within shrimp ponds

(Lemonnier et al., 2016). Tho et al. (2012) demonstrated that sediment provided a

bettermicroenvironment for Vibrio thanwater and,particularly in the rainy season,


36

pondsedimentprovidedthebesthabitatforV.nigripulchritudo(Goarantetal.,2006;

Wallingetal.,2010)andV.cholerae(Lekshmyetal.,2014).Asanattempttodecrease

thelargeamountofpondsediment,therefore,generallyearthenpondshavebeenlined

withplasticsheets.

A linedpond is considered tobebetterpractice in termsof reducing thevolumeof

sediment.Thisisatechniquethatisgenerallyapplicabletointensiveshrimpfarming.

ReillyandTwiddy(1992),however,havedemonstratedthatintensivefarmingsystems

increasedtheriskofV.cholerae infectioninfarmedshrimp.Rearinghighnumbersof

shrimpwithinthelinedpondsledtoahighnutrientconcentrationinthewaterpond,

whichisimportantasariskfactorforvibriosis(Funge-SmithandBriggs,1998).Thathigh

abundance of V. choleraandV. parahaemolyticuswas caused by the heavy organic

matterwithinshrimppondswasproposedbyGaneshetal.(2010).HighpHlevels(>7),

highsalinity(>0.5%)andhighammoniaarealsorelatedtoVibrioabundanceinshrimp

ponds (Heenatigala and Fernando, 2016; Lekshmy et al., 2014; Liu and Chen, 2004;

LokkhumlueandPrakitchaiwattana,2013).AhighriskofVibriodiseaseswasdetectedwhenever shrimp ponds lacked diversity of Vibrio communities (Sung et al., 1999).

Comparedwithlinedponds,earthenpondscouldbefullypreparedbydrying,harrowing

andfillingwithprobioticBacillus,toreducetheriskofVibrio(Moriarty,1998;Nimratet

al.,2008).

Vibriosisisawaterbornediseaseandextractionofculturewaterfromtheseahasbeen

identified as a risk factor in terms of increasing the presence of V. harveyi and

V. splendidus (Lavilla-Pitogo et al., 1990), demonstrating the importance of water

treatmentat thebeginningof stocking.Although farmsoftenuse recirculatedwater

systems(meaningnoextractionofculturewaterfromthesea),Vibriocouldstillgrow

rapidly,principallyduetothepoorqualityofthewaterreusedinthefarms(Colt,2006).

Thisispossiblyrelatedtothepoorplanningoftherecirculatingsystems,whichaidsthe

transmissionofdiseasesfrominfectedpondstoothers(Funge-SmithandBriggs,1998;

Mugnieretal.,2013).

Vibriosisinshrimpfarmingoftenco-occurswithotherpathogens.Whenfarmedshrimp

areinfectedbyWSD,Kannapiranetal.(2009)aswellasSelvinandLipton(2003)found


37

thattheriskofV.harveyiandV.alginolyticusinfectionisincreased.Recently,therehas

beenhighprevalenceofwhitefecesdisease(WFD;aprotozoaninfection)andAHPND

incidenceatthesameshrimpfarmingsites(Limsuwan,2010;Sriurairatanaetal.,2014).

Somboonetal.(2012)indicatedthatthehaemolymphandintestineofWFD-infected

shrimphashighnumbersofV.vulnificus.

Anthropogenic factors are shown to be associated with vibriosis in farmed shrimp.

Examplesoftheseanthropogenicfactorsincludethelackofpathogen-freebroodstock

screening,andtheuseofequipmentorfacilitieswithoutdisinfection(Chrisoliteetal.,

2008).

Theriskfactorsforvibriosisoutbreaksinfarmedshrimpdescibedabovewillbeusedin

designingourwork(Chapter3).


38

Table 1.1 Reviews of risk factors for vibriosis in shrimp farming

Riskfactor Hostspecies Pathogenspecies

(1)Farminglocation

Differentgeographiclocation(betweenthewestandtheeastcoastalareaofIndia)

Penaeusspp. Vibriospp.(Gopal,2005)

Farminglocationnearhumancommunities

Penaeusspp. V.parahaemolyticus,V.vulnificus,V.alginolyticus(Mohneyetal.,1994),andV.cholerae(ReillyandTwiddy,1992)

Farminglocatedclosetoestuaries L.vannamei V.alginolyticus(WangandChen,2005)

Farminglocatedinorclosetoagriculturalareas

L.vannamei V.parahaemolyticus(Roqueetal.,2005)

(2)Climate

Cooltemperature(at20°C) P.stylirostris V.penaeicida(Goarantetal.,2000)

Warmtemperature(at32and34°C) L.vannamei V.alginolyticus(Chengetal.,2005)

Wetseason Penaeusspp. V.cholerae(ReillyandTwiddy,1992)

(3)Farmingmanagementpractices

FeedingshrimpwithuntreatedArtemia

Penaeusspp. V.parahaemolyticusandV.harveyi(Quiroz-Guzmánetal.,2013)

Highphytoplanktondynamics P.stylirostris Vibriospp.(Lemonnieretal.,2016)

Intensiveshrimpaquaculture Penaeusspp. V.cholerae(ReillyandTwiddy,1992)

Stressedshrimpandlargeamountsofthepathogeninponds

P.stylirostris V.nigripulchritudo(Mugnieretal.,2013)

Largeamountsofpondsediment Penaeusspp. V.nigripulchritudo(Goarantetal.,2006;Wallingetal.,2010),V.cholerae(Lekshmyetal.,2014),andVibriospp.(Thoetal.,2012)

Heavyorganicmatter Penaeusspp. V.choleraeandV.parahaemolyticus(Ganeshetal.,2010)

HighpH(>7)andhighsalinityandammonialevels

P.monodon

V.alginolyticus,V.parahaemolyticus,V.damsela,andV.anguillarum(HeenatigalaandFernando,2016)


39

Table 1. 1 (cont.)

Riskfactor Hostspecies Pathogenspecies

Highersalinitylevel L.vannamei V.parahaemolyticus(LokkhumlueandPrakitchaiwattana,2013)

Highconcentrationofammoniainwater

L.vannamei V.alginolyticus(LiuandChen,2004)

Noapplicationsofeitherponddryinginsunlightorpondharrowing

Penaeusspp. Vibriospp.(Nimratetal.,2008)

DecreasesinthediversityoftheVibriocommunity

P.monodon Vibriospp.(Sungetal.,1999)

Usingwatersourcedfromthesea P.monodon V.harveyiandV.splendidus(Lavilla-Pitogoetal.,1990)

(4)Viralandprotozoandiseaseoutbreaks

OutbreaksofWSD P.monodon V.harveyi(Kannapiranetal.,2009),andV.alginolyticus(SelvinandLipton,2003)

OutbreaksofWFD L.vannamei V.vulnificus(Somboonetal.,2012)

(5)Anthropogenicactivities

Transmissionofpathogensviabroodstock,maturationandspawningfacilitiesintheshrimphatchery

P.monodon V.harveyi(Chrisoliteetal.,2008)


40

1.5.3 Site-to-sitetransmissionofshrimpdiseases

Inshrimpfarming,thefollowingmodesappeartobetheimportantroutesinsite-to-site

diseasetransmission:long-distanceviamovementsofliveshrimp(eitherbroodstockor

shrimp seed), local spread via closeproximityof sites, and sharingofwater courses

(Figure1.7).However,notalltransmissioneventsfitthispatternandotherroutesof

transmissionexist.

Figure 1.7 Major routes in site-to-site transmission of shrimp diseases

Themost common route for transmitting diseases from site to site is long-distance

transmission.Withlong-distancetransmission,asusceptiblesitecanbeinfectedviathe

importationofinfectedshrimp.ThepandemicsofWSD,YHD,IHHNandTS(Diseasebox

2–4and6,respectively)infarmedshrimpintheUSAwereobviousexamplesofnational

epizooticsresultingfromthemovementsofliveinfectedshrimpseedandbroodstock

(Lightner,2003;Lightneretal.,1997).ItincludesthepotentialtransmissionofAHPND,

anewemergingoutbreakinshrimpfarming(OIE,2013a).

Note that this research (Chapters 4, 5 and 6) focuses on the domestic epizootics in

shrimp farming, where all site-to-site movements of live shrimp can serve as long-

distance transmission routes at the country scale. Although movements of frozen

shrimpproductscanleadtosomeinfections,e.g.WSSVandYHV(Nunanetal.,1998),

thesearenotincludedinthisresearch.

Hatcherysite

Nurserysite

Ongrowing site

SeaLong-distancetransmission

viashrimpseedandbroodstock movements

Unknowntransmission

Localtransmission

Localtransmission

Unknowntransmission


41

Diseasebox6:Taurasyndrome

Taurasyndrome(TS)istheinfectionofshrimpbytaurasyndromevirus(OIE,2013b).Itwas

firstdetectedinfarmedshrimpinEcuadorin1992(Chaivisuthangkuraetal.,2016).TSwas

observedinTaiwanin1998duetointroducingTS-infectedshrimpfromepizooticcountries

(Tuetal.,1999).Diseasedshrimphaveapalereddishcolouration,witharedtailfanand

pleopods,andaresoft-shelled (Bonamietal.,1997;Lightneretal.,1995;Songetal.,2003).

Acutemortalityofshrimpispossiblewithinthreedays(ChunIandYenLing,2000).

Anotherpathwayforsite-to-sitediseasetransmission isvia localspread,whichoften

occursduetoanthropogenicactivities.InThailand,afewlarge-scalefarmsapplyclosed

recirculatingsystems inshrimprearing.Theremaining farms,however,mayconduct

waterdischargeorwaterexchange(Boydetal.,2017;Flahertyetal.,2000;Yaemkasem

et al., 2017). Although water exchange results in a decrease in the ammonia

concentration in shrimp ponds (Hopkins et al., 1993), where the exchange occurs

directlybetweenshrimppondsandnaturalwatercourses(e.g.canals,lakes,riversand

thesea),withoutproperwatertreatment, thiscancontributetowidespreaddisease

through hydrological connectivity (Anh et al., 2010; Pruder, 2004; Tendencia et al.,

2011). Additionally, the absence of installation of crab and bird fencing in shrimp

farmingcanaidlocalspreadofdiseaseduetophysicalproximityofsites,asdescribed

inBalakrishnanetal.(2011)andKumaran(2009).

Themovementof fomites (inanimateobjects such as farming facilities, vehicles and

staff’s clothes) aids local and long-distance transmission by introducing diseases to

susceptible sites (Rodgers etal., 2011).Corsin etal. (2005) found that therewasno

strongassociationbetweenpotentialfomites(sharingfarmingfacilitiesandstaff)and

disease spread in shrimp farming, despite fomites often being subject to disease

mitigationmeasures, such as the sanitation of incoming vehicles anddisinfection of

facilitiesinfarming(Bondad-Reantaso,2016;Dvorak,2009;Mohanetal.,2004;Yanong

andErlacher-Reid,2012).


42

For modelling purposes, the final mode for disease transmission between sites is

unknown transmission with a usual exposure to risk factor but in which a major

transmissionpathwayisnotidentified.

Thenatureofshrimpdiseasesandtheirtransmission,aspresentedinthissection,are

importantinforminganepizootiologicalstudyforAHPNDandotherdiseases.Crucially,

a better understanding of the occurrence and transmission of infectious diseases in

farmedshrimpcanbeachievedwithmodellingapproaches.Amongthesemodels,this

studyisinterestedinnetworkmodelsandcompartmentalepidemicmodels,whichare

discussedinChapter2“Epidemiologicalandepizootiologicaltoolsfordesignofdisease

preventionandcontrolstrategies”.

1.6 Researchoutline

Thepurposeofthisresearchistodevelopepizootiologicaltoolstoevaluatethespread

ofAHPNDintheThaishrimpfarmingindustry.Itisimportanttoknowtheeconomicand

sociologicalimportanceofshrimpfarming,thecharacteristicsofThaishrimpfarming,

the occurrence of diseases and their transmission, and the disease prevention and

controlmeasuresthathavebeendeveloped.Thethesis,therefore,includesareviewof

the literature on the epidemiological and epizootiological tools for various sectors,

allowingthedevelopmentofeffectivetoolstopreventandcontrolthespreadofdisease

spreadindividualsiteandcountrylevels.

This first chapterhasbeenwritten to give a general introduction toepizootiological

toolsforAHPNDandotherknowndiseasesintheThaishrimpfarmingsector,andto

demonstratetheresearchoutlinehere.Theremainderoftheresearchaimsto:

- Investigate the risk factors for acute hepatopancreatic necrosis disease

(AHPND);

- Demonstratethestructureoftheliveshrimpmovementnetwork(LSMN),which

poses potential for site-to-site transmission of AHPND and other known

infectiousdiseases;


43

- Identify connections in the LSMNposing ahigh risk for disease transmission,

leadingtowardsthedevelopmentofdiseasesurveillanceandcontrolalgorithms

forThaishrimpfarming;and,

- Model thedynamicsofAHPNDepizootics in shrimp farming sites,where the

modelconsidersbothlong-distanceandlocaltransmission.

Theresearchhasbeendivided intosevenchapters;anoverviewofthechaptersand

theirlinkageisgiveninFigure1.8.

Thefollowingchapter(Chapter2)discussestherelevantliterature.Itexplores:(1)study

designsinepidemiologyandepizoology,(2)graphornetworktheory,and(3)epidemic

modelsandnetworkmodels.

Inthethirdchapter,asarecentoutbreakinshrimpfarming,theriskfactorsforAHPND

wereinvestigatedbyanepizootiologicalsurveyatfarmlevel.Across-sectionalstudyis

described.ThisstudylinkedwiththedatafromtheSustainingEthicalAquacultureTrade

(SEAT),EUFP7researchproject. Importantly, thissurveydata, i.e.diseasemitigation

measures, farmingmanagementpracticesand thecumulativeAHPND incidence,has

alsobeenusedininterpretingresultsofChapter4andinmodellingthespreadofAHPND

ofChapter6.

Inthefourthchapter,graphornetworktheoryisappliedforthefirsttimetotheThai

shrimpfarmingsector.Thechapterdemonstratestheindustrysusceptibilitytoinfection

vialong-distancetransmission(i.e.liveshrimpmovement)basedontherealnetworkof

liveshrimpmovements inThailand (LSMN). It containsaquantitativeanalysisof the

LSMNincludingpropertiessuchasdegrees,averagepathlength,clusteringcoefficients

andassortativity.Theresultsofthischapterinformthefifthchapter.

Disease-controlalgorithmsaredevelopedtoidentifyhigh-riskconnectionsintheThai

shrimp farming network (Chapter 5). These algorithms include various network

centrality measurements (e.g. betweenness, eigenvector and degree), and their

capacities in reducing thepotential epizootic size in the LSMNaremeasuredby the

reachabilityofsites(innetworkterminology,nodes)andconnectedcomponents.The


44

bestalgorithmcanbeusedasacontrolstrategy.Moreover,diseaseoutbreaksdonot

only leadtofinancial lossesforproducersandinterruptionofbusiness,buttheyalso

influenceannualgovernmentexpenditure(FAO,2013b).Thedisease-controlalgorithm

developedinthischapterwillthereforebeconcernedwithoperatingcostsaswellas

effectiveness.

Inthesixthchapter,thedynamicsofAHPNDepizooticsareexplainedusinganetwork-

basedepidemicmodel.TheresultsindicatetheseasonalityofAHPNDspread,andthe

effect of long-distance and local transmission on the AHPND epizootic dynamics in

Thailand;theyalsosuggestdiseasepreventionandcontrolmeasurestoexplore.

All the results of the research are discussed in the final chapter, including the

contributionof the research to theThai shrimp farming sector, andpotential future

work.


45

Figure 1.8 Outline of the “Epizootiological tools for AHPND in Thai shrimp farming” research.

Networkmodelsandcompartmentalepidemicmodels

Diseaseepizootics

Shrimpfarming

Epidemiologicalapproaches

Epidemiologicalandepizootiological study

designs

Networkstructureanalysis

Controlstrategies


Chapter2Literaturereviews

Chapter5Disease-controlalgorithms

Chapter3RiskfactorsforAHPND

Chapter4Structureofliveshrimpmovementnetwork

Chapter6Network-basedepizootic

modelforAHPND

Chapter7Generaldiscussion


46

1.7 References

Aguirre-Guzmán,G.,Vázquez-Juárez,R.andAscencio,F.(2001)DifferencesinthesusceptibilityofAmericanwhiteshrimplarvalsubstages(Litopenaeusvannamei)tofourVibriospecies.JournalofInvertebratePathology,78(4),pp.215-219.

Akazawa,N.andEguchi,M.(2013)EnvironmentaltriggerforEMS/AHPNSidentifiedinAgrobestshrimpponds.GlobalAquacultureAdvocateJuly/August2013,pp.16-17.

Alam,S.N.(2015)Safetyintheshrimpsupplychain.In:P.Vishweshwaraiahetal.,ed.Regulatingsafetyoftraditionalandethnicfoods.Oxford:AcademicPress,pp.99-124.

Anderson,R.M.andMay,R.M.(1981)Thepopulationdynamicsofmicroparasitesandtheirinvertebratehosts.PhilosophicalTransactionsoftheRoyalSocietyofLondonB:BiologicalSciences,291(1054),pp.451-524.

Anh,P.T.,Kroeze,C.,Bush,S.R.andMol,A.P.J.(2010)Waterpollutionbyintensivebrackishshrimpfarminginsouth-eastVietnam:causesandoptionsforcontrol.AgriculturalWaterManagement,97(6),pp.872-882.

Balakrishnan,G.,Peyail,S.,Ramachandran,K.,Theivasigamani,A.,Savji,K.A.,Chokkaiah,M.andNataraj,P.(2011)GrowthofculturedwhitelegshrimpLitopenaeusvannamei(Boone1931)indifferentstockingdensity.AdvancesinAppliedScienceResearch,2(3),pp.107-113.

Barman,D.,Kumar,V.,Roy,S.andMandal,S.C.(2012)Specificpathogenfreeshrimps:Theirscopeinaquaculture.WorldAquaculture,43(1),pp.67.

Barraza-Guardado,R.H.,Arreola-Lizarraga,J.A.,Lopez-Torres,M.A.,Casillas-Hernandez,R.,Miranda-Baeza,A.,Magallon-Barrajas,F.andIbarra-Gamez,C.(2013)EffluentsofshrimpfarmsanditsinfluenceonthecoastalecosystemsofBahiadeKino,Mexico.TheScientificWorldJournal,2013,pp.1-8.

Bell,T.A.andLightner,D.V.(1984)IHHNvirus:infectivityandpathogenicitystudiesinPenaeusstylirostrisandPenaeusvannamei.Aquaculture,38(3),pp.185-194.

Bhattacharya,P.andNinan,K.N.(2011)Socialcost-benefitanalysisofintensiveversustraditionalshrimpfarming:acasestudyfromIndia.NaturalResourcesForum,35(4),pp.321–333.

Bonami,J.R.,Hasson,K.W.,Mari,J.,Poulos,B.T.andLightner,D.V.(1997)Taurasyndromeofmarinepenaeidshrimp:characterisationoftheviralagent.TheJournalofGeneralVirology,78(Pt2),pp.313–319.


47

Bondad-Reantaso,M.G.(2016)Acutehepatopancreaticnecrosisdisease(AHPND)ofpenaeidshrimps:Globalperspective.In:R.V.PakingkingJr.andE.G.T.deJesus-Ayson,ed.ProceedingsoftheASEANRegionalTechnicalConsultationonEMS/AHPNDandOtherTransboundaryDiseasesforImprovedAquaticAnimalHealthinSoutheastAsia,22ndto24thFebruary2016.AquacultureDepartment,SoutheastAsianFisheriesDevelopmentCenter,pp.16–23.

Boonyawiwat,V.(2009)TraditionalandmolecularepidemiologytodetermineriskfactorsforoutbreaksofshrimpwhitespotdiseaseinThailand.PhD,KasetsartUniversity.

Boyd,C.E.,McNevin,A.A.,Racine,P.,Tinh,H.Q.,Minh,H.N.,Viriyatum,R.,Paungkaew,D.andEngle,C.(2017)Resourceuseassessmentofshrimp,LitopenaeusvannameiandPenaeusmonodon,productioninThailandandVietnam.JournaloftheWorldAquacultureSociety,48(2),pp.201–226.

Brugere,C.,Onuigbo,D.M.andMorgan,K.L.(2017)Peoplematterinanimaldiseasesurveillance:challengesandopportunitiesfortheaquaculturesector.Aquaculture,467,pp.158–169.

Campos,L.N.S.,Herrera,F.D.,Araujo,A.D.R.andSánchez,R.A.G.(2014)LitopenaeusvannameiimmunestimulatedwithMacrocystispyriferaextract:improvingtheimmuneresponseagainstVibriocampbellii.JournalofCoastalLifeMedicine,2(8),pp.617–624.

Cao,H.,An,J.,Zheng,W.andHe,S.(2015)Vibriocholeraepathogenfromthefreshwater-culturedwhitelegshrimpPenaeusvannameiandcontrolwithBdellovibriobacteriovorus.JournalofInvertebratePathology,130,pp.13–20.

Chaivisuthangkura,P.,Vaniksampanna,A.,Pasookhush,P.,Longyant,S.andSithigorngul,P.(2016)Taurasyndromevirus.In:D.Liu,ed.Moleculardetectionofanimalviralpathogens.NewYork:CRCPress,pp.17–25.

Chayaburakul,K.,Lightner,D.V.,Sriurairattana,S.,Nelson,K.T.andWithyachumnarnkul,B.(2005)Differentresponsestoinfectioushypodermalandhematopoieticnecrosisvirus(IHHNV)inPenaeusmonodonandP.vannamei.DiseasesofAquaticOrganisms,67(3),pp.191–200.

Chen,H.,Liu,S.,Xu,X.,Liu,S.,Zhou,G.,Sun,K.,Zhao,J.andYing,G.(2015)AntibioticsintypicalmarineaquaculturefarmssurroundingHailingIsland,SouthChina:Occurrence,bioaccumulationandhumandietaryexposure.MarinePollutionBulletin,90(1–2),pp.181–187.

Cheng,L.,Lin,W.,Wang,P.,Tsai,M.,Hsu,J.andChen,S.(2013)WhitespotsyndromevirusepizooticinculturedPacificwhiteshrimpLitopenaeusvannamei(Boone)inTaiwan.JournalofFishDiseases,36(12),pp.977–985.


48

Cheng,W.,Wang,L.andChen,J.(2005)EffectofwatertemperatureontheimmuneresponseofwhiteshrimpLitopenaeusvannameitoVibrioalginolyticus.Aquaculture,250(3–4),pp.592-601.

Chinabut,S.andPuttinaowarat,S.(2005)Thechoiceofdiseasecontrolstrategiestosecureinternationalmarketaccessforaquacultureproducts.DevelopmentsinBiologicals,121,pp.255–261.

Chou,H.,Huang,C.,Wang,C.,Chiang,H.andLo,C.(1995)PathogenicityofabaculovirusinfectioncausingwhitespotsyndromeinculturedpenaeidshrimpinTaiwan.DiseasesofAquaticOrganisms,23(3),pp.165–173.

ChunI,Y.andYenLing,S.(2000)OutbreaksoftaurasyndromeinPacificwhiteshrimpPenaeusvannameiculturedinTaiwan.FishPathology,35(1),pp.21–24.

Cock,J.,Gitterle,T.,Salazar,M.andRye,M.(2009)Breedingfordiseaseresistanceofpenaeidshrimps.Aquaculture,286(1–2),pp.1–11.

Colt,J.(2006)Waterqualityrequirementsforreusesystems.AquaculturalEngineering,34(3),pp.143-156.

Corsin,F.,Funge-Smith,S.andClausen,J.(2007)AqualitativeassessmentofstandardsandcertificationschemesapplicabletoaquacultureintheAsia–Pacificregion.RAPPublication2007/25.Bangkok:FAO.Available:http://www.fao.org/3/a-ai388e.pdf[Accessed:18March2017].

Corsin,F.,Turnbull,J.F.,Mohan,C.V.,Hao,N.V.andMorgan,K.L.(2005)Pond-levelriskfactorsforwhitespotdiseaseoutbreaks.DiseasesinAsianAquacultureV,pp.75–92.

Dvorak,G.(2009)Biosecurityforaquaculturefacilitiesinthenorthcentralregion.NCRACFactSheet.Paper8.IowaState:NorthCentralRegionalAquacultureCenter.Available:http://lib.dr.iastate.edu/cgi/viewcontent.cgi?article=1007&context=ncrac_factsheets[Accessed:13March2015].

Eduardo,M.L.andMohan,C.V.(2012)Earlymortalitysyndrome(EMS)/Acutehepatopancreaticnecrosissyndrome(AHPNS):anemergingthreatintheAsianshrimpindustry.Bangkok:NACA.Available:http://library.enaca.org/Health/DiseaseLibrary/disease-advisory-ems-ahpns.pdf[Accessed:15August2014].

Escobedo-Bonilla,C.M.,Alday-Sanz,V.,Wille,M.,Sorgeloos,P.,Pensaert,M.andNauwynck,H.(2008)Areviewonthemorphology,molecularcharacterization,morphogenesisandpathogenesisofwhitespotsyndromevirus.JournalofFishDiseases,31(1),pp.1–18.


49

FAO(2013a)GLOBEFISH-Analysisandinformationonworldfishtrade(Shrimp-September2013).Available:http://www.fao.org/in-action/globefish/market-reports/resource-detail/en/c/338065/[Accessed:12November2015].

FAO(2013b)ReportoftheFAO/MARDtechnicalworkshoponearlymortalitysyndrome(EMS)oracutehepatopancreaticnecrosissyndrome(AHPNS)ofculturedshrimp(underTCP/VIE/3304).FAOFisheriesandAquacultureReportNo.1053.Rome:FAO.Available:http://www.fao.org/docrep/018/i3422e/i3422e00.htm[Accessed:11October2013].

FAO(2014)CulturedaquaticspeciesinformationprogrammePenaeusvannamei(Boone,1931).Rome:FAO.Available:http://www.fao.org/fishery/culturedspecies/Litopenaeus_vannamei/en[Accessed:29April2017].

FAO(2015)Aquarterlyupdateonworldseafoodmarkets.Issue2/2015.Rome:FAO.Available:http://www.fao.org/3/a-bc009e.pdf[Accessed:29April2017].

FAO(2016a)Foodoutlook,biannualreportonglobalfoodmarkets.ISSN0251-1959.Rome:FAO.Available:http://www.fao.org/3/a-I5703E.pdf[Accessed:29April2017].

FAO(2016b)Thestateofworldfisheriesandaquaculture2016.ISBN978-92-5-109185-2.Rome:FAO.Available:http://www.fao.org/3/a-i5555e.pdf[Accessed:8July2016].

FAO(2017)Europeanpricereport.Issue1/2017.Rome:FAO.Available:http://www.fao.org/3/a-br288e.pdf[Accessed:29April2017].

FDA(2016)FDAissuesimportalertonimportedshrimpandprawnsfromPeninsularMalaysia.Available:http://www.fda.gov/food/newsevents/constituentupdates/ucm496475.htm[Accessed:22June2016].

Filose,J.(1995)Factorsaffectingtheprocessingandmarketingoffarmraisedshrimp.In:C.L.BrowdyandJ.S.Hopkins,ed.SwimmingthroughTroubledWater:ProceedingsoftheSpecialSessiononShrimpFarming.California,1stto4thFeburary1995.BatonRouge,Louisiana:WorldAquacultureSociety,pp.227–234.

Flaherty,M.,Szuster,B.andMiller,P.(2000)LowsalinityinlandshrimpfarminginThailand.Ambio,29(3),pp.174–179.

Flegel,T.W.(2012)Historicemergence,impactandcurrentstatusofshrimppathogensinAsia.JournalofInvertebratePathology,110(2),pp.166–173.

Flegel,T.W.,Nielsen,L.,Thamavit,V.,Kongtim,S.andPasharawipas,T.(2004)Presenceofmultiplevirusesinnon-diseased,cultivatedshrimpatharvest.Aquaculture,240(1–4),pp.55–68.

Funge-Smith,S.J.andBriggs,M.R.P.(1998)Nutrientbudgetsinintensiveshrimpponds:implicationsforsustainability.Aquaculture,164(1–4),pp.117-133.


50

Ganesh,E.A.,Das,S.,Chandrasekar,K.,Arun,G.andBalamurugan,S.(2010)MonitoringoftotalheterotrophicbacteriaandVibriospp.inanaquaculturepond.Curr.Res.J.Biol.Sci,2(1),pp.48-52.

Gerland,P.,Raftery,A.E.,Sevcikova,H.,Li,N.,Gu,D.,Spoorenberg,T.,Alkema,L.,Fosdick,B.K.,Chunn,J.,Lalic,N.,Bay,G.,Buettner,T.,Heilig,G.K.andWilmoth,J.(2014)Worldpopulationstabilizationunlikelythiscentury.Science(NewYork,N.Y.),346(6206),pp.234–237.

Goarant,C.,Ansquer,D.,Herlin,J.,Domalain,D.,Imbert,F.andDeDecker,S.(2006)“SummerSyndrome”inLitopenaeusstylirostrisinNewCaledonia:Pathologyandepidemiologyoftheetiologicalagent,Vibrionigripulchritudo.Aquaculture,253(1–4),pp.105-113.

Goarant,C.,Herlin,J.,Brizard,R.,Marteau,A.,Martin,C.andMartin,B.(2000)ToxicfactorsofVibriostrainspathogenictoshrimp.DiseasesofAquaticOrganisms,40(2),pp.101-107.

Gopal,S.(2005)TheoccurrenceofVibriospeciesintropicalshrimpcultureenvironments;implicationsforfoodsafety.InternationalJournalofFoodMicrobiology,102(2),pp.151-159.

Gräslund,S.andBengtsson,B.(2001)Chemicalsandbiologicalproductsusedinsouth-eastAsianshrimpfarming,andtheirpotentialimpactontheenvironment—areview.ScienceoftheTotalEnvironment,280(1–3),pp.93–131.

Gräslund,S.,Holmström,K.andWahlström,A.(2003)Afieldsurveyofchemicalsandbiologicalproductsusedinshrimpfarming.MarinePollutionBulletin,46(1),pp.81–90.

Green,D.M.,Werkman,M.andMunro,L.A.(2012)ThepotentialfortargetedsurveillanceoflivefishmovementsinScotland.JournalofFishDiseases,35(1),pp.29–37.

Hameed,A.S.,Charles,M.X.andAnilkumar,M.(2000)ToleranceofMacrobrachiumrosenbergiitowhitespotsyndromevirus.Aquaculture,183(3),pp.207–213.

He,F.andKwang,J.(2008)IdentificationandcharacterizationofanewE3ubiquitinligaseinwhitespotsyndromevirusinvolvedinviruslatency.VirologyJournal,5(1),pp.151.

Heenatigala,P.andFernando,M.(2016)OccurrenceofbacteriaspeciesresponsibleforvibriosisinshrimppondculturesystemsinSriLankaandassessmentofthesuitablecontrolmeasures.SriLankaJournalofAquaticSciences,21(1),pp.1–17.

Holmström,K.,Gräslund,S.,Wahlström,A.,Poungshompoo,S.,Bengtsson,B.andKautsky,N.(2003)Antibioticuseinshrimpfarmingandimplicationsforenvironmentalimpactsandhumanhealth.InternationalJournalofFoodScience&Technology,38(3),pp.255–266.


51

Hopkins,J.S.,Hamilton,R.D.,Sandier,P.A.,Browdy,C.L.andStokes,A.D.(1993)Effectofwaterexchangerateonproduction,waterquality,effluentcharacteristicsandnitrogenbudgetsofintensiveshrimpponds.JournaloftheWorldAquacultureSociety,24(3),pp.304–320.

Johnson,K.N.,vanHulten,M.C.W.andBarnes,A.C.(2008)“Vaccination”ofshrimpagainstviralpathogens:Phenomenologyandunderlyingmechanisms.Vaccine,26(38),pp.4885–4892.

Kalagayan,H.,Godin,D.,Kanna,R.,Hagino,G.,Sweeney,J.,Wyban,J.andBrock,J.(1991)IHHNvirusasanetiologicalfactorinrunt-deformitysyndrome(RDS)ofjuvenilePenaeusvannameiculturedinHawaii.JournaloftheWorldAquacultureSociety,22(4),pp.235–243.

Kannapiran,E.,Ravindran,J.,Chandrasekar,R.andKalaiarasi,A.(2009)Studiesonluminous,VibrioharveyiassociatedwithshrimpculturesystemrearingPenaeusmonodon.JournalofEnvironmentalBiology,30(5),pp.791-795.

Kasornchandra,J.(2014)Themajordiseasesofshrimpandmarinefishandtheepidemicprevention.Bangkok:ThailandDepartmentofFisheries(DoF).

Kassam,L.,Subasinghe,R.andPhilips,M.(2011)Aquaculturefarmerorganizationsandclustermanagement.FAOFisheriesTechnicalPaperNo.563.Rome:FAO.Available:http://www.fao.org/docrep/014/i2275e/i2275e.pdf[Accessed:29April2017].

Kautsky,N.,Rönnbäck,P.,Tedengren,M.andTroell,M.(2000)Ecosystemperspectivesonmanagementofdiseaseinshrimppondfarming.Aquaculture,191(1–3),pp.145-161.

Keeling,M.J.andEames,K.T.(2005)Networksandepidemicmodels.JournaloftheRoyalSocietyInterface,2(4),pp.295–307.

Khadijah,S.,Neo,S.Y.,Hossain,M.S.,Miller,L.D.,Mathavan,S.andKwang,J.(2003)Identificationofwhitespotsyndromeviruslatency-relatedgenesinspecific-pathogen-freeshrimpsbyuseofamicroarray.JournalofVirology,77(18),pp.10162–10167.

Kongkeo,H.andDavy,F.B.(2010)BackyardhatcheriesandsmallscaleshrimpandprawnfarminginThailand.In:S.S.DeSilvaandF.B.Davy,eds.SuccessstoriesinAsianaquaculture.Dordrecht:Springer,pp.67–83.

Kumaran,M.(2009)Groupapproachtoshrimpfarming:thekeytosustainability.AquacAsia,14(3),pp.18–21.

Lai,H.,Ng,T.H.,Ando,M.,Lee,C.,Chen,I.,Chuang,J.,Mavichak,R.,Chang,S.,Yeh,M.,Chiang,Y.,Takeyama,H.,Hamaguchi,H.,Lo,C.,Aoki,T.andWang,H.(2015)Pathogenesisofacutehepatopancreaticnecrosisdisease(AHPND)inshrimp.Fish&ShellfishImmunology,47(2),pp.1006–1014.


52

Lavilla-Pitogo,C.R.,Baticados,M.C.L.,Cruz-Lacierda,E.R.anddelaPena,L.D.(1990)OccurrenceofluminousbacterialdiseaseofPenaeusmonodonlarvaeinthePhilippines.Aquaculture,91(1),pp.1-13.

Le,T.X.andMunekage,Y.(2004)ResiduesofselectedantibioticsinwaterandmudfromshrimppondsinmangroveareasinVietnam.MarinePollutionBulletin,49(11–12),pp.922–929.

Lebel,L.,Mungkung,R.,Gheewala,S.H.andLebel,P.(2010)Innovationcycles,nichesandsustainabilityintheshrimpaquacultureindustryinThailand.EnvironmentalScience&Policy,13(4),pp.291-302.

Lekshmy,S.,Nansimole,A.,Mini,M.,Athira,N.andRadhakrishnan,T.(2014)OccurrenceofVibriocholeraeinshrimpcultureenvironmentsofKerala,India.IndianJournalofScientificResearch,5(2),pp.151.

Lemonnier,H.,Lantoine,F.,Courties,C.,Guillebault,D.,Nézan,E.,Chomérat,N.,Escoubeyrou,K.,Galinié,C.,Blockmans,B.andLaugier,T.(2016)Dynamicsofphytoplanktoncommunitiesineutrophyingtropicalshrimppondsaffectedbyvibriosis.MarinePollutionBulletin,110(1),pp.449-459.

Lightner,D.V.(1999)ThepenaeidshrimpvirusesTSV,IHHNV,WSSV,andYHV.JournalofAppliedAquaculture,9(2),pp.27–52.

Lightner,D.V.(2003)ThepenaeidshrimpviralpandemicsduetoIHHNV,WSSV,TSVandYHV:historyintheAmericasandcurrentstatus.UJNRAquaculturePanelSymposium.California,USA,17thto18thand20thNovember2003,pp.17–20.

Lightner,D.V.(2011)Virusdiseasesoffarmedshrimpinthewesternhemisphere(theAmericas):areview.JournalofInvertebratePathology,106(1),pp.110–130.

Lightner,D.V.,Hasson,K.,White,B.andRedman,R.(1998)ExperimentalinfectionofwesternhemispherepenaeidshrimpwithAsianwhitespotsyndromevirusandAsianyellowheadvirus.JournalofAquaticAnimalHealth,10(3),pp.271–281.

Lightner,D.V.,Redman,R.M.andBell,T.A.(1983b)Infectioushypodermalandhematopoieticnecrosis,anewlyrecognizedvirusdiseaseofpenaeidshrimp.JournalofInvertebratePathology,42(1),pp.62–70.

Lightner,D.V.,Redman,R.M.,Arce,S.andMoss,S.M.(2009)Specificpathogen-freeshrimpstocksinshrimpfarmingfacilitiesasanovelmethodfordiseasecontrolincrustaceans.ShellfishSafetyandQuality,pp.384–424.

Lightner,D.V.,Redman,R.M.,Bell,T.A.andBrock,J.A.(1983a)DetectionofIHHNvirusinPenaeusstylirostrisandP.vannameiimportedintoHawaii.JournaloftheWorldMaricultureSociety,14(1–4),pp.212–225.


53

Lightner,D.V.,Redman,R.M.,Hasson,K.W.andPantoja,C.R.(1995)TaurasyndromeinPenaeusvannamei(Crustacea:Decapoda):grosssigns,histopathologyandultrastructure.DiseasesofAquaticOrganisms,21(1),pp.53–59.

Lightner,D.V.,Redman,R.M.,Poulos,B.T.,Nunan,L.M.,Mari,J.L.andHasson,K.W.(1997)RiskofspreadofpenaeidshrimpvirusesintheAmericasbytheinternationalmovementofliveandfrozenshrimp.RevueScientifiqueEtTechnique(InternationalOfficeofEpizootics),16(1),pp.146–160.

Limsuwan,C.(2009)ExperiencescultivatingwhiteshrimpinThailand,summaryofpresentationsinPeru,Colombia,andGuatemala,inFebruary2009.Nicovita-AlicorpTechnicalService.Available:http://www.nicovita.com.pe/en/extranet/Boletines/Abr-Jun_09.pdf[Accessed:3March2015].

Limsuwan,C.(2010)WhitefecesdiseaseinThailand.Nicovita-AlicorpTechnicalService.Available:http://www.nicovita.com.pe/cdn/Content/CMS/Archivos/Documentos/DOC_271_2.pdf[Accessed:27April2017].

Lio-Po,G.D.,Lavilla,C.R.andCruz-Lacierda,E.R.(2001)HealthManagementinaquaculture.Iloilo,Philippines:SEAFDEC.

Liu,C.andChen,J.(2004)EffectofammoniaontheimmuneresponseofwhiteshrimpLitopenaeusvannameianditssusceptibilitytoVibrioalginolyticus.Fish&ShellfishImmunology,16(3),pp.321-334.

Liu,P.,Lee,K.,Yii,K.,Kou,G.andChen,S.(1996)News&Notes:isolationofVibrioharveyifromdiseasedkurumaprawnsPenaeusjaponicus.CurrentMicrobiology,33(2),pp.129–132.

Lokkhumlue,M.andPrakitchaiwattana,C.(2013)InfluencesofcultivationconditionsonmicrobialprofilesofPacificwhiteshrimp(Litopenaeusvannamei)harvestedfromeasternandcentralThailand.ChiangMaiUniversityJournalofNaturalScience,pp.1-9.

Longyant,S.,Rukpratanporn,S.,Chaivisuthangkura,P.,Suksawad,P.,Srisuk,C.,Sithigorngul,W.,Piyatiratitivorakul,S.andSithigorngul,P.(2008)IdentificationofVibriospp.invibriosisPenaeusvannameiusingdevelopedmonoclonalantibodies.JournalofInvertebratePathology,98(1),pp.63–68.

Marchand,C.,Molnar,N.,Deborde,J.,Patrona,L.andMeziane,T.(2014)Seasonalpatternofthebiogeochemicalpropertiesofmangrovesedimentsreceivingshrimpfarmeffluents(NewCaledonia).JournalofAquacultureResearch&Development,5(5),pp.1–13.

Mohan,C.,Corsin,F.andPadiyar,P.(2004)Farm-levelbiosecurityandwhitespotdisease(WSD)ofshrimp.AquacultureHealthInternational,3,pp.16–20.


54

Mohney,L.L.,Lightner,D.V.andBell,T.A.(1994)Anepizooticofvibriosisinecuadorianpond-rearedPenaeusvannameiBoone(Crustacea:Decapoda).JournaloftheWorldAquacultureSociety,25(1),pp.116-125.

Montgomery-Brock,D.,Tacon,A.G.J.,Poulos,B.andLightner,D.V.(2007)Reducedreplicationofinfectioushypodermalandhematopoieticnecrosisvirus(IHHNV)inLitopenaeusvannameiheldinwarmwater.Aquaculture,265(1–4),pp.41–48.

Moriarty,D.J.(1998)ControlofluminousVibriospeciesinpenaeidaquacultureponds.Aquaculture,164(1),pp.351-358.

Moss,S.M.,Moss,D.R.,Arce,S.M.,Lightner,D.V.andLotz,J.M.(2012)Theroleofselectivebreedingandbiosecurityinthepreventionofdiseaseinpenaeidshrimpaquaculture.JournalofInvertebratePathology,110(2),pp.247–250.

Mugnier,C.,Justou,C.,Lemonnier,H.,Patrois,J.,Ansquer,D.,Goarant,C.andLecoz,J.(2013)Biological,physiological,immunologicalandnutritionalassessmentoffarm-rearedLitopenaeusstylirostrisshrimpaffectedorunaffectedbyvibriosis.Aquaculture,388–391,pp.105-114.

Muniesa,A.,Perez-Enriquez,R.,Cabanillas-Ramos,J.,Magallón-Barajas,F.J.,Chávez-Sánchez,C.,Esparza-Leal,H.andBlas,I.(2015)IdentifyingriskfactorsassociatedwithwhitespotdiseaseoutbreaksofshrimpsintheGulfofCalifornia(Mexico)throughexpertopinionandsurveys.ReviewsinAquaculture,pp.1–9.

NACA(2014)Diseasesofcrustaceans:acutehepatopancreaticnecrosissyndrome(AHPNS).Available:http://www.enaca.org/publications/health/disease-cards/ahpnd-disease-card-2014.pdf[Accessed:1May2015].

NACA(2017)Quarterlyaquaticanimaldiseasereport(AsiaandPacificRegion)1998-2016.Thailand:NACA.Available:http://www.enaca.org/modules/library/publication.php?tag_id=279&label_type=1&title=quarterly-aquatic-animal-disease-report[Accessed:29April2017].

Namikoshi,A.,Wu,J.L.,Yamashita,T.,Nishizawa,T.,Nishioka,T.,Arimoto,M.andMuroga,K.(2004)VaccinationtrialswithPenaeusjaponicustoinduceresistancetowhitespotsyndromevirus.Aquaculture,229(1–4),pp.25–35.

NationalBureauofAgriculturalCommodityandFoodStandards(2014)Goodaquaculturepracticesformarineshrimpfarm.Available:http://www.acfs.go.th/standard/download/GAP-FOR-MARINE-SHRIMP-FARM.pdf[Accessed:10March2017].


55

NationalEconomicandSocialDevelopmentBoard(2017)Grossdomesticproduct,chainvolumemeasures:Q4/2016.Bangkok:OfficeoftheNationalEconomicandSocialDevelopmentBoard.Available:http://www.nesdb.go.th/nesdb_en/ewt_news.php?nid=4340&filename=index[Accessed:20March2017].

Nimrat,S.,Suksawat,S.,Maleeweach,P.andVuthiphandchai,V.(2008)Effectofdifferentshrimppondbottomsoiltreatmentsonthechangeofphysicalcharacteristicsandpathogenicbacteriainpondbottomsoil.Aquaculture,285(1),pp.123-129.

Nunan,L.M.,Poulos,B.T.andLightner,D.V.(1998)Thedetectionofwhitespotsyndromevirus(WSSV)andyellowheadvirus(YHV)inimportedcommodityshrimp.Aquaculture,160(1),pp.19–30.

OIE(2013a)Acutehepatopancreaticnecrosisdisease,aetiologyepidemiologydiagnosispreventionandcontrolreferences.Available:http://www.oie.int/fileadmin/Home/eng/Internationa_Standard_Setting/docs/pdf/Aquatic_Commission/AHPND_DEC_2013.pdf[Accessed:12August2014].

OIE(2013b)Aquaticanimalhealthcode.Section9:diseasesofcrustaceans.WorldOrganisationforAnimalHealth(OIE).Available:http://www.oie.int/fileadmin/Home/eng/Health_standards/aahc/2010/en_titre_1.9.htm[Accessed:12August2014].

Overstreet,R.M.,Lightner,D.V.,Hasson,K.W.,McIlwain,S.andLotz,J.M.(1997)SusceptibilitytoTaurasyndromevirusofsomepenaeidshrimpspeciesnativetotheGulfofMexicoandthesoutheasternUnitedStates.JournalofInvertebratePathology,69(2),pp.165–176.

Piamsomboon,P.,Inchaisri,C.andWongtavatchai,J.(2016)ClimatefactorsinfluencetheoccurrenceofwhitespotdiseaseinculturedpenaeidshrimpinChanthaburiprovince,Thailand.AquacultureEnvironmentInteractions,8,pp.331–337.

Piumsombun,S.,Rab,M.A.,Dey,M.M.andSrichantuk,N.(2005)ThefarmingpracticesandeconomicsofaquacultureinThailand.AquacultureEconomics&Management,9(1–2),pp.265–287.

Pongthanapanich,T.andRoth,E.(2006)VoluntarymanagementinThaishrimpfarming.AquacultureEconomics&Management,10(3),pp.265–287.

Pruder,G.D.(2004)Biosecurity:applicationinaquaculture.AquaculturalEngineering,32(1),pp.3–10.

Quiroz-Guzmán,E.,Balcázar,J.L.,Vázquez-Juárez,R.,Cruz-Villacorta,A.A.andMartínez-Díaz,S.F.(2013)Proliferation,colonization,anddetrimentaleffectsofVibrioparahaemolyticusandVibrioharveyiduringbrineshrimphatching.Aquaculture,406,pp.85-90.


56

Rfoundationforstatisticalcomputing(2015)R:alanguageandenvironmentforstatisticalcomputing.Available:https://www.R-project.org/[Accessed:5December2015].

Rahman,M.M.,Corteel,M.,Dantas-Lima,J.J.,Wille,M.,Alday-Sanz,V.,Pensaert,M.B.,Sorgeloos,P.andNauwynck,H.J.(2007)Impactofdailyfluctuationsofoptimum(27°C)andhighwatertemperature(33°C)onPenaeusvannameijuvenilesinfectedwithwhitespotsyndromevirus(WSSV).Aquaculture,269(1–4),pp.107–113.

Rahman,M.M.,Escobedo-Bonilla,C.M.,Corteel,M.,Dantas-Lima,J.J.,Wille,M.,Sanz,V.A.,Pensaert,M.B.,Sorgeloos,P.andNauwynck,H.J.(2006)Effectofhighwatertemperature(33°C)ontheclinicalandvirologicaloutcomeofexperimentalinfectionswithwhitespotsyndromevirus(WSSV)inspecificpathogen-free(SPF)Litopenaeusvannamei.Aquaculture,261(3),pp.842–849.

Rajendran,K.V.,Vijayan,K.K.,Santiago,T.C.andKrol,R.M.(1999)Experimentalhostrangeandhistopathologyofwhitespotsyndromevirus(WSSV)infectioninshrimp,prawns,crabsandlobstersfromIndia.JournalofFishDiseases,22(3),pp.183–191.

Ramos-Carreño,S.,Valencia-Yáñez,R.,Correa-Sandoval,F.,Ruíz-García,N.,Díaz-Herrera,F.andGiffard-Mena,I.(2014)Whitespotsyndromevirus(WSSV)infectioninshrimp(Litopenaeusvannamei)exposedtolowandhighsalinity.ArchivesofVirology,159(9),pp.2213–2222.

Reilly,P.J.A.andTwiddy,D.R.(1992)SalmonellaandVibriocholeraeinbrackishwaterculturedtropicalprawns.InternationalJournalofFoodMicrobiology,16(4),pp.293-301.

Rico,A.,Satapornvanit,K.,Haque,M.M.,Min,J.,Nguyen,P.T.,Telfer,T.C.andVanDenBrink,P.J.(2012)UseofchemicalsandbiologicalproductsinAsianaquacultureandtheirpotentialenvironmentalrisks:acriticalreview.ReviewsinAquaculture,4(2),pp.75–93.

Rocha,R.d.S.,Sousa,O.V.d.andVieira,RegineHelenaSilvadosFernandes.(2016)Multidrug-resistantVibrioassociatedwithanestuaryaffectedbyshrimpfarminginNortheasternBrazil.MarinePollutionBulletin,105(1),pp.337-340.

Rockett,I.R.(1999)Populationandhealth:anintroductiontoepidemiology.PopulationBulletin,54(4),pp.1.

Rodgers,C.,Mohan,C.andPeeler,E.(2011)Thespreadofpathogensthroughtradeinaquaticanimalsandtheirproducts.RevSciTechOffIntEpiz,30(1),pp.241–256.

Rodríguez,J.,Bayot,B.,Amano,Y.,Panchana,F.,DeBlas,I.,Alday,V.andCalderón,J.(2003)WhitespotsyndromevirusinfectioninculturedPenaeusvannamei(Boone)inEcuadorwithemphasisonhistopathologyandultrastructure.JournalofFishDiseases,26(8),pp.439–450.


57

Roque,A.,Abad,S.,Betancourt-Lozano,M.,delaParra,L.M.G.,Baird,D.,Guerra-Flores,A.L.andGomez-Gil,B.(2005)EvaluationofthesusceptibilityoftheculturedshrimpLitopenaeusvannameitovibriosiswhenorallyexposedtotheinsecticidemethylparathion.Chemosphere,60(1),pp.126-134.

Saulnier,D.,Haffner,P.,Goarant,C.,Levy,P.andAnsquer,D.(2000)Experimentalinfectionmodelsforshrimpvibriosisstudies:areview.Aquaculture,191(1),pp.133–144.

Selvin,J.andLipton,A.(2003)VibrioalginolyticusassociatedwithwhitespotdiseaseofPenaeusmonodon.DiseasesofAquaticOrganisms,57(1),pp.147-150.

Senapin,S.,Thaowbut,Y.,Gangnonngiw,W.,Chuchird,N.,Sriurairatana,S.andFlegel,T.W.(2010)Impactofyellowheadvirusoutbreaksinthewhitelegshrimp,Penaeusvannamei(Boone),inThailand.JournalofFishDiseases,33(5),pp.421–430.

Somboon,M.,Purivirojkul,W.,Limsuwan,C.andChuchird,N.(2012)EffectofVibriospp.inwhitefecesinfectedshrimpinChantaburi,Thailand.KasetsartUniversityFisheriesResearchBulletin,36(1),pp.7-15.

Somjetlerdcharoen,A.(2002)Chloramphenicolconcernsinshrimpculture.AquacultureAsia,7(1),pp.51-55.

Song,Y.,Yu,C.,Lien,T.,Huang,C.andLin,M.(2003)HaemolymphparametersofPacificwhiteshrimp(Litopenaeusvannamei)infectedwithtaurasyndromevirus.Fish&ShellfishImmunology,14(4),pp.317–331.

Songsanjinda,P.(2013)TheapplicationofaquaticanimalmovementdocumentdatabaseinThailand.

Soto-Rodriguez,S.A.,Gomez-Gil,B.andLozano,R.(2010)‘Bright-red’syndromeinPacificwhiteshrimpLitopenaeusvannameiiscausedbyVibrioharveyi.DiseasesofAquaticOrganisms,92(1),pp.11–19.

Sriurairatana,S.,Boonyawiwat,V.,Gangnonngiw,W.,Laosutthipong,C.,Hiranchan,J.andFlegel,T.W.(2014)Whitefecessyndromeofshrimparisesfromtransformation,sloughingandaggregationofhepatopancreaticmicrovilliintovermiformbodiessuperficiallyresemblinggregarines.PloSOne,9(6),e99170.Available:http://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0099170&type=printable[Accessed:1March2015].

Sudha,P.M.,Mohan,C.V.,Shankar,K.M.andHegde,A.(1998)RelationshipbetweenwhitespotsyndromevirusinfectionandclinicalmanifestationinIndianculturedpenaeidshrimp.Aquaculture,167(1–2),pp.95–101.


58

Sung,H.,Li,H.,Tsai,F.,Ting,Y.andChao,W.(1999)ChangesinthecompositionofVibriocommunitiesinpondwaterduringtigershrimp(Penaeusmonodon)cultivationandinthehepatopancreasofhealthyanddiseasedshrimp.JournalofExperimentalMarineBiologyandEcology,236(2),pp.261-271.

Supungul,P.,Jaree,P.,Somboonwiwat,K.,Junprung,W.,Proespraiwong,P.,Mavichak,R.andTassanakajon,A.(2015)Apotentialapplicationofshrimpantilipopolysaccharidefactorindiseasecontrolinaquaculture.AquacultureResearch,43(3),pp.1–13.

SuratthaniShrimpFarmersClub(2014)SuratthaniShrimpFarmersClub.Available:http://suratthanishrimp.com.

Tendencia,E.A.,Bosma,R.H.andVerreth,J.A.J.(2011)Whitespotsyndromevirus(WSSV)riskfactorsassociatedwithshrimpfarmingpracticesinpolycultureandmonoculturefarmsinthePhilippines.Aquaculture,311(1–4),pp.87–93.

ThailandDoF(2016)Numberoffarms,areaundercultureandyieldbyspeciesandprovinces,1999-2013.Bangkok:ThailandDepartmentofFisheries(DoF).Available:http://www.fisheries.go.th/it-stat/yearbook/Index.htm[Accessed:27June2016].

TheWorldBank(2015)Employmentinagriculture(%oftotalemployment).Available:http://data.worldbank.org/indicator/SL.AGR.EMPL.ZS?locations=TH[Accessed:26April2017].

Thitamadee,S.,Prachumwat,A.,Srisala,J.,Jaroenlak,P.,Salachan,P.V.,Sritunyalucksana,K.,Flegel,T.W.andItsathitphaisarn,O.(2016)ReviewofcurrentdiseasethreatsforcultivatedpenaeidshrimpinAsia.Aquaculture,452,pp.69–87.

Tho,N.,Merckx,R.andUt,V.N.(2012)BiologicalcharacteristicsoftheimprovedextensiveshrimpsystemintheMekongdeltaofVietnam.AquacultureResearch,43(4),pp.526-537.

Tookwinas,S.,Chiyakum,K.andSomsueb,S.(2005)AquacultureofwhiteshrimpPenaeusvannameiinThailand.Iloilo,Philippines:SEAFDEC.Available:https://repository.seafdec.org.ph/bitstream/handle/10862/855/RTC-P.vannamei_p74-80.pdf?sequence=1&isAllowed=y[Accessed:1April2014].

Tran,L.,Nunan,L.,Redman,R.M.,Mohney,L.L.,Pantoja,C.R.,Fitzsimmons,K.andLightner,D.V.(2013)Determinationoftheinfectiousnatureoftheagentofacutehepatopancreaticnecrosissyndromeaffectingpenaeidshrimp.DiseasesofAquaticOrganisms,105(1),pp.45–55.

Tu,C.,Huang,H.,Chuang,S.,Hsu,J.,Kuo,S.,Li,N.,Hsu,T.,Li,M.andLin,S.(1999)taurasyndromeinPacificwhiteshrimpPenaeusvannameiculturedinTaiwan.DiseasesofAquaticOrganisms,38(2),pp.159–161.


59

Uchida,K.,Konishi,Y.,Harada,K.,Okihashi,M.,Yamaguchi,T.,Do,M.H.N.,ThiBui,L.,DucNugyen,T.,DoNguyen,P.andThiKhong,D.(2016)MonitoringofantibioticresiduesinaquaticproductsinurbanandruralareasofVietnam.JournalofAgriculturalandFoodChemistry,64(31),pp.6133-6138.

Uddin,M.T.(2008)Valuechainsandstandardsinshrimpexport.Available:http://ir.nul.nagoya-u.ac.jp/jspui/bitstream/2237/10939/1/169.pdf.

UndercurrentNews(2014)Thaishrimpexportvolumesdown32%throughAugust.UK:UndercurrentNews.Available:https://www.undercurrentnews.com/2014/10/27/thai-shrimp-export-volumes-down-32-through-august[Accessed:18June2016].

Vandenberghe,J.,Li,Y.,Verdonck,L.,Li,J.,Sorgeloos,P.,Xu,H.andSwings,J.(1998)VibriosassociatedwithPenaeuschinensis(Crustacea:Decapoda)larvaeinChineseshrimphatcheries.Aquaculture,169(1),pp.121–132.

Vaseeharan,B.,PremAnand,T.,Murugan,T.andChen,J.(2006)ShrimpvaccinationtrialswiththeVP292proteinofwhitespotsyndromevirus.LettersinAppliedMicrobiology,43(2),pp.137–142.

Walker,P.J.andWinton,J.R.(2010)Emergingviraldiseasesoffishandshrimp.VeterinaryResearch,41(6),pp.51.

Walling,E.,Vourey,E.,Ansquer,D.,Beliaeff,B.andGoarant,C.(2010)Vibrionigripulchritudomonitoringandstraindynamicsinshrimppondsediments.JournalofAppliedMicrobiology,108(6),pp.2003-2011.

Wang,F.andChen,J.(2006)EffectofsalinityontheimmuneresponseoftigershrimpPenaeusmonodonanditssusceptibilitytoPhotobacteriumdamselaesubsp.damselae.Fish&ShellfishImmunology,20(5),pp.671–681.

Wang,L.andChen,J.(2005)TheimmuneresponseofwhiteshrimpLitopenaeusvannameianditssusceptibilitytoVibrioalginolyticusatdifferentsalinitylevels.Fish&ShellfishImmunology,18(4),pp.269–278.

Werkman,M.,Green,D.M.,Murray,A.G.andTurnbull,J.F.(2011)Theeffectivenessoffallowingstrategiesindiseasecontrolinsalmonaquacultureassessedwithan𝑆𝐼𝑆model.PreventiveVeterinaryMedicine,98(1),pp.64–73.

Witteveldt,J.,Vlak,J.M.andvanHulten,M.C.W.(2006)IncreasedtoleranceofLitopenaeusvannameitowhitespotsyndromevirus(WSSV)infectionafteroralapplicationoftheviralenvelopeproteinVP28.DiseasesofAquaticOrganisms,70(1/2),pp.167–170.

Wu,W.,Wang,L.andZhang,X.(2005)Identificationofwhitespotsyndromevirus(WSSV)envelopeproteinsinvolvedinshrimpinfection.Virology,332(2),pp.578–583.


60

Wyban,J.(2007)Domesticationofpacificwhiteshrimprevolutionizesaquaculture.GlobalAquacultureAdvocate,10(4),pp.42–44.

Yaemkasem,S.,Boonyawiwat,V.,Kasornchandra,J.andPoolkhet,C.(2017)RiskfactorsassociatedwithwhitespotsyndromevirusoutbreaksinmarineshrimpfarmsinRayongProvince,Thailand.DiseasesofAquaticOrganisms,124(3),pp.193–199.

Yamprayoon,J.andSukhumparnich,K.(2010)Thaiaquaculture:achievingqualityandsafetythroughmanagementandsustainability.JournaloftheWorldAquacultureSociety,41(2),pp.274–280.

Yanong,R.P.andErlacher-Reid,C.(2012)BiosecurityinAquaculture,Part1:AnOverview.USDASouthernRegionalAquacultureCenter,4707,pp.1–16.

Zhang,J.,Li,Z.,Wen,G.,Wang,Y.,Luo,L.,Zhang,H.andDong,H.(2016)Relationshipbetweenwhitespotsyndromevirus(WSSV)loadsandcharacterizationsofwaterqualityin Litopenaeus vannamei culture ponds during the tropical storm. Iranian Journal ofVeterinaryResearch,17(3),pp.210.

Zhu,C.andDong,S.(2013)AquaculturesiteselectionandcarryingcapacitymanagementinthePeople’sRepublicofChina.In:L.G.Ross,T.C.Telfer,L.Falconer,D.SotoandJ.Aguilar-Manjarrez,ed.SiteSelectionandCarryingCapacitiesforInlandandCoastalAquaculture.Stirling,UK,6thto8thDecember2010.Rome:FAO,pp.219-230.Available:http://www.fao.org/3/contents/7358f8b4-00c1-55bd-83fd-01f93bc68ac1/i3322e.pdf#page=231[Accessed:15January2015].

Chapter2Epidemiologicalandepizootiologicaltools

61

Chapter 2 - Epidemiological and epizootiological tools for design of disease prevention and control strategies

N.Saleetid;D.M.Green;F.J.Murray

Preface

Thischapterdescribesthetheoreticalliteratureandempiricalevidencethatrelatesto

the development of disease prevention and control strategies. The chapter covers

epidemiological and epizootiological study designs: experimental, observational and

theoreticalstudies.Theseapproacheswillallowthe(1)investigationoftheriskfactors

foracutehepatopancreaticnecrosisdisease(AHPND)inChapter3(mostofthesehave

beenpreviouslybeenshowntobeusefulhistoricalinshrimpdiseases);(2)studyofthe

structureoftheliveshrimpmovementnetworkinChapter4;(3)designofthealgorithms

fortargeteddiseasesurveillanceandcontrol inChapter5;and(4)runningofnetwork-

basedmodelsofAHPNDepizooticdynamics inChapter6.The finalpartdescribes the

applicationtoAHPND.


62

Chapter 2 - Epidemiological and epizootiological tools for design of disease prevention and control strategies

Advances intechnologyandinnovationhave ledtoan increase inshrimpfarming. In

turn, dramatic decreases in global shrimp production due to widespread infectious

diseaseshavebeenwidelyreported(FAO,2013;FAO,2016).Therehavethusbeenmany

attempts to design disease prevention and control measures, both for the early

detection of diseases, and for maintaining the sustainability of the shrimp farming

sector (Bondad-Reantaso et al., 2005). This chapter is intended to: (1) review the

epidemiologicalandepizootiologicaltoolsthathavebeenthemostusefulhistoricallyto

examine the distribution (pattern and frequency) and risk factors for disease

transmission among a population, and (2) identify gaps in current epizootiological

research into disease outbreaks in shrimp farming. Epidemiological studies cover

experimental,observationalandtheoreticalapproaches,however,thisthesisfocuses

on the two latter typesofepidemiological studiesbecause theyaremoreuseful for

determiningdiseasetransmissioninthewholeshrimpfarmingsites.

2.1 Experimentalstudies

Evaluatingriskfactorsfordiseaseisoneofthemajortasksinepidemiology(WHO,2016)

that can be conducted with experiments. Experimental studies are used where the

effect of exposure to a risk factor is evaluatedby assigning that exposure alongside

controlstoastudypopulationsuchasaclinicaltrialofnewdrugforwhitespotdisease

(WSD)inshrimp(Ocampoetal.,2014).Muchoftheexperimentalstudyofdiseaseshas

been done in the context of testing a particular hypothesis in the laboratory. For

example, theoccurrenceofWSD in farmed shrimpwas found tobeassociatedwith

chemicalcomponents,suchasanexposureof20mgL-1totalammonianitrogen(Liang

etal.,2016),alowwatertemperatureof27°C(Rahmanetal.,2006),andarapidchange

insalinityfrom22to14ppt(Liuetal.,2006).Theepizoologyofvibriosisinshrimpwas

studiedexperimentallytoassesstheirassociationwithpesticide-contaminatedshrimp

(Labrieetal.,2003),andthepresenceofWSDbeforevibriosis(Phuocetal.,2008).From

thisliterature,itisobservedthatsuchexperimentalstudieskeepotherenvironmental


63

factorsconstant,while in fact thecomplexityof thenaturalenvironmenteffects the

occurrenceofdiseases.

2.2 Observationalstudies

Thesecondtypeofepidemiologicalstudiesisobservationalstudies(wheretheeffectof

exposuretoariskfactorisobservedbyexperimenterswithoutpriorassignmentofany

exposuretoastudypopulation)(Jepsenetal.,2004).Observationalstudiesoftentake

place in a natural environment to test multiple hypothesises, for example, where

multipleriskfactorsarethoughttobeassociatedwiththeoccurrenceofdisease.An

exampleofthiscanbeseeninCorsinetal.(2001).Usingacohortstudydesign,about

100 potential risk factors have been investigated for their association with WSD

occurringinfarmedshrimp(Corsinetal.,2001).Thisisoneofthethreemaintypesof

observationalstudiesasdescribedfollowing.

The threemain types of observational studies are cross-sectional, cohort and case–

controlstudies.Withdifferenttemporaldesigns,cross-sectionalstudiesaremainlyused

to determine prevalence, meaning the number of cases (diseased individuals) in a

population at a particular point in time. Cohort studies may be prospective or

retrospective.Whereas,case–controlstudiesaregenerallyretrospective(Mann,2003;

SongandChung, 2010).Among these, if apopulation sampled is alreadydefinedas

disease cases and controls, case–control studies show comparative advantages,

particularlythroughbeingabletolimittheeffectofconfoundingfactors(riskfactorsfor

adiseaseinnon-exposedindividuals,whichareassociatedwiththeexposureofinterest

inthestudiedpopulation)(Mann,2003;Salasetal.,1999).Todealwithanyconfounding

factor,casesandcontrolsarematchedusingspecificcriteriasuchastheageandgender

ofparticipants (McCreadieandScottishComorbidityStudyGroup,2002;Yusufetal.,

2004),orthegeographiclocationoffarmingsites(Boonyawiwatetal.,2016).

Investigations with case–control studies also minimise the incompleteness of data

collectionthat isoftencausedbydeathor inabilitytocontactcasesduringfollow-up

periods.Thisproblemhappensinprospectivecohortstudieswithalargesamplesize

(Holme et al., 2016; Swerdlow et al., 2015). In addition, since finding cases ismore


64

expensiveincross-sectionalstudies,somebiasmayariseduetotherecruitmenteffort

needed to complete those studies, as demonstrated in Van Schayck et al. (2002).

Regardingtheselimitations,case–controlstudiesareaneffectiveandfeasiblestrategy

tofindriskfactorsfordiseases,comparedwithotherstudydesigns(Diomidus,2002;

Schlesselman,1982;SchulzandGrimes,2002).

Toillustratethisfurther,thefollowingthreeexamplesdrawuponthethreemaintypes

ofobservationalstudies:cross-sectional,cohortandcase–controlstudies.Thefirsttwo

examplesstudiedWSD;anotherstudiedAHPND.Lookingatotherdiseasesandtherisk

factorsforthem,however,willprovideideasandhypothesesforthiswork,aswellas

forotheremergingordifferentdiseaseslikeAHPND.

Across-sectionalstudywasconductedbyTendenciaetal.(2011)toinvestigatetherisk

factorsforWSDinfectioninshrimpfarmsinthePhilippinesduringa14-monthsurvey

of the disease. Cases (WSD-infected farms) were identified by the

principalclinicalsignofWSD,i.e.whitespotsappearedonthebodyoffarmedshrimp,

or by a polymerase chain reaction (PCR) test for confirmation of WSD. The result

stronglysuggestedthatfeedingfreshmolluscstoshrimpwasamajorriskfactorofWSD

infection.Itwasexpectedthat,asfilterfeeders,molluscscouldactasWSSVcarriersby

obtainingthepathogensinsoilandwaterwithshrimppredationofthemolluscsleading

to the transferof thesepathogens. This researchhasaweakly temporal association

betweenthepresenceofdiseaseandtheriskfactorbecausebothdataweremeasured

simultaneously. Thus, itmaydifficult to determinewhether thepresenceof disease

followedexposure to the risk factor in timeor exposure to thepotential risk factor

resulted from the presence of disease (Song and Chung, 2010). However, there are

manyotherpublishedworksthatsupportthefindingthattheuseoffreshmolluscsand

fishinshrimprearingisapotentialriskfactorforWSDandotherviraldiseases(Hameed

etal.,2002;Vijayanetal.,2005)

As a forward-looking approach, a prospective cohort study investigates individual

groups moving forwards from exposure to a risk factor to the later presence (or

absence)ofastudydisease(GrimesandSchulz,2002).Inshrimpfarming,aprospective

cohortstudywasconductedbyCorsinetal. (2001)to investigatetherisk factors for


65

WSD.Theauthorsdesignedthecohortstudywiththe24shrimppondsofaVietnamese

farm, and a six-month follow-up of the study disease. Cases (infected ponds) were

definedbyaPCRtest.Themainresultindicatedthatearthenshrimppondsestablished

closetotheseawereassociatedwithanincreasedriskofWSDpresence.Thisresearch

couldutilise a cohort studybecauseof the shortproduction cycleof farmed shrimp

(aroundsixmonths)andtheuseofasmallnumberofepizootiologicalunits(24shrimp

pondsofafarm).

In contrast to prospective cohort studies and cross-sectional studies, a case–control

study is a retrospective or backward-looking approach (Hoffmann and Lim, 2007;

Pearce,2012).Study individualsarealreadydivided into twogroups:cases (denoted

diseased individuals) and controls (referred non-diseased individuals), and then

exposuretocandidateriskfactorsinthepastineachgroupisexamined(Mann,2003).

Recently, a case–control study was designed to identify the risk factors for AHPND

(Boonyawiwatetal.,2016).Caseswereobtainedfromreportingofdiseaseoccurrences

by farmers to local staff of the Thailand Department of Fisheries. During the study

period, the pathogenic agent of this disease remained unknown (the PCR test was

unavailable), but the cases were confirmed by histopathology based on the major

AHPNDsignsgiveninNACA(2014).Thisresearchsuggestedfivefactorswhichrelated

toincreasedriskofAHPNDoccurrenceatshrimppondlevel,i.e.theuseofchlorinefor

water treatment, theavailabilityofawater reservoir, thecultureofpredator fish in

waterpreparationponds,thestockingofmultipleshrimpspecies,andincreaseddensity

ofshrimpstocking.Acase–controlstudydesignwasmostefficientforthisresearchfor

tworeasons.First,beinganewdiseaseatthetime,AHPNDcasesremainedrareorhard

toconfirm.Second,thepresenceofAHPNDamongshrimpfarmingsiteswassuspected

to be related to multiple risk factors, especially in respect to farm management

practices.

This section has described the main types of observational study designs for

epidemiologythatareusedtoevaluatetheriskfactorsforshrimpdiseases.Togainmore

knowledgeofdisease spreadand todesigneffectivediseasepreventionand control

measures,many recent epidemiological studies have alsoused theoretical approach


66

withmathematicalmodelstoenhancetheunderstandingofdiseaseepidemicdynamics.

Thesewillbediscussednext.

2.3 Theoreticalstudies

In theoretical study design of epidemiology, the dynamics of disease spread can be

determinedbymeansofawidevarietyofmathematicalmodelssuchascompartmental

epidemicmodels,mass-actionmodels,networkmodelsandindividual-basedmodels.

Thissection,therefore,givesdetailsofthemathematicalmodellingapproaches.These

providemoreunderstandingofthechallenges instudyingthedynamicsof infectious

diseases transmitted among shrimp farming sites in Thailand, since these epizootic

dynamics remain poorly understood. Some of the discussed epidemic modelling

approacheshaveneverbeenappliedtoshrimpepizoology.

2.3.1 Compartmentalepidemicmodelsformicroparasitesversusmacroparasites

Mostdiseaseoutbreaksinshrimpfarmingarecausedbymicroparasites(Flegeletal.,

2008). Severe infections frommacroparasites occur less frequently due to chemical

treatmentandgoodmanagementpractices in farming.Nevertheless,macroparasites

canserveasavectorinthetransmissionofmicroparasites,leadingtolargeepidemics

(Loetal.,1996;Vijayanetal.,2005;Zhangetal.,2008).Tomodeltheirspreads,the

natureofthelifecycleandhowitisappropriatelymodelledareimportant(Anderson

andMay,1991).

Thelifecyclesofmicroparasitescontrastwiththoseofmacroparasites.Amicroparasite

cancomplete its lifecycle(multiply)withinan individualhost,whileamacroparasite

utilisesoneormorehosts,orispartlyfreeliving.Amicroparasite(viruses,bacteria,fungi

and protozoans) is a very small organism but amacroparasite (e.g. arthropods and

worms) is largerandcanbecounted.Nevertheless, itshouldbenotedthatalthough

somemicroparasitesare“micro”inscale,theirnaturetendstobe“macro”inmodelling.

Anexampleisfree-livingciliateprotozoans.Recently,theyhavebeenshowntocause

ectoparasitic diseases of farmed shrimp in Iran, i.e.mostly by species of the genus

Zoothamnium (Afsharnasab, 2015), and of farmed Nile tilapia in Saudi Arabia, i.e.

Trichodinamaritinkae,T.centrostrigeataandT.frenata(Abdel-Bakietal.,2017).


67

To model the dynamics of microparasite transmission, the Kermack-McKendrick

mathematicalmodeliswidelyused(Diekmannetal.,1995;KermackandMcKendrick,

1927). Table 2.1 shows three well-known mathematical model types and their

applicationtomicroparasiteinfections(ofteninhumandiseases),inwhichanindividual

can be in one of three compartments:𝑆 (susceptible), 𝐼 (infectious) or𝑅 (recovery,

removed,deathorquarantine).Thus,theyarealsocalledcompartmentalmodels.

Table 2.1 Three well-known compartmental models and their application to microparasite infections

Model Host Microparasitedisease

𝑆𝐼𝑅 Human Measles(Bjørnstadetal.,2002;SattenspielandDietz,1995;Shulginetal., 1998), influenza (Brauer, 2008), dengue fever (Feng and Velasco-Hernández, 1997), chickenpox,measles,mumps and influenza (Allen,2008;Coburnetal.,2009;FengandVelasco-Hernández,1997),andZikavirus(Bewicketal.,2016)

Livestock Foot-and-mouth disease (Hagenaars et al., 2011; Heath et al., 2008;Keeling,2005b)

Salmon Furunculosis(Ogutetal.,2005)

Shrimp Whitespotdisease(Hernandez-Llamasetal.,2013;LotzandSoto,2002),andtaurasyndrome(Lotzetal.,2003)

𝑆𝐼𝑆 Human Sexually transmitted diseases, i.e. hepatitis B (Kribs-Zaleta andMartcheva,2002)andGonorrhoea(Grayetal.,2011)

𝑆𝐼 Human SexuallytransmitteddiseasesHIV/AIDS(BozkurtandPeker,2014;Lloyd-Smithetal.,2004)

𝑆𝐼𝑅modelsareappropriatefordiseaseswithfullimmunityorpermanentremovalby

death while 𝑆𝐼𝑆 models (susceptible-infectious-susceptible) have been used for

diseaseswithlittleornoimmunity(Allen,1994;BlyussandKyrychko,2005;Fengand

Velasco-Hernández, 1997; Keeling and Eames, 2005). Hence, in the 𝑆𝐼𝑆 models an

individualcanbecomesusceptibleafterthatindividualrecoversfromaninitialinfection.

Finally, 𝑆𝐼 models apply to diseases that remain infectious for life and are never

“removed”(Kimetal.,2013;Tassier,2013).Becausethesecompartmentalmodelsapply

toindividuals,theinterpretationofthematanindividualscalediffersbetweensmaller

andlargerscales.


68

A unique compartmental model may not be sufficient to model all the various

mechanisms for disease transmission. In many cases of studying epidemiological

patterns (e.g. spreadbymultiple transmission routes), researchershaveaddedextra

compartments intothestandardmathematicalmodeltogainmoreunderstandingof

therealisticdiseasetransmission(Johnsonetal.,2016;Ngetal.,2003;TienandEarn,

2010). Two examples using 𝑆𝐼𝑅𝑆 models where an individual achieves temporary

immunity,and𝑆𝐸𝐼𝑅models(withcompartmentterm“𝐸”denoted“exposed”)where

duringanincubationperiodanindividualhasthediseasebutisnotyetinfectious,such

asforinfluenza(DegliAttietal.,2008),tuberculosis(Maraisetal.,2004)andmeasles

(Robinsonetal.,2014).Theseadditionalcompartmentsextendtheentirelifecyclesof

pathogensintothemodellingofdiseaseoutbreaks.

Inanothertypeofmodellingshrimpdiseases,macroparasiteinfectionsprimarilycausea

decreaseingrowthrateandstrength,andanincreaseintheoccurrenceofdeformityinfish

(WilliamsandBunkley-Williams,2000).Macroparasitesthatarecommonlyfoundinfarmed

fish and shrimp include helminths, e.g. monogeneans, cestodes and nematodes

(Domínguez-Machínetal.,2011;Soler-Jiménezetal.,2016),andcopepods,e.g.sea lice

(Igboelietal.,2014).Duetothecomplexlifecyclesoftheseorganisms,MayandAnderson

(1979)proposethatcompartmentalmodelsareunlikely tobehelpful forstudyingtheir

transmission. They state that thedynamicsofmacroparasite infectionsaremuchmore

dependentonthenumberofparasitesinahost.Thus,modellingbasedonthenumberof

hosts, parasites, and free-living infective stages becomesmore efficient than standard

compartmentalmodels(MayandAnderson,1979).

For the outbreak of AHPND (a microparasite infection) in Thai shrimp farming, the

𝑆𝐸𝐼𝑅𝑆compartmentalmodelisausefultooltomodeltheepizooticdynamicsatsite

level.Fourcompartmentsinthe𝑆𝐸𝐼𝑅𝑆model:susceptible(𝑆),exposed(𝐸),infected(𝐼)

and removed (𝑅) cover the actual patterns of AHPNDoccurrence in shrimp farming

sites.Othermodelsmayworkwell,however.Forexample,macroparasitemodelscan

workatsitelevelifthewithin-siteprevalencevariesalot.Alternatively,simplermodels,

suchas𝑆𝐼𝑅𝑆models,canbeused.Additionally,compartmentalmodelsofthespread

of infectious disease that utilise mass-action or network approaches might be

applicable.Thesearediscussedinthefollowingsub-section.


69

2.3.2 Mass-actionversusnetworkmodels

Mass-action and network models have important roles in modelling and assessing

diseasetransmissionbothbetweenpersonsoranimalfarmingsites,andwithinapond.

Thedistinguishingcontrastbetweenmass-actionandnetworkmodelsisoutlinedhere.

2.3.2.1 Mass-actionmodels

Mass-action models assume disease transmission in a homogenous population of

humans, animals or farming sites (Hethcote and Van Ark, 1987; Nold, 1980). For

example,thefollowing𝑆𝐼epidemicmodeldivides𝑆and𝐼intotwocompartmentsthat

are differentially susceptible and infectious. The rate of individual in class 𝑖 being

infectedattime𝑡is:

where𝑆> + 𝐼> = 𝑁> atalltimes,theparameters𝛼and𝛽areaconstantcontactrateand

transmissionrateperunittime,respectively.

In addition, the direction of disease transmission is neglected in most mass-action

models. This does not include themass-actionmodel that is naturally directed. The

directionof transmissioneffectsepidemicdynamicsbecausean individualcanactas

eitherasourceorasinkofinfection,orasbothasourceandsinkofinfection(Chaves

andHernandez,2004;Wesolowskietal.,2012).Forexample,intheThaishrimpfarming,

thepotentialpathwayoftransmissionthroughliveshrimpmovementsisfromhatchery

sites(sourcesof infection)tonurserysitesortoongrowingsites(sinksfor infection),

althoughtherearealsoasmallnumberofupstreammovements(fromongrowingsites

to hatchery sites) to reproduce new generations of shrimp, such as for genetic

improvementprogrammes(seeFigure2.1:thisstructurehasbeenappliedtoChapters

4,5and6).Thus,neglectoftheserealepidemiologicalpatternsmayleadtoover-or

under-estimationofdiseaseepidemicsinmass-actionmodels(Meyers,2007).

dSi

dt= �↵iSi

X

j

�jIj


70

Figure 2.1 Potential pathway for disease transmission via live shrimp movements in Thailand. The main pathway is downstream towards the ongrowing sites. There are a small number of upstream movements (from ongrowing sites to seed-producing sites) for reproducing new generations of shrimp.

2.3.2.2 Networkmodels

Owingtomodel-basedapproaches,severalrecentepidemicmodelsondiseaseshave

takenthereal-lifediseasespread,i.e.theheterogeneityofinfectioninapopulation,and

thedirectionoftransmissionintoconsideration.Thesearetypicallyknownasnetwork

models (Funk et al., 2010; Perisic and Bauch, 2009). The major difference between

networkmodelsandmass-actionmodelsisthatthenetworkmodelscancapturethe

reality that the population tends to be heterogeneous with non-random mixing

(HethcoteandVanArk,1987;Nold,1980).Keeling(2005a)statedthatinmass-action

modelseachinfectedindividualcantransmittoallotherindividuals,whilethedisease

spreadinnetworkmodelsrequiresconnectionsbetweenindividualsthatrepresenta

transmissionrouteofdiseasestobespecified.Hence,thechanceoftransmissionfrom

infected individuals in network models is dependent on contact networks that are

generallydegree-heterogeneous(Keeling,2005a;Volzetal.,2011).

Insupportofthisview,Woolhouseetal.(1997)explainedtheheterogeneityofinfection

withthe80/20rule.Therulesuggeststhatoften20%ofindividualscontributeatleast

80%oftheinfectioninthewholepopulation.Therefore,whereasthenetworkmodels

are estimated using the 80/20 rule, the mass-model assumption disregards the

heterogeneityofinfectioninapopulation,i.e.inthenumberofconnectionsbetween

Hatcherysites(1st seed-producingsites)

Nurserysites(2nd seed-producingsites)

Ongrowing sites


71

individuals(Finkenstadtetal.,2002;Germannetal.,2006;Kaneetal.,1999),andthe

timeofhost-parasiteinteractions(Milleretal.,2012).

Basically,themodellingofanetworkisbasedongraphtheory,inwhichanetworkrefers

to a group of sites (network terminology, nodes) connected by direct or indirect

connections. Networks with direct connections potentially allow the modelling of

diseasespreadtoaddressthesourcesandsinksfordiseasetransmissionbecausethe

directionofconnectionsisalreadyknown(GatesandWoolhouse,2015).Inmanycases,

networkmodelsarereconstructedthroughsimulatednetworks,whilerealnetworksare

lessoftenmodelledduetotheunavailabilityofreliablenetworkdata,orlackofadata

recordingsystem(KeelingandEames,2005).Asimplesimulatednetworkisthatofa

nodeconnectingwithtwoneighboursinaring.Additionally,WittenandPoulter(2007)

presented network types that have often been used in epidemicmodelling: random,

lattice, Watt–Strogatz small world and Barabasí–Albert scale-free networks.

DemonstrationsofthesesimulatednetworksareshowninFigure2.2.

Figure 2.2 Epidemic network models often are often characterised by these five simulated networks. All networks contain 20 nodes. The probability of each connection is 0.4 for drawing the random network, and 0.1 for drawing the small-world network.

Thetheoreticaldistinguishingpropertiesinsimulatednetworksareusefultodetermine

disease spread inmany real networks including farmed animalmovement networks

Ringnetwork Randomnetwork

Latticenetwork Small-worldnetwork

Scale-freenetwork



Scale-freenetwork



Scale-freenetwork



Scale-freenetwork


72

(Christley et al., 2005a; Kiss et al., 2006). Poisson degree distribution, a major

characteristicofrandomnetworks,islessoftenfoundinreality(Newmanetal.,2002),

butmanyrealnetworksarefoundtohaveotherproperties,suchasthoseofsmall-world

networks and scale-free networks. A small-world network is characterised by high

clustering,andshortpathlengths(distancesbetweensites)(WattsandStrogatz,1998).

As measured by a clustering coefficient, clustering is defined as the tendency for

trianglesofconnections toexist innetworks (Newman,2008).Small-worldnetworks

withshortmeanpathlengthsalsolinktothesmall-worldexperimentofMilgrametal.

(1992)whostudiedthe‘sixdegreesofseparation’theory.Theyproposethatindividuals

can get a piece of information (or a disease) via a connection of nomore than six

intermediates (Milgram et al., 1992; Watts, 1999). Infection within small-world

networksiscommonlyfast,mainlyduetotheirshortpathlengths(thenumberofpaths

traversed between a site pair) (Kiss et al., 2006; Newman, 2008). For scale-free

networks, degree distributions lie on a power-law form𝑃(𝑘)~𝑘/0 (Barabasi, 2009;

Pastor-SatorrasandVespignani,2001).Mostsitesinascale-freenetworkhaveasmall

number of connections, but a few sites have a large numbers of connections. The

heterogeneity in these site degrees becomes more interesting in terms of disease

preventionandcontrol,particularlyindesigningacontrolstrategyatthemosthighly

connectedsites(Barthélemyetal.,2005).

In termsof theapplicationof theoretical studies inepidemiology,mostof themcan

representdiseasespreadinwholepopulationsandleadtothedevelopmentofefficient

diseasesurveillanceandcontrolmeasures. Inordertoachievetheaimofthisthesis,

therefore, network models for targeted disease surveillance and control will be

addressedinthenextsub-section.

2.3.2.3 Networkmodelsfortargeteddiseasesurveillanceandcontrol

Targeted disease surveillance and control using a risk-based approach is one of the

potentialstrategies tomakethe farmingsectormoresustainable (PeelerandTaylor,

2011).Forfarmedshrimp,accordingtoNACA(2017),theAsianaquacultureindustryhas

establishedtwoimportanttypesofsurveillancetoprotectthehealthoffarmedaquatic

animals:activeandpassivesurveillance.Activesurveillanceisundertakenwiththegoal


73

of investigating targeted pathogens, while passive surveillance utilises laboratory

samplessubmittedfordiseasetestingpurposes(BurgessandMorley,2015).Thereare

several limitations with these surveillance programmes, however, such as a lack of

suitable resources for surveillance, and the effective diagnosis of diseases (Bondad-

Reantasoetal.,2005).Cost-effectiveinterventionsfordiseasecontrolscanbeimproved

iftargetedsurveillanceandcontrolisdevelopedandimplemented.

Forthespreadofdiseaseonsocialoranimalmovementnetworks,theideaoftargeted

surveillance and control is that high-risk connections serve strongly as a potential

transmission route for infectiousdiseases; and thus their removal from thenetwork

leads to a decrease in transmission (Duan et al., 2005; Green et al., 2012; Lou and

Ruggeri,2010).Bajardietal.(2012)arguethattargetingdiseasesurveillanceandcontrol

approaches based on centralitymeasuresmay fail tominimise the epidemic in the

network due to temporal fluctuations. Measures of centrality, however, play an

importantroleintargetingdiseasesurveillanceandcontrolforfarmedanimaldiseases,

asshowninTable2.2.

Table2.2alsodemonstratesthatanoptimalcentralitymeasureforonenetworkmay

not performwell for others, because such networks have particular structures. For

example, thealgorithmbasedonbetweenness centrality relatedwell to the specific

properties of the network of livestock movements in France, which displayed a

scale-free network and a large connected component (a giant strongly connected

component or GSCC) over the network (Rautureau et al., 2011). Accordingly, these

centralitymeasures,whichalreadyapplytofarmedanimalnetworks,areusedtoform

potentialcandidatedisease-controlalgorithmsintheLSMN(Chapter5).


74

Table 2.2 Centrality measures studied in five networks of farmed animal movements resulting in optimal strategies for targeted disease surveillance and control

Centralitymeasure LivefishinScotlanda

LivestockintheUKb

LivestockinItalyc

LivestockinArgentinad

LivestockinFrancee

Degree x x x* x* x

Betweenness x x* x x x*

Community-bridging x

Greedy x*

Eigenvector x x

Closeness x

xExaminedforthisnetwork*TheoptimalcentralitymeasureaLivefishmovementsinScotland(Greenetal.,2012).bLivestockmovementsintheUK(Ortiz-Pelaezetal.,2006).cLivestockmovementsinItaly(Nataleetal.,2009).dLivestockmovementsinArgentina(Aznaretal.,2011).eLivestockmovementsinFrance(Rautureauetal.,2011).Degree=thenumberofconnectionsofeachnodeBetweenness=thenumberofshortestpathsthatgothroughnodes,oralternatively,connectionsCommunity-bridging=theconnectionsthatlinkbetweentwosub-networksGreedy=theabilityofconnectionsintermsofthegreatestreductiontoeithermeanandmaximumreachEigenvector=thedegreecentralitiesofallneighbouringnodesconnectedtoaspecifiednodeCloseness=thenumberofshortestpathsfromaspecifiednodetoallneighbouringnodes

WefinishthissectionbyremarkingthatinThailandnoworkhasbeendonetostudy

shrimpepizoologywithnetworkmodelling.Tofillthisgap,thenetworkstructureoflive

shrimpmovementsofThailandisanalysed,andthepotentialofthenetworkinrespect

todiseasespreadevaluatedinChapter4.Inaddition,disease-controlalgorithmsforthe

liveshrimpmovementnetworkofThailandareexaminedinChapter5intermsoftheir

efficacyinreducingthepotentialepizooticsize.

While themodellingof thedynamicsofdiseaseepidemicsbecomesmoreandmore

complex, another epidemiological tool, individual-based modelling, might be an

appropriatetool.Thistypeofmodelisdiscussedinthenextsub-section.


75

2.3.3 Individual-basedsimulationmodels

Individual-basedmodellingisarobusttoolforepizoology.Thekeydistinguishingfeature

of individual-based modelling is the ability to track individuals in a population. In

individual-based models, the transmission of diseases can be tracked at microscale

interactions (e.g.disease spreadbetween infected sitesandsusceptible sites,where

“individual”referstothesite)inordertointerprettheepidemicdynamicsofawhole

populationinaspecifictimingandlocation(Al-Mamunetal.,2016;Kisjesetal.,2014).

Individual-basedmodelsareincreasinglyimportantforexaminingmanyseveredisease

outbreaks,andsupportingdecision-makers in thedevelopmentofcontrol strategies.

Theyhavebeenmainly applied in recent epidemiological studiesof humandiseases

suchastuberculosis(CardonaandPrats,2016),andanimaldiseasessuchasvibriosis

(Paillardetal.,2014)

An important reason for choosing individual-basedmodels is that the modelling of

infectious disease transmission has become very complex (Gu et al., 2003). The

complexitiesinmodellingepidemicdynamicsareillustratedinthefollowingexamples.

Demographics (the age or gender of persons) affect the spread of influenza type A

(Arimaetal.,2013;Donaldsonetal.,2009;Quandelacyetal.,2014).Pathogenssuchas

influenza,malariaordenguecontainmultiplestrainscausingdiverseepidemiological

patternsinpopulations(LourençoandRecker,2013).Individualsmaycarryoutvarious

behaviourssuchasavoidingcontactwithinfectedpersons(VanSegbroecketal.,2010),

and choosing longer distances of animalmovements for aquaculture (Keeling et al.,

2001).

Inaddition,manybiologicalprocesseshaveastochasticnature.Forexample,asiteis

exposedtoadisease,butisnotinfected.BlackandMcKane(2012)indicatethatthere

hasbeena recent increase in thenumberof individual-basedmodelsandstochastic

models for predicting infectious disease dynamics. To incorporate stochasticity in

individual-basedmodels,Rattanaetal.(2013)usedtheGillespiealgorithmtodetermine

nextevents according toevent timesdistributedexponentially. Insteadofusing this

exponentialdistribution,Verguetal.(2010)usedagammadistribution(i.e.apositively


76

skeweddistribution)toevaluatethechangesfrominfectioustorecoveredstates.This

servedexplicitlytosimulatethebiologyofindividuals.

An important concept that is established inmany individual-basedmodels is thatof

metapopulationconcept.Thisconceptemphasisesthat individualsoftencharacterise

twoormore structuredpopulations (Levin,1974).Keeling etal. (2010) indicate that

individual-basedmodelsthatdonotmaintainthis individuals’structureoverestimate

thespatialspreadandpotentialsizeoftheepidemic.Anexampleofindividual-based

metapopulationmodelsisdiseasestransmissioninfishfarmingsitesmodelledontwo

structuredpopulations:thesiteandthefish(Green,2010).Thisstructurealsoappears

infarmedshrimpproduction,inwhichsitestuctureissimplified,givingrisetoasingle

levelmodel,asintheChapter6ofthisthesis.

Insummary,thedynamicsofdiseaseepidemicsareaffectedbytheheterogeneityof

individualsandthestochasticnatureofthetransmissionprocess.Thesedynamicsare

difficult to determine by using equation-basedmodels, but it can be achievedwith

individual-basedmodels.

2.4 Applicationtoacutehepatopancreaticnecrosisdisease(AHPND)

ThischapterpresentstoolsthatarepotentiallyusefulforexaminingAHPNDspreadin

Thailand.AHPND,anewdiseasewithhighmortalitiesoffarmedshrimp,hasoccurred

ineasternThailandsince2011/2012fromwhenceitwastransmittedtoothershrimp

farmingareas.Theseepizooticsresultedinthelowestproductionoffarmedshrimpin

thetenyearsfrom2003to2013(ThailandDoF,2016).

InordertoinvestigatetheriskfactorsforAHPNDthatstillhadanunknownetiologyat

the time of commencement of this study, farmers' knowledge of shrimpdiseases is

directlyrelevantinordertodevelopacasedefinitionforAHPND.Thaifarmershavelong

experienceinfarmingshrimp,combinedwiththeexperienceoflossesduetodiseases

(Tookwinas et al., 2005).When infections occur, farmers are able to submit farmed

shrimp fordiagnosisofdiseases inmanyaquaticanimalhealthservicesbothprivate

and governmental laboratories; the laboratory results give the farmers better


77

knowledge about shrimp diseases. Thai farmers are therefore able to differentiate

AHPNDcasesfromothershrimpdiseases.

For Thai shrimp farming, data on live shrimp movements from source sites to

destinationsitesareavailable(SEAFDEC/MFRDMD,2016).Thiskindofdataisimportant

for studying recent epizoology. The potential transmission of AHPND, and other

infectious diseases of farmed shrimp, can therefore bemodelled as a network that

containsall sitesandtheirconnections.Theanalysisof thenetworkstructureallows

diseaseepizooticstobeevaluated,andassistsinthedevelopmentofcontrolstrategies

atacountrylevel,suchastargeteddiseasesurveillanceandcontrol.

IntherealpatternofAHPNDspread,latentperiodsareevidentinsite-to-siteAHPND

transmission. During these periods, farmed shrimp are infected by AHPND, but no

clinicalsignsofthatdiseaseappearatsitelevel(Tranetal.,2013).Nonetheless,these

exposed sites remain infectious. After a site is infected with the disease, different

fallowing periods are presented given different farming management practices for

removingthedisease.Then,thesitesstartanewcropandbecomeatriskofre-infection

(thisevidencewasobservedintheepizootiologicalsurveyreportedinChapter3).This

indicates that the 𝑆𝐼 model for AHPND epidemic dynamics requires extra

compartments,i.e.exposed(𝐸)andremoved(𝑅)states.

Another epidemiological tool that can be applied for studying AHPND epizootic

dynamics in Thai shrimp farming is the compartmental, individual-based epidemic

model.Thismodelishelpfulintestingcontrolstrategiesbychangingparameterssuch

as lower rates of long-distance transmission, denoting better disease control

arrangementsinthecountry(Rileyetal.,2003).

To conclude, the epidemiological and epizootiological tools that we present in this

chapter are useful for shrimp disease epizoology in Thailand. Not only are these

epizootiologicaltoolssuitableforAHPND,buttheyarealsousefulforotherinfectious

diseasesoffarmedshrimpsuchasWSD,YHDandothervibriosis.


78

2.5 References

Abdel-Baki,A.S.,AlGhamdi,A.andAl-Quraishy,S.(2017)FirstrecordofthreeAfricantrichodinids(Ciliophora:Peritrichida)inculturedNiletilapia(Oreochromisniloticus)inSaudiArabiawithre-evaluationoftheirhostspecificity.ParasitologyResearch,116(4),pp.1–7.

Afsharnasab,M.(2015)Prevalenceandintensityofprotozoanectoparasiteofthewhitelegshrimp(Penaeusindicus)inHellehsite,SouthofIran.IranianJournalofAquaticAnimalHealth,2(1),pp.17–23.

Allen,L.J.(1994)Somediscrete-time𝑆𝐼,𝑆𝐼𝑅,and𝑆𝐼𝑆epidemicmodels.MathematicalBiosciences,124(1),pp.83–105.

Allen,L.J.(2008)Anintroductiontostochasticepidemicmodels.In:F.Braueretal.,ed.MathematicalEpidemiology.Berlin:Springer,pp.81–130.

Al-Mamun,M.A.,Smith,R.L.,Schukken,Y.H.andGröhn,Y.T.(2016)ModelingofMycobacteriumaviumsubsp.paratuberculosisdynamicsinadairyherd:Anindividualbasedapproach.JournalofTheoreticalBiology,408,pp.105-117.

Anderson,R.M.andMay,R.M.(1991)Infectiousdiseasesofhumans.Oxford:Oxforduniversitypress.

Arima,Y.,Zu,R.,Murhekar,M.,Vong,S.andShimada,T.(2013)HumaninfectionswithavianinfluenzaA(H7N9)virusinChina:preliminaryassessmentsoftheageandsexdistribution.WesternPacificSurveillanceandResponse,4(2),pp.1-3.

Aznar,M.N.,Stevenson,M.A.,Zarich,L.andLeón,E.A.(2011)AnalysisofcattlemovementsinArgentina,2005.PreventiveVeterinaryMedicine,98(2–3),pp.119-127.

Bajardi,P.,Barrat,A.,Savini,L.andColizza,V.(2012)Optimisingsurveillanceforlivestockdiseasespreadingthroughanimalmovements.JournaloftheRoyalSociety,Interface,9(76),pp.2814-2825.

Barabasi,A.L.(2009)Scale-freenetworks:adecadeandbeyond.Science(NewYork,N.Y.),325(5939),pp.412-413.

Barthélemy,M.,Barrat,A.,Pastor-Satorras,R.andVespignani,A.(2005)Dynamicalpatternsofepidemicoutbreaksincomplexheterogeneousnetworks.JournalofTheoreticalBiology,235(2),pp.275-288.

Bewick,S.,Fagan,W.F.,Calabrese,J.M.andAgusto,F.(2016)Zikavirus:endemicversusepidemicdynamicsandimplicationsfordiseasespreadintheAmericas.bioRxiv,pp.041897.


79

Bjørnstad,O.N.,Finkenstädt,B.F.andGrenfell,B.T.(2002)Dynamicsofmeaslesepidemics:estimatingscalingoftransmissionratesusingatimeseries𝑆𝐼𝑅model.EcologicalMonographs,72(2),pp.169-184.

Black,A.J.andMcKane,A.J.(2012)Stochasticformulationofecologicalmodelsandtheirapplications.TrendsinEcology&Evolution,27(6),pp.337-345.

Blyuss,K.B.andKyrychko,Y.N.(2005)Onabasicmodelofatwo-diseaseepidemic.AppliedMathematicsandComputation,160(1),pp.177-187.

Bondad-Reantaso,M.G.,Subasinghe,R.P.,Arthur,J.R.,Ogawa,K.,Chinabut,S.,Adlard,R.,Tan,Z.andShariff,M.(2005)DiseaseandhealthmanagementinAsianaquaculture.VeterinaryParasitology,132(3–4),pp.249-272.

Boonyawiwat,V.,Patanasatienkul,T.,Kasornchandra,J.,Poolkhet,C.,Yaemkasem,S.,Hammell,L.andDavidson,J.(2016)Impactoffarmmanagementonexpressionofearlymortalitysyndrome/acutehepatopancreaticnecrosisdisease(EMS/AHPND)onpenaeidshrimpfarmsinThailand.JournalofFishDiseases,40(5),pp.649-659.

Bozkurt,F.andPeker,F.(2014)MathematicalmodellingofHIVepidemicandstabilityanalysis.AdvancesinDifferenceEquations,2014(1),pp.1.

Brauer,F.(2008)Epidemicmodelswithheterogeneousmixingandtreatment.BulletinofMathematicalBiology,70(7),pp.1869-1885.

Burgess,B.A.andMorley,P.S.(2015)Veterinaryhospitalsurveillancesystems.VeterinaryClinicsofNorthAmerica:SmallAnimalPractice,45(2),pp.235-242.

Cardona,P.andPrats,C.(2016)ThesmallbreathingamplitudeattheupperlobesfavorstheattractionofpolymorphonuclearneutrophilstoMycobacteriumtuberculosislesionsandhelpstounderstandtheevolutiontowardactivediseaseinanindividual-basedmodel.FrontiersinMicrobiology,7,pp.354.

Chaves,L.F.andHernandez,M.(2004)MathematicalmodellingofAmericancutaneousleishmaniasis:incidentalhostsandthresholdconditionsforinfectionpersistence.ActaTropica,92(3),pp.245-252.

Christley,R.,Robinson,S.,Lysons,R.andFrench,N.(2005)NetworkanalysisofcattlemovementinGreatBritain.Proc.Soc.Vet.Epidemiol.Prev.Med,pp.234-243.

Coburn,B.J.,Wagner,B.G.andBlower,S.(2009)Modelinginfluenzaepidemicsandpandemics:insightsintothefutureofswineflu(H1N1).BMCMedicine,7(1),pp.30.

Corsin,F.,Turnbull,J.,Hao,N.,Mohan,C.,Phi,T.,Phuoc,L.,Tinh,N.andMorgan,K.(2001)RiskfactorsassociatedwithwhitespotsyndromevirusinfectioninaVietnameserice-shrimpfarmingsystem.DiseasesofAquaticOrganisms,47(1),pp.1-12.


80

DegliAtti,M.L.C.,Merler,S.,Rizzo,C.,Ajelli,M.,Massari,M.,Manfredi,P.,Furlanello,C.,Tomba,G.S.andIannelli,M.(2008)MitigationmeasuresforpandemicinfluenzainItaly:anindividualbasedmodelconsideringdifferentscenarios.PloSOne,3(3).

Diekmann,O.,Heesterbeek,J.andMetz,J.(1995)ThelegacyofKermackandMcKendrick.EpidemicModels:TheirStructureandRelationtoData(D.Mollison,Ed.),pp.95-115.

Diomidus,M.(2002)Epidemiologicalstudydesigns.StudiesinHealthTechnologyandInformatics,65,pp.126-135.

Domínguez-Machín,M.E.,Hernández-Vergara,M.P.,Jiménez-García,I.,Simá-Álvarez,R.andRodríguez-Canul,R.(2011)Surveyofprotozoan,helminthandviralinfectionsinshrimpLitopenaeussetiferusandprawnMacrobrachiumacanthurusnativetotheJamapaRiverregion,Mexico.DiseasesofAquaticOrganisms,96(2),pp.97-103.

Donaldson,L.J.,Rutter,P.D.,Ellis,B.M.,Greaves,F.E.,Mytton,O.T.,Pebody,R.G.andYardley,I.E.(2009)MortalityfrompandemicA/H1N12009influenzainEngland:publichealthsurveillancestudy.BMJ(ClinicalResearchEd.),339,pp.b5213.

Duan,W.,Chen,Z.,Liu,Z.andJin,W.(2005)Efficienttargetstrategiesforcontagioninscale-freenetworks.PhysicalReviewE,72(2),pp.026133.

FAO(2013)ReportoftheFAO/MARDtechnicalworkshoponearlymortalitysyndrome(EMS)oracutehepatopancreaticnecrosissyndrome(AHPNS)ofculturedshrimp(underTCP/VIE/3304).FAOFisheriesandAquacultureReportNo.1053.Rome:FAO.Available:http://www.fao.org/docrep/018/i3422e/i3422e00.htm[Accessed:11October2013].

FAO(2016)Thestateofworldfisheriesandaquaculture2016.ISBN978-92-5-109185-2.Rome:FAO.Available:http://www.fao.org/3/a-i5555e.pdf[Accessed:8July2016].

Feng,Z.andVelasco-Hernández,J.X.(1997)Competitiveexclusioninavector-hostmodelforthedenguefever.JournalofMathematicalBiology,35(5),pp.523-544.

Finkenstadt,B.F.,Bjornstad,O.N.andGrenfell,B.T.(2002)Astochasticmodelforextinctionandrecurrenceofepidemics:estimationandinferenceformeaslesoutbreaks.Biostatistics(Oxford,England),3(4),pp.493-510.

Flegel,T.W.,Lightner,D.V.,Lo,C.F.andOwens,L.(2008)Shrimpdiseasecontrol:past,presentandfuture.In:M.G.Bondad-Reantaso,C.V.Mohan,M.CrumlishandR.P.Subasinghe,ed.DiseasesinAsianAquaculture.SriLanka,25thto28thOctober2005.Manila,Philippines:AsianFisheriesSociety,pp.355-378.

Funk,S.,Salathe,M.andJansen,V.A.(2010)Modellingtheinfluenceofhumanbehaviouronthespreadofinfectiousdiseases:areview.JournaloftheRoyalSociety,Interface,7(50),pp.1247-1256.


81

Gates,M.C.andWoolhouse,M.E.(2015)Controllinginfectiousdiseasethroughthetargetedmanipulationofcontactnetworkstructure.Epidemics,12,pp.11–19.

Germann,T.C.,Kadau,K.,Longini,I.M.,JrandMacken,C.A.(2006)MitigationstrategiesforpandemicinfluenzaintheUnitedStates.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,103(15),pp.5935–5940.

Gray,A.,Greenhalgh,D.,Hu,L.,Mao,X.,&Pan,J.(2011)Astochasticdifferentialequation𝑆𝐼𝑆epidemicmodel.SIAMJournalonAppliedMathematics,71(3),pp.876–902.

Green,D.M.(2010)Astrategicmodelforepidemiccontrolinaquaculture.PreventiveVeterinaryMedicine,94(1),pp.119–127.

Green,D.M.,Werkman,M.andMunro,L.A.(2012)ThepotentialfortargetedsurveillanceoflivefishmovementsinScotland.JournalofFishDiseases,35(1),pp.29–37.

Grimes,D.A.andSchulz,K.F.(2002)Cohortstudies:marchingtowardsoutcomes.TheLancet,359(9303),pp.341–345.

Gu,W.,Killeen,G.F.,Mbogo,C.M.,Regens,J.L.,Githure,J.I.andBeier,J.C.(2003)Anindividual-basedmodelofPlasmodiumfalciparummalariatransmissiononthecoastofKenya.TransactionsoftheRoyalSocietyofTropicalMedicineandHygiene,97(1),pp.43–50.

Hagenaars,T.J.,Dekker,A.,deJong,M.C.andEble,P.L.(2011)Estimationoffootandmouthdiseasetransmissionparameters,usingoutbreakdataandtransmissionexperiments.RevueScientifiqueEtTechnique(InternationalOfficeofEpizootics),30(2),pp.467–477.

Hameed,A.S.,Murthi,B.L.M.,Rasheed,M.,Sathish,S.,Yoganandhan,K.,Murugan,V.andJayaraman,K.(2002)AninvestigationofArtemiaasapossiblevectorforwhitespotsyndromevirus(WSSV)transmissiontoPenaeusindicus.Aquaculture,204(1),pp.1–10.

Heath,M.F.,Vernon,M.C.andWebb,C.R.(2008)ConstructionofnetworkswithintrinsictemporalstructurefromUKcattlemovementdata.BMCVeterinaryResearch,4,pp.11.

Hernandez-Llamas,A.,Magallon-Barajas,F.J.,Perez-Enriquez,R.,Cabanillas-Ramos,J.,Esparza-Leal,H.M.andPortillo-Clark,G.(2013)PondshutdownasastrategyforpreventingoutbreaksofwhitespotdiseaseinshrimpfarmsinMexico.ReviewsinAquaculture,5,pp.1–8.

Hethcote,H.W.andVanArk,J.W.(1987)Epidemiologicalmodelsforheterogeneouspopulations:proportionatemixing,parameterestimation,andimmunizationprograms.MathematicalBiosciences,84(1),pp.85–118.


82

Hoffmann,R.G.andLim,H.J.(2007)Observationalstudydesign.In:W.T.Ambrosius,ed.TopicsinBiostatisticsNewJersey:HumanaPress,pp.19–31.

Holme,I.,Retterstøl,K.,Norum,K.andHjermann,I.(2016)Lifelongbenefitsonmyocardialinfarctionmortality:40-yearfollow-upoftherandomizedOslodietandantismokingstudy.JournalofInternalMedicine,280,pp.221–227.

Igboeli,O.O.,Burka,J.F.andFast,M.D.(2014)Lepeophtheirussalmonis:apersistingchallengeforsalmonaquaculture.AnimFront,4(1),pp.22–32.

Jepsen,P.,Johnsen,S.P.,Gillman,M.W.andSorensen,H.T.(2004)Interpretationofobservationalstudies.Heart(BritishCardiacSociety),90(8),pp.956–960.

Johnson,T.L.,Landguth,E.L.andStone,E.F.(2016)Modelingrelapsingdiseasedynamicsinahost-vectorcommunity.PLoSNeglTropDis,10(2),e0004428.Available:http://journals.plos.org/plosntds/article/file?id=10.1371/journal.pntd.0004428&type=printable[Accessed:1February2017].

Kane,A.,Lloyd,J.,Zaffran,M.,Simonsen,L.andKane,M.(1999)TransmissionofhepatitisB,hepatitisCandhumanimmunodeficiencyvirusesthroughunsafeinjectionsinthedevelopingworld:model-basedregionalestimates.BulletinoftheWorldHealthOrganization,77(10),pp.801–807.

Keeling,M.(2005a)Theimplicationsofnetworkstructureforepidemicdynamics.TheoreticalPopulationBiology,67(1),pp.1–8.

Keeling,M.J.andEames,K.T.(2005)Networksandepidemicmodels.JournaloftheRoyalSocietyInterface,2(4),pp.295–307.

Keeling,M.J.(2005b)Modelsoffoot-and-mouthdisease.ProceedingsoftheBiologicalSciences/theRoyalSociety,272(1569),pp.1195–1202.

Keeling,M.J.,Danon,L.,Vernon,M.C.andHouse,T.A.(2010)Individualidentityandmovementnetworksfordiseasemetapopulations.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,107(19),pp.8866–8870.

Keeling,M.J.,Woolhouse,M.E.,Shaw,D.J.,Matthews,L.,Chase-Topping,M.,Haydon,D.T.,Cornell,S.J.,Kappey,J.,Wilesmith,J.andGrenfell,B.T.(2001)Dynamicsofthe2001UKfootandmouthepidemic:stochasticdispersalinaheterogeneouslandscape.Science(NewYork,N.Y.),294(5543),pp.813–817.

Kermack,W.O.andMcKendrick,A.G.(1927)Acontributiontothemathematicaltheoryofepidemics.Proc.R.Soc.Lond.A,115(772),pp.700–721.

Kim,K.I.,Lin,Z.andZhang,Q.(2013)An𝑆𝐼𝑅epidemicmodelwithfreeboundary.NonlinearAnalysis:RealWorldApplications,14(5),pp.1992–2001.


83

Kisjes,K.H.,DuintjerTebbens,R.J.,Wallace,G.S.,Pallansch,M.A.,Cochi,S.L.,Wassilak,S.G.andThompson,K.M.(2014)Individual-basedmodelingofpotentialpoliovirustransmissioninconnectedreligiouscommunitiesinNorthAmericawithlowuptakeofvaccination.TheJournalofInfectiousDiseases,210Suppl1,pp.S424-S433.

Kiss,I.Z.,Green,D.M.andKao,R.R.(2006)ThenetworkofsheepmovementswithinGreatBritain:Networkpropertiesandtheirimplicationsforinfectiousdiseasespread.JournaloftheRoyalSociety,Interface/theRoyalSociety,3(10),pp.669-677.

Kribs-Zaleta,C.M.andMartcheva,M.(2002)Vaccinationstrategiesandbackwardbifurcationinanage-since-infectionstructuredmodel.MathematicalBiosciences,177–178,pp.317-332.

Labrie,L.,Roque,A.,Gomez-Gil,B.andTurnbull,J.F.(2003)EffectofmethylparathiononthesusceptibilityofshrimpLitopenaeusvannameitoexperimentalvibriosis.DiseasesofAquaticOrganisms,57(3),pp.265-270.

Levin,S.A.(1974)DispersionandPopulationInteractions.TheAmericanNaturalist,108(960),pp.207-228.

Liang,Z.,Liu,R.,Zhao,D.,Wang,L.,Sun,M.,Wang,M.andSong,L.(2016)Ammoniaexposureinducesoxidativestress,endoplasmicreticulumstressandapoptosisinhepatopancreasofPacificwhiteshrimp(Litopenaeusvannamei).Fish&ShellfishImmunology,54,pp.523-528.

Liu,B.,Yu,Z.,Song,X.,Guan,Y.,Jian,X.andHe,J.(2006)Theeffectofacutesalinitychangeonwhitespotsyndrome(WSS)outbreaksinFenneropenaeuschinensis.Aquaculture,253(1–4),pp.163-170.

Lloyd-Smith,J.O.,Getz,W.M.andWesterhoff,H.V.(2004)Frequency-dependentincidenceinmodelsofsexuallytransmitteddiseases:portrayalofpair-basedtransmissionandeffectsofillnessoncontactbehaviour.BiologicalSciences,271(1539),pp.625-634.

Lo,C.,Ho,C.,Peng,S.,Chen,C.,Hsu,H.,Chiu,Y.,Chang,C.,Liu,K.,Su,M.andWang,C.(1996)Whitespotsyndromebaculovirus(WSBV)detectedinculturedandcapturedshrimp,crabsandotherarthropods.DiseasesofAquaticOrganisms,27(3),pp.215-225.

Lotz,J.M.andSoto,M.A.(2002)Modelofwhitespotsyndromevirus(WSSV)epidemicsinLitopenaeusvannamei.DiseasesofAquaticOrganisms,50(3),pp.199-209.

Lotz,J.M.,Flowers,A.M.andBreland,V.(2003)Amodeloftaurasyndromevirus(TSV)epidemicsinLitopenaeusvannamei.JournalofInvertebratePathology,83(2),pp.168-176.


84

Lou,J.andRuggeri,T.(2010)Thedynamicsofspreadingandimmunestrategiesofsexuallytransmitteddiseasesonscale-freenetwork.JournalofMathematicalAnalysisandApplications,365(1),pp.210-219.

Lourenço,J.andRecker,M.(2013)Natural,persistentoscillationsinaspatialmulti-straindiseasesystemwithapplicationtodengue.PLoSComputBiol,9(10),e1003308.Available:http://journals.plos.org/ploscompbiol/article/file?id=10.1371/journal.pcbi.1003308&type=printable[Accessed:20August2015].

Mann,C.J.(2003)Observationalresearchmethods.ResearchdesignII:cohort,crosssectional,andcase-controlstudies.EmergencyMedicineJournal:EMJ,20(1),pp.54-60.

Marais,B.,Gie,R.,Schaaf,H.,Hesseling,A.,Obihara,C.,Starke,J.,Enarson,D.,Donald,P.andBeyers,N.(2004)Thenaturalhistoryofchildhoodintra-thoracictuberculosis:acriticalreviewofliteraturefromthepre-chemotherapyera[StateoftheArt].TheInternationalJournalofTuberculosisandLungDisease,8(4),pp.392-402.

May,R.M.andAnderson,R.M.(1979)Populationbiologyofinfectiousdiseases:PartII.Nature,280(5722),pp.455-461.

McCreadie,R.G.andScottishComorbidityStudyGroup.(2002)Useofdrugs,alcoholandtobaccobypeoplewithschizophrenia:case-controlstudy.TheBritishJournalofPsychiatry:TheJournalofMentalScience,181,pp.321-325.

Meyers,L.(2007)Contactnetworkepidemiology:Bondpercolationappliedtoinfectiousdiseasepredictionandcontrol.BulletinoftheAmericanMathematicalSociety,44(1),pp.63-86.

Milgram,S.,Sabini,J.E.andSilver,M.E.(1992)Theindividualinasocialworld:essaysandexperiments.NewYork:Mcgraw-HillBookCompany.

Miller,J.C.,Slim,A.C.andVolz,E.M.(2012)Edge-basedcompartmentalmodellingforinfectiousdiseasespread.JournaloftheRoyalSociety,Interface/theRoyalSociety,9(70),pp.890-906.




85

Natale,F.,Giovannini,A.,Savini,L.,Palma,D.,Possenti,L.,Fiore,G.andCalistri,P.(2009)NetworkanalysisofItaliancattletradepatternsandevaluationofrisksforpotentialdiseasespread.PreventiveVeterinaryMedicine,92(4),pp.341-350.

Newman,M.E.(2008)Themathematicsofnetworks.TheNewPalgraveEncyclopediaofEconomics,2(2008),pp.1-12.

Newman,M.E.,Watts,D.J.andStrogatz,S.H.(2002)Randomgraphmodelsofsocialnetworks.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,99Suppl1,pp.2566-2572.

Ng,T.W.,Turinici,G.andDanchin,A.(2003)AdoubleepidemicmodelfortheSARSpropagation.BMCInfectiousDiseases,3(1),pp.1.

Nold,A.(1980)Heterogeneityindisease-transmissionmodeling.MathematicalBiosciences,52(3),pp.227-240.

Ocampo,L.,Chavez,B.,Tapia,G.,Ibarra,C.andSumano,H.,2014.Efficacyofapharmaceuticalpreparationbasedonglycyrrhizicacidinachallengestudyofwhitespotsyndromeinwhiteshrimp(Litopenaeusvannamei).Aquaculture,428,pp.280-283.

Ogut,H.,LaPatra,S.andReno,P.(2005)Effectsofhostdensityonfurunculosisepidemicsdeterminedbythesimple𝑆𝐼𝑅model.PreventiveVeterinaryMedicine,71(1),pp.83-90.

Ortiz-Pelaez,A.,Pfeiffer,D.,Soares-Magalhaes,R.andGuitian,F.(2006)Useofsocialnetworkanalysistocharacterizethepatternofanimalmovementsintheinitialphasesofthe2001footandmouthdisease(FMD)epidemicintheUK.PreventiveVeterinaryMedicine,76(1),pp.40-55.

Paillard,C.,Jean,F.,Ford,S.E.,Powell,E.N.,Klinck,J.M.,Hofmann,E.E.andFlye-Sainte-Marie,J.(2014)Atheoreticalindividual-basedmodelofbrownringdiseaseinManilaclams,Venerupisphilippinarum.JournalofSeaResearch,91,pp.15-34.

Pastor-Satorras,R.andVespignani,A.(2001)Epidemicspreadinginscale-freenetworks.PhysicalReviewLetters,86(14),pp.3200.

Pearce,N.(2012)Classificationofepidemiologicalstudydesigns.InternationalJournalofEpidemiology,41(2),pp.393-397.

Peeler,E.J.andTaylor,N.G.(2011)Theapplicationofepidemiologyinaquaticanimalhealth-opportunitiesandchallenges.VeterinaryResearch,42(1),pp.94.


86

Perisic,A.andBauch,C.T.(2009)Socialcontactnetworksanddiseaseeradicabilityundervoluntaryvaccination.PLoSComputBiol,5(2),e1000280.Available:http://journals.plos.org/ploscompbiol/article/file?id=10.1371/journal.pcbi.1000280&type=printable[Accessed:3May2016].

Phuoc,L.,Corteel,M.,Nauwynck,H.,Pensaert,M.,Alday-Sanz,V.,VanDenBroeck,W.,Sorgeloos,P.andBossier,P.(2008)Increasedsusceptibilityofwhitespotsyndromevirus-infectedLitopenaeusvannameitoVibriocampbellii.EnvironmentalMicrobiology,10(10),pp.2718-2727.

Quandelacy,T.M.,Viboud,C.,Charu,V.,Lipsitch,M.andGoldstein,E.(2014)Age-andsex-relatedriskfactorsforinfluenza-associatedmortalityintheUnitedStatesbetween1997-2007.AmericanJournalofEpidemiology,179(2),pp.156-167.

Rahman,M.M.,Escobedo-Bonilla,C.M.,Corteel,M.,Dantas-Lima,J.J.,Wille,M.,Sanz,V.A.,Pensaert,M.B.,Sorgeloos,P.andNauwynck,H.J.(2006)Effectofhighwatertemperature(33°C)ontheclinicalandvirologicaloutcomeofexperimentalinfectionswithwhitespotsyndromevirus(WSSV)inspecificpathogen-free(SPF)Litopenaeusvannamei.Aquaculture,261(3),pp.842-849.

Rattana,P.,Blyuss,K.B.,Eames,K.T.andKiss,I.Z.(2013)Aclassofpairwisemodelsforepidemicdynamicsonweightednetworks.BulletinofMathematicalBiology,75(3),pp.466-490.

Rautureau,S.,Dufour,B.andDurand,B.(2011)Vulnerabilityofanimaltradenetworkstothespreadofinfectiousdiseases:amethodologicalapproachappliedtoevaluationandemergencycontrolstrategiesincattle,France,2005.TransboundaryandEmergingDiseases,58(2),pp.110-120.

Robinson,M.,Conan,A.,Duong,V.,Ly,S.,Ngan,C.,Buchy,P.,Tarantola,A.andRodo,X.(2014)AmodelforachikungunyaoutbreakinaruralCambodiansetting:implicationsfordiseasecontrolinuninfectedareas.PLoSNeglTropDis,8(9),e3120.Available:http://journals.plos.org/plosntds/article/file?id=10.1371/journal.pntd.0003120&type=printable[Accessed:4November2016].

Salas,M.,Hotman,A.andStricker,B.H.(1999)Confoundingbyindication:anexampleofvariationintheuseofepidemiologicterminology.AmericanJournalofEpidemiology,149(11),pp.981-983.

Sattenspiel,L.andDietz,K.(1995)Astructuredepidemicmodelincorporatinggeographicmobilityamongregions.MathematicalBiosciences,128(1),pp.71-91.

Schlesselman,J.J.(1982)Case-controlstudies:design,conduct,analysis.NewYork:OxfordUniversityPress.

Schulz,K.F.andGrimes,D.A.(2002)Case-controlstudies:researchinreverse.TheLancet,359(9304),pp.431-434.


87

SEAFDEC/MFRDMD(2016)Traceabilityoffishandfisheryproducts:capturefisheryproducts.Available:http://www.seafdec.org/documents/sc16_wp10.pdf[Accessed:3March2016].

Shulgin,B.,Stone,L.andAgur,Z.(1998)Pulsevaccinationstrategyinthe𝑆𝐼𝑅epidemicmodel.BulletinofMathematicalBiology,60(6),pp.1123-1148.

Soler-Jiménez,L.C.,Paredes-Trujillo,A.I.andVidal-Martínez,V.M.(2016)HelminthparasitesoffinfishcommercialaquacultureinLatinAmerica.JournalofHelminthology,91,pp.1-27.

Song,J.W.andChung,K.C.(2010)Observationalstudies:cohortandcase-controlstudies.PlasticandReconstructiveSurgery,126(6),pp.2234-2242.

Swerdlow,A.J.,Cooke,R.,Albertsson-Wikland,K.,Borgstrom,B.,Butler,G.,Cianfarani,S.,Clayton,P.,Coste,J.,Deodati,A.,Ecosse,E.,Gausche,R.,Giacomozzi,C.,Kiess,W.,Hokken-Koelega,A.C.,Kuehni,C.E.,Landier,F.,Maes,M.,Mullis,P.E.,Pfaffle,R.,Savendahl,L.,Sommer,G.,Thomas,M.,Tollerfield,S.,Zandwijken,G.R.andCarel,J.C.(2015)DescriptionoftheSAGhEcohort:alargeEuropeanstudyofmortalityandcancerincidencerisksafterchildhoodtreatmentwithrecombinantgrowthhormone.HormoneResearchinPaediatrics,84(3),pp.172-183.

Tassier,T.(2013)Theeconomicsofepidemiology.NewYork:Springer.

Tendencia,E.A.,Bosma,R.H.andVerreth,J.A.J.(2011)Whitespotsyndromevirus(WSSV)riskfactorsassociatedwithshrimpfarmingpracticesinpolycultureandmonoculturefarmsinthePhilippines.Aquaculture,311(1–4),pp.87-93.


Tien,J.H.andEarn,D.J.(2010)Multipletransmissionpathwaysanddiseasedynamicsinawaterbornepathogenmodel.BulletinofMathematicalBiology,72(6),pp.1506-1533.

Tookwinas,S.,Chiyakum,K.andSomsueb,S.(2005)AquacultureofwhiteshrimpPenaeusvannameiinThailand.Iloilo,Philippines:SEAFDEC.Available:https://repository.seafdec.org.ph/bitstream/handle/10862/855/RTC-P.vannamei_p74-80.pdf?sequence=1&isAllowed=y[Accessed:1April2014].

Tran,L.,Nunan,L.,Redman,R.M.,Mohney,L.L.,Pantoja,C.R.,Fitzsimmons,K.andLightner,D.V.(2013)Determinationoftheinfectiousnatureoftheagentofacutehepatopancreaticnecrosissyndromeaffectingpenaeidshrimp.DiseasesofAquaticOrganisms,105(1),pp.45-55.


88

VanSchayck,C.P.,Loozen,J.M.,Wagena,E.,Akkermans,R.P.andWesseling,G.J.(2002)Detectingpatientsatahighriskofdevelopingchronicobstructivepulmonarydiseaseingeneralpractice:crosssectionalcasefindingstudy.BMJ(ClinicalResearchEd.),324(7350),pp.1370.

VanSegbroeck,S.,Santos,F.C.andPacheco,J.M.(2010)Adaptivecontactnetworkschangeeffectivediseaseinfectiousnessanddynamics.PLoSComputBiol,6(8),e1000895.Available:http://journals.plos.org/ploscompbiol/article/file?id=10.1371/journal.pcbi.1000895&type=printable[Accessed:12May2016].

Vergu,E.,Busson,H.andEzanno,P.(2010)Impactoftheinfectionperioddistributionontheepidemicspreadinametapopulationmodel.PloSOne,5(2),pp.e9371.

Vijayan,K.,Raj,V.S.,Balasubramanian,C.,Alavandi,S.,Sekhar,V.T.andSantiago,T.(2005)Polychaeteworms—avectorforwhitespotsyndromevirus(WSSV).DiseasesofAquaticOrganisms,63(2-3),pp.107-111.

Volz,E.M.,Miller,J.C.,Galvani,A.andMeyers,L.A.(2011)Effectsofheterogeneousandclusteredcontactpatternsoninfectiousdiseasedynamics.PLoSComputBiol,7(6),e1002042.Available:http://journals.plos.org/ploscompbiol/article/file?id=10.1371/journal.pcbi.1002042&type=printable[Accessed:12May2015].

Watts,D.J.(1999)Networks,dynamics,andthesmall-worldphenomenon1.AmericanJournalofSociology,105(2),pp.493-527.

Watts,D.J.andStrogatz,S.H.(1998)Collectivedynamicsof‘small-world’networks.Nature,393(6684),pp.440-442.

Wesolowski,A.,Eagle,N.,Tatem,A.J.,Smith,D.L.,Noor,A.M.,Snow,R.W.andBuckee,C.O.(2012)Quantifyingtheimpactofhumanmobilityonmalaria.Science(NewYork,N.Y.),338(6104),pp.267-270.

WHO(2016)Theterminologyofepidemiology.Available:http://www.who.int/topics/epidemiology/en/[Accessed:14July2016].

Williams,E.andBunkley-Williams,L.(2000)Multicellularparasite(macroparasite)problemsinaquaculture.EncyclopediaofAquaculture.NewYork:Wiley,pp.562-579.

Witten,G.andPoulter,G.(2007)Simulationsofinfectiousdiseasesonnetworks.ComputersinBiologyandMedicine,37(2),pp.195-205.

Woodward,M.(2013)Epidemiology:studydesignanddataanalysis.Thirded.NewYork:CRCpress.


89

Woolhouse,M.E.,Dye,C.,Etard,J.F.,Smith,T.,Charlwood,J.D.,Garnett,G.P.,Hagan,P.,Hii,J.L.,Ndhlovu,P.D.,Quinnell,R.J.,Watts,C.H.,Chandiwana,S.K.andAnderson,R.M.(1997)Heterogeneitiesinthetransmissionofinfectiousagents:implicationsforthedesignofcontrolprograms.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,94(1),pp.338-342.

Yusuf,S.,Hawken,S.,Ôunpuu,S.,Dans,T.,Avezum,A.,Lanas,F.,McQueen,M.,Budaj,A.,Pais,P.,Varigos,J.andLisheng,L.(2004)Effectofpotentiallymodifiableriskfactorsassociatedwithmyocardialinfarctionin52countries(theINTERHEARTstudy):case–controlstudy.TheLancet,364(9438),pp.937-952.

Zhang,J.,Dong,S.,Dong,Y.,Tian,X.andHou,C.(2008)BioassayevidenceforthetransmissionofWSSVbytheharpacticoidcopepodNitocrasp.JournalofInvertebratePathology,97(1),pp.33-39.

Chapter3Cross-sectionalstudy

90

Chapter 3 - Evaluating risk factors for transmission of acute hepatopancreatic necrosis disease (AHPND) in the Thai shrimp farming sector


Preface

Buildingontheauthor’smaster’sdegreestudyin2012/2013,theinvestigationofrisk

factorsforsite-to-sitetransmissionofAHPNDisintensivelyassessedinthischapter.An

extra phase of data collection (Phase 4: Cross-checking of the risk factors) was

implementedandmoreadvancedstatisticalapproacheswereapplied,insofarasthese

aretechnicallyadmissible.Intheextraphase,thedatawerecollectedusingface-to-face

interviewswithninekey informants.Moreadvancedstatisticalanalyses(i.e.receiver

operating characteristic curveandcross-validationanalysis)wereapplied than those

usedintheoriginalmaster’sthesis.Thesurveydata,i.e.farmingmanagementpractices

and the cumulative AHPND incidence, has also been used in interpreting results of

Chapter4andinmodellingthespreadofAHPNDofChapter6.

Atthetimeofdatacollectionforthisobservationalepizootiologicalstudy,thecauseof

theinitialoutbreakofAHPNDwasunknown.Thus,thedesignationofcasesandcontrols

wasbasedontheAHPNDdecisiontree,whichwasdevelopedasanepizootiologicaltool

inthecross-sectionalstudy.Notethatthischapterisdesignedforpublication.Thisstudy

utilisedthedatafromtheSustainingEthicalAquacultureTrade(SEAT),EUFP7research

project.


91

Chapter 3 - Evaluating risk factors for transmission of acute hepatopancreatic necrosis disease (AHPND) in the Thai shrimp farming sector

3.1 Abstract

Inthisstudy,theriskfactorsforsite-to-sitetransmissionofAHPNDwereevaluatedin

Thailandusinga cross-sectional approach.Anunbiased sample frameof206 shrimp

farms (previouslyparticipated in theSustainingEthicalAquacultureTrade (SEAT),EU

FP7 researchproject)wereengaged in four consecutive structured surveys in2013–

2014. The outcome led to the development of a decision tree for AHPND case

determinationandriskestimationusingunivariateandunconditionallogisticregression

analysis.

Interviewsweresuccessfullyperformedwith143oftheabove206shrimpfarms(70%).

35%ofthe143mettheAHPNDcasedefinition,withahigherproportionintheeast.

Southern farms showed a delay in AHPND onset, and large-scale farms that usually

invested inmore biosecurity resources than others also showed a delayed onset of

AHPND. The cumulative incidence of AHPND in southern and large-scale farms

increased sharply after the first occurrence AHPND, however. Two risk factors for

AHPNDtransmissionwerefound:earthenpondswerelessriskywithanoddsratio(𝑂𝑅)

of0.25(95%CI0.06–0.8;P-value=0.02)comparedwithshrimprearinginfullyplastic-

lined ponds; and the absence of pond harrowingwas higher riskywith an𝑂𝑅 of 3.9

(95 % CI1.3–12.6; P-value=0.01) compared with the presence of pond harrowing.

ThesefindingsimplythattheThaishrimpfarmingindustryshouldenhancebiosecurity

systems,andthatasimplegoodfarmingmanagementpractice,suchasharrowingpond

bottomwhichareacommonpracticeofshrimpfarminginearthernponds,mayprotect

farmsagainstAHPND.

3.2 Introduction

Acutehepatopancreaticnecrosisdisease(AHPND)isthemosteconomicallydamaging

epizooticpandemictoaffectthefarmedpenaeidshrimpsectorsincetheoutbreakof

whitespotsyndromeinthe1990’s(Flegel,1997).In2013thetotalannuallosseswere


92

estimatedasUSD5billion(Rosenberry,2013).AHPND,alsodescribedasearlymortality

syndrome (EMS)prior to the recentdetectionofabacterialpathologicalagent (OIE,

2013;Thitamadeeetal.,2016)wasfirstdetectedinChinain2009.Itwassubsequently

reported in Vietnam in 2010,Malaysia and Thailand in 2011 (Eduardo andMohan,

2012),Mexicoin2013(Nunanetal.,2014)andthePhilippinesin2014(Leobertetal.,

2015),whereasmassmortalities attributed toVibrio species in Indian shrimp farms

werenon-AHPND(Kumaretal.,2014).

AHPND(mostThaifarmerscalledthisdisease"EMS"or"EMS/AHPND")wasreportedin

Thailand at first time in the east central provinces in late 2011 (FAO, 2013) and in

southernproducerprovincesinlate2012.Ineachinstance,theincidencerosesharply

in the months following the first detection, resulting in widespread precautionary

fallowingofpondsastheprimaryfarmerresponse(Flegel,2012).Althoughthedisease

affects a wide range of important commercial penaeid shrimp species, including

Litopenaeus vannamei, Penaeus monodon and P. chinensis farmed over a range of

productionintensities(OIE,2013),thehighestriskoflossappearstobeassociatedwith

moreintensivefarmingpractices(FAO,2013),thenormforL.vannameiproduction.

Theformer“EMS”titledescribestheoccurrenceofmassmortalities(>70%)duringthe

first 35 days of culture in newly prepared ponds (FAO, 2013b; Flegel and Lo, 2014;

Lightner et al., 2012; Thitamadee et al., 2016). AHPND, meanwhile, describes the

characteristicdegenerativepathologyinthevitaldigestiveandglandularmidgut-organ.

Toxin-producingstrainsofVibrioparahaemolyticushavebeenconfirmedasaprimary

causativeagentTranetal.(2013).Thesepathogensalsohadahighchanceofantibiotic

resistance(Hanetal.,2015a;Hanetal.,2015b;Laietal.,2015).SeveralVibriospecies

are zoonotic (Austin, 2010). Fortunately, the three strains of V. parahaemolyticus

implicatedinAHPNDinfectioninThailanddonotappeartopresentapublichealth-risk

(Chonsinetal.,2016;Kondoetal.,2014).

TheclinicalsignsofAHPNDdonotdiffersignificantlyinsomecasesfromthoseofshrimp

infectedwithdifferenttypesofvibriosis.Forexample,asahepatopancreaticinfection,

V.harveyicausesnecrosisofthehepatopancreaticcells,andathickerabnormalbasal


93

laminathaninnormalshrimp(Jiravanichpaisaletal.,1994;Tranetal.,2013).Clearcase

definitionisthereforeveryimportanttoevaluatetheriskfactorsforAHPNDoccurrence.

Arangeofstudieshaspositivelycorrelatedloss-severitywithvariousenvironmentalrisk

factors.ThegeneralabundanceofVibriospp.andotherpathogensinshrimphasbeen

linkedtohighphytoplanktonlevels(Petersonetal.,2010)andpHlevel(>8.2,Costaet

al.,2010),andAkazawaandEguchi(2013)foundsuchacorrelationbetweenelevated

pHandAHPNDinMalaysia.InChina,outbreaksofAHPNDhavebeenassociatedwith

elevatedsalinitylevelsandolderponds(Panakorn,2012).Co-locationofsemi-intensive

andintensivefarmingsystemsinsouthernVietnampointedtoanincreasedprobability

ofAHPNDmortalitiesinintensivesystems(FAO,2013).Therehasbeenagenerallackof

systematicepizootiologicalstudiesunderpinningsuchobservations,however.

Takingacross-sectionalapproach,theThaishrimpfarmingsitesweresampledovera

studyperiodtoseekpairedgroups,withorwithoutAHPNDclinicalsigns.Bothgroups

were identified and comparedon thebasis of potential causal attributes. The study

aimedtoassess(1)theassociationofenvironmentalandfarm-managementriskfactors

withthegeographicalprevalenceandincidenceofputativeAHPNDcasesinThailandin

order to (2) draw inferences regarding its transmission and to suggest possible

mitigationstrategies.

3.3 Methods

3.3.1 DevelopmentofacasedefinitionanddecisiontreeforAHPND

SincethecausalagentofAHPNDwasnotknownatthistime,casedefinitions inthis

studywerebasedonclinicalsignsdescribed intheAHPNDdiseaseadvisoryofNACA

(2014),throughkeyinformantandfarmerperceptions(Section3.3.4),andareviewof

other secondary literature on Thai and global AHPND outbreaks. Based on this

information,an‘AHPNDcasedecision-tree’wasconstructedaroundfourspecificand

measurableindicators:(1)dateofonsetofthefirstclinicalsignsofAHPND,(2)theage

of the affected shrimp, (3) mortality rates, and (4) three characteristic and easily

observedclinicalsignsofAHPNDinfection:i.e.locationofdeadshrimp,diseasedshrimp

behaviourandthepresenceofapale/whitishhepatopancreas.Toassistdifferentiation


94

in this final step, picture cards showing gross AHPND whitish and atrophied

hepatopancreas pathology (FAO, 2013; NACA, 2014) were shown alongside those

showing clinical signs of other high-prevalence shrimp diseases (white spot disease,

yellow head disease, taura syndrome, vibriosis and infectious hypodermal

haematopoietic necrosis virus:Murray et al., 2013). The picture cards are shown in

AppendixAattheendofthisthesis.

3.3.2 CandidateriskfactorsforAHPNDoccurrenceatfarmlevel

Candidate farm-level risk (i.e. independent) factors for AHPND occurrence were

reviewedfromsecondarydata(AkazawaandEguchi,2013;FAO,2013;Panakorn,2012)

and key informant opinion (Thai Department of Fisheries or DoF staff and farmers:

Section 3.3.4). In addition, potential risk factors were also mined from an earlier

‘IntegratedFarmerSurvey’(Section3.3.3).Adiverserangeof24factors(Table3.1)was

short-listed covering farm scale and location, farming experience, management

practices,farminfrastructureandwatermanagementcharacteristics.

Table 3.1 Candidate risk factors for AHPND occurrence at farm level

Mainfactor Candidateriskfactors

Farmscaleandlocation Farmscaleandregion

Farmingexperience Ageoffarm

Managementpractices Ageofpondaffected,ponddryingduration,stockingdensity,pondharrowingbeforestocking,effluenttreatment,sedimentremovalandsedimentfate

Farminfrastructure Ongrowingpondtype,maximumwaterdepth,cleanwaterstorage,alternativecleanwaterstorageandgrowingpond,andsedimentpond

Watermanagementcharacteristics

Watersource,dischargemethod,waterstoragemethod,recirculateandreusewater,maximumwaterreplacement,waterexchangefrequency,maximumsalinity,averagepHandaveragealkalinity


95

3.3.3 Sampledesign

An ‘Integrated Farmer Survey’ (IFS: Murray et al., 2013), separately enumerated

betweenDecember2010andJuly2011(i.e.immediatelypriortotheonsetofAHPND),

providedasampleframeof206farmsforthisresearch.Theprimaryselectionphaseof

206samplefarmswasbasedondistrict-wiseproductiondatacollectedfromprovincial

DoF offices between 2009 and 2014 (Thailand DoF, 2016), followed by randomised

selectionoffarmclustersandindividualfarmswithinclusters.Theresultsofthecurrent

study are therefore expected to be generalisable to wider Thai shrimp production

conditions.

Inmore detail, this framewas based on amulti-phase, random sampling approach

stratified on farm-scale and location, i.e. being conducted across the principal Thai

farming areas in east central (Chachoengsao and Chanthaburi provinces) and south

Thailand (mainly Suratthani province and some areas in Songkhla province). Farm

selection was stratified on three scale levels—large, medium and small—based on

indicatorsofbusinessownership,labourpattern,farmmanagementandthenumberof

ongrowingponds,asshowninTable3.2(Murrayetal.,2013).

Table 3.2 Criteria used for classifying Thai shrimp farms into three scales: small, medium and large (from Murray et al., 2013)

CriteriaFarmscale

Small Medium Large

Ownershipofbusiness Householdorextendedfamily

Householdorexternalowner

Corporate(i.e.jointstockcompany)

Full-timelabour(non-family)

<3Full-timelabour Yes Yes

Management Householdorextendedfamily

Householdorexternalsalariedmanager

Salariedmanager

Numberofongrowingponds

<3 >2 >2


96

AsakeyvariabletodetermineAHPNDspread,thelocationoffarmscorrespondedto

differentonset-timesforAHPND,i.e.firstintheeastfollowedbythesouth,andalsoco-

variedwithscaletosomedegree,asfarmstendedtobelargeronaverageinthesouth.

Informalorganisationoffarmersthroughshrimpfarmerclubsalsotendedtobemore

advancedinthesouth(Kassametal.,2011)and,therefore,hypothetically,alsotheir

abilitytocoordinatecollectiveactioninresponsetodiseaseoutbreaks.

3.3.4 Surveydesign

Primarydatacollectionwasbasedonfouriterativesurveyphasesconductedbetween

March2013andDecember2014,asdescribedbelow.All interviewswereconducted

with permission of the farmers in the sample-frame. The interviews took place at a

convenienttimeforthefarmers,andwereconductedintheThailanguagebythemain

authorwhoisaThainativespeaker.

Phase1Brieftelephonesurvey.Anattemptwasmadetocontactbytelephoneall

206 farmers in the sample-frame. With a short 15–20 minute structured survey,

respondentswereaskedaboutdiseaseproblemsandpossibleassociatedriskfactorson

theirfarmssincethefirstreportsofAHPNDinThailandatthelateof2011.Questions

coveredthetiming(month)ofthefirstAHPND-typelosses,clinicalsigns,numbersof

pondsaffected,mortality rates, ageofdiseased shrimp,managementpractices, and

measurestakentomitigatediseasetransmissionorrecurrence,asshowninAppendixB

attheendofthisthesis.

Phase2In-depthface-to-faceinterview. This more in-depth structured survey

incorporatedthepre-prepareddecision-treeanddiseasecardsdescribedabove.Atotal

of143farmerswhohasrespondedatPhase-1farmersprovidedthesample-framefor

thisphase.Arandom-selectionmethodwasusedtoselect14–15farmsforinterviewin

eachofthethreefarm-scaleandthreeprovinces(i.e.twoeasternandonesouthern).

An interview was arranged in their farm. It took approximately 30–45minutes per

interview.Notethatthenumberoffarmsineachscalecategorywaslessbalanceddue

tolimitsontotalnumbersineacharea.


97

Phase3Follow-upgap-fillinganddatavalidation. Thisphasewasconductedwith

Phase-2farmersbyphonebetweenJuneandSeptember2013.Itincludedcollectionof

additionaldataonstockingdensitycharacteristics,ageofaffectedponds,ponddrainage

andsludgeremovalpractices,andwaterqualityconditions(pH,salinityandalkalinity)

Phase4Cross-checkingoftheriskfactors.Thefinalphasewastoexplorefarmers’

perspectivesontheriskfactorsinrespecttoAHPNDobtainedfromthecross-sectional

analysis,andanychangeinlocalshrimpfarmingpractices.Severalopen-questionswere

usedtoasknineskilledpersonsfromwithinthesampleframeinDecember2014.An

interviewwasarrangedintheirfarmandtookapproximately50minutesto1hour.This

interviewallowedthefarmers’opinionsandbeliefsinallrespects.

3.4 DataAnalysis

3.4.1 DescriptiveanalysisoftheoccurrenceofAHPNDandotherdiseases

TheprevalenceofputativeAHPNDcases,andotherdiseases,wasplottedagainstthe

locationand farm-scale stratificationvariables. The cumulativemonthly incidenceof

newAHPNDcaseswasalsoplottedagainstthesamevariables,calculatedasthenumber

ofnewcasespermonthtimedividedbythenumberofthetotalrespondingpopulation

atrisk ineachstratificationcategory.Basedonthetimingofthefirstreportedcases

within the sample frame, we used two location-specific exposure times for case

definitions.First,casesintheeasternprovincesrangedfromJanuary2012toMay2013.

Second,casesinthesouthernprovincesrangedfromDecember2012toMay2013.

3.4.2 StatisticalanalysisoftheriskfactorsforAHPND

TheaimofthestatisticalanalysiswastoevaluatetheriskfactorsforAHPNDoccurrence

at farm level. All statistical analyses were conducted within the 𝑅 Programme

Environment (R foundation for statistical computing, 2015). The risk of AHPNDwas

estimated as an odds ratio (𝑂𝑅); this measure is often used to investigate risks

associatedwithrarediseases(SchmidtandKohlmann,2008):


98

𝑅GHI7JGK = CaseswithriskfactorNon-caseswithit

𝑅;78[\76 = CaseswithoutriskfactorNon-caseswithoutit

𝑂𝑅 = 𝑅GHI7JGK𝑅;78[\76

where𝑂𝑅 >1:riskfactorforincreasedriskofAHPND

𝑂𝑅 <1:riskfactorfordecreasedriskofAHPND

𝑂𝑅 =1:riskfactorisindependentofAHPNDpresence.

Usingtwo-stepmethods,the24candidatevariables(Section3.3.2)werescreenedusing

univariatetestsfollowedbymultivariatetests.Fortheunivariatetests,thesignificance

ofthe𝑂𝑅wasassessedthrough95%confidenceintervals(CI),Fisher’sExactTestor

binary logistic regression. Multivariate tests were done using unconditional logistic

regressionmodelstoevaluatethepotentialriskfactorsthatwereobtainedfromthe

univariatetestsabove.AllvariableswithaP-value<0.1inunivariatemodelswerekept

totestusingbackwardstepwise(variableswereremovedfromthemodel)regression

procedure.Givenseveral competingmodels, theAIC (Akaike’s InformationCriterion)

wasusedasacomparisontool,inwhichthebetter-fittingmodelshowedalowervalue

ofAIC(Akaike,1974).

Thepredictiveperformanceofthemodelswasevaluatedintwoways:

(1) Cross-validation technique:Using a random-selectionmethod, farm samples in a

groupofcasesandcontrolsweredividedintotwoapproximatelyequal-sizesubsets:D1

andD2.ModelswereconstructedonD1asatrainingset,andthepredictiveabilityofthe

modelswastestedusingD2asatestingset(Morietal.,1999).Then,wecomputedthe

proportion of observations, which were correctly predicted. Cross validation was

performedbothwaysround(swappingthesets),andanaverageoftheaccuracieswas

calculatedandreported.


99

(2)Receiveroperatingcharacteristic(ROC)curve:TheROCapproachwasusedtojudge

theperformanceofmodelsintermsofpredictingAHPNDoccurrencewithtwopossible

outcomes:cases(diseasedfarms)andcontrols(non-diseasefarms).Weusedarangeof

different cut-off points obtained from estimated probabilities (log odds) to define

predictedcasesandnoncases.TheycoordinatesontheROCcurveforamodelwere

derived as true positives (sensitivity), and the x coordinates were derived as false

positives(1-specificity)ateachcut-offpoint.BasedontheareaundertheROCcurve

(AUC), themodel was classified to be either an informativemodel (AUC>0.5; the

predictiveresultofmodelisbetterthanarandommodel)oranuninformativemodel

(AUC=0.5;thepredictiveresultofmodelisnotdifferentfromarandommodel)(Alonzo

andPepe,2002;KumarandIndrayan,2011).

Toaidunderstanding,theoverallmethodologyisshowninFigure3.1.


100

Figure 3.1 A flow chart of the methodology used in evaluating risk factors for transmission of acute hepatopancreatic necrosis disease (AHPND) in Thai shrimp farming. The survey design contained four major phases. Data analysis evaluated the risk factors for AHPND in three steps.

3.5 Results

3.5.1 IdentificationofAHPNDcasesandcontrols

3.5.1.1Case-identificationAHPNDdecisiontree

Acase-identificationAHPNDdecisiontreewasdevelopedforthisstudybecause,atthe

time,AHPNDwasadiseaseofunknownetiology.Four indicatorsofputativeAHPND

infectionwereincorporatedintothisdecisiontree,resultinginfarmsbeingassignedto

one of three mutually exclusive infection categories: higher or lower probability of

Surveydesignandenumeration Dataanalysis

Phase1 Brieftelephonesurvey

Phase2 In-depthface-to-faceinterviews

Phase3 Telephonesurveytofollow-upgap-fillinganddatavalidation

Step1:Caseandcontrolclassificationusingacase-identificationAHPNDdecisiontreeStep2:Descriptiveanalysis(diseasestatusreport,cumulativeincidencetrends)Step3:Statisticalanalysis

- Calculationofoddsratio(!")andconfidenceinterval(CI)

- Unconditionallogisticregressionanalysis- Modelvalidations(crossvalidationandROC

approach)

Phase4 Cross-checkingoftheriskfactors


101

AHPNDinfection,ornoAHPND(Figure3.2).Justificationsfortheindicators,constituting

fourtiersinthedecision-treewereasfollows:

Tier1DateofonsetofthefirstAHPNDclinicalsigns.Theonsettimingofputative

caseshad tooccurafter the first reportedcasesofAHPNDofficially reportedby the

ThailandDepartmentofFisheries(DoF)indifferentregions, i.e.thefourthquarterof

2011(OctobertoDecember)ineasternprovinces(FAO,2013),andthefourthquarter

of2012insouthernprovinces,accordingtooursurvey.

Tier2Ageofaffectedshrimp.BasedonthereportsofFlegelandLo(2014)andFAO

(2013),any infectionthatoccurredbeyond35daysafter firststockingofpost-larvae

wasexcludedfrombeingaputativeAHPNDcase.

Tier3Mortalityratesofinfectedfarm.AkazawaandEguchi(2013)reportedaverage

mortalityratesof70–80%in‘AHPNDaffectedpondsinalargecommercialshrimpfarm

inMalaysia(AkazawaandEguchi,2013)whilstmortalitiesapproaching100%havebeen

observedinothercountries(EduardoandMohan,2012).Consequently,caseswith70–

100%mortalityratewereclassedashavinghigherprobabilityofAPHNDinfectionwhilst

farmswithratesbetween10–70%wereclassifiedashavinglowerAHPNDprobability.

Tier4MultiplevisibleclinicalsignsoftheAHPNDpathology.BasedontheNACA

advisory (2014),casesclassifiedas lowerorhigherprobabilityAHPNDcases inTier3

were subjected to further confirmation according to the following additional

behaviouralgrosspathologicalsigns:

(1) Mortality location: Mortalities concentrated around the pond edge and pond

bottom.

(2)Shrimpbehaviour(spiralorswirlingswimmingtoedgeofpondandturningbelly-up),

and

(3)Pale/whitish/atrophyofthehepatopancreas(HP;alsoHPhardtocrushbyhand),

anddiscolourationofabdominalmuscle(opaque;pinkishorwhite).


102

Havingdefinedhigherand lowerprobabilitycases fromTier3, thehigherprobability

casesneededtohaveonly(3)describedabovetobeconfirmed,whilsttwo,oneofthese

was(3),ormoreclinicalsignswererequiredtoconfirmlowerprobabilitycases.Inboth

instances,failuretorecogniseanyoftheseclinicalsignsresultedindemotionofthecase

toanon-AHPNDlosscause.Thecasesmayhaveco-infectionofothershrimpdiseases

suchaswhitespotdiseaseandyellowheaddisease.

Figure 3.2 The AHPND decision tree for determination of higher AHPND probability, lower AHPND probability, and no AHPND. The decision was based upon four tiers including the significant behavioural gross pathological clinical signs of AHPND.

3.5.1.2Completedinterviews

Interviewswere completed for 143 of the 206 farms in the original sample (70%).

Reasonsfornon-responsewereasfollows:48farms(76%)didnotrespondtocallsor

Timingoffirstsymptoms East- BeginningatOctober2011orSouth- BeginningatOctober2012

Ageofaffectedshrimp≤35dayspoststocking

Mortalityrate70–100%

Mortalityrate<70%

Mortalityrate<10%

Ifthepathological

condition(3)met

If≥ 2pathological

conditions met,oneoftheseistheclinicalsign

(3)

HigherAHPNDprobability

LowerAHPNDprobability

NoAHPND

Tier1

Tier2

Tier3

Tier4

Outcomes

NoNo

Pathological clinical signs(1)Mortalitylocation:Mortalitiesconcentratedaround thepondedgeandpondbottom.(2)Shrimpbehaviour (spiralorswirlingswimmingtoedgeofpondandturningbelly-up)(3)Pale/whitish/atrophyofthehepatopancreas (HP;alsoHPhardtocrushbyhand),anddiscolourationofabdominalmuscle(opaque;pinkishorwhite)


103

theirnumberswerenolongervalid;15farms(24%)hadceasedtoengageinshrimp

farming(Table3.3).

Table 3.3 The outcome from the telephone survey (Phase 1) followed by face-to-face interviews (Phase 2)

Region East South

TotalProvince Chanthaburi Chachoengsao Suratthani Songkhla

farmscale S M L S M L S M L S M L

Completedinterview

22 7 3 43 9 0 27 25 4 0 0 3 143

Non-responseforcalling

4 2 0 7 2 0 6 10 0 0 0 4 35

Invalidcontactnumber

3 0 1 3 1 0 4 1 0 0 0 0 13

Notengagedinshrimpfarming

3 2 0 3 1 0 5 0 0 0 0 1 15

Total 32 11 4 56 13 0 42 36 4 0 0 8 206

S=smallfarm;M=mediumfarm;L=largefarm

3.5.1.3Caseandcontrolsamples

Basedonthedecision-treeoutcomes,the143respondingfarmswereeachassignedto

oneofthreegroups:(1)51wereclassedashavingahighprobabilityofAHPND,(2)55

farmsreportednodiseaselossorclinicalsignsconsistentwithotherdiseases,whilst(3)

theremaining37farmshadanindeterminatestatus,i.e.low-probabilityofAHPND.To

increasestatisticalpower,weomittedthese37farmswiththeindeterminatestatusof

group3fromfurtheranalysis.Thefirsttwogroupswereassignedascaseandcontrol

groups,respectively(Table3.4).


104

Table 3.4 Cross-tabulation of outcomes for case and control samples

FarmScale

GeographicLocationTotal

Eastregion Southregion

Case Control Case Control Case Control

Small 26 24 5 14 31 38

Medium 8 3 9 12 17 15

Large 2 0 1 2 3 2

Total 36 27 15 28 51 55

3.5.2 DescriptiveepizoologyofAHPND

3.5.2.1Reportofdiseasestatus

ThediseasestatusofoursamplefarmsisshowninFigure3.3.AHPNDpresentedamajor

diseaseproblemforshrimpfarmsinthesampleframeinbothregionsofThailand,with

ahigherAHPNDoccurrenceof43%(36of84)ofsamplefarmsintheeastcomparedto

25%(15of59)ofsamplefarmsinthesouth.Consistentwiththeearlieronsetofthe

disease in the east, Chanthaburi province had the highest AHPND prevalence,

accountingfor44%(14of32)ofsamplefarmsinChanthaburi,followedby42%(22of

52)ofsamplefarmsinChachoengsao,and37%(15of56)ofsamplefarmsinSuratthani.

The figure also shows that farms infected with white spot disease also had a high

probabilityofAHPNDinfection,asfoundinsamplefarmsinChanthaburiandSuratthani.

FarmswithoutAHPNDclinicalsignswereinfectedwithwhitespotdisease,yellowhead

disease,taurasyndromeandwhitefecessyndrome.Theresults,therefore,providedan

overview of the disease problems facing the Thai shrimp farming sector during this

epizootiologicalsurvey.


105

Figure 3.3 Report of disease status stratified according to geographic location and farm-scale between January 2012 and May 2013. AHPND was the main disease problem for Thai shrimp farming. Other diseases included white spot disease, yellow head disease, taura syndrome and white feces syndrome.

3.5.2.2Cumulativeincidencetrendsgraphed

Intheeasterndistricts,thecumulativeAHPNDincidencewas43casesper100farmsin

17monthsortwocasesper100farm-months,andinthesouthwas25casesper100

farmsinsixmonthsorfourcasesper100farm-months(Figure3.4).Itwasseenclearly

thatboththesouthernlocationandthelarge-scalevariablesshowedadelayofAHPND

onset compared with remaining factors (the eastern location, and the small- and

medium-scalevariables),buttheircumulativeincidenceincreasedsharplyafterthefirst

incidenceofAHPND.

0%

20%

40%

60%

80%

100%

Small Medium Large Small Medium Large Small Medium Large Small Medium Large

Chanthaburi Chachoengsao Suratthani Songkhla

East South

Numbe

roffarms(%)

Farmlocationandscale

22%

39%

6%

33%

17%

83%

100% 16% 40% 5% 29% 50% 100%

25%

59% 60%

69%

26%

29%

29%

13%

50%


106

(a) (b)

Figure 3.4 The cumulative incidence of AHPND between January 2012 and May 2013, accounting to two regions (a) and three farm-scales (b).

3.5.3 RiskfactorsforAHPNDtransmissionatfarmlevel

The results of the univariate analysis are summarised in Table 3.5. Two of the 24

variableswereassociatedwithAHPNDatfarm-levelatthe0.05significancelevel.The

firstvariablewastheearthenpond–asignificantriskfactorforAHPNDwithan𝑂𝑅of0.25

(CI 0.06–0.8; P-value=0.02); the second was the absence of pond harrowing–a

significantriskfactorforAHPNDwithan𝑂𝑅of3.9(CI1.3–12.6;P-value=0.01).These

twovariablesweremodelled in furtheranalysis,whereas theremaining22variables

whichwerenotpredictiveofAHPNDaccordingtotheseunivariatetests(P-value>0.1)

wereexcluded.

JAN DEC2012

JAN MAY2013

051015202530354045 East

South

Rateper100

JAN DEC2012

JAN MAY2013

Small-scaleMedium-scaleLarge-scale

051015202530354045


107

Table 3.5 The statistically significant risk factors for AHPND with odds ratios (𝑶𝑹s) and 95 % confidence intervals

Variable CaseNo. ControlNo. P-value 𝑶𝑹(CI)

Ongrowingpondtype(1)

Earthenpond(2) 32 460.02 0.25(0.06–0.8)

Linedpond 14 5

Pondmanagement(1)

Nopondharrowing(2) 33 240.01 3.9(1.3–12.6)

Pondharrowing 7 20

Note–(1)Fisher’sexacttestand(2)Referencelevel

FurtheranalysistoevaluateriskfactorsforAHPNDwasperformedusingunconditional

logistic regression models. Variables with a P-value<0.1 were kept to be run in

backwardstepwiseunconditionallogisticregressionmodels.Themodelsonlyincluded

casesinthedatawherealldatafieldsofthetwovariables(ongrowingpondtypeand

pondmanagement)werecompleted.

Theresultsoftheunconditionallogisticregressionmodelsthatfittedthedatabestare

showninTable3.6.ThefirstunconditionallogisticregressionmodelhasthelowerAIC

value.Thefirstnestedmodelcontainedongrowingpondtypeandpondmanagement

(AIC=102.19);thesecondmodelcontainedpondmanagement(AIC=107.29).


108

Table 3.6 Unconditional logistic regression analysis of risk factors for AHPND

Model Variable CaseNo. ControlNo. Exp(coefficient)(1) SE(coef)

Model1

AIC=102.19

Ongrowingpondtype

Pondmanagement

39 41

0.19

4.35

0.67

0.47

Model2

AIC=107.29

Pondmanagement

3.95 0.52

Note–(1)Maximumlikelihoodestimationoddsratios

Formodelvalidation,thecross-validationanalysisispresentedinTable3.7.Themean

percentage correct of the two models obtained from the unconditional logistic

regressionisaround65%accordingtoourdatasets(referredtoas“D1”and“D2”).

Table 3.7 Cross-validation results on the AHPND models obtained from unconditional logistic regression

Model TestingonD1 TestingonD2 Mean

Model1:Ongrowingpondtypeandpondmanagement

68% 63% 66%

Model2:Pondmanagement 65% 63% 64%

UsingtheROCapproach,bothmodelswereplottedfortheirabilitytopredictAHPND

presenceat farm level.Theproportions inabinaryoutcome(diseaseornon-disease

farm)weremodelledbyusingapredictionof the logoddsandusingvariouscut-off

pointstodefinethepredictedoutcomes.Thetruepositives(sensitivity)ofthemodel

werepresentedontheverticalaxis,andthefalsepositives(1-specificity)wereshown

onthehorizontalaxis.

TheROCresultsarepresentedinFigure3.5.TheareaundertheROCcurve(AUC)ofboth

modelswascirca68%,i.e.ahighervaluethananuninformativemodel(AUCof0.5).


109

Figure 3.5 ROC curves for AHPND models. Model 1 with two variables: ongrowing pond type and pond management. Model 2 with one variable: pond management. Both models obtain the AUC > 0.5 (better results than random). The diagonal line indicates the prediction from a random model.

TheseriskmodelsofAHPNDandtheirvalidationimpliedthat,inthecompletesubset,

thereweresignificantinteractionsbetweenpondharrowingpractice,ongrowingpond

typeandtheincidenceofputativeAHPNDcasesoccurringatfarm-level.Withthesetwo

riskfactors, itcanbesuggestedthatshrimpfarmersshouldapplypondharrowingas

oneofthemostimportantfarmingpractices.Thissuggestionfollowsthecomputation

ofthe𝑂𝑅whichinterpretsthatthesamplefarmswheretheabsenceofpondharrowing

beforestockingwereinfectedwithAHPND3.9timesmoreoftenthanthesamplefarms

thatappliedpondharrowing.

3.6 Discussion

The initial AHPND distribution in Thailand during January 2012 and May 2013 was

examinedinthesampleframeofthisobservationalepizoologystudy.Thefirstincidence

occurredintheeasternprovincesinJanuary2012,anddelayedincidenceinthesouth

in December 2012, at a higher cumulative incidence. A need for improvements in

biosecurity in Thai shrimp farming is implied through this research, given that the

incidenceofAHPNDoccurredinallfarmscales,eventhelargecommercialfarmswhich

generallyinvestmorebiosecurityresourcesinshrimpfarmingthanothers.

In this research, the identified risk factors for AHPND transmission emphasise the

importance of environmental farming managements. One of these risk factors is

1- Specificity 1- Specificity

Sensitivity

Model1:AUC=0.71 Model2:AUC=0.64


110

earthenponds to raiseshrimp.Earthenpondswitha largeareaofpondsoilplayan

important role in shrimp farming production. For example, they provide a higher

capacity toaccumulateandabsorbnutrients (nitrogenandphosphorus) andorganic

matter (i.e.uneaten feed, faecesandmoribundshrimp),comparedwith linedponds

(Burfordetal.,2003;Funge-Smith,1996).Moriarty(1997)proposesthatearthenponds

providealargerhabitatformicroorganisms,whichisamajormechanismtoenhance

thefoodwebwithinponds,whilelinedpondshavealimitedcapacityinthisregard.The

accumulatedsediment,however,mayexceedthecapacityofpondstodecomposethe

nutrients,causingpoorwaterquality,andtoxicitytofarmedshrimp,i.e.fromnitriteand

ammonia (Avnimelech and Ritvo, 2003; Boyd et al., 2002; Hargreaves, 1998).

Predominantly, shrimp farm sediment is one of the important habitats for vibrios

(Lekshmyetal.,2014;Thoetal.,2012;Wallingetal.,2010),includingthepathogenic

agentofAHPND(Kongkumnerd,2014).

Incontrast,linedpondsincreasetheriskofAHPNDinfection.Thitamadeeetal.(2016)

notethattheutilisationofpondliningisaneffectivetoolfordiseaseprevention,but

fully lined ponds may contain gaps in the plastic sheets from installing aerators or

feedingtrays,andduetotheshortlifetimeofplasticsofaround2–5years.Theseleaks

allowanaerobicorganismstogrowunderneaththeplasticsheetsandincrementtherisk

of AHPND. The disadvantage of pond lining has been stated by Boyd (2014), i.e.

phytoplankton blooms, low-alkalinity water, and high amounts of sludge (organic

matter).Hence,linedpondsarealsolikelytoberelatedtoenvironmentalproblemsin

shrimpfarming.

Withearthenponds,thefarmerscanfullymanagetheanaerobicconditionsofAHPND

pathogens, and other comparative organisms through proper pond management

techniques,suchasliming,ponddrying,andpondbottomharrowing.Accordingtoour

research,notperformingharrowingbeforestockingwasafactorthatincreasedtherisk

of AHPND transmission. This finding is explained by the advantages of harrowing in

enhancing the shrimppondenvironment.Whenpond soil is exposed to theair, soil

respirationincreases(BoydandPippopinyo,1994;EgnaandBoyd,1997;Xinglongand

Boyd,2006).Predominantly, itcontributestoa loweramountoftoxicgases,suchas

hydrogensulphideandnitrite, inpondsoilbecausethesetoxicgasesareoxidisedto


111

non-toxicforms(Adhikarietal.,2012;Boydetal.,2002).Therefore,bothahabitatof

facultative anaerobic bacteria such asV. parahaemolyticus (Youngren-Grimes et al.,

1988) and toxic gases should be decreased in shrimp ponds that are harrowed. In

addition,pondharrowingwithponddryingandprobiotics loadingisabletoenhance

decompositioninsediment,andtoeliminatethenumbersofVibriolivinginsediment

efficiently(Boyd,2003;Moriarty,1998;Moriarty,1999;Nimratetal.,2008).DeSchryver

etal.(2014)alsosuggestthatgoodmicrobialmanagementwithinshrimppondscanbe

anexcellentstrategyforAHPNDpreventionratherthanusingdisinfecting.

Theuseofacase-identificationAHPNDdecisiontreetodeterminecaseandcontrolwas

discussedhere.Thecase-identificationAHPNDdecisiontreeisaflexibleandintelligent

epizootiological tool: it supports epizoology in terms of being a quicker and less

expensiveanalysis. Furthermore, the case-identificationAHPNDdecision tree canbe

adaptedwhenthereisrecurrenceofAHPND,orthegrosssignsofdiseasepathologycan

bechangedwhenthere is incidenceofothernewdiseases. Identifyingcaseswithan

AHPNDdecisiontreemaybelessaccuratethanbyusinglaboratoryhistopathologyand

PCRtestingasdiagnostictools.Themainreasonisthatthiscase-identificationAHPND

decisiontreewasdevelopedforadiseaseofunknownetiology,thuswehavetohave

anidentifiedpathogeninordertohaveadefinitivecasedefinition.

Thisresearchhasseverallimitations.Thepossibilityforself-selectionbiasmayarisein

the survey design because the sample farms may have lower operating costs,

lower quality facilities, or misunderstanding of appropriate biosecurity practices,

meaning that the sample farms are more likely to get infections because they are

inherently riskier. We found, however, that self-selection according to biosecurity

practiceshadonlyasmallbiasonriskestimations.Asseenbythecumulativeincidence

ofAHPND(Figure3.4),large-scalefarmstendedtohaveaslowerincidenceofAHPND

thanmedium-andsmall-scalefarms.Thisshowedthatstratificationcouldminimisethe

self-selectionbiasintheresearch.Moreover,withpriorbusinessinThaishrimpfarming

andexperiencewithdiseaseproblems, thePhase1population shouldprovide good

representatives in Phase 2 in terms of providing information about the association

betweenAHPNDoccurrenceandriskfactors.


112

Alltheidentifiedriskfactorsconveytheneedforchangesinfarmmanagementpractices

at shrimp farming sites. Other risk factors for AHPND may be included in further

researchandcross-sectionalstudies,however.Effectivestrategiestodevelopdisease

prevention and control at the country level are also needed. This can be achieved

throughanalysing the structureof the live shrimpmovementnetwork to gainmore

understanding of howdisease epizootic dynamics play out across thewhole shrimp

farmingsectorinThailand.


113

3.7 References

Adhikari,S.,Lal,R.andSahu,B.C.(2012)CarbonsequestrationinthebottomsedimentsofaquaculturepondsofOrissa,India.EcologicalEngineering,47,pp.198-202.

Akaike,H.(1974)Anewlookatthestatisticalmodelidentification.AutomaticControl,IEEETransactionsOn,19(6),pp.716-723.

Akazawa,N.andEguchi,M.(2013)EnvironmentaltriggerforEMS/AHPNSidentifiedinAgrobestshrimpponds.GlobalAquacultureAdvocateJuly/August2013,pp.16-17.

Alonzo,T.A.andPepe,M.S.(2002)Distribution-freeROCanalysisusingbinaryregressiontechniques.Biostatistics(Oxford,England),3(3),pp.421-432.

Al-Tawfiq,J.A.,Hinedi,K.,Ghandour,J.,Khairalla,H.,Musleh,S.,Ujayli,A.andMemish,Z.A.(2014)MiddleEastrespiratorysyndromecoronavirus:acase-controlstudyofhospitalizedpatients.ClinicalInfectiousDiseases:AnOfficialPublicationoftheInfectiousDiseasesSocietyofAmerica,59(2),pp.160-165.

Austin,B.(2010)Vibriosascausalagentsofzoonoses.VeterinaryMicrobiology,140(3–4),pp.310-317.

Avnimelech,Y.andRitvo,G.(2003)Shrimpandfishpondsoils:processesandmanagement.Aquaculture,220(1–4),pp.549-567.

Boyd,C.E.andPippopinyo,S.(1994)Factorsaffectingrespirationindrypondbottomsoils.Aquaculture,120(3–4),pp.283-293.

Boyd,C.E.(2003)Bottomsoilandwaterqualitymanagementinshrimpponds.JournalofAppliedAquaculture,13(1-2),pp.11-33.

Boyd,C.E.(2014)Relationshipsbetweenpondbottomsoilmanagementandwaterqualityinaquacultureponds.Adelaide,SouthAustralia,7-11June2014.Adelaide:WorldAquacultureSociety,pp.1-29.Available:https://www.was.org/documents/MeetingPresentations/WA2014/WA2014_0263.pdf[Accessed:11April2017].

Boyd,C.E.,Wood,C.andThunjai,T.(2002)Aquaculturepondbottomsoilqualitymanagement.Oregon:PondDynamics/AquacultureCollaborativeResearchSupportProgram,OregonStateUniversity.

Burford,M.A.,Thompson,P.J.,McIntosh,R.P.,Bauman,R.H.andPearson,D.C.(2003)Nutrientandmicrobialdynamicsinhigh-intensity,zero-exchangeshrimppondsinBelize.Aquaculture,219(1),pp.393-411.


114

Chonsin,K.,Matsuda,S.,Theethakaew,C.,Kodama,T.,Junjhon,J.,Suzuki,Y.,Suthienkul,O.andIida,T.(2016)GeneticdiversityofVibrioparahaemolyticusstrainsisolatedfromfarmedPacificwhiteshrimpandambientpondwateraffectedbyacutehepatopancreaticnecrosisdiseaseoutbreakinThailand.FEMSMicrobiologyLetters,363(2),pp.1-8.

Chrisolite,B.,Thiyagarajan,S.,Alavandi,S.V.,Abhilash,E.C.,Kalaimani,N.,Vijayan,K.K.andSantiago,T.C.(2008)DistributionofluminescentVibrioharveyiandtheirbacteriophagesinacommercialshrimphatcheryinSouthIndia.Aquaculture,275(1–4),pp.13-19.

Costa,R.A.,Silva,G.C.,Peixoto,J.R.,Vieira,G.H.andVieira,R.H.(2010)QuantificationanddistributionofVibriospeciesinwaterfromanestuaryinCeará-Brazilimpactedbyshrimpfarming.BrazilianJournalofOceanography,58(3),pp.183-188.

DeSchryver,P.,Defoirdt,T.andSorgeloos,P.(2014)Earlymortalitysyndromeoutbreaks:amicrobialmanagementissueinshrimpfarming?PLoSPathogens,10(4),pp.e1003919.


Egna,H.S.andBoyd,C.E.(1997)Dynamicsofpondaquaculture.CRCpress.


Flegel,T.W.(1997)Majorviraldiseasesoftheblacktigerprawn(Penaeusmonodon)inThailand.WorldJournalofMicrobiologyandBiotechnology,13(4),pp.433-442.

Flegel,T.W.andLo,C.(2014)Announcementregardingfreereleaseofprimersforspecificdetectionofbacterialisolatesthatcauseacutehepatopancreaticnecrosisdisease(AHPND).Bangkok:NACA.Available:http://www.enaca.org/publications/health/disease-cards/ahpnd-detection-method-announcement.pdf[Accessed:5December2014].

Flegel,T.W.(2012)Historicemergence,impactandcurrentstatusofshrimppathogensinAsia.JournalofInvertebratePathology,110(2),pp.166-173.

Funge-Smith,S.J.(1996)WaterandsedimentqualityindifferentintensiveshrimpculturesystemsinsouthernThailand.CoastalAquacultureandEnvironment:StrategiesforSustainability.ODAResearchProject,6011,pp.1-27.


115

Han,J.E.,Tang,K.F.,Tran,L.H.andLightner,D.V.(2015a)Photorhabdusinsect-related(Pir)toxin-likegenesinaplasmidofVibrioparahaemolyticus,thecausativeagentofacutehepatopancreaticnecrosisdisease(AHPND)ofshrimp.Dis.Aquat.Org,113,pp.33-40.

Han,J.E.,Mohney,L.L.,Tang,K.F.J.,Pantoja,C.R.andLightner,D.V.(2015b)PlasmidmediatedtetracyclineresistanceofVibrioparahaemolyticusassociatedwithacutehepatopancreaticnecrosisdisease(AHPND)inshrimps.AquacultureReports,2,pp.17-21.

Hargreaves,J.A.(1998)Nitrogenbiogeochemistryofaquacultureponds.Aquaculture,166(3–4),pp.181-212.

Jiravanichpaisal,P.,Miyazaki,T.andLimsuwan,C.(1994)Histopathology,biochemistry,andpathogenicityofVibrioharveyiinfectingblacktigerprawnPenaeusmonodon.JournalofAquaticAnimalHealth,6(1),pp.27-35.

Kassam,L.,Subasinghe,R.andPhilips,M.(2011)Aquaculturefarmerorganizationsandclustermanagement.FAOFisheriesTechnicalPaperNo.563.Rome:FAO.Available:http://www.fao.org/docrep/014/i2275e/i2275e.pdf[Accessed:29April2017].

Kautsky,N.,Rönnbäck,P.,Tedengren,M.andTroell,M.(2000)Ecosystemperspectivesonmanagementofdiseaseinshrimppondfarming.Aquaculture,191(1–3),pp.145-161.

Kondo,H.,Tinwongger,S.,Proespraiwong,P.,Mavichak,R.,Unajak,S.,Nozaki,R.andHirono,I.(2014)DraftgenomesequencesofsixstrainsofVibrioparahaemolyticusisolatedfromearlymortalitysyndrome/acutehepatopancreaticnecrosisdiseaseshrimpinThailand.GenomeAnnouncements,2(2),e00221-14.Available:http://genomea.asm.org/content/2/2/e00221-14.full[Accessed:2May2014].

Kongkumnerd,J.(2014)ThecurrentstatusofEMSoutbreakinThaimarineshrimpfarming.Bangkok:ThailandDepartmentofFisheries(DoF).Available:http://www.shrimpaqua.com/download/EMS/situation-EMS.pdf[Accessed:10December2014].

Kumar,R.andIndrayan,A.(2011)Receiveroperatingcharacteristic(ROC)curveformedicalresearchers.IndianPediatrics,48(4),pp.277-287.

Kumar,B.K.,Deekshit,V.K.,Raj,J.R.M.,Rai,P.,Shivanagowda,B.M.,Karunasagar,I.andKarunasagar,I.(2014)DiversityofVibrioparahaemolyticusassociatedwithdiseaseoutbreakamongculturedLitopenaeusvannamei(Pacificwhiteshrimp)inIndia.Aquaculture,433,pp.247-251.

Lai,H.,Ng,T.H.,Ando,M.,Lee,C.,Chen,I.,Chuang,J.,Mavichak,R.,Chang,S.,Yeh,M.,Chiang,Y.,Takeyama,H.,Hamaguchi,H.,Lo,C.,Aoki,T.andWang,H.(2015)


116

Pathogenesisofacutehepatopancreaticnecrosisdisease(AHPND)inshrimp.Fish&ShellfishImmunology,47(2),pp.1006-1014.

Le,T.X.andMunekage,Y.(2004)ResiduesofselectedantibioticsinwaterandmudfromshrimppondsinmangroveareasinVietnam.MarinePollutionBulletin,49(11–12),pp.922-929.

Lekshmy,S.,Nansimole,A.,Mini,M.,Athira,N.andRadhakrishnan,T.(2014)OccurrenceofVibriocholeraeinshrimpcultureenvironmentsofKerala,India.IndianJournalofScientificResearch,5(2),pp.151.

Leobert,D.,Cabillon,N.A.R.,Catedral,D.D.,Amar,E.C.,Usero,R.C.,Monotilla,W.D.,Calpe,A.T.,Fernandez,D.D.andSaloma,C.P.(2015)Acutehepatopancreaticnecrosisdisease(AHPND)outbreaksinPenaeusvannameiandP.monodonculturedinthePhilippines.DisAquatOrg,116,pp.251-254.

Lightner,D.V.,Redman,R.M.,Pantoja,C.R.,Noble,B.I.andTran,L.(2012)EarlymortalitysyndromeaffectsshrimpinAsia.GlobalAquacultureAdvocateMagazine,40Available:https://pdf.gaalliance.org/pdf/GAA-Lightner-Jan12.pdf[Accessed:31March2013].

Mori,Y.,Takahashi,H.andOka,R.(1999)Image-to-wordtransformationbasedondividingandvectorquantizingimageswithwords.FirstInternationalWorkshoponMultimediaIntelligentStorageandRetrievalManagement.Citeseer,pp.1-9.

Moriarty,D.J.(1997)Theroleofmicroorganismsinaquacultureponds.Aquaculture,151(1),pp.333-349.

Moriarty,D.J.(1998)ControlofluminousVibriospeciesinpenaeidaquacultureponds.Aquaculture,164(1),pp.351-358.

Moriarty,D.J.(1999)Diseasecontrolinshrimpaquaculturewithprobioticbacteria.In:C.R.Bell,M.BrylinskyandP.Johnson-Green,ed.InternationalSymposiumonMicrobialEcology.Halifax,Canada,9thto14thAugust1998.KentvilleN.S.:AtlanticCanadaSocietyforMicrobialEcology,pp.237-243.Available:http://socrates.acadiau.ca/isme/symposium08/moriarty.pdf[Accessed:25March2016].

Murray,F.J.,Haque,M.M.,Zhang,W.,Thanh,L.P.,NietesSatapornvanit,A.andLittle,D.C.(2013)DefiningboundariestowardsunderstandingsustainableethicalaquaculturetradebetweenAsiaandEurope.SEATProjectReportRef:D2.82.8.Stirling,UK.www.SEATglobal.eu:UniversityofStirling.



117

Nimrat,S.,Suksawat,S.,Maleeweach,P.andVuthiphandchai,V.(2008)Effectofdifferentshrimppondbottomsoiltreatmentsonthechangeofphysicalcharacteristicsandpathogenicbacteriainpondbottomsoil.Aquaculture,285(1),pp.123-129.

Nunan,L.,Lightner,D.V.,Pantoja,C.andGomez-Jimenez,S.(2014)Detectionofacutehepatopancreaticnecrosisdisease(AHPND)inMexico.DiseasesofAquaticOrganisms,111(1),pp.81-86.

OIE(2013)Acutehepatopancreaticnecrosisdisease,aetiologyepidemiologydiagnosispreventionandcontrolreferences.Available:http://www.oie.int/fileadmin/Home/eng/Internationa_Standard_Setting/docs/pdf/Aquatic_Commission/AHPND_DEC_2013.pdf[Accessed:12August2014].

Panakorn,S.(2012)Opinionarticle:moreonearlymortalitysyndromeinshrimp.AquacultureAsiaPacific,8(1),pp.8-10.

Pearce,N.(2016)Analysisofmatchedcase-controlstudies.BMJ(ClinicalResearchEd.),352,i969.Available:http://dx.doi.org/10.1136/bmj.i969[Accessed:19June2017].

Peterson,O.,Asplund,M.,Karunasagar,I.andGodhe,A.(2010)PhytoplanktoncommunitycompositionanddiversityeffectsonthegrowthofmarineVibriobacteria.In:P.PagouandG.Hallegraeff,ed.InternationalConferenceonHarmfulAlgae.Crete,Greece,1stto5thNovember2010.InternationalSocietyfortheStudyofHarmfulAlgaeandIntergovernmentalOceanographicCommissionofUNESCO2013,pp.147-151.


Rosenberry,B.(2013)Earlymortalitysyndrome:managingtheperfectkiller,awebinarorganizedbytheGlobalAquacultureAllianceandsponsoredbySeafoodSource.com.HoChiMinhCity,Vietnam.December10–13,2013.ShrimpNewsInternational.Available:http://www.shrimpnews.com/FreeReportsFolder/NewsReportsFolder/VietnamGAAwebinarEMS.html[Accessed:15April2014].

Schmidt,C.O.andKohlmann,T.(2008)Whentousetheoddsratioortherelativerisk?InternationalJournalofPublicHealth,53(3),pp.165-167.

Schwartz,J.(2004)Istheassociationofairborneparticleswithdailydeathsconfoundedbygaseousairpollutants?Anapproachtocontrolbymatching.EnvironmentalHealthPerspective,112(5),pp.557-561.



118

Thitamadee,S.,Prachumwat,A.,Srisala,J.,Jaroenlak,P.,Salachan,P.V.,Sritunyalucksana,K.,Flegel,T.W.andItsathitphaisarn,O.(2016)ReviewofcurrentdiseasethreatsforcultivatedpenaeidshrimpinAsia.Aquaculture,452,pp.69-87.

Tho,N.,Merckx,R.andUt,V.N.(2012)BiologicalcharacteristicsoftheimprovedextensiveshrimpsystemintheMekongdeltaofVietnam.AquacultureResearch,43(4),pp.526-537.

Tran,L.,Nunan,L.,Redman,R.M.,Mohney,L.L.,Pantoja,C.R.,Fitzsimmons,K.andLightner,D.V.(2013)Determinationoftheinfectiousnatureoftheagentofacutehepatopancreaticnecrosissyndromeaffectingpenaeidshrimp.DiseasesofAquaticOrganisms,105(1),pp.45-55.

VanDenEeden,S.K.,Tanner,C.M.,Bernstein,A.L.,Fross,R.D.,Leimpeter,A.,Bloch,D.A.andNelson,L.M.(2003)Incidenceofparkinson’sdisease:variationbyage,gender,andrace/ethnicity.AmericanJournalofEpidemiology,157(11),pp.1015-1022.

Xinglong,J.andBoyd,C.E.(2006)Relationshipbetweenorganiccarbonconcentrationandpotentialpondbottomsoilrespiration.AquaculturalEngineering,35(2),pp.147-151.

Xu,S.,Shetterly,S.,Cook,A.J.,Raebel,M.A.,Goonesekera,S.,Shoaibi,A.,Roy,J.andFireman,B.(2016)Evaluationofpropensityscores,diseaseriskscores,andregressioninconfounderadjustmentforthesafetyofemergingtreatmentwithgroupsequentialmonitoring.PharmacoepidemiologyandDrugSafety,25(4),pp.453-461.

Youngren-Grimes,B.L.,Grimes,D.J.andColwell,R.R.(1988)GrowthofVibriocholerae,Vibrioparahaemolyticus,andVibriovulnificusunderstrictanaerobicconditions.SystematicandAppliedMicrobiology,11(1),pp.13-15.

Chapter4Networkanalysis

119

Chapter 4 - Analysis of the network structure of the live shrimp movements relevant to AHPND epizootic


Preface

The fourth chapter analyses the structure of the live shrimpmovement network of

Thailand(LSMN)usinggraphtheoryandnetworkapproaches.Theanalysisisaimedat

finding the real structure of the LSMN and to thereby suggest potential disease

transmission mechanisms, which is an important step towards the prevention and

control of a disease epizootic. The epizootiological survey conducted in Chapter 3

providesinformationondiseasemitigationmeasuresasanaidininterpretingresultsin

this chapter. The chapter is designed for publication, and thus, the spread of acute

hepatopancreatic necrosis disease (AHPND) is described again in the introduction

section.Thefirstpartdescribesthegeneralcharacteristicsof theLSMN.Second, the

network is visualised according to the provincial borders of Thailand. Then it is

quantified using statistical andmathematicalmeasures. The final part discusses the

network structure relevant to the spreadofAHPND.This chapter contributes to the

developmentofacontrolstrategyinthefifthchapterandepizooticmodelsofAHPND

inthesixthchapter.

Impactstatement

TheLSMNarewellrecorded,buthavebeenusedinalimitedextenttoassessthespread

ofdisease.To increaseunderstandingof the spreadofdisease fromsite to site, the

LSMNwasmodelledand its structureexamined in termsof relating these factors to

potential disease epizootics between sites. This is the first project to use network

modellingtocharacterisepotentialdiseasetransmissionintheshrimpfarmingsector

and provides authoritative information to the Thailand Department of Fisheries for

targeteddiseasesurveillanceandcontrolwithlimitedoperatingresources.


120

Chapter 4 - Analysis of the network structure of the live shrimp movements relevant to AHPND epizootic

4.1 Abstract

This research models and analyses the live shrimp movement network of Thailand

(LSMN),whichhasapotentialeffectonsite-to-sitediseasetransmission.Themovement

data were collected over a 13-month period from March 2013 to March 2014.

Importantly,thespreadofacutehepatopancreaticnecrosisdisease(AHPND),andother

known diseases, occurred during the period covered. Results show that large-scale

connectivityintheLSMNtypicallyreliesoninter-provincemovementswithanaverage

distanceofaround200km.TheLSMNwasexaminedbynetworkmodellingandfound

to have a mixture of characteristics both hindering and aiding disease spread. For

hindering transmission, the correlation between 𝑖𝑛 and 𝑜𝑢𝑡degrees was weakly

positive,i.e.siteswithahighriskofcatchingdiseaseposedalowriskfortransmitting

the disease (assuming solely network spread), and the LSMN showed disassortative

mixingin𝑟(𝑖𝑛, 𝑜𝑢𝑡),i.e.alowpreferenceforconnectionsjoinsiteswithhigh𝑖𝑛degree

link toconnectionswithhigh𝑜𝑢𝑡degree.However, therewere lowvalues formean

shortestpathlengthandclusteringcoefficient.Theselattercharacteristicstendtobe

associated with the potential for disease epizootics. In addition to the small-world

property(i.e.shortmeanpathlength)presentedintheLSMN,thenetworkexhibited

power-law distributions of 𝑖𝑛 and 𝑜𝑢𝑡 degrees with exponents of 2.87 and 2.17,

respectively, indicating the scale-free phenomenon. This result showing the

heterogeneity in site degrees demonstrates that a targeted strategy can potentially

performwelltominimisethescaleofadiseaseepizootic.

4.2 Introduction

Shrimpfarminghasbeeninvolvedinthesocio-economicgrowthofThailandsincethe

1980s(Szuster,2006).Nevertheless,theinvasionofinfectiousdiseasespresentsasthe

majorbarriertosuccessinthesector.ThehistoricaldiseasesofAsianshrimpfarming

overthepast30yearshadbeenreviewedinFlegel(2012).Microparasites,i.e.viruses

andbacteria,areamajorcauseofAsianfarmedshrimpdeaths,representingarelatively


121

huge economic loss (Flegel, 2012). This is a reasonwhy theWorldOrganisation for

AnimalHealth(OIE)requiresallmembercountriestodevelopdiseasesurveillanceand

controlmeasuresforaquaticanimals.

Intermsofpathwaysofshrimpdiseasetransmission, long-distancetransmission(i.e.

liveshrimpmovements)posesahighpotentialforsite-to-sitediseasespread.Totakea

recentexample,acutehepatopancreaticnecrosisdisease(AHPND,alsoknownasEMS)

hashitThaishrimpfarmingsincelate2011,andthediseasehasbeentransmittedacross

Thairegionsthroughthemovementsofinfectiousliveshrimp(FAO,2013b;OIE,2013a).

ThisepizooticdiseasedamagedThaishrimpproductionbyanestimated500000tonnes

of shrimpduring a 3-yearperiod from2011 to2014 (Songsanjinda, 2015).Although

there have been many worldwide efforts to stop the spread of AHPND, such as

movement restrictions, biofloc technology, genetics improvements, and enhanced

breeding techniques (Hong et al., 2016; Pakingking Jr et al., 2016), no such control

strategiesforAHPNDemergefromnetworkmodelling.

Network modelling is playing an increasingly important role in epizoology. Its

application relies on graph theory. A graph, or network, includes a set of sites (in

network terminology: nodes) and their connections. Most often, weights of

connections,suchasthefrequenciesofconnectionsbetweenthesamepairsofsitesare

ignored, by analysing non-weighted networks. There are two reasons for this: (1)

weightednetworksaremorecomplextoanalyseand(2)therehasbeenalackofoff-

the-shelf tools that can be used to analyse them, whereas many tools have been

designedfornon-weightednetworks(Newman,2004;Robinaughetal.,2016;Weiet

al., 2013; Yang et al., 2012). However, weighted networks generate an improved

representationandamore realistic structure thannon-weightedones (Barrat et al.,

2004;Wiedermannetal.,2013).Consequently, thereareagrowingnumberof tools

usedforweightednetworks.Thismeansthatthisstudycanevaluatetwoformsoflive

shrimpmovementnetwork:non-weightedandweightednetworks.

Networkmodel approaches can be helpful conveniently to plot andmathematically

describetheconnectionswithincontactnetworks,andtopredictdiseasedynamicsand

persistencefromtheirderivedstructure(Kurversetal.,2014;Meyersetal.,2003).Many


122

studies of disease outbreaks have used both simulated and real networkmodels to

evaluateandcomparetheeffectivenessofcontrolstrategies(Caietal.,2014;Guinatet

al.,2016;Hartvigsenetal.,2007;Maetal.,2013;Werkmanetal.,2011).Huangetal.

(2016),however,suggestedthatrealnetworksreconstructedfromreal-lifeconnections

facilitatedmorereliablepredictionofdiseaseepidemicsthanusingsimulatednetworks.

Hence,therealnetworkmodel isbecoming increasingly important invarioussectors

suchasaquaculture(Greenetal.,2012;MunroandGregory,2009;Tayloretal.,2010),

livestockproduction(Aznaretal.,2011;Büttneretal.,2013;Kissetal.,2006),andplant

trade(Moslonka-Lefebvreetal.,2011).

The structure of the live shrimp movement network of Thailand (LSMN) has been

examined in this research, to explain its susceptibility to disease transmission. The

computer-based recordingof real live shrimpmovementsbetween sites servedas a

data source here, following from the aquatic animal trade regulation of Thailand,

B.E.2553(2010).Therecordingofmovementsisoperatedbyauthorisedusers,providing

uniquerecordsinthatthesourcesanddestinationsofdailyshrimpbatchmovements

can illustrate the spread of diseases from site to site. Furthermore, this research

indicatesthatmovementrecordshaveakeyroleinshrimpepizoology,contributingto

the enhancement of disease surveillance and control strategies in the Thai shrimp

farmingsector.

4.3 Methods

4.3.1 Datasources

Theliveshrimpmovementnetwork(LSMN)data—anelectronicandThaigovernment

database—wasprovidedbytheThailandDepartmentofFisheries.Itcontainsrecords

collectedfromdailybatchmovementsconsistingof:(1)thefarmregistrationnumber

(thefirsttwodigitsdenoteprovincialsitelocationasshowninAppendixCattheendof

thisthesis;inthethirdandfourthpositions,01denoteongrowingsiteand02denote

seed-producingsite),(2)thesourcesiteanddestinationsiteofliveshrimp,(3)thedate

ofmovement,and(4)theseedquantity.Importantly,thedataincludedthe13-month

periodofAHPNDspreadfromMarch2013toMarch2014duringwhichAHPNDspread


123

aroundthecountry(NACA,2017),andcovereduptothreeproductioncyclesoffarmed

shrimp(Flahertyetal.,2000).Unfortunately,unreportedmovements,mostlygenerated

from non-commercial farming with low productivity and for breeding improvement

purposeswerenotavailable.Thisrepresentedalimitationofthestudy,althoughthe

expectednumbersaffectedareverysmall.

Microsoft Access was used to combine circa 99 000 available records, combining

multiplerecordsofabatchmovedwithinadayasoneconnection.Weomittedrecords

havingnorecordeddata inanyof the fields.Furthermore,167recordswerechosen

randomlyfromtheentiredatasetfordatavalidationintermsofthephysicallocationof

sourceanddestinationbecausefailuresinthisregardcouldaffectthereliabilityofthe

networkmodelling.Thissamplesizeofthe167recordswasobtainedfromtheonline

EpiToolsepidemiologicalcalculators(http://epitools.ausvet.com.au).Thenwechecked

thephysicallocationofsourceanddestinationinsuchchosenrecordsagainsttheThai

shrimp farm registration database obtained from Coastal Aquaculture Research and

DevelopmentDivision(2013),resultinginfourerrorsatvillagelevel(moredetailofthe

ThailocalgovernmentalunitswasgivenbyNagaietal.,2008).Thus,the95%confidence

intervalforthepopulationproportionofsucherrorswascalculatedas0.9%to6%,which

was computed by using the 𝑅 Programme Environment with the Hmisc (binconf)

package (Harrell, 2015; R foundation for statistical computing, 2015). These errors

woulddecreasereliabilityinourresultsifvillageswereusedtoidentifysitelocations,

butarenotimportantinthischapter.

4.3.2 Identificationofliveshrimpmovementtypesbyprovincialscale

UsingtheprovincialbordersofThailand,theliveshrimpmovementscouldbedividedinto

twotypes.Theuseofprovincialborderscorrespondswiththeprovincialadministration

of Thailand Department of Fisheries (www4.fisheries.go.th). Firstly, intra-province

movementisaconnectionbetweentwositeswithinthesameprovince;secondly,inter-

provincemovementisaconnectionbetweentwodifferentprovinces.Furthermore,the

distancebetweentwosites—𝑖and𝑗—wascomputedasstraight-linedistance(𝑑)basing

on the geographic coordinates of the Thai sub-districts available fromGoogle Earth

(2015).Theobtainedresultsrepresentedtheexpecteddistancesofinfectionbetween


124

sitesasproposedinBogeletal.(1976).Theformulaforthestraight-linedistance(4.1)

showninDubéetal.(2008)is:

(4.1)

where𝑡isthecircumferenceoftheearth(=40075km).

4.3.3 Provincialvisualisationfortheliveshrimpmovementnetwork(LSMN)

Accordingtothetwomovementtypes(intra-andinter-provincemovements),theLSMN

provincialmovementsfor37provinceswerevisualisedbyusingPajekSoftware(Mrvar

and Batagelj, 1996). The frequency of connections for the intra-province and inter-

provincemovementswerepresentedwithdifferentlinecoloursandscaledlinewidths

onalogscaleduetothewidevariationinsizes.Inanefforttoillustratethemovement

networkbasedonnationaladministrativeboundaries,thepositionof37provinceswas

approximatedtothegeographiclocationsonamapofThailand(Kaesler,N.A.),using

Inkscapesoftware(www.inkscape.org).

4.3.4 LSMNadjacencymatrixfornetworkrepresentationandanalysisatsitelevel

Thequantitativeanalysisof theLSMNstructurewasperformed in the𝑅Programme

Environment(Rfoundationforstatisticalcomputing,2015).TheLSMNwasrepresented

byanadjacencymatrix.The igraph softwarepackage(CsardiandNepusz,2006)was

used since it is more flexible for analysing a large complex network with a sparse

adjacencymatrix (a squarematrixused to representanetworkwhoseelementsare

mostly zeros). The package provides many network analysis tools, e.g. matrix

eigenvalues,andnetworkrewiringalgorithmsusedinthisresearch(CsardiandNepusz,

2006).Inadditiontothe igraphpackage,wealsousedthetnetpackagefromOpsahl

(2009)tocalculateweightedshortestpathsandweightedclustering.

dij =t

360⇥ (x2 + y

2)12

x = (longitudesite j � longitudesite i)⇥ cos(latitudesite i)

y = latitudesite j � latitudesite i


125

InthecaseoftheLSMN,theweightsoftheseconnectionsweretakenintoaccount.A

weight𝑤𝑖𝑗impliedthefrequencyofconnectionsthatvariedinsuchsitepairs(𝑖,𝑗).Thus,

the LSMNwas represented by the adjacencymatrixℎ,whichwas the element-wise

multiplicationofmatrix𝑎bymatrix𝑤asexplainedabove.𝑎𝑖𝑗tookthevalue0ifthere

wasnoconnectionfromsite𝑖tosite𝑗and𝑎𝑖𝑗tookthevalue1otherwise,representing

apathwayofdiseasetransmissionfrom𝑖to𝑗.About300self-loopconnections(𝑎>> =1)

wereremovedfromtheanalysisbecausetheseself-loopsdidnotcontributetofurther

spreadofdiseases(Brittonetal.,2011;Draiefetal.,2008).

Adirectednetworkcanbeusedtoquantifythelocalnetworkstructureofeachsiteby

calculating𝑖𝑛and𝑜𝑢𝑡degrees,andundirecteddegrees,asin(4.2)–(4.4),respectively.

𝐼𝑛degreesindicatethenumberofconnectionsmoveintoeachsiteand𝑜𝑢𝑡degrees

indicatethenumberofconnectionsmovefromeachsite.Undirecteddegreesdenote

thenumberofconnectionsineitherdirectionofeachsite.Thesenon-weighteddegree

calculations(norepeatedconnections)areshowninBarratetal.(2004)andGreenet

al.(2009).

(4.2)

(4.3)

(4.4)

Additionally,weightedmeasureswerecarriedoutforcomparisonproposes,asin(4.5)–

(4.7).

(4.5)

(4.6)

(4.7)

kini

=X

j

aji

kouti

=X

j

aij

kundi

= kini

+ kouti

�X

j

(aij

)(aji

)

kini

=X

j

aji

wji

kouti

=X

j

aij

wij

kundi

= kini

+ kouti


126

Thetotalnumberofconnections(𝑀)wasequaltothetotalnumberofeither𝑖𝑛degrees

or𝑜𝑢𝑡degrees(4.8).

(4.8)

ThedegreedistributionsoftheLSMN(theweightednetwork)werestatisticallyanalysed

for a power-law characteristic and exponents were estimated, using the

Kolmogorov-Smirmovtestwiththeigraphsoftwarepackage(CsardiandNepusz,2006).

ItsimplementationusesthemethodofClausetetal.(2009)andNewman(2005);the

nullhypothesisisthattheLSMNisgeneratedfromapower-lawdistribution.Otherbasic

statisticsofthedegrees,i.e.summary,mean,maximumandminimum,andcoefficient

ofvariation,werealsopresentedintheresults.

Theshortestpath𝐿>? denotesaconnection𝑖 → 𝑗inanetwork.The𝐿>? fornon-weighted

directednetworkswascalculatedbyusingtheDijkstra'salgorithm(CsardiandNepusz,

2006; Dijkstra, 1959). Weighted directed networks require more computations,

however; for these, the 𝐿>? was computed as in Opsahl (2009): first the 𝐿>? was

calculatedbyusingtheDijkstra'salgorithm,thenitwasdividedbythemeanconnection

weightofthenetwork(thetotalnumberofrepeatedconnectionsdividedbythetotal

numberofconnections),givinganadjacencymatrixoftheweightedshortestpathsfor

theLSMN.ToexplainOpsahl’salgorithm,asmallnetworkisshowninFigure4.1.Noted

that,insomecases,theweightedshortestpathsdonothaveaclearmeaninginreallife

suchasgivingthedecimalvaluesofshortestpathlengths.

M =X

j

kini

=X

j

kouti


127

(a) (b)

Figure 4.1 A small weighted directed network (a) and its matrix of the shortest paths 𝑳𝒊𝒋 (b) computed by the algorithm of Opsahl (2009). The average shortest path length ⟨𝑳⟩of this networkis1.84 (= 36.8/20).

Theaverageshortestpathlength⟨𝐿⟩wascomputedfollowingMaoandZhang(2013)

(4.9):

(4.9)

where[𝑋]istheIversonbracketdenoting1wherecondition𝑋istrueand0otherwise.

Clusteringreferstothepresenceofgroupedsitesthatthediseasecantransmitthrough

(ShirleyandRushton,2005).Thecharacteristicof clusteringwasdemonstratedby the

clusteringcoefficient(𝐶)derivedfromtheconceptof‘anyfriendofyoursisafriendof

mine’.ThiswassimilartothestudyofGreenetal.(2012).𝐶wascalculatedasaratioof

thenumberoftrianglestothenumberoftriples.AhnertandFink(2008)definedatriangle

asasetofthreesiteswherewith{𝑆o → 𝑆p → 𝑆q, 𝑆o → 𝑆q},meaningthatbothdirectand

indirect routes for𝑆o → 𝑆qexist,andwheresucha trianglecorrespondedtoa triple

S1 S2 S3 S4 S5S1 NA 2.4 1.2 3.0 3.2S2 0.6 NA 1.8 0.6 0.8S3 1.2 2.4 NA 3.0 3.2S4 1.2 0.6 2.4 NA 1.4S5 1.8 1.2 3.0 1.8 NA

1

4

3 24

4

112

2

1.Calculatethemeanedgeweight

24/10 2.4

2.Evaluatetheshortestpaths(Anexampleforasite‘ ’)=2.4/1=2.4=2.4/2=1.2=2.4/1+2.4/4=3.0=2.4/1+2.4/3=3.2

3.Givethematrixofshortestpaths

From

To


128

𝑆o → 𝑆pand𝑆o → 𝑆q.Thus,the𝐶 istheproportionoftripleswhereadirectrouteof

transmissionalsoexists.

Wecomputed theassortativity coefficient (𝑟) to represent theassortativemixingby

degreeintheLSMN.Assortativemixingbydegreeiscommonincontactnetworksof

persons and animals. In addition to being a network property that aids disease

transmission, assortative mixing inhibits the effectiveness of targeted vaccination

strategiesduetothepersistenceofgiantcomponents(thelargestnumberofsitesina

network thatare interconnectedbydirectedconnections) in thenetwork (Newman,

2003). Theextentofassortativemixing shows the tendencyof sites inanetwork to

connecttoothersiteswithsimilardegrees,i.e.high-degreesitestendtobeconnected

to other high-degree sites (Newman, 2003). In the context of epizoology, this can

indicate thatdisease ismoreeasily spread in thenetwork, leading toahigherbasic

reproduction number (𝑅r) compared with a network that has a negative value of

assortativitycoefficient(disassortativity).Fosteretal.(2010)statethattheassortativity

of directed networks can be represented by four measures: 𝑟(𝑜𝑢𝑡, 𝑖𝑛), 𝑟(𝑖𝑛, 𝑜𝑢𝑡),

𝑟(𝑜𝑢𝑡, 𝑜𝑢𝑡),and𝑟(𝑖𝑛, 𝑖𝑛).Amongthesefourmeasures,however,themostinteresting

forepidemiologicalstudyisthe𝑟(𝑖𝑛, 𝑜𝑢𝑡)—i.e.directedconnectionsjoiningsiteswith

a high 𝑖𝑛 degree link to directed connections joining sites with high 𝑜𝑢𝑡 degree—

because thismeasure describes the patterns of network connections,which lead to

epidemics.ItsequationisshowninGreenetal.(2009),andiswrittenbelowby(4.10):

(4.10)

where𝑘>>8and𝑘?7s[denotethe𝑖𝑛degreeofsourceand𝑜𝑢𝑡degreeofdestinationfor

connection𝑖 → 𝑗respectively,and𝑀isthetotalnumberofconnectionsinthenetwork.

HeesterbeekandDietz(1996)definethebasicreproductionratio(𝑅r)astheexpected

numberofsecondarycasesgeneratedbyatypicalcaseduringitstransmissionperiodin

aparticulargroupofsusceptibleindividuals.Inordertoestimatetheepidemicthreshold

r(in, out) =M

Pi!j

k

in

i

k

out

j

�⇣P

i!j

k

in

i

⌘⇣Pi!j

k

out

j

⌘

rhM

Pi!j

(kini

)2 �⇣P

i!j

k

in

i

⌘2ihM

Pi!j

(koutj

)2 �⇣P

i!j

k

out

j

⌘2i


129

basedonnetwork topology,oneestimate for𝑅r fromadegree-basedestimation is

giveninGreenetal.(2012)(4.11):

(4.11)

where⟨𝑋⟩representsanaverageofthe𝑋value.

Anestimated𝑅r >1indicatesthatanoutbreakcanspreadthoughthenetworkupon

introduction (Heesterbeek and Dietz, 1996; Jones, 2007). In addition, the largest

eigenvalueloftheLSMN’sadjacencymatrixℎwascalculated,sincethiscloselyrelates

to theepidemic threshold in thenetwork (BeckerandHall, 1996;Chakrabarti etal.,

2008; Prakash et al., 2010). With few closed cycles in the network, however, this

measure could easily be zero, or highly non-representative of the network. Thus, a

simple adjustment was made, following Green et al. (2009), i.e. adding a constant

number(𝐾=0.5)toallconnections,asin(4.12).

(4.12)

Then,thelargesteigenvaluelwascomputedwiththeeigen_centralityfunctioninthe

igraph package (Csardi andNepusz, 2006). This resultwas obtained by solving the

equation ℎ𝑉 = 𝑉l , where 𝑉 is called an eigenvector of ℎ corresponding to an

eigenvalue(Restrepoetal.,2007).

4.3.5 Rewiringthenetwork

Toevaluatethesmall-worldpropertyofthenetwork,astructuralcomparisonwasmade

fortheLSMN,foreither𝐶or⟨𝐿⟩,bycomparingitwithrandomlyrewirednetworks,while

preservingthenumberofsitesanddegreedistributionfromtheoriginalone(Evans,

2007; Kiss andGreen, 2008;Maslov and Sneppen, 2002;Noldus andVanMieghem,

2013). One thousand rewired networks were developed, where in such rewired

networks the probability of rewiring was set at one, resulting in all the two-pair

connectionsintheLSMNbeingswapped.Forexample,twoconnections𝑆o → 𝑆pand

𝑆q → 𝑆v are replacedby thesimulatedconnections𝑆o → 𝑆v and𝑆q → 𝑆p (seeFigure

R0 ⇠ hkinkoutiphkinihkouti

hadjustij =

Kwij

N+ hij


130

4.2).TheMann-Whitney𝑈testwasusedtocomparefor⟨𝐿⟩intheoriginalnetworkwith

thedistributionofrewiredvalues.Theresultaddressedthequestionofwhetherthe

originalnetworkisconsistentwiththerewiredones.

Figure 4.2 An example of a rewiring process which generates a new network by swapping the endpoints of two-pair connections in a network.

Connectionscanberewiredlocallyornon-locally,resultinginchangingthesusceptibility

ofthenetworkto infection.Thus,theeffectoftherewiringprocessonthepotential

epizootic size was also evaluated here. The effect of the rewiring process on the

potential epizootic size in the 1 000 rewired LSMNswas assessed by themeasures

explainedbelow.

Sitereach.ThisnetworkmeasurewasproposedinGreenetal.(2012).Thenumber

ofsitesreachablefromanyother(𝑅>)inanadjacencymatrixofshortestpaths(𝐿>?)was

countedasin(4.13):

(4.13)

where[𝑋]istheIversonbracketagain.

Themaximumreachineachrewirednetworkwasusedtorepresentanestimateofthe

worst-caseepidemicsize,andthemeanreachimpliedatypicalepidemicsize(Greenet

al.,2012).Bothmaximumandmeanreacharedefinedin(4.14)and(4.15),respectively:

(4.14)

(4.15)

where𝑁isthetotalnumberofsitesinthenetwork.

Max reach = maxi(Ri)

Mean reach =

PRi

N


131

Giantstronglyconnectedcomponents(GSCC)fordirectednetworks.AGSCCrefers

to the largest number of sites in a network that are interconnected by directed

connections.Thishasoftenbeenusedtoassessthepotentialepidemicsizeofnetworks

(Kaoetal.,2007;Kissetal.,2006;Rautureauetal.,2011).Danonetal.(2011)explained

thatallsitesinaGSCCwereabletocatchthediseaseiftheinfectionwasseededbyany

ofthosesites.Inotherwords,allsitesintheGSCCwereequallyatriskfromtransmission

(Greenetal.,2009;Yatabeetal.,2015).Inadditiontostronglyconnectedcomponents,

all connectionswere also considered as bidirectional in order tomeasure the giant

weaklyconnectedcomponentorGWCC(thelargestnumberofsitesinanetworkthat

areinterconnectedbyundirectedconnections)(Pastor-Satorrasetal.,2015).

TheGSCCwassoughtbytwoconsecutivedepth-firstsearchesbasedonthemethodof

Tarjan(1972).TheGWCC,meanwhile,wassearchedbyasimplebreadth-firstalgorithm

(Korf, 1985). Both implementations were run on the 𝑅 Programme Environment

packageigraph(CsardiandNepusz,2006;Rfoundationforstatisticalcomputing,2015).

To provide a better understanding of the possible outcomes, examples of the

implementationofthealgorithmsaregiveninFigure4.3(forSCC)andFigure4.4(for

WCC).

Figure 4.3 Strongly connected component of a directed network with eight sites. This small network has five strongly connected components (SCCs), which are shown by grey shading. The size of the giant strongly connected component (GSCC) is equal to three and contains sites 𝑺𝟓, 𝑺𝟔 and 𝑺𝟕.


132

Figure 4.4 Weakly connected component of a bidirectional network with eight sites. This small network has two weakly connected components (WCCs), which are shown by grey shading. The size of the giant weakly connected component (WSCC) is equal to four, with a tie between the two sets of sites {𝑺𝟏, 𝑺𝟐, 𝑺𝟑, 𝑺𝟒} and {𝑺𝟓, 𝑺𝟔, 𝑺𝟕, 𝑺𝟖}.

4.4 Results

Thenetworkmodellingreportedinthissectionconsiderstheliveshrimpmovementdata

overa13-monthstudyperiodfromMarch2013toMarch2014.

4.4.1 GeneralcharacteristicsoftheliveshrimpmovementnetworkofThailand(LSMN)

4.4.1.1Thenumberofsites

Figure4.5displaysthegeneralcharacteristicsoftheLSMN.Circa13800shrimpfarming

siteswere located in37provincesof five regions, i.e. south (5665sites;41%ofthe

total),east(4874sites;35%),central(1949sites;14%),west(1312sites;9%),andone

site in the northeast. The highest number of seed-producing sites denoted both

hatcheriesandnurserieswasintheeasternregion(379sites;47%ofthetotal804seed

producingsites).Whereas11provinceshavenoseed-producingsites.Therangeand

mean of the ratio of seed producing sites to ongrowing siteswas 0–1.22 and 0.09,

respectively.


133

Figure 4.5 Circa 13 800 shrimp farming sites located in five regions and 37 provinces of Thailand. Values in brackets show the number of seed-producing sites and ongrowing sites, respectively. Among regions, the highest number of sites is in south ( ) and the lowest number is in northeast ( ). These data were collected from the live shrimp movements in Thailand from March 2013 to March 2014.

4.4.1.2Thecharacteristicsofliveshrimpmovements

The diagrammatic representation of LSMN, which demonstrates the Thai shrimp

farmingindustrystructure,isshowninFigure4.6.Afewhatcheriesobtainbroodstock

L.vannameifromongrowingsiteswithbroodstockimprovementprogrammes.Instead

ofdirectsellingtheshrimpseedatPL10totheongrowingsites,thehatcheriespass

someoftheirproductiontothenurseriesatnaupliusstage.Then,manynurseriesrear

theseedfromthenaupliusuntilPL10beforesellingtheproductiontotheongrowing

sites.Thefigurealsoindicatesthatthereareafewnumberofshrimpseed(PL12–14)

movementsbetweentheon-growingsitesasthesesitesareconductedbyrelativesof

family.


134

Figure 4.6 Diagrammatic representation of LSMN demonstrating the Thai shrimp farming industry structure.

Thecharacteristicsof liveshrimpmovementsaresummarised inFigures4.7and4.8.

Overall,circa13800siteswereinvolvedin33720site-to-sitemovements.Asshownin

Figure 4.7, thesemovements contained 74 462 repeated connections that included

57 281 connections entailing inter-provincemovements (77%of the total repeated

connections),andthemonthlymaximumwasreachedatabout6000connectionsin

March2014.Theremainingconnectionswereintra-provincemovements(23%),with

themaximumforthesebeingreachedatabout2000connectionsinSeptember2013.

Hatcherysite2

Nurserysite

Ongrowing site PL12-14

Hatcherysite1


135

(a)

(b) (c)

Figure 4.7 Distribution of the number of repeated connections over the 13-month study period (March 2013–March 2014) of live shrimp movements in Thailand. (a) The total number of connections are stratified by two movement types: inter- and intra-province movements. (b) The monthly distribution by inter-province movements. (c) The monthly distribution by intra-province movements.

Figure4.8givesthenumberofshrimpmoved.Mostshrimpweremovedbetweensites

viatheinter-provincemovements(83%ofthe161133×10�shrimp).Themaximum

wasreachedatabout13000billionshrimpinMarch2014.Theremainingshrimpwere

movedbytheintra-provincemovements,countingaround17%,withthemaximumwas

reachedatabout3000billionshrimpinJanuary2014.

The results suggest that the movements, whether counted by the number of

connectionsorthenumberofshrimpmoved,cancontributetothespreadofdiseases

across provinces. Importantly, the provincial controls of inter- and intra-province

Num

bero

fcon

nections

10000

20000

30000

40000

50000

60000

Inter-provincemovements

Intra-provincemovements

57281(77%)

17181(23%)

0

100020003000400050006000

MAR DEC2013

JAN MAR2014

0Num

bero

fcon

nections

Maximum(MAR2014)

100020003000400050006000

MAR DEC2013

JAN MAR2014

Num

bero

fcon

nections

Maximum(SEP2013)

0


136

movements (i.e. movement restriction) have the potential for preventing and

controllingdiseasespreadinThaishrimpfarming.

(a)

(b) (c)

Figure 4.8 Distribution of the number of shrimp moved over the 13-month study period (March 2013–March 2014) of live shrimp movements in Thailand. (a) The total number of shrimp moved were stratified by two movement types: inter- and intra-province movements. (b) The monthly distribution by inter-province movements. (c) The monthly distribution by intra-province movements.

AsmeasuredgeographicallyontheThaimap,themeanstraight-linedistancesofthe

intra- and inter-province movements were 24 and 192 km, respectively. The 75th

percentile of straight-line distance (km) for the inter-province movements was

approximately 200 km, and around 30 km for the intra-provincemovements. These

characteristicsofliveshrimpmovementscanbeusedtodescribediseaseepizooticsdue

tolong-distancetransmission.

0Num

bero

fshrim

pmoved

(

)

Inter-provincemovements

Intra-provincemovements

20000400006000080000100000120000140000 133990(83%)

27143(17%)

2000400060008000100001200014000

Jan-00

Jan-00

Jan-00

Jan-00

Jan-00

Jan-00

Jan-00

Jan-00

Jan-00

Jan-00

Jan-00

Jan-00

Jan-00

Num

bero

fshrim

pmoved

(

)

MAR DEC2013

JAN MAR2014

Maximum(MAR2014)

02000400060008000100001200014000

1 2 3 4 5 6 7 8 9 10 11 12 13

Num

bero

fshrim

pmoved

(

)

MAR DEC2013

JAN MAR2014

Maximum(JAN2014)

0


137

4.4.2 VisualisingtheLSMNbasedonnationalprovincialcentres

TheLSMNvisualisationbasedonnationalprovincialcentresispresentedinFigure4.9.

The figure also displays a subset of an example of provincial dataset, which is the

movementsofliveshrimpfromallthreehatcheriesinSamutprakan(SPK)toongrowing

sites,asshowninthebox.Thelinesandarrowsinthefigureindicatetheconnectionsof

liveshrimpbetweensourcesitesanddestinationsiteswithfrequenciesofconnections

corresponding to the colour and width (on a log scale) of the line. For disease

surveillanceandcontrolintheThaishrimpfarming,theprovincialcontrolofmovements

haspotential in termsofpracticalitybecauseThailandhas localDoFoffices inall77

provinces, which play important roles in disease prevention and control

implementation.Thiscanbesupportedbyabetterunderstandingoftwomaintypesof

themovements(inter-andintra-provincemovements).

In Figure 4.9, The inter-province movements with the highest number of repeated

connections (> 1 500 connections) are displayed in the orange and thicker lines,

correspondingtoChonburi(CBI)→Chanthaburi(CTI),Chonburi(CBI)→Chachoengsao

(CCO),Chachoengsao(CCO)→Chanthaburi(CTI),Trat(TRT)→Chanthaburi(CTI),and

Chumphon(CPN)→Suratthani(SNI).Usingthesamekey,theintra-provincemovements

(i.e.aloopfromaprovincetoitself)withthehighestnumberofconnectionswerefound

inChachoengsao (CCO),Nakhonsithammarat (NRT),Phuket (PKT),Songkhla (SKA),and

Prachuapkhirikhan(PKN).

The visualised and quantitative outcomes of the LSMN bolsters understanding of

diseasetransmissionvialong-distancetransmissionintheThaishrimpfarmingsector,

andprovideinformationfortheplanninganddesigningofbasicdiseasesurveillanceand

controlmeasures.Forexample,inthecontextofnetworkmodels,provinceswithahigh

numberofconnectionsoutoftheprovincialboundariesmayhaveanimportantrolein

transmittingdiseasestootherprovinces.Hence,regulatorscanusetheresultsofthis

worktoallocatealargeramountofsurveillanceresourcestothoseprovinces,suchas

toChonburi(CBI)andTrat(TRT).


138

Figure 4.9 The provincial structure of the live shrimp movement network of Thailand (LSMN) over a 13-month period (March 2013–March 2014). The shrimp farming sites are located in 37 provinces of Thailand. The line width is plotted on a log scale according to the actual number of repeated connections. Four classifications of connections are shown by the different line colours: (1) 1–500 connections with grey, (2) 501–1 000 connections with dark grey, (3) 1 001–1 500 connections in blue, and (4) > 1 500 connections in orange. The figure also displays a subset of an example of provincial dataset, which is the movements of live shrimp from all three hatcheries in Samutprakan (SPK) to ongrowing sites, as shown in the box. The abbreviation list of national provincial centres is shown in Appendix C.

NumberofrepeatedconnectionsSNI

1 501 1001 >1500

Liveshrimpmovementsfromthreehatcheries(H)inSPKto

ongrowing sites

Pajek

H H

H


139

4.4.3 Descriptiveanalysisoftheliveshrimpmovementnetwork(LSMN)atsitelevel

ThesitedegreesoftheLSMNweremeasured,andtheresultsaresummarisedinTable

4.1.Itappearsthattherearedifferencesbetweenusingthenon-weighteddegrees(not

including repeated connections) and the weighted degrees (including repeated

connections).

ComputingtheLSMNasnon-weighted,themean(andcoefficientofvariation)forthe

𝑖𝑛degree𝑘>8was2.4(0.9)and2.4(9.2)forthe𝑜𝑢𝑡degree𝑘7s[.Theundirecteddegree

𝑘s8K was4.9(4.6).𝐼𝑛degree𝑘>8hadanarrowrangeofdegreesbetween0and30.In

contrast, 𝑘7s[had a larger range between 0 and 932. Using the Pearson product-

momentcorrelation,thecorrelationbetweenthesitedegrees𝑘>8and𝑘7s[wasweakly

positivewithavalueof0.03(𝑁=13801andP-value<0.01).Theweakpositivevalue

ofdegreecorrelationindicatedthatsiteswithahighriskofcatchingdiseaseposeda

low risk for transmitting the disease (assuming solely network spread). This also

reflectedthefragmentednatureofthenetworkforhinderingdiseasetransmission(Kiss

etal.,2006).

Examining the sitedegrees from theweightednetwork, thedegreepropertieswere

abouttwotimes largerthanforthenon-weightedone.Themean(andcoefficientof

variation)for𝑘>8was5.4(1.7)andfor𝑘7s[was5.4(13.4).Thevaluefor𝑘s8K was10.8

(6.9).𝐼𝑛degree𝑘>8hadanarrowrangeofdegreesbetween0and232. Incontrast,

𝑘7s[hadalargerrangebetween0and5839.Thecorrelationbetweenthesitedegrees

𝑘>8and𝑘7s[wasweaklypositivewithavalueof0.24(𝑁=13801andP-value<0.01).

The fact thatamuchweakercorrelationbetweenthesitedegrees𝑘>8and𝑘7s[was

observed with the non-weighted LSMN than with the weighted one suggests that

studyingthenon-weightednetworkwouldmissthiscorrelation,anditwouldleadto

underestimationofepizooticspread.


140

Table 4.1 Degree properties of the live shrimp movement network of Thailand (LSMN). With two types of site degree calculations (non-weighted and weighted). The properties measured are total number of sites, total degrees, mean degrees, variation coefficient of degrees, and degree correlation.

Property Non-weighteddegree Weighteddegree

Totalnumberofsites(𝑁) 13801

Totaldegrees

- 𝑖𝑛degree

- 𝑜𝑢𝑡degree

- 𝑢𝑛𝑑𝑖𝑟𝑒𝑐𝑡𝑒𝑑degree

33720

33720

67414

74462

74462

148924

Meandegree

- 𝑖𝑛degree



2.4

2.4

4.9

5.4

5.4

10.8

Coefficientofvariation

- 𝑖𝑛degree



0.9

9.2

4.6

1.7

13.4

6.9

Degreecorrelation 0.03 0.24

Consideringweighteddegreedistribution,itwasfoundthattheLSMNdemonstrateda

power-law𝑃(𝑘)~𝑘/0forboth𝑖𝑛and𝑜𝑢𝑡degreedistributionswiththeexponents(𝛾)

2.87and2.17,respectively(Figure4.10).Theexponentsofsitedegreedistributionsin

theLSMNwereintherangeoftwotothree,similartomanyscale-freenetworks,as

proposedinGohetal.(2002).Additionally,theKolmogorov-Smirmovtestacceptedthe

power-law as a plausible model with large P-values>0.05 (P-values of 𝑘>8 = 0.7

and𝑘7s[=0.54).TheseresultsindicatethattheLSMNdisplaysascale-freetopology.

Ascale-freepropertyisofmajorinterestforepizoology(Bogunáetal.,2003;Shirleyand

Rushton,2005).InrespecttotheLSMN,thescale-freepropertyindicatedthatmostsites

hadverylowdegreesofconnectionsandfewsiteshadhighdegreesofconnections.The

transmission capacityof this small group couldbe comparedwith the80/20 rule as


141

proposed in Woolhouse et al. (1997). These authors suggested that, given the

heterogeneityinsitedegrees,infectionin20%ofthetotalsitesissufficienttoleadto

theinfectionoftheremainingsites.

(a) (b)

𝒊𝒏degrees(logscale) 𝒐𝒖𝒕degrees(logscale)

Figure 4.10 The weighted degree distributions for the LSMN plotted on a log–log scale. The sites with higher 𝒊𝒏 degrees have a greater chance of being infected (a), and the sites with higher 𝒐𝒖𝒕 degrees have a greater risk for transmitting disease (b).

BasedontheweightedLSMN,the𝑅restimatedbythedegree-basedcalculation,was

high (~ 34.5), compared to the largest eigenvalue= 16.2. Thehigh valueof𝑅r can

indicate that the connectivity of the LSMNobeys a bipartite structure: the network

consists of two types of sites, i.e. seed-producing sites and ongrowing sites. To aid

understandingofthebipartitestructure,Table4.2showsthedescriptionofthetotal

numberofconnectionsbetweentwosite types in theweightedLSMN.Thebipartite

structureisalsoevidentinthenon-weightedLSMN(Table4.3).Fromthesetables,both

the non-weighted and weighted LSMN demonstrate a high number of connections

between different site types (> 80 % of the total connections). Nevertheless, the

remainingconnectionsjoinsiteswithinthesametypeofsite.Sinceasmallnumberof

connectionsdisobeyingsuchthisbipartitestructure,thismaygiverisetosomedifficulty

in interpreting𝑅r and indescribingwhat𝑅rmeans toseed-producingorongrowing

sites.

Freq

uency(lo

gscale) 𝑃>8~𝑘/p.�� 𝑃7s[~𝑘/p.o�


142

Table 4.2 Description of the number of connections between seed-producing sites and ongrowing sites based on the weighted degree of the LSMN. Their proportions are given in brackets.

Weighteddegree Destination

Seed-producingsite Ongrowingsite

SourceSeed-producingsite 10775 (14.5%) 63596 (85.4%)

Ongrowingsite 7 (0%) 84 (0.1%)

Table 4.3 Description of the number of connections between seed-producing sites and ongrowing sites based on the non-weighted degree of the LSMN. Their proportions are given in brackets.

Non-weighteddegree Destination


SourceSeed-producingsite 2047 (6.1%) 31589 (93.7%)

Ongrowingsite 7 (0%) 77 (0.2%)

Asmeasuredbytheweighteddegree,theLSMNshowed𝑖𝑛-𝑜𝑢𝑡disassortativemixing

withavalueof-0.09.ThisimpliesalowpreferenceforconnectionsintheLSMNthatjoin

siteswith ahigh 𝑖𝑛 degree link to thosewith ahigh𝑜𝑢𝑡 degree. This results in less

extensivetransmissionofthediseasetoothersites.Thisfindingalsoconfirmsthatthe

LSMNisahierarchicalnetwork(e.g.afewsitesaresourcesformanyconnectionsinthe

LSMN),similartowhathasbeenobservedinrelationtopigproductionintheUnited

States (Lee et al., 2017). According to Barthélemy et al. (2004), these few sites can

becomesuperspreadersfortransmittingdiseasesandinducefastepidemics.Inaddition,

usingthesameassortativitymeasure,theresultwassimilar intherewirednetworks.

This result demonstrates that the dynamics of connections do not affect assortative

mixingpatternintheLSMN.

Akeypropertyofsmall-worldnetworks(i.e.smallmeanpathlengths)isshowninboth

the non-weighted andweighted LSMN. Focusing on the non-weighted network, the

valueofthemeanpathlength 𝐿 wassmall(3.47),with0.5%ofpotentialtotalpaths


143

𝑁(𝑁 − 1).FortheweightedLSMN, 𝐿 wasequalto2.99,with0.14%ofpotentialtotal

paths.Thesmallervalueof 𝐿 intheweightednetworkprovidesagoodexplanationfor

the speed of disease transmission in the network. For example, if the connections

betweentwosites(𝑖, 𝑗)aremorefrequent,asmeasuredbytheweightednetwork,a

diseasemight be transmitted quicker through this network structure than the non-

weightedone(Opsahl,2009),andthisisreflectedinthelowvalueof𝐿>?.

Tofurtherinvestigatetheemergenceofthesmall-worldproperty,thedistributionof

weighted path lengths is plotted in Figure 4.11. This clearly demonstrates that the

networkincludesmanyshortpathsandfewlongpaths.Inaddition,theresultsof 𝐿

werecomparedwiththesmall-worldexperimentofMilgrametal.(1992)whostudied

the‘sixdegreesofseparation’theory.Thissuggeststhatsitescangetadiseaseviaa

connection of no more than six intermediates, showing a small-world network. In

contrastwiththesmall-worldnetworksstudiedinWattsandStrogatz(1998),theLSMN

characterised a small 𝐿 but with low clustering as measured by a non-weighted

clusteringcoefficient𝐶=0.0051(theweighted𝐶=0.1).Thislowvalueof𝐶impliesthat

therearefewstrongties(triangles)intheLSMN.

Figure 4.11 The distribution of weighted path lengths in the live shrimp movement network of Thailand (LSMN) is shown as a fraction of total connections.

Pathlength

Prop

ortio

n

0.0020

0.0015

0.0010

0.0005

0

1⎼10

11⎼20

21⎼50

51⎼100

101⎼200

201⎼500

501⎼10

00

1001⎼2

000

2001⎼5000

5001⎼7290


144

To provide more evidence in respect to this small-world property, the important

propertiesoftheLSMNcomparedwithrewirednetworksaresummarisedinTable4.4.

Usingthenon-parametricMann–Whitney𝑈testcomparingthemeanpathlength 𝐿

for theoriginalnetworkandrewiredones,weconcludedthat the 𝐿 of theoriginal

networkandtherewiredonesdidnotdifferata0.05significancelevel(P-value=0.84).

For1000randomlyrewirednetworkswithrewiringprobabilities=1 (conservingthe

originalLSMN’sweighteddegreedistributionandthenumberofsites𝑁),theaverage

clustering coefficient 𝐶was 0.06 (SD=0.008), and the average of 𝐿 was 3.84

(SD=2.12),with0.5%ofpotentialpathsexisting.These indicate that theclustering

coefficient(𝐶),andtheaverageshortestpathlength 𝐿 withintheLSMNarecloseto

valuescomputedfromthecorrespondingrandomlyrewirednetworks.Thus,eveninthe

eventoftheconnectionsbeingchanged,theLSMNstillhasapropertyofasmall-world

networkandasmallvalueofclusteringcoefficient.

Both properties of the LSMN were also compared with two networks of animal

movements.ThefirstwasthenetworkstructureofScottishlivefishmovements(561

sitesand1340connections)thathad𝐶=0.07and 𝐿 =5.92(Greenetal.,2012).The

secondwasthestaticnetworkofpigmovementsinGermany(97980sitesand315333

connections)thathad 𝐿 =5.5(Lentzetal.,2016).Asmallcharacteristicpathlength

andalowclusteringcoefficientdenotethatsite-to-sitetransmissionofdiseaseinthe

LSMNismucheasierandquickerthaninthesetwoalternativeexamples.

Table4.4alsoshowsthattheestimatedmaximumpotentialepizooticsizeintheLSMN

was smaller than in the 1 000 rewired networks. This result was assessed by two

networkmeasures: (1) site reach, including repeated connections, and (2) the giant

strongly connected component (GSCC) and the giant weakly connected component

(GWCC).

Withtheworst-caseepizootics,theaveragemaximumreachforallfullyrewiredLSMNs

washigherthanintheoriginalLSMN,anincreasefrom7290to11000(SD=449.7).As

measuredbymeanreach,theestimatedtypicalepizooticsizealsoincreasedfrom19.5

to72.5(SD=20.2).Similarly,themeansizeofGSCCwasgreaterfortherewiredLSMNs

comparedwiththeoriginalone,anincreasefrom5to27.4(SD=17.8).Incontrast,there


145

wasnodifferenceinthesizeofGWCCbetweentheoriginalnetworkandtherewired

LSMNs.

Table 4.4 Estimated maximum and mean reach, size of giant strongly connected components (GSCCs), and size of giant weakly connected component (GWCCs) for both the LSMN and the rewired LSMNs

Networkmeasure LSMNFullyrewiredLSMNs

Average(SD)

Averagepathlength 𝐿 2.99 3.84(2.12)

Clusteringcoefficient𝐶 0.1 0.06(0.008)

Totalreach 268000 1005000(284000)

Maximumreach 7290 11000(450)

Meanreach 19.5 72.5(20.2)

SizeofGSCC 5 27.4(17.8)

SizeofGWCC 13000 13000(3.5)

4.5 Discussion

Many infectious diseases transmit among populations via network spread, such as

denguediseaseinhumans(Reineretal.,2014;Stoddardetal.,2013),foot-and-mouth

disease in domestic and wild animals (Sobrino and Domingo, 2017), and pancreas

disease in farmed fish (Stene et al., 2014). Thus, using graph theory and a network

approach,thestructureoftheliveshrimpmovementnetworkofThailand(LSMN)over

the13-monthstudyperiodwasinvestigatedinrespecttothepotentialspreadofshrimp

diseases,andtheimportantepizootiologicalpropertiesbothaidingandhinderingthe

spreadofdiseaseswerecharacterised.Becausethemovementsof liveshrimpplaya

crucialroleininfectiousdiseasetransmissionfromsitetosite,particularlyinrespectto

therecentoutbreakofAHPND(OIE,2013)andotherknownshrimpdiseases(Lightner,

1983),theseresultsareasteptowardsdesigninganeffectivediseasesurveillanceand

controlprogrammeforThaishrimpfarming.


146

Networkvisualisationprovidesagoodsourceforstudyingdiseasespreadwithinreal

complexnetworks.VisualisingH1N1 influenzapandemicworldwide (Brockmannand

Helbing, 2013), for example, gives a better understanding of influenza spread via

personswhotravelledacrossc.4000airportsin2009.TheLSMNasvisualisedusingthe

37provincialborders,givesagood illustrationof theconnections in theThaishrimp

farming.Visualising the LSMNas adirectmovementnetwork, it canbe clearly seen

whichprovinceshavehighconnectionsinandout.Thenodesdenotinghatcherieshave

highriskofbeingsourceof infection; thenodesdenotingnurserieshavehighriskof

beingsourceandsinkofinfection;thenodesdenotingongrowingsiteshavehighriskof

being sinkof infection.Additionally, theprovincial visualisationwith twomovement

types (intra- and inter-provincemovements) emphasises that the regulators should

increase efforts in respect to movement controls even during normal ‘peacetime’

situation.

TheresultsalsodemonstratethesepropertiesoftheLSMNthateitherhinderoraidthe

spread of disease. For hindering the transmission, the LSMN had weak correlation

betweensitedegrees.Foraidingtransmission,theLSMNdisplayedasmallcharacteristic

pathlengthandlowclustering.Itshouldbenotedthatclusteringisaverylocalclustering

measurement,asmeasuredbytheclusteringcoefficient.Notallclusteringcoefficients

arenecessaryallthatlocal,whilethereislotsofclusteringatlargescalesduetolocal

tradingbeingcommon intheLSMN(e.g.withinprovince).Thus,whentheclustering

wasremovedbyrewiring, itwasnotsurprisingthattheestimatedpotentialepizooic

sizealsoincreased.

For Thai shrimp farming, repeated connections between sites are common because

therearea few shrimp seedproducers inThailand. Thepatternof connections that

often repeat may relate to the rapid transmission of AHPND and other infectious

diseases in thenetwork.Asmeasuredbymeanpath length 𝐿 , theweighted LSMN

(includingrepeatedconnections)displayedashortervalueof 𝐿 thanthenon-weighted

one (not including repeated connections). In this case, Shirley and Rushton (2005)

describethatanetworkwiththeshortestvalueof 𝐿 ismorelikelytohavethefastest

rateofinfectionduringperiodsofepizootic.Fastdecisionmakingisthereforerequired


147

topreventdiseaseepizooticsintheThaishrimpfarming,inparticularwhenthenumber

ofrepeatedconnectionsincreases.

The structure of the LSMN is scale-free (exhibiting power-law 𝑖𝑛-and 𝑜𝑢𝑡-degree

distributions with the exponents𝛾 = 2.87 and 2.17, respectively). These results

correspondedtothestudiesofPastor-SatorrasandVespignani(2002).Theseauthors

demonstratedthatthevariancesofpower-lawdistributions0< 𝛾 ≤3becameinfinite,

resultinginzeroepidemicthresholdsanddiseasespreadandpersistenceinscale-free

networks at any transmission rate, while exponents above three indicated finite

variances.ChatterjeeandDurrett(2009)alsofoundanon-zeroepidemicthresholdin

random networks with power-law distributions 𝛾 >3. A disease-control strategy

focusedon reducing the transmissionprobabilitywouldprobably apply to the cases

where𝛾 >3(Newman,2002).Topreventdiseasespreadinsuchpower-lawnetworks,

therefore,controlstrategiesfocusedonkeepingnon-zeroepidemicthreshold,suchas

by targeting highly connected sites, aremore effective (Dezső and Barabási, 2002).

Nevertheless,thesearetheoreticalconsiderations.Anyactualdatasethasafinitemean

andvariance.

TheimportantpropertiesofLSMNthatinfluenceonthespreadofdiseasesaredetected

bymodellingthe1000rewirednetworksinordertoexaminesmall-worldphenomenon

and estimate epizootic sizes. In these rewired networks, themean path length 𝐿 ,

reachability,andgiantstronglyconnectedcomponents(GSCC)tendedtobelargerthan

intheoriginalnetwork,whiletheclusteringcoefficient(𝐶)becamesmaller.Inaddition,

theassortativemixingremainedconstantthroughoutalltherewirednetworks.Among

these properties, the increases in either the maximum reach or giant connected

components afternetwork rewiring canbeexplained through thebehaviourof Thai

shrimpfarmersthatwasobservedduringtheface-to-faceinterviewsintheChapter3

“EvaluatingriskfactorsforAHPNDtransmissionintheThaishrimpfarmingsector”.The

ongrowingfarmersarelikelytomakeanewcontactwhentheyfeeldissatisfiedwiththe

seedquality.Consequently,thisbehaviourcontributestoanincreaseintheestimated

epizooticsizeintheThaishrimpfarming.


148

TheLSMNthatwestudyheredoesnotcontainbothunreportedmovementsandabout

300self-loopconnections.Thisisanacknowledgedlimitationofthestudy.Unreported

movementsaretypicallygeneratedfromnon-commercialfarmingwithlowproductivity

andforbreedingimprovementpurposes,whereasself-loopsarecausedbysitesthatact

as both seed-producing and ongrowing sites, which hold same farm registration

number.ThestructureoftheLSMNmaychangeifthenetworkanalysisincludesdata

setinrespecttounreportedmovementsandself-loops.

Themovementsbetweenseedproducingsitesrepresentedthemovementsofshrimp

seedwithnaupliusorinitialPLstagesfromhatcheriestonurseries.Thedatacovereda

fewmovementsfromongrowingsitestoseedproducingsites(hatcheries)forbreeding

improvement purposes. The movements between ongrowing sites appeared to be

sharingshrimpseedbetweenrelativesoffamilysuchasfather’sandson’sfarms,for

example.

Insummary,ournetworkanalysisdescribesthescopeofpotentialdiseasetransmission

among the Thai shrimp farming sites via live shrimp movements. The LSMN is

characterisedbyimportantepizootiologicalpropertiesthatbothaidandhinderdisease

transmission.Becausescale-freepropertiesarefoundintheLSMN,weemphasisethat

optimaltargeteddiseasesurveillanceandcontrolcanreducethespreadofepizooticsin

the Thai shrimp farming. Moreover, not only can a targeted strategy offer more

effectivepreventionofdiseaseepizootics,butitalsoprovidesamoreefficientuseof

limitedresourcesfordiseasesurveillanceandcontrol.Theseconsiderationsleadusto

furtherresearchinthenextchapter.


149

4.6 References

Ahnert,S.E.andFink,T.(2008)Clusteringsignaturesclassifydirectednetworks.PhysicalReviewE,78(3),pp.036112.

Aznar,M.N.,Stevenson,M.A.,Zarich,L.andLeón,E.A.(2011)AnalysisofcattlemovementsinArgentina,2005.PreventiveVeterinaryMedicine,98(2–3),pp.119-127.

Barrat,A.,Barthelemy,M.,Pastor-Satorras,R.andVespignani,A.(2004)Thearchitectureofcomplexweightednetworks.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,101(11),pp.3747-3752.

Barthélemy,M.,Barrat,A.,Pastor-Satorras,R.andVespignani,A.(2004)Velocityandhierarchicalspreadofepidemicoutbreaksinscale-freenetworks.PhysicalReviewLetters,92(17),pp.1-4.

Becker,N.G.andHall,R.(1996)Immunizationlevelsforpreventingepidemicsinacommunityofhouseholdsmadeupofindividualsofvarioustypes.MathematicalBiosciences,132(2),pp.205-216.

Bogel,K.,Moegle,H.,Knorpp,F.,Arata,A.,Dietz,K.andDiethelm,P.(1976)CharacteristicsofthespreadofawildliferabiesepidemicinEurope.BulletinoftheWorldHealthOrganization,54(4),pp.433-447.

Boguná,M.,Pastor-Satorras,R.andVespignani,A.(2003)Absenceofepidemicthresholdinscale-freenetworkswithdegreecorrelations.PhysicalReviewLetters,90(2),pp.028701.

Britton,T.,Deijfen,M.andLiljeros,F.(2011)Aweightedconfigurationmodelandinhomogeneousepidemics.JournalofStatisticalPhysics,145(5),pp.1368-1384.

Brockmann,D.andHelbing,D.(2013)Thehiddengeometryofcomplex,network-drivencontagionphenomena.Science(NewYork,N.Y.),342(6164),pp.1337-1342.

Büttner,K.,Krieter,J.,Traulsen,A.andTraulsen,I.(2013)StaticnetworkanalysisofaporksupplychaininNorthernGermany—Characterisationofthepotentialspreadofinfectiousdiseasesviaanimalmovements.PreventiveVeterinaryMedicine,110(3–4),pp.418-428.

Cai,C.,Wu,Z.andGuan,J.(2014)Effectofvaccinationstrategiesonthedynamicbehaviorofepidemicspreadingandvaccinecoverage.Chaos,Solitons&Fractals,62,pp.36-43.

Chakrabarti,D.,Wang,Y.,Wang,C.,Leskovec,J.andFaloutsos,C.(2008)Epidemicthresholdsinrealnetworks.ACMTransactionsonInformationandSystemSecurity(TISSEC),10(4),pp.1.


150

Chatterjee,S.andDurrett,R.(2009)Contactprocessesonrandomgraphswithpowerlawdegreedistributionshavecriticalvalue0.TheAnnalsofProbability,37(6),pp.2332-2356.

Clauset,A.,Shalizi,C.R.andNewman,M.E.(2009)Power-lawdistributionsinempiricaldata.SIAMReview,51(4),pp.661-703.

CoastalAquacultureResearchandDevelopmentDivision(2013)Thaishrimpfarmregistrationdatabase.

Csardi,G.andNepusz,T.(2006)Theigraphsoftwarepackageforcomplexnetworkresearch.InterJournal,ComplexSystems(http://igraph.org),pp.1695.

Danon,L.,Ford,A.P.,House,T.,Jewell,C.P.,Keeling,M.J.,Roberts,G.O.,Ross,J.V.andVernon,M.C.(2011)Networksandtheepidemiologyofinfectiousdisease.InterdisciplinaryPerspectivesonInfectiousDiseases,2011,pp.1-28.

Dezső,Z.andBarabási,A.(2002)Haltingvirusesinscale-freenetworks.PhysicalReviewE,65(5),pp.055103.

Dijkstra,E.W.(1959)Anoteontwoproblemsinconnexionwithgraphs.NumerischeMathematik,1(1),pp.269-271.

Draief,M.,Ganesh,A.andMassoulié,L.(2008)Thresholdsforvirusspreadonnetworks.TheAnnalsofAppliedProbability,18(2),pp.359-378.

Dubé,C.,Ribble,C.,Kelton,D.andMcNab,B.(2008)ComparingnetworkanalysismeasurestodeterminepotentialepidemicsizeofhighlycontagiousexoticdiseasesinfragmentedmonthlynetworksofdairycattlemovementsinOntario,Canada.TransboundaryandEmergingDiseases,55(9-10),pp.382-392.

Evans,T.(2007)Exactsolutionsfornetworkrewiringmodels.TheEuropeanPhysicalJournalB,56(1),pp.65-69.


Flaherty,M.,Szuster,B.andMiller,P.(2000)LowsalinityinlandshrimpfarminginThailand.Ambio,29(3),pp.174-179.

Flegel,T.W.(2012)Historicemergence,impactandcurrentstatusofshrimppathogensinAsia.JournalofInvertebratePathology,110(2),pp.166-173.


151

Foster,J.G.,Foster,D.V.,Grassberger,P.andPaczuski,M.(2010)Edgedirectionandthestructureofnetworks.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,107(24),pp.10815-10820.

Goh,K.I.,Oh,E.,Jeong,H.,Kahng,B.andKim,D.(2002)Classificationofscale-freenetworks.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,99(20),pp.12583-12588.

GoogleEarth(2015)Geographiccoordinatesofthesub-districtsacrossThailand.GoogleInc.Available:http://www.google.com/earth/index.html[Accessed:3March2015].

Green,D.M.,Werkman,M.andMunro,L.A.(2012)ThepotentialfortargetedsurveillanceoflivefishmovementsinScotland.JournalofFishDiseases,35(1),pp.29-37.

Green,D.M.,Gregory,A.andMunro,L.A.(2009)Small-andlarge-scalenetworkstructureoflivefishmovementsinScotland.PreventiveVeterinaryMedicine,91(2–4),pp.261-269.

Guinat,C.,Relun,A.,Wall,B.,Morris,A.,Dixon,L.andPfeiffer,D.U.(2016)Exploringpigtradepatternstoinformthedesignofrisk-baseddiseasesurveillanceandcontrolstrategies.ScientificReports,6,pp.28429.

Harrell,F.E.(2015)Hmisc:Harrellmiscellaneous.Rpackageversion3.16-0.Available:http://CRAN.R-project.org/package=Hmisc[Accessed:6December2015].

Hartvigsen,G.,Dresch,J.M.,Zielinski,A.L.,Macula,A.J.andLeary,C.C.(2007)Networkstructure,andvaccinationstrategyandeffortinteracttoaffectthedynamicsofinfluenzaepidemics.JournalofTheoreticalBiology,246(2),pp.205-213.

Heesterbeek,J.A.P.andDietz,K.(1996)TheconceptofR0inepidemictheory.StatisticaNeerlandica,50(1),pp.89-110.

Hong,X.,Lu,L.andXu,D.(2016)Progressinresearchonacutehepatopancreaticnecrosisdisease(AHPND).AquacultInt,24,pp.577-593.

Huang,C.,Liu,X.,Sun,S.,Li,S.C.,Deng,M.,He,G.,Zhang,H.,Wang,C.,Zhou,Y.andZhao,Y.(2016)Insightsintothetransmissionofrespiratoryinfectiousdiseasesthroughempiricalhumancontactnetworks.ScientificReports,6,pp.31484.

Jones,J.H.(2007)NotesonR0.Califonia:DepartmentofAnthropologicalSciences.Available:https://people.stanford.edu/jhj1/sites/default/files/file/jones-r0-notes.pdf[Accessed:12May2015].

Kaesler,D.(N.A.)MapofThailand(License:royaltyfree).Available:http://www.dreamstime.com[Accessed:19August2014].


152

Kao,R.R.,Green,D.M.,Johnson,J.andKiss,I.Z.(2007)Diseasedynamicsoververydifferenttime-scales:foot-and-mouthdiseaseandscrapieonthenetworkoflivestockmovementsintheUK.JournaloftheRoyalSociety,Interface/theRoyalSociety,4(16),pp.907-916.

Kiss,I.Z.andGreen,D.M.(2008)Commenton“Propertiesofhighlyclusterednetworks”.PhysicalReviewE,78(4),pp.048101.


Korf,R.E.(1985)Depth-firstiterative-deepening:Anoptimaladmissibletreesearch.ArtificialIntelligence,27(1),pp.97-109.

Kurvers,R.H.J.M.,Krause,J.,Croft,D.P.,Wilson,A.D.M.andWolf,M.(2014)Theevolutionaryandecologicalconsequencesofanimalsocialnetworks:emergingissues.TrendsinEcology&Evolution,29(6),pp.326-335.

Lee,K.,Polson,D.,Lowe,E.,Main,R.,Holtkamp,D.andMartínez-López,B.(2017)UnravelingthecontactpatternsandnetworkstructureofpigshipmentsintheUnitedStatesanditsassociationwithporcinereproductiveandrespiratorysyndromevirus(PRRSV)outbreaks.PreventiveVeterinaryMedicine,138,pp.113-123.

Lentz,H.H.,Koher,A.,Hövel,P.,Gethmann,J.,Sauter-Louis,C.,Selhorst,T.andConraths,F.J.(2016)Diseasespreadthroughanimalmovements:astaticandtemporalnetworkanalysisofpigtradeinGermany.PloSOne,11(5),e0155196.Available:http://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0155196&type=printable[Accessed:12January2017].

Lightner,D.V.,Redman,R.M.,Bell,T.A.andBrock,J.A.(1983a)DetectionofIHHNvirusinPenaeusstylirostrisandP.vannameiimportedintoHawaii.JournaloftheWorldMaricultureSociety,14(1–4),pp.212–225.

Ma,J.,vandenDriessche,P.andWilleboordse,F.H.(2013)Theimportanceofcontactnetworktopologyforthesuccessofvaccinationstrategies.JournalofTheoreticalBiology,325,pp.12-21.

Mao,G.andZhang,N.(2013)Analysisofaverageshortest-pathlengthofscale-freenetwork.JournalofAppliedMathematics,2013,pp.1-5.

Maslov,S.andSneppen,K.(2002)Specificityandstabilityintopologyofproteinnetworks.Science(NewYork,N.Y.),296(5569),pp.910-913.

Meyers,L.A.,Newman,M.,Martin,M.andSchrag,S.(2003)Applyingnetworktheorytoepidemics:controlmeasuresforMycoplasmapneumoniaeoutbreaks.EmergingInfectiousDiseases,9(2),pp.204-210.


153

Milgram,S.,Sabini,J.E.andSilver,M.E.(1992)Theindividualinasocialworld:essaysandexperiments.NewYork:Mcgraw-HillBookCompany.

Moslonka-Lefebvre,M.,Finley,A.,Dorigatti,I.,Dehnen-Schmutz,K.,Harwood,T.,Jeger,M.J.,Xu,X.,Holdenrieder,O.andPautasso,M.(2011)Networksinplantepidemiology:fromgenestolandscapes,countries,andcontinents.Phytopathology,101(4),pp.392-403.

Mrvar,A.andBatagelj,V.(1996)Pajek:AnalysisandVisulisationofLargeNetworks.Available:http://pajek.imfm.si[Accessed:20August2014].

Munro,L.andGregory,A.(2009)ApplicationofnetworkanalysistofarmedsalmonidmovementdatafromScotland.JournalofFishDiseases,32(7),pp.641-644.


Nagai,F.,Mektrairat,N.andFunatsu,T.(2008)LocalgovernmentinThailand:analysisofthelocaladministrativeorganizationsurvey.Chiba,Japan:Instituteofdevelopingeconomies.

Newman,M.E.(2002)Spreadofepidemicdiseaseonnetworks.PhysicalReviewE,66(1),pp.016128.

Newman,M.E.(2003)Mixingpatternsinnetworks.PhysicalReviewE,67(2),pp.026126.

Newman,M.E.(2004)Analysisofweightednetworks.PhysicalReviewE,70(5),pp.056131.

Newman,M.E.(2005)Powerlaws,ParetodistributionsandZipf'slaw.ContemporaryPhysics,46(5),pp.323-351.

Noldus,R.andVanMieghem,P.(2013)Effectofdegree-preserving,assortativerewiringonOSPFrouterconfiguration.TeletrafficCongress(ITC),201325thInternational.IEEE,pp.1-4.


Opsahl,T.(2009)Structureandevolutionofweightednetworks.UniversityofLondon(QueenMaryCollege),London,UK.Available:https://toreopsahl.com/tnet[Accessed:1December2015].


154

PakingkingJr,R.V.,deJesus-Ayson,EvelynGraceTandAcosta,B.O.(2016)Addressingacutehepatopancreaticnecrosisdisease(AHPND)andothertransboundarydiseasesforimprovedaquaticanimalhealthinSoutheastAsia.ProceedingsoftheASEANRegionalTechnicalConsultationonEMS/AHPNDandOtherTransboundaryDiseasesforImprovedAquaticAnimalHealthinSoutheastAsia.MakatiCity,Philippines,22ndto24thFebruary2016.Iloilo,Philippines:SEAFDECAvailable:https://repository.seafdec.org.ph/handle/10862/3096[Accessed:20February2017].

Pastor-Satorras,R.,Castellano,C.,VanMieghem,P.andVespignani,A.(2015)Epidemicprocessesincomplexnetworks.ReviewsofModernPhysics,87(3),pp.925.

Pastor-Satorras,R.andVespignani,A.(2002)Epidemicdynamicsinfinitesizescale-freenetworks.PhysicalReviewE,65(3),pp.035108.

Prakash,B.A.,Chakrabarti,D.,Faloutsos,M.,Valler,N.andFaloutsos,C.(2010)Gottheflu(ormumps)?checktheeigenvalue!arXivPreprintarXiv:1004.0060,pp.1-26.



Reiner,R.C.,Stoddard,S.T.andScott,T.W.(2014)Sociallystructuredhumanmovementshapesdenguetransmissiondespitethediffusiveeffectofmosquitodispersal.Epidemics,6,pp.30-36.

Restrepo,J.G.,Ott,E.andHunt,B.R.(2007)Approximatingthelargesteigenvalueofnetworkadjacencymatrices.PhysicalReviewE,76(5),pp.056119.

Robinaugh,D.J.,Millner,A.J.andMcNally,R.J.(2016)Identifyinghighlyinfluentialnodesinthecomplicatedgriefnetwork.JournalofAbnormalPsychology,125(6),pp.747.

Shirley,M.D.F.andRushton,S.P.(2005)Theimpactsofnetworktopologyondiseasespread.EcologicalComplexity,2(3),pp.287-299.

Sobrino,F.andDomingo,E.(2017)Foot-and-mouthdiseasevirus:currentresearchandemergingtrends.Norfolk,UK:CaisterAcademicPress.

Songsanjinda,P.(2015)CrisisofThaishrimpproduction.Bangkok,16thto20thAugust2015.ThailandResearchExpo2015:NationalResearchCouncilofThailand,pp.1-35.


155

Stene,A.,Viljugrein,H.,Yndestad,H.,Tavornpanich,S.andSkjerve,E.(2014)Transmissiondynamicsofpancreasdisease(PD)inaNorwegianfjord:aspectsofwatertransport,contactnetworksandinfectionpressureamongsalmonfarms.JournalofFishDiseases,37(2),pp.123-134.

Stoddard,S.T.,Forshey,B.M.,Morrison,A.C.,Paz-Soldan,V.A.,Vazquez-Prokopec,G.M.,Astete,H.,Reiner,R.C.,Jr,Vilcarromero,S.,Elder,J.P.,Halsey,E.S.,Kochel,T.J.,Kitron,U.andScott,T.W.(2013)House-to-househumanmovementdrivesdenguevirustransmission.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,110(3),pp.994-999.

Szuster,B.(2006)CoastalshrimpfarminginThailand:searchingforsustainability.In:T.H.Chu,ed.Environmentandlivelihoodsintropicalcoastalzones:managingagriculture-fishery-aquacultureconflicts.Norfolk,UK:CABInternational,pp.86-97.

Tarjan,R.(1972)Depth-firstsearchandlineargraphalgorithms.SIAMJournalonComputing,1(2),pp.146-160.

Taylor,N.,Way,K.,Jeffery,K.andPeeler,E.(2010)TheroleoflivefishmovementsinspreadingkoiherpesvirusthroughoutEnglandandWales.JournalofFishDiseases,33(12),pp.1005.

Watts,D.J.andStrogatz,S.H.(1998)Collectivedynamicsof‘small-world’networks.Nature,393(6684),pp.440-442.

Wei,D.,Deng,X.,Zhang,X.,Deng,Y.andMahadevan,S.(2013)Identifyinginfluentialnodesinweightednetworksbasedonevidencetheory.PhysicaA:StatisticalMechanicsanditsApplications,392(10),pp.2564-2575.

Werkman,M.,Green,D.M.,Murray,A.G.andTurnbull,J.F.(2011)Theeffectivenessoffallowingstrategiesindiseasecontrolinsalmonaquacultureassessedwithan𝑆𝐼𝑆model.PreventiveVeterinaryMedicine,98(1),pp.64-73.

Wiedermann,M.,Donges,J.F.,Heitzig,J.andKurths,J.(2013)Node-weightedinteractingnetworkmeasuresimprovetherepresentationofreal-worldcomplexsystems.EPL(EurophysicsLetters),102(2),pp.28007.


Yang,Z.,Fu,D.,Tang,Y.,Zhang,Y.,Hao,Y.,Gui,C.,Ji,X.andYue,X.(2012)Linkpredictionbasedonweightednetworks.AsiaSim2012.Berlin:Springer,pp.119-126.


156

Yatabe,T.,More,S.,Geoghegan,F.,McManus,C.,Hill,A.E.andMartínez-López,B.(2015)CharacterizationofthelivesalmonidmovementnetworkinIreland:implicationsfordiseasepreventionandcontrol.PreventiveVeterinaryMedicine,122(1),pp.195-204.

Chapter5Disease-controlalgorithm

157

Chapter 5 - Target priority for targeted disease surveillance and control in the live shrimp movement network of Thailand


Preface

Chapter5usesthedataintroducedinChapter4.Theaimofthischapteristodevelopa

targeted disease-control algorithm for the Thai shrimp farming sector based on the

analysisofthestructureoftheliveshrimpmovementnetwork(LSMN)undertakenin

Chapter 4. This chapter is designed for publication; thus, in its introduction part

informationonthespreadofAHPNDinshrimpfarmingsitesisgivenagain.Oneofthe

keystoworkingsuccessfullywiththealgorithmiswriting𝑅scripts.These𝑅scriptswere

developedbyprogrammingworkdonebytheauthor,butwithadvicefromD.M.Green,

andaredifferentfromthoseinpreviousstudies.

Impactstatement

Diseasesurveillanceandcontrolcanleadtoreducedpopulationlossesfrominfectious

diseasesbydetectingandtreatingepizooticsatanearlystage.InthecaseofThaishrimp

farming,unlesstargeteddiseasesurveillanceandcontrolisestablished,diseasescould

spread and persist among sites, due to the structure of the LSMN. Thus, this study

adoptednetworkapproachestoidentifyhigh-riskconnectionswhoseremovalfromthe

network could reduce epizootics. The targeted disease-control algorithms obtained

fromthisstudyprovidetheThailandDepartmentofFisherieswithagoodstrategyfor

planningtheannualdiseasesurveillanceandcontrolprogrammeforfarmedshrimpand

otheraquaticspecies.


158

Chapter 5 - Target priority for targeted disease surveillance in the live shrimp movement network of Thailand

5.1 Abstract

Targeteddiseasesurveillanceandcontrolbyusingnetworkapproacheshasbeenshown

tocontributetoreducingthespreadofdiseasesinvariousfarmanimalsectorssuchas

cattle,pigandfish,buthasneverbeenpreviouslyassessedinshrimp,eventhoughthe

liveshrimpmovementnetworkmakessitesveryvulnerabletodiseaseepizootics.To

implement targeted disease surveillance and control, therefore, five disease-control

algorithmswereevaluatedandtheirefficacycomparedintermsofreducingapotential

epizooticinthelarge-scalenetworkofliveshrimpmovementsofThailand(LSMN).The

resultsofthesealgorithmsshowthataneffectivestrategytocontroldiseasespreadin

theThaishrimpfarmingcanbeachievedbyremovingasmallnumberoftargetedhigh-

risk connections. Specially, two disease-control algorithms based on betweenness

centrality(thenumberofshortestpathsbetweenpossiblepairsofsitesthatrunalong

it) and subnet-crossing (connections crossing subnets within large network) are

proposedforusetoprioritisetargetsfordiseasesurveillanceandcontrolmeasuresin

Thaishrimpfarming.

5.2 Introduction

Infectiousdiseaseshavecontinuallyhitshrimpfarmingsectors.Overthelast30years,

it ismicroparasitesthathavemainlycausedsevereinfectionsinshrimp(Flegeletal.,

2008;Thitamadeeetal.,2016).Particularly,outbreaksofviralwhitespotdiseaseinthe

1990’s did not only resulted in socio-economic losses, but also contributed to the

enhancementofmanymitigationmeasures such as biosecurity (Lightner, 2005) and

diagnostictechniques(Flegeletal.,2008).Consequently,ahugefinancial investment

hasbeenmadeinthedevelopmentofdiseasesurveillanceprogrammesatnationaland

internationalscales(Bondad-Reantasoetal.,2005).Nonetheless,noapproachhasbeen

proposedtotargethigh-riskconnectionsinliveshrimpmovementnetworks.


159

Recently,farmedshrimphavebecomevulnerabletoabacterialdiseasenamedacute

hepatopancreaticnecrosis(AHPND,alsoknownasEMS).Itsoutbreakcausedastrong

decreaseinshrimpexportsfromtwomajorshrimp-producingcountries(i.e.Chinaand

Thailand)ina5-yearperiodfrom2009to2013(Portley,2016).Mostconcernedly,

AHPNDcausedtheThaishrimpproductionsectortodeclinefromaround500000

tonnesin2011to200000tonnesin2014(a56%decrease;Songsanjinda,()2015).In

addition to an annual operating cost of about USD 2 Million for shrimp disease

surveillanceandcontrol(PlanningDivision,2016),in2014/2015theThaigovernment

usedanewUSD2.7million investment for themitigationofAHPND (Kongkumnerd,

2014).Duetothehighcostofdiseasesurveillanceandcontrol,regulatorsshoulddesign

disease surveillance and control programmes strategically, i.e. identifying high-risk

shrimpfarmingsitesorconnections,andthiscanbedonewithnetworkapproaches.

Manyepidemiologicalstudieshaveperformedcentralitymeasuresfortheidentification

ofhigh-risk individualswhoseremoval fromthenetworkminimisespotentialdisease

spread.Theidentificationofhigh-riskindividualsinanetworkrelatestothe80/20rule

proposedbyWoolhouseetal. (1997).Theauthorsdevelopedthisruletoexplainthe

effectsofheterogeneities insite(innetworkterminology,nodes)degreesondisease

transmission, in that generally 20%of infectious sites corresponded to80%of the

transmission.Thus,diseasesurveillanceandcontroltargetedatthe“core”20%group

canbethemosteffectivestrategy(Woolhouseetal.,2005).

Due to the rangeofnetwork structures,however,anygivenmeasure for identifying

influentialsitesorconnectionsmaynotsuitableforallnetworks(NewmanandPark,

2003). Degree centrality is proposed by many epidemiologists as a means to

demonstratethemostimportantindividualsinnetworks(Bohmetal.,2009;Christley

et al., 2005b; Zhang et al., 2010).Algorithmsbasedonbetweenness centrality have

beenshowntobethemostefficientwaytoreducepotentialepidemicsizeintermsof

component structure in the case of the network of cattle movements in France

(Rautureauetal.,2011),andthenetworkofpigmovementsinGermany(Lentzetal.,

2016).Thealgorithmbasedonbetweennesscentralityalsoreducespotentialepidemic

sizeintermsofsitereachabilityinthelivefishmovementnetworkinScotland(Green

etal.,2009).Forcomparison,othermeasuresofcentralityincludeclosenesscentrality


160

(Fournie et al., 2013) and eigenvector centrality (Herrera et al., 2016). Hence, it is

importanttoaddressthreekeyquestionsinthisresearch:(1)doesanalgorithmwork

foragivensystem?(2)howdoesthisalgorithmrelatetothepropertiesofthenetwork?

and(3)whichisappropriateforThaishrimpfarming?

Thisresearch,therefore,hasaimedtofindtheoptimaldisease-controlalgorithmforthe

Thai shrimp farming system. Disease-control algorithms with and without targeted

approachesareevaluatedandcomparedusingthemovementrecordsof liveshrimp

movementsinThailand(LSMN)overthe13-monthperiodfromMarch2013toMarch

2014. The outcome of this research can form part of the process of implementing

mitigationmeasuresformanagementareasinrealtimeduringanepizooticperiod,and

for developing disease surveillance and control programmes in conventional non-

epizooticperiods,notonly for theThai shrimp farming sectorbutalsoother shrimp

producingcountries.

5.3 Materialsandmethods

5.3.1 Datasourcefortheliveshrimpmovementnetwork(LSMN)

AswithChapter4,theLSMNwithitsc.74400repeatedconnectionsintheperiodfrom

March2013andMarch2014wasthedatasourceforthisresearch.Theseofficialdata

wererecordedbyauthorisersof theThailandDepartmentofFisheries, following the

aquatic animal trade regulation of Thailand, B.E.2553 (2010). The recorded data

indicatedthefarmregistrationnumber,thesourceanddestinationoftheliveshrimp

movement,thedateofthemovement,andtheseedquantity.

The LSMN was represented by a connection-weighted adjacency matrix ℎ, an

element-wisemultiplicationofmatrix𝑎bymatrix𝑤.Theterm“site”refersto“node”in

conventional network terminology. All connections between sites were shown in a

matrix𝑎>?.Anelement𝑎𝑖𝑗tookthevalue0iftherewasnoconnectionfromsite𝑖to𝑗

and 1 otherwise,which in this case represented a potential pathway for site-to-site

diseasetransmission.Theweight𝑤𝑖𝑗denotedthefrequencyofconnectionsofsuch𝑎>?.

About 300 self-loop connections (𝑎>> = 1)were removed from the analysis because


161

theseself-loopshadnoeffectonsite-to-sitediseasetransmission(Brittonetal.,2011;

Draiefetal.,2008).

5.3.2 Disease-controlalgorithmsfortargeteddiseasesurveillanceandcontrol

Disease-control algorithmswere developed to target high-risk connections between

twosites(𝑖and𝑗)intheLSMN,whoseremovalfromthenetworkreducedthepotential

transmission of disease. The algorithms used in this researchwere developed from

thosedescribedbyGreenetal.(2012),whichcontainfourtargetedapproachesbased

on betweenness, connection weight, eigenvector centrality, and subnet-crossing to

identifytargetsforremovaland—asacontrol—anon-targetedapproach.Aneffective

algorithmshouldhavehighperformanceinreducingthesusceptibilityofanetworkto

adiseaseepizooticwithrelativelyfewremovals.

Figure5.1demonstratesthedisease-controlalgorithm.Thealgorithmbeginswiththe

whole network. Then, the connections 𝑎 from 𝑖 to 𝑗 in the network are listed in

descendingorderaccordingtooneofthefollowingcriteria.Ifconnectionshaveequal

values,theseconnectionsareselectedatrandomforeachreplicate.


162

Figure 5.1 Schematic explaining disease-control algorithms with and without targeted approaches for targeted disease surveillance and control for the live shrimp movement network of Thailand (LSMN). The process is stopped when 1 000 targeted connections are removed.

Betweenness.Thebetweennesscentralityvaluewascalculatedforconnections𝑎>? in

thenetwork.Thebetweennessofaconnection𝑎>? wasdefinedbyGirvanandNewman

(2002)asthenumberofshortestpathsbetweenpossiblepairsofsitesthatrunalongit

(not compute repeated connections). The authors proposed themeasure to answer

which connections inanetworkweremost “between”otherpairsof sites.High-risk

connectionsweredetectedusingabetweennesscentralitycriterioncontained inthe

Startwiththewholenetwork

Choosethetargetaccordingtocertaincriteria: 1.Targetedapproaches 1.1Betweenness 1.2Connectionweight 1.3Eigenvector 1.4Subnet-crossing

2.Non-targetedapproach

Removetargetsbydifferentstepsizesofremovals

rangingfrom1–500removals

Evaluateimpactonthenetworkwithsitereachand

connectedcomponentsuntil1000targetsremoved

Stop

Repeated


163

edge-betweennessfunctionintheigraphpackage(Brandes,2001;CsardiandNepusz,

2006).

Connectionweight.Withsimpledegreecentrality,asitewithahighdegreecentrality

score (high number of connections 𝑎>?) is considered more important for disease

transmissioninanetworkthanothersiteswithalowerdegreecentrality(Christleyet

al.,2005b;Tanakaetal.,2014).Sincethisresearchfocusedonconnections(𝑖,𝑗)witha

highweight𝑤>?,theconnection𝑎>? withhighestweightwasconsideredasatargetfor

removalfromthenetwork.

Eigenvectorcentrality.Eigenvectorcentralityhasbeenfoundtobeworthstudying

a network where some high degree sites are connected to many low degree sites

(Bonacich,2007).Theeigenvectorcentralityvalueofasitecouldbeexpressedbythe

matrix form ℎ𝑉 = 𝑉l, wherel =an eigenvalue of the network (a constant),𝑉=an

eigenvector corresponding with the eigenvalue l, and ℎ = a large, sparse, non-

symmetricmatrix (LehoucqandScott, 1996). For the LSMN, thehigh-risk siteswere

detectedusinganeigenvectorcriterionwithintheevcentfunctionintheigraphpackage

(CsardiandNepusz,2006).Ingeneral,thesitewiththehighesteigenvectorcentrality

valuecontainslotsofconnections.Inthiscase,targetedconnections𝑎>? belongingto

thissitewerechosenatrandom.

Subnet-crossing.Asubnet(subnetwork)correspondstoagroupofsitesthataremore

denselyinterconnectedthanbetweengroups.Theconceptbehindthesubnet-crossing

algorithms-basedapproachistoremoveconnections𝑎>? crossingsubnetswithinlarge

networks. Here, the fastgreedy.community subnet detection function in the igraph

packagewasused(Clausetetal.,2004;CsardiandNepusz,2006).Then,connections

thatlinktwodifferentsubnetswereindicatedwiththecommandcrossing(Clausetet

al.,2004;CsardiandNepusz,2006).Often,subnet-crossingconnectionswerehigh in

number,thusarandommethodwasusedtoselectatargetforremoval.

Non-targetedapproach(ascontrol).Inthisresearch,anon-targetedapproachwas

used to comparewith the targeted approaches. A connection𝑎>? in the LSMNwas

targetedsimplybyusingarandomisedselection.


164

In order to demonstrate the effect of the removal of connections on the network

structure, after a connection removal the site reach andnetwork componentswere

calculated.

Sitereach.Thenumberofsitesreachablefromothers𝑅> representpotentialtargets

fordisease spread (Green etal., 2012). Site reachwascalculated fromanadjacency

matrix of shortest paths (𝐿>?), here computed by Dijkstra’s algorithm (Csardi and

Nepusz,2006;Dijkstra,1959).Ifthereisapathfromafocalsite𝑖toanother𝑗,thematrix

𝐿>? givesapositivevalue.Ontheotherhand,ifthereisnopathfromafocalsitetoother

the𝐿>? isdefinedasinfinity.FortheLSMN,theadjacencymatrixofshortestpaths(𝐿>?)

wascalculatedthroughtheshortest.paths functioninthe igraphpackage(Csardiand

Nepusz,2006).Then, thenumberofsites that reached𝑅> isequal to thenumberof

positivevaluesofthatfocalsite𝑖asdefinedin(5.1):

(5.1)

where[𝑋]istheIversonbracketdenoting1wherecondition𝑋istrueand0otherwise.

The maximum value of𝑅> across all sites served as an estimate of the worst-case

epidemicsize,andthemeannumberof𝑅> servesasanestimateoftypicalepidemic

size,definedin(5.2)and(5.3),respectively.Thesemeasureswereusedtoquantifythe

susceptibilityofthenetworktoanepidemic(Greenetal.,2012):

(5.2)

(5.3)

where𝑁isthetotalnumberofsitesinthenetwork.

Connectedcomponent.Wecomputedthemaximumsizeofthestronglyconnected

component(SCC),calledthegiantstronglyconnectedcomponent(GSCC),andamean

sizeofSCCforthewholenetwork.TheGSCCdeterminedthelargestnumberofsites

connectedbydirectconnectionswithinthenetwork.Theweaklyconnectedcomponent

Max reach = maxi(Ri)

Mean reach =

PRi

N


165

sizes(WCC)werecalculatedinthesamewayaseitherthemaximumormeansize,in

whichthenetworkwasconsideredasundirected(Pastor-Satorrasetal.,2015).

TheSCCwasimplementedbytwoconsecutivedepth-firstsearches,andtheWCCwas

searchedviaasimplebreadth-firstalgorithm.Bothkindsofnetworkcomponentswere

computedwiththeclusters function(mode“strong”usedforSCC,andmode“weak”

usedforWCC)intheigraphpackage(CsardiandNepusz,2006).

Inthecalculationofthemeanconnectedcomponentsize,thecontraharmonicmean

wasused,insteadofthearithmeticmean.Thecontraharmonicmean𝐿p(ℂ)computes

asthearithmeticmeanofthesquaresofthevaluesdividedbythearithmeticmeanof

thevalues,definedin(5.4,Pahikkala,2010):

𝐿p ℂ =ℂ>p>

ℂ>> (5.4)

whereℂ> isthenumberofsitesincomponent𝑖.

Thisformofaveragebehavesbetterintermsofnotbeingsoaffectedbysmallisolated

groupsofsites(Moskovitz,1933).Anexampleofthearithmeticmeanof3,30and50

is (3+30+50)/3 = 27.7, whereas the contraharmonic mean 𝐿p(ℂ) of those values is

(9+900+2500)/(3+30+50)=41.

5.3.3 Usingthedisease-controlalgorithms

Eachalgorithmwasrepeatedover1000connectionremovalsusingthe𝑅Programme

Environment(Rfoundationforstatisticalcomputing,2015).Allalgorithmsweresetto

recalculatethenetworkpropertiesatthestartofeachiteration.Inaddition,duetothe

stochasticnatureofalltargetedapproachalgorithms,suchalgorithmswererepeated

1000timesandaverageresultswerepresentedintheresultssection.

Someofthesealgorithmsarecomplexandrequirerecalculationofnetworkproperties

ateachstep.Forexample,astepsizeofonewouldnaivelybeexpectedtoworkbest,

but its computation was costly, i.e. needed much computer memory or slowly

computed. Thus, as an attempt to minimise its computational complexity, in each


166

approach,thehighest-rankedconnectionswereremovedfromthenetwork,withthe

numberofconnections removedat each step varied fromone to500. Forexample,

1 000 removals were performed for the step size of one connection removal at a

removal,and100removalswereperformedforthestepsizeof10connectionremovals

ataremoval.

5.3.4 Characterisingthetargetedconnections

Alltargetswerecharacterisedintermsoftheconnectionlength(km)asthestraight-line

distancebetweentwosites(Dubéetal.,2008).Basingonthegeographiccoordinatesof

theThaisub-districtsavailablefromGoogleEarth(2015),thestraight-linedistance(𝑑)

wascalculatedaccordingtothe(4.1)inChapter4.

5.4 Results

5.4.1 Thenumberofsitesreachedinthenetwork

Theresultsofusingthefivedisease-controlalgorithmswithstepsizeoneareshownin

Figure5.2.Eachalgorithmdemonstratesdifferentabilitiestoreduceboththemaximum

reachandmeanreachofsitesintheLSMN.Oncethetotalof1000connections𝑎>? was

removed, the betweenness-based algorithm performed well for both network

measures.Themaximumandmeanreachwerereducedto50%after400targetsfrom

thenetwork.Theotheralgorithms(connectionweight-,eigenvector-,subnetcrossing-,

andrandom-based)performedrelativelypoorly.Thebetweenness-basedalgorithmstill

performedwellwhenastepsizeofmorethanoneremovalwasapplied (resultsnot

shown).


167

(a)

(b)

Figure 5.2 Evaluating the disease-control algorithms against the network reachability. The betweenness algorithm performs well for both measures: (a) maximum reach and (b) mean reach. The graphs (y-axis) are plotted on a square-root (SQRT) scale to aid reading.

With regards to using different step sizes, the network snapshot at 250-removed

connectionswiththebetweennessalgorithmandtherandomalgorithmarepresented

andcomparedinFigure5.3.Thefigureshowslittleornodifferenceinbothmaximum

reachandmeanreachwhetherastepsizeofoneisusedoralargerstepsize.

0

10

20

30

40

50

60

70

80

90

0 100 200 300 400 500 600 700 800 900 1000

SQRT

(maxim

umre

ach)

Numberofremovals

BetweennessConnectionweightEigenvectorSubnetcrossingRandom

0

1

2

3

4

5

0 100 200 300 400 500 600 700 800 900 1000

SQRT

(meanreach)

Numberofremovals



168

(a)

(b)

Figure 5.3 Results of different step sizes of the betweenness algorithm compared to the random algorithm at 250 removals. The estimated epizootic sizes were measured by (a) maximum reach measure and (b) mean reach. Note that the results of 20-, 100-, 200-, and 500-step sizes with (*) are 240, 200, 200, and 500 removals, as these do not divide neatly into 250.

5.4.2 Reducingconnectedcomponentsinthenetwork

Intermsofreducingconnectedcomponents,theresultsofusingthefivedisease-control

algorithmswithastepsizeofoneareshowninFigure5.4.Theeffectsofthealgorithms

onthestronglyconnectedcomponent(SCC)arenotshownhere,however,sincethis

networkmeasuregaveverysmallvaluesfortheGSCCrangingfromonetofive(witha

meanSCCofaroundone),whereastheLSMNhadagiantweaklyconnectedcomponent

(GWCC)composedofalmostallsites(13786of13801sites;meanWCC=13771).Thus,

wefocusedontheWCCinboththeGWCCandmeanWCC.

2988 2987 2987 2977 2976 2975 3201 3580

2164

7202 7219 7205 7257 7174 7205 7244 7252 7178

0

2000

4000

6000

8000

1 2 5 10 20 50 100 200 500

8000

6000

4000

2000

01 2 5 10 20* 50 200* 500*

Betweenness

RandomMaxim

umre

ach

Stepsizes100*

6 6 6 6 6 6 7 86

19 19 19 19 19 19 19 19 19

0

10

20

30

1 2 5 10 20 50 100 200 500

30

20

10

01 2 5 10 20* 50 200* 500*

Betweenness

Random

Meanreach

Stepsizes100*


169

Withthetotalof1000connectionsremoved,thesubnet-crossingalgorithmperformed

welltoreducetheGWCCandmeanWCCintheLSMN.TheGWCCslightlydecreased

from13786to13509,whilethemeanWCCdecreasedfrom13771to13490.Theother

algorithms (betweenness-, connection weight-, eigenvector-, and random-based)

performedrelativelylesswellintermsofreducingtheGWCCandthemeanWCC.The

results of other step sizes also demonstrated the good performance of the

subset-crossingalgorithms(resultsnotshown).

Interestingly,thebetweennessalgorithmhadaverysmallimpactinrespecttoreducing

boththeGWCCandmeanWCCintheLSMN.Thiscontrastswiththeearlierresultinthe

Section 5.4.1. This is presumably a result of the GWCC being generated by a large

numberof sites, and lots of connections in theGWCC showing similar betweenness

centralityscores.

Figure5.5showsthenetworksnapshotsat250connectionsremoved. Itcanbeseen

that, overall, with the subnet-crossing and random algorithms there is only a small

difference in both GWCC andmeanWCCwhether using the 1-removal step size or

biggerstepsizes.


170

(a)

(b)

Figure 5.4 Evaluating the disease-control algorithms against the weakly connected components (WCC). The subnet-crossing algorithm performed well for both measures: (a) GWCC and (b) mean WCC. The graphs (y-axis) are plotted on a square-root (SQRT) scale to aid reading. Note that the y-axis does not start at zero.

115

116

117

118

0 100 200 300 400 500 600 700 800 900 1000

SQRT

(GWCC

)

Numberofremovals


115

116

117

118

0 100 200 300 400 500 600 700 800 900 1000

SQRT

(meanWCC

)

Numberofremovals



171

(a)

(b)

Figure 5.5 Results of different step sizes when comparing the subnet-crossing algorithm and random algorithm at 250 removals. The estimated epizootic sizes were measured by (a) GWCC and (b) mean WCC. Note that the y-axis does not start at zero, and the results of 20-, 100-, 200-, and 500-step sizes with (*) are 240, 200, 200, and 500 removals, as these do not divide neatly into 250.

5.4.3 Thecharacteristicsoftargetedconnections

Thetargetedconnectionsfromthedisease-controlalgorithmsabovearedenotedasthe

high-risk connections fordisease spread in the LSMN. Inorder to characterise these

high-riskconnections,alltargetsfromthebetweennessandsubnet-crossingalgorithms

werecharacterisedintermsofconnectionlengths(km).Wefoundthat,whenusingthe

betweenness-basedalgorithm, thegeographicdistancesof the targeted connections

were (on average) longer, compared to the mean connection length in the whole

connections(around200km).Whenusingthesubnet-crossingalgorithm,however,the

13717 13717 13715 13717 13720 1372213734 13729

13634

13748 13746 13737 13742 1374413734

13745 13756

13690

13600

13650

13700

13750

13800

1 2 5 10 20 50 100 200 500

13800

13750

13700

13650

1 2 5 10 20* 50 200* 500*

Subnetcrossing

Random

GWCC

Stepsizes100*

13600

13702 13702 13700 13702 13705 1370713719 13714

13619

13733 13732 13723 13727 13729 1371913730 13741

13675

13600

13650

13700

13750

13800

1 2 5 10 20 50 100 200 5001 2 5 10 20* 50 200* 500*

Subnetcrossing

Random

Stepsizes100*

13800

13750

13700

13650MeanWCC

13600


172

geographicdistancesofthetargetedconnectionswere(onaverage)shorter.Withthe

betweennessalgorithm,themeanconnectionlengthwas271km(SD=281),compared

to170km(SD=180)withthesubnetcrossing-based

Moreover, the types of source-destination pairs for the 1 000 removals, are quite

different between using the two algorithms. Tables 5.1 shows that, based on the

betweenness algorithm,most targeted connections join other siteswithin the same

typeofsite, i.e. seed-producingsites.Table5.2demonstrates thatahighnumberof

connectionsbetweendifferentsitetypes,i.e.fromseed-producingsitestoongrowing

sites,arethepriorityofthesubnet-crossingalgorithm.

Table 5.1 Source and destination site types of the top 1 000 removals from the betweenness-based algorithm shown by probabilities (in percentages) in the total number of removals, and in the whole connections (values in brackets).

Betweennessalgorithm Destination


SourceSeed-producingsite 99.3% (2.9%) 0%

Ongrowingsite 7% (0%) 0%

Table 5.2 Source and destination site types of the top 1 000 removals from the subnet-crossing based algorithm shown by probabilities (in percentages) in the total number of removals, and in the whole connections (values in brackets).

Subnet-crossingalgorithm Destination


SourceSeed-producingsite 7.5% (0.2%) 92.1% (2.7%)

Ongrowingsite 0.1% (0%) 0.3% (0%)

5.5 Discussion

Several approachesbasedon real networkmodelsof animalmovementshavebeen

widelyproposedforexaminingthepotentialfordiseasetransmissionandfordesigning

control strategies. A recent example is the network of livestockmovements inNew


173

Zealand,whichshowsthatremovingasmallproportionofsitesdecreasestheestimated

epizooticsize(Marquetouxetal.,2016),whereasthenetworkmodelapproachhasnow

beenimplementedinshrimpfarmingsectorinthisresearch(Chapter4“Analysisofthe

networkstructureoftheliveshrimpmovementsrelevanttoAHPNDepizootic”).

AsthemainresultofChapter4,theliveshrimpmovementnetworkofThailand(LSMN)

displayssmall-worldandscale-freeproperties.Itsstructuredemonstratesthatrandom

connectionremovaltendstobecomehighlyinefficientandcostlyasacontrolstrategy

(DezsőandBarabási,2002;Eubanketal.,2004;MayandLloyd,2001;Pastor-Satorras

andVespignani,2002b).Hence,fivedisease-controlalgorithmswereevaluatedinour

researchandtheireffectivenesscomparedontherealstructureoftheLSMN.Inorder

to limit the amount of surveillance resources, each algorithm is specified to allow a

maximumof1000removals,accountingfor3%ofallsite-to-siteconnections𝑎>? inthe

LSMN. The optimal algorithm will provide a smaller scale of estimated potential

epizootic indicated by two network measures, i.e. site reach (maximum and mean

reach) and connected components (SCC and WCC). Essentially, maximum reach is

almost like a compromise between GSCC and GWCC, and in some ways more

epizootiologicallyuseful.

In terms of site reach, the betweenness-based algorithm had high performance in

reducing the susceptibility of our studied network to a disease epizootic with few

removals (this is network structive dependent; since if the structureof thenetwork

changes,othercentralitymeasuresmaybecomemoreeffective).Theremovalofthe

3%ofconnections targetedby thebetweennesscriterionstronglycorresponds toa

decreaseinsitereachofatleast80%intheLSMN.GirvanandNewman(2002)defined

thebetweennesscentralityofaconnectionasthenumberofshortestpathsbetween

pairs of sites that passes through it. Connection removals with high betweenness

centralityscoresmeanthatthepreferentialpathwaysofmanypairsofsitesdisappear

fromthenetwork.Consequently,thesitereachintheLSMNbecomessmaller.

Considering the LSMN data,most of those high-betweenness connectionswere the

connections between seed-producing sites, which played an important role in

distributingshrimpseed(withtheirpathogens)tomanyongrowingsitesinthenetwork.


174

It implies that further connections can occur shortly after the first connection,

correspondingtothelifecycleofshrimp(Quispeetal.,2016).Forexample,nurserysites

reartheseedfromnaupliusuntilpostlarvalstagewithashortperiodof20daysbefore

selling theproductionto theongrowingsites (FAO,2014).Giventhisshortperiodof

shrimpseedproduction,theoperationofthediseasesurveillanceandcontrolmeasures

alongthesehigh-betweennessconnections(e.g.testsforshrimpdiseases)needstobe

fasttoachieveanearlydetectionofdiseaseoutbreaks.

AlthoughtheconceptsofreachabilityandWCCarenotdifferent,inournetworkstudy

thetwosetsofresultsaredifferent.IntermsofWCC,targetingbasedonthesubnet-

crossingalgorithmwassuitableforreducingtheGWCCsandthemeansizeofWCCsin

theLSMN.Itcanbepresumedthatthecomponent-basedmeasuresarebeingtested

againstthemeasuresthemselves.Thus,itislikelytoberatherefficientinthatcase.As

seen,however,giventhatwith1000targetedconnectionsremovedthereisonlyasmall

decrease ofGWCCs andmeanWCCs, the component-basedmeasuremay require a

highernumberoftargetedconnectionsifitistoreduceahighestimatedepizooticsize.

Theneedfortheremovalofmoretargetedconnectionswouldresultinahighercostof

interventionintermsofdiseasesurveillanceandcontrols,andthusthesubnet-crossing

algorithmmaynotbeasuitablestrategy.

AlthoughtheGSCCwaschosenasanindicatorofmaximumpotentialepizooticsizein

thenetworkanalysisofMarquetouxetal.(2016),itisnotagoodmeasureintheLSMN.

Marquetouxetal.analysedthelivestockmovementsfromsitetositeinNewZealand,

witha largenumberof sites contained in itsGSCC (accounting for79%of the total

numberofsites;129of164).Incontrast,theGSCCintheLSMNonlyincluded0.04%of

allsites.Thestudyofepidemicsindirectednetworks,however,oftenusestheGSCCin

ordertodetermineanestimatedepidemicsizewithinthenetworks,suchaswithhuman

socialnetworks(Eubanketal.,2004),andanimalcontactnetworks(Kissetal.,2006;

Rautureauetal.,2011).OnemajoradvantageoftheGSCCmeasurewasproposedby

Volkovaetal.(2010)andKaoetal.(2007).TheydescribedthattheGSCCapproached

thelowerboundofthefinalepidemicsizeintheabsenceofcontrolstrategies,aswith

thelatedevelopmentofpathogendiagnosisinthecaseofAHPND.Thus,ifthestructure

of LSMN possesses a large strongly connected component, the GSCC may become


175

compelling in designing targeted disease surveillance and controls for Thai shrimp

farming.

Although, the use of a larger step size had little or no effect on the reduction of

estimated epidemic sizes, whether measured by site reach and WCC, it gave an

importantadvantage,allowingfastcomputationsforalarge-complexnetworklikethe

LSMN. A few fast algorithms for computing network centrality have already been

developed (Bader et al., 2007; Brandes, 2001;Madduri et al., 2009; Shi and Zhang,

2011). It appears, however, that most algorithms in the literature use one-step

removals.ThestudyofGreenetal.(2012),forexample,conductedone-stepremovalto

evaluate the effects of targeted removal for the directed network of fish farms. By

removingmorethanoneconnection,acloselyrelatedexampleispresentedinNatale

et al. (2009). They specified the number of targets before simulating the epidemic

networkmodelssuchasthe1%ofsiteswiththehighestcentralityscores.Aswellas

allowingfastercomputations,theuseofbiggerstepsizes(morethanoneremoval)may

helptodesigndisease-controlalgorithms.

Insummary,inanattempttoreducethespreadofdiseasesintheLSMNbyremovinga

limited number of connections between sites, the targeted strategies performed

relativelywellcomparedtoanon-targetedapproach.Thisresearchwasdoneonthe

single network layer of live shrimpmovement. Different resultsmight arise, if local

contact or water-borne contact were included. This complex mode of disease

transmissionleadsustouseepizooticnetworkmodelsintoAHPND,asaddressedinthe

followingchapter.


176

5.6 References

Bader,D.A.,Kintali,S.,Madduri,K.andMihail,M.(2007)Approximatingbetweennesscentrality.In:A.BonatoandF.R.K.Chung,eds.InternationalWorkshoponAlgorithmsandModelsfortheWeb-Graph.Germany:Springer,pp.124-137.

Bohm,M.,Hutchings,M.R.andWhite,P.C.(2009)Contactnetworksinawildlife-livestockhostcommunity:identifyinghigh-riskindividualsinthetransmissionofbovineTBamongbadgersandcattle.PloSOne,4(4),e5016.Available:http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0005016[Accessed:29April2009].

Bonacich,P.(2007)Someuniquepropertiesofeigenvectorcentrality.SocialNetworks,29(4),pp.555-564.

Bondad-Reantaso,M.G.,Subasinghe,R.P.,Arthur,J.R.,Ogawa,K.,Chinabut,S.,Adlard,R.,Tan,Z.andShariff,M.(2005)DiseaseandhealthmanagementinAsianaquaculture.VeterinaryParasitology,132(3–4),pp.249-272.

Brandes,U.(2001)Afasteralgorithmforbetweennesscentrality*.JournalofMathematicalSociology,25(2),pp.163-177.


Christley,R.M.,Pinchbeck,G.L.,Bowers,R.G.,Clancy,D.,French,N.P.,Bennett,R.andTurner,J.(2005)Infectioninsocialnetworks:usingnetworkanalysistoidentifyhigh-riskindividuals.AmericanJournalofEpidemiology,162(10),pp.1024-1031.

Clauset,A.,Newman,M.E.andMoore,C.(2004)Findingcommunitystructureinverylargenetworks.PhysicalReviewE,70(6).

Csardi,G.andNepusz,T.(2006)Theigraphsoftwarepackageforcomplexnetworkresearch.InterJournal,ComplexSystems(http://igraph.org),pp.1695.

Dezső,Z.andBarabási,A.(2002)Haltingvirusesinscale-freenetworks.PhysicalReviewE,65(5),pp.055103.

Dijkstra,E.W.(1959)Anoteontwoproblemsinconnexionwithgraphs.NumerischeMathematik,1(1),pp.269-271.



177

Dubé,C.,Ribble,C.,Kelton,D.andMcNab,B.(2008)ComparingnetworkanalysismeasurestodeterminepotentialepidemicsizeofhighlycontagiousexoticdiseasesinfragmentedmonthlynetworksofdairycattlemovementsinOntario,Canada.TransboundaryandEmergingDiseases,55(9-10),pp.382-392.

Eubank,S.,Guclu,H.,Kumar,V.A.,Marathe,M.V.,Srinivasan,A.,Toroczkai,Z.andWang,N.(2004)Modellingdiseaseoutbreaksinrealisticurbansocialnetworks.Nature,429(6988),pp.180-184.

FAO(2014)CulturedaquaticspeciesinformationprogrammePenaeusvannamei(Boone,1931).Rome:FAO.Available:http://www.fao.org/fishery/culturedspecies/Litopenaeus_vannamei/en[Accessed:29April2017].

Flegel,T.W.,Lightner,D.V.,Lo,C.F.andOwens,L.(2008)Shrimpdiseasecontrol:past,presentandfuture.In:M.G.Bondad-Reantaso,C.V.Mohan,M.CrumlishandR.P.Subasinghe,ed.DiseasesinAsianAquaculture.SriLanka,25thto28thOctober2005.Manila,Philippines:AsianFisheriesSociety,pp.355-378.

Fournie,G.,Guitian,J.,Desvaux,S.,Cuong,V.C.,Dungdo,H.,Pfeiffer,D.U.,Mangtani,P.andGhani,A.C.(2013)InterventionsforavianinfluenzaA(H5N1)riskmanagementinlivebirdmarketnetworks.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,110(22),pp.9177-9182.

Girvan,M.andNewman,M.E.(2002)Communitystructureinsocialandbiologicalnetworks.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,99(12),pp.7821-7826.

GoogleEarth(2015)Geographiccoordinatesofthesub-districtsacrossThailand.GoogleInc.Available:http://www.google.com/earth/index.html[Accessed:3March2015].


Green,D.M.,Gregory,A.andMunro,L.A.(2009)Small-andlarge-scalenetworkstructureoflivefishmovementsinScotland.PreventiveVeterinaryMedicine,91(2–4),pp.261-269.

Herrera,J.L.,Srinivasan,R.,Brownstein,J.S.,Galvani,A.P.andMeyers,L.A.(2016)Diseasesurveillanceoncomplexsocialnetworks.PLoSComputBiol,12(7),e1004928.Available:http://journals.plos.org/ploscompbiol/article/file?id=10.1371/journal.pcbi.1004928&type=printable[Accessed:30January2017].


178

Kao,R.R.,Green,D.M.,Johnson,J.andKiss,I.Z.(2007)Diseasedynamicsoververydifferenttime-scales:foot-and-mouthdiseaseandscrapieonthenetworkoflivestockmovementsintheUK.JournaloftheRoyalSociety,Interface/theRoyalSociety,4(16),pp.907-916.



Lehoucq,R.andScott,J.(1996)Anevaluationofsoftwareforcomputingeigenvaluesofsparsenonsymmetricmatrices.PreprintMCS-P547,1195,pp.5.

Lentz,H.H.,Koher,A.,Hövel,P.,Gethmann,J.,Sauter-Louis,C.,Selhorst,T.andConraths,F.J.(2016)Diseasespreadthroughanimalmovements:astaticandtemporalnetworkanalysisofpigtradeinGermany.PloSOne,11(5),e0155196.Available:http://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0155196&type=printable[Accessed:12January2017].

Lightner,D.V.(2005)Biosecurityinshrimpfarming:pathogenexclusionthroughuseofSPFstockandroutinesurveillance.JournaloftheWorldAquacultureSociety,36(3),pp.229-248.

Madduri,K.,Ediger,D.,Jiang,K.,Bader,D.A.andChavarria-Miranda,D.(2009)Afasterparallelalgorithmandefficientmultithreadedimplementationsforevaluatingbetweennesscentralityonmassivedatasets.Parallel&DistributedProcessing,2009.IPDPS2009.IEEEInternationalSymposiumOn.IEEE,pp.1-8.

Marquetoux,N.,Stevenson,M.A.,Wilson,P.,Ridler,A.andHeuer,C.(2016)Usingsocialnetworkanalysistoinformdiseasecontrolinterventions.PreventiveVeterinaryMedicine,126,pp.94-104.

May,R.M.andLloyd,A.L.(2001)Infectiondynamicsonscale-freenetworks.PhysicalReviewE,64(6),pp.066112.

Moskovitz,D.(1933)Analignmentchartforvariousmeans.TheAmericanMathematicalMonthly,40(10),pp.592-596.

Natale,F.,Giovannini,A.,Savini,L.,Palma,D.,Possenti,L.,Fiore,G.andCalistri,P.(2009)NetworkanalysisofItaliancattletradepatternsandevaluationofrisksforpotentialdiseasespread.PreventiveVeterinaryMedicine,92(4),pp.341-350.


179

Newman,M.E.andPark,J.(2003)Whysocialnetworksaredifferentfromothertypesofnetworks.PhysicalReviewE,68(3),pp.036122.

Pahikkala,J.(2010)OncontraharmonicmeanandPythagoreantriples.ElementeDerMathematik,65(2),pp.62-67.

Pastor-Satorras,R.,Castellano,C.,VanMieghem,P.andVespignani,A.(2015)Epidemicprocessesincomplexnetworks.ReviewsofModernPhysics,87(3),pp.925.

Pastor-Satorras,R.andVespignani,A.(2002)Immunizationofcomplexnetworks.PhysicalReviewE,65(3),pp.036104.

PlanningDivision(2016)Annualplans2016.Bangkok:ThailandDepartmentofFisheries(DoF).Available:http://www.fisheries.go.th/planning/index.php?option=com_content&view=article&id=68:2010-11-30-09-28-07&catid=9:2009-09-12-10-16-07&Itemid=5[Accessed:23September2016].

Portley,N.Portley,N.(2016)SFPreportontheshrimpsector:Asianfarmedshrimptradeandsustainability.SustainableFisheriesPartnershipFoundation.Available:www.sustainablefish.org[Accessed:21September2016].

Quispe,R.L.,Justino,E.B.,Vieira,F.N.,Jaramillo,M.L.,Rosa,R.D.andPerazzolo,L.M.(2016)Transcriptionalprofilingofimmune-relatedgenesinPacificwhiteshrimp(Litopenaeusvannamei)duringontogenesis.Fish&ShellfishImmunology,58,pp.103-107.



Shi,Z.andZhang,B.(2011)FastnetworkcentralityanalysisusingGPUs.BMCBioinformatics,12(1),pp.1.

Songsanjinda,P.(2015)CrisisofThaishrimpproduction.Bangkok,16thto20thAugust2015.ThailandResearchExpo2015:NationalResearchCouncilofThailand,pp.1-35.

Tanaka,G.,Urabe,C.andAihara,K.(2014)Randomandtargetedinterventionsforepidemiccontrolinmetapopulationmodels.ScientificReports,4,pp.1-8.


180


Volkova,V.V.,Howey,R.,Savill,N.J.andWoolhouse,M.E.(2010)Sheepmovementnetworksandthetransmissionofinfectiousdiseases.PLoSOne,5(6),e11185.Available:http://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0011185&type=printable[Accessed:23April2015].


Woolhouse,M.E.,Shaw,D.J.,Matthews,L.,Liu,W.C.,Mellor,D.J.andThomas,M.R.(2005)Epidemiologicalimplicationsofthecontactnetworkstructureforcattlefarmsandthe20-80rule.BiologyLetters,1(3),pp.350-352.

Zhang,H.,Zhang,J.,Zhou,C.,Small,M.andWang,B.(2010)Hubnodesinhibittheoutbreakofepidemicundervoluntaryvaccination.NewJournalofPhysics,12(2),pp.023015.

Chapter6𝑆𝐸𝐼𝑅𝑆network-basedmodel

181

Chapter 6 - Epizootic disease modelling in farmed shrimp using compartmental epizootic network-based simulations


Preface

The dynamics of an acute hepatopancreatic necrosis disease (AHPND) epizootic are

modelledinordertoacquireabetterunderstandingofitsepizoologyandtoexamine

the effect of two potential pathways for AHPND epizootics (long-distance and local

spread)withintheThaishrimpfarmingindustry.Theliveshrimpmovementdata(LSMN)

usedinChapters4and5alsoservesasamajordatasourceinthischapter,aswellas

thesurveydatacollectedinChapter3providesinformationonincubationandfallow

periodsforAHPND,andachanceofinfectionvialocalspread.TheRcodesforthemodel

simulationwerewritten by the author,with input fromD.M.Green. The chapter is

designedintheformatofpublicationstohelpusprepareapotentialjournalpaper,and

thus,natureofAHPNDisdescribedagaininthisintroductionsection.

Impactstatement

Althoughtestingforinfectiousdiseasesbeforemovementsoffarmedshrimpisawell-

established principle in Thailand, in some cases diseasesmay not be detected. This

makes it more difficult to design disease controls, and provides a motivation for

modellingAHPNDepizooticdynamics,especiallysinceoutbreaksofAHPNDhavecaused

massiveshrimpproductionlossesinThailand.Thepurposeofthismodelistoevaluate

theeffectoflong-distanceandlocalspreadrelativetoAHPNDepizooticsinThailandand

tosuggestmitigationmeasures.Poorbiosecuritypractices(e.g.waterdischargeandlow

efficiencyofdiseasescreeningofliveshrimpmovements)arethereforeassumedinthe

model. The results of the model could emphasise the importance of biosecurity in

stoppingthespreadofdiseaseacrossshrimpfarmingsites.


182

Chapter 6 - Epizootic disease modelling in farmed shrimp using compartmental epizootic network-based simulations

6.1 Abstract

Sincethefirstoutbreakofacutehepatopancreaticnecrosisdisease(AHPND)inThailand

in2011/2012,therehasbeenlittleinthewayofepizootiologicalstudyofitsepizootic

dynamics.Here,therefore,an𝑆𝐸𝐼𝑅𝑆compartmental,individual-basedepizooticmodel

is used to explore potential AHPND spread within the real live shrimp movement

networkofThailand(LSMN)ina13-monthperiod.Inaddition,themodelexaminesthe

effectoftwopotentialpathwaysforthespreadofAHPND:long-distancetransmission

vialiveshrimpmovementsandlocaltransmission(e.g.sharedwaterbodiesforfarming

and sites located in close proximity). The results reveal that AHPND epizootics in

Thailand aremore likely to occur during hot and rainy seasons (between April and

August). The inter-province movements are potentially a major source of AHPND

epizootics.Withlowerratesoflong-distancetransmissioninthemodel(𝛽6789 <1),the

numberofinfectedsiteswassmaller.Inthemodel,localtransmissionalone(𝛽6789was

0 and𝛽67;<6 was 0.002) was responsible for very small epizootic sizes. Themodels

thereforesuggest that theAHPNDepizooticdynamics inThai shrimpfarmingcanbe

minimisedbyenhancingbiosecuritymeasuresinrespecttoliveshrimpmovements,and

interventions to reduce local spreadshouldbeconsideredaspartofcurrentdisease

surveillanceandcontrolmeasures.

6.2 Introduction

Acutehepatopancreaticnecrosisdisease(AHPND)isanewepizooticbacterialdisease

infarmedshrimp,firstidentifiedbyLightneretal.(2012).Itcausesseveremortalityof

shrimpofupto100%ininfectedsites.Diseasedshrimphaveobviousvisiblesignsof

AHPND,i.e.emptygutandstomach,andpaleoratrophiedhepatopancreas.Mostsites

infectedbyAHPNDhavebeenfoundtohaveincreasedshrimpmortalitywithin10–35

dayspoststockingofpostlarvae(FAO,2013;NACA,2014).


183

Thescaleofbacterialepizooticsinshrimpfarmingisnotlessthanofthatviralepizootic.

AHPNDhascausedhugeglobaleconomiclossesduetothefactthatitsepizooticsoccur

throughout the major producing countries, i.e. China, Vietnam, Thailand, Malaysia,

MexicoandthePhilippines(Dabuetal.,2015;EduardoandMohan,2012;Nunanetal.,

2014). Since AHPND first appeared in Southern China in 2009, it still persists in the

shrimp farming sector (NACA, 2017; Zorriehzahra and Banaederakhshan, 2015).

Importantly,theemergenceofAHPNDisencouragingshrimpstakeholdersthroughout

theworldtoundergoadramatictransitioninfarmingpractices(EduardoandMohan,

2012;FAO,2013).TheeraoftheAHPNDepizooticisthereforewellrecognisedbymany

shrimpstakeholdersasbeingeitheracrisisforshrimpaquacultureorthebeginningof

much-neededimprovementsinshrimpfarming

MovementsofinfectedshrimpareamajorrouteforAHPNDtransmission(OIE,2013a;

Tran etal., 2013b).Alternatively, the specific strainsofVibrioparahaemolyticus, the

causative agent of AHPND, can spread via the live feeds commonly used in

seed-producingsitessuchaspolychaetewormsandbivalves(Ushijima,2014).Thecase–

controlstudyofBoonyawiwatetal.(2016)alsofoundthatthesourceofshrimpseed

relatedtotheoccurrenceofAHPNDatsite level.Shrimpfarmershavealsoobserved

thatmassmortalitiesofshrimphaveoccurredinAHPND-infectedsiteshaveoccurred

eveninnewlypreparedponds(FAO,2013).Accordingtothisevidence,manyfarmers

suspect that seed-producing sites generate long-distance transmissionofAHPND, as

describedaboveinChapter3.

Diseaseoutbreaksviaphysicalproximityofsitesandtheirhydrologicalconnectivityare

classified as local transmission with non-contact network spread. This follows from

commonactivitiessuchassharingofwatersupplies (GalappaththiandBerkes,2015;

Smith,1999)anddischargingwaterfromculturedponds(Muniesaetal.,2015;Patilet

al.,2002;Vandergeestetal.,1999).Importantly,pondwaterandsoilserveasagood

mediumforthecausativeagentofAHPNDtogrow(Chonsinetal.,2016;Kongkumnerd,

2014;Tranetal.,2013a).Hence,thereislikelytobemorethanonepathwayinfluencing

thedynamicsofAHPNDepizooticdynamics,makingitmoredifficultforregulatorsto

designdiseasecontrols.


184

Inmanycases,efficientdiseasecontrolscanbeachievedbyusingepidemicandnetwork

modelling approaches. Epidemic network models attempt to capture the epidemic

dynamicsinapopulationinwhichrealconnectionsexistbetweenindividualsandallow

potentialdiseasetransmission(KeelingandEames,2005).Inhumandiseases,the𝑆𝐼𝑅

(susceptible-infectious-recovered)epidemicmodelhasbeenrecentlymodelledforthe

spreadofEbolainLiberia(Rizzoetal.,2016).Epidemicmodelsforthespreadoffoot-

and-mouthdiseaseintherealnetworksof livestockmovements(DurandandMahul,

2000; Rvachev and Longini, 1985) have also provided many new insights and

epizootiologicaltoolsthatcanbeappliedtootherfarmedanimaldiseases,suchascarp

(Tayloretal.,2011)andsalmon(Werkmanetal.,2011).Importantly,thesemodelshave

resultedinthedevelopmentofeffectivediseasepreventionandcontrolmethods.An

AHPNDmodelthereforerepresentsaverypowerfultooltoassistinthedevelopmentof

suitablemeasurestopreventlargeAHPNDoutbreaks.

An improvedunderstanding of the dynamics of theAHPNDepizootic among sites is

requiredinordertosupportdiseasepreventionandcontrol.Inaddition,althoughwe

knowmuch about the infection at site level, this does notmean thatwe know the

epizootic at country scale. Thus, we have developed an 𝑆𝐸𝐼𝑅𝑆 compartmental,

individual-based epizootic model in this research to: (1) examine AHPND epizootic

dynamicsinThaishrimpfarmingsites,and(2)analysetheeffectoflong-distanceand

local transmission, including the effect of established biosecurity measures on live

shrimpmovements.Thisisthefirstepizooticmodellingofbacterialdiseasebasedon

therealnetworkofliveshrimpmovements.

6.3 Materialsandmethod

6.3.1 Theliveshrimpmovementnetwork(LSMN)

A traceability system for farmed shrimp developed by the Thailand Department of

Fisheriesprovidedaccesstotherecordsofdailyliveshrimpmovementsamong13801

sites𝑁[7[<6 (seed-producingsites=804,andongrowingsites=12997)inthe396-day

periodfrom1March2013to31March2014.Overthisperiodtherehadbeenonsetsof

AHPNDandothershrimpdiseasessuchaswhitespotdisease,yellowheaddisease,and


185

taura syndrome (NACA, 2017). The records contained a source site of shrimp, a

destinationsiteofshrimp,amovementdate,andaquantityofshrimpseed,usefulin

determining the country-wide dynamics of infectious diseases. Multiple records

representingabatchmovedwithinadaywerecombinedasonerecord.Intotal,our

modelwasbasedoncirca74000records.

Inthisresearch,anadjacencymatrixℎ>?[ofsize𝑁[7[<6×𝑁[7[<6×396dayswasusedto

representtheLSMNmathematicallysuchthatthetimingwasanactualconnectiondate

fromsite𝑖tosite𝑗onday𝑡.Theℎ>?[wastheelement-wisemultiplicationofamatrix

𝑎>?[bymatrix𝑤>?[onday𝑡.Element𝑎𝑖𝑗𝑡tookthevalue1iftherewasaconnectionfrom

asite𝑖toasite𝑗implyingapossiblepathfordiseasetransmissionfromsite𝑖to𝑗,and

𝑎𝑖𝑗𝑡tookthevalue0otherwise.About300self-loopconnections(𝑎>>[=1)wereremoved

from the analysis because these self-loops did not contribute to further spread of

diseases(Brittonetal.,2011;Draiefetal.,2008).

UnlikeinChapters4and5,wheretheweight𝑤𝑖𝑗𝑡denotedthefrequencyofconnections,

eachelementaddressedinthispartwasthenumberofshrimpseedintheconnections

betweenthesamepairsofsites(inunitsofabillionshrimp)underanassumptionthat

aconnectionwithalargernumberofinfectedshrimpwouldoffermoreriskofpathogen

transmissionthanconnectionswithfewerinfectedshrimp.

6.3.2 Localcontactsbetweenshrimpfarmingsites

Eitherphysicalproximityofsitesortheirhydrologicalconnectivitywasassumedtopose

ariskoflocaltransmissionforAHPND.Thesub-districtareawasusedtodeterminethe

locationofsites(moredetailoftheThailocalgovernmentalunitswasgivenbyNagaiet

al., 2008).Anundirectedblockmatrix𝑏>? of size𝑁[7[<6×𝑁[7[<6 determined the local

connectionsbetweensitesinthesamesub-district(notincludingself-loops).Themodel

assumedthatthelocalnon-networkspreadwithinsub-districtswasofamixingrandom

type:eachsitecouldinfectallmembers,exceptingforsub-districtswithonlyonesite.

Asitelocatedinasub-districtwithahighernumberofinfectedandexposedsiteswould

presumablyhavemorechanceofbecominginfectedthanothers.


186

Thefrequencyofthenumberofsitemembersforthe748sub-districtsrelevanttothe

LSMNisillustratedinFigure6.1.75%ofthesub-districtshadequaltoorlessthan21

sitemembers(i.e.the75thpercentilefor748sub-districts=21).

Figure 6.1 Frequency of number of site members per sub-district. Most sub-districts had ≤ 21 site members (the 75th percentile for 748 sub-districts = 21).

6.3.3 An𝑺𝑬𝑰𝑹𝑺compartmental,individual-basedepizooticmodelforacutehepatopancreaticnecrosisdisease(AHPND)

An𝑆𝐸𝐼𝑅𝑆compartmental,individual-basedepizooticmodelwasconstructedtoexplore

AHPND transmission among the Thai shrimp farming sites. Two constructs were

involved in themodel: (1) all shrimp farming sites were taken from the LSMN, i.e.

seed-producingsitesandongrowingsites,and(2)thedirectednetworkofliveshrimp

movements(Section6.3.1)andtheundirectednetworkoflocalspread(Section6.3.2)

withtheformerencompassingtwoscenarios:scenarioA(bothintra-andinter-province

movements)andscenarioB(intra-provincemovementsalone).

Allsiteswereclassifiedintooneoffourstates:susceptible(𝑆),exposed(𝐸),infected(𝐼)

andremoved(𝑅);theirstatesvariedovertime.The‘𝑆’statewasnotinfected,butwas

stockedandcouldbeinfected.The‘𝐸’statewasinfectedbutwithouteithervisiblesigns

ofAHPNDorunusualshrimpmortality.Sitesat‘𝐸’state,however,couldposeariskto

othersites.Thenextstate, ‘𝐼’,wasinfectedandhadincreasedshrimpmortalitywith

clinicalsignsofAHPND.Similartotheexposedsites,the‘𝐼’couldposearisktoothers.

149

120

177 172

130

0

50

100

150

200

1 2-3 4-10 11-30 31-100

Freq

uency

Numberofsitememberspersub-district


187

Thefinalstate‘𝑅’wasfallow,meaningthatthesitewasnotstockedandinusefora

shrimpproduction.Accordingly,the‘𝑅’sitecouldnotbeinfected.

Tosimulateanepizootic,themodelbuildingconsistedofthefollowingthreesteps:

6.3.3.1 Initialisationofmodelsimulation

Epizooticswere started in eachmonth to explore the seasonal effects. During each

simulation,theinitialinfectedsites(seeds)withafinitefractionofseedsat0.02were

selectedat randomfromthe804seed-producingsites in theLSMN(=16seeds).All

othersiteswereassumedtobesusceptible.Hence,avector𝐸>[representedthestate

of site 𝑖 at initialisation 𝑡 = 𝑡r: 0 for susceptible, infected and removed, and 1 for

exposed.

ModelparameterswereestimatedbasedonthedetailsofthespreadofAHPNDand

Thaishrimpfarmingactivitiesreportedintheliterature,andthesurveyeddatathatwas

collectedintheChapter3.Theprobabilitiesofinfectionviaaliveshrimpmovement(𝑓)

andlocalspread(𝑔)aredescribedinsection6.3.3.2.

Incubation,infectious,andfallowperiodsforeachsiteweredifferent.Tocharacterise

twooftheseperiods(incubationandfallow)threedistributionswerefittedtothedata

collectedduring theepizootiological survey in theChapter3 (Weibull, lognormal, and

gamma)(Figure6.2).Thesecandidatedistributionshaverecentlybeenemployedinthe

researchofTojinbaraetal.(2016).Thelognormaldistributionwasfoundtobethebest

fit to the observed incubation periods of a microparasitic disease, followed by the

gamma distribution, whereas the Weibull did not fit well (Bénet et al., 2013).

Nevertheless, the Weibull distribution is usually found to be the best fit to the

emergenceofbacterialdiseases(Chen,2007;Henryonetal.,2005;Laietal.,2016).


188

(a) (b)

Figure 6.2 Density plots of fitted distributions of the data for incubation periods (a) and fallow periods (b) in days.

ThebestfittingdistributionprovidedalowerAkaike'sInformationCriterion(AIC)value

asinLeclercetal.(2014)andBénetetal.(2013).Table6.1showstheresultsoffitting

ofthethreedistributionstothedatainthisstudybymaximumlikelihood.Considering

the AIC values, the Weibull distribution was the best fit to the observed AHPND

incubation periods (AIC = 339), and the gamma distributionwas the best fit to the

observed fallow periods (AIC = 576). In the modelling, therefore, the sites were

randomly given different periods, distributed according to these appropriate

distributions.Thesestepswerecomputedusingthe𝑅ProgrammeEnvironmentwith

the fitdistrplus package (Delignette-Muller and Dutang, 2015; R foundation for

statisticalcomputing,2015).

Table 6.1 Akaike's Information Criterion (AIC) values of fitted distributions. The Weibull distribution gives the smallest AIC value for incubation period data while the gamma distribution gives the smallest AIC value for fallow period data.

Observeddata Weibull Lognormal Gamma

Incubationperiod 339 351 346

Fallowperiod 578 582 576

When the sites were infectious, over half of the farmers (65 %, 61of94) in the

epizootiological survey of the Chapter3 indicated their intention to stop stocking

suddenlyafterinfection(herewecalledthisfarmerbehaviour“typeI”).Thisbehavioural

Incubationperiod(days) Fallowperiod(days)

Density


189

intentionimpliedashort-terminfectionatthesites.Incontrast,theremainingfarmers

triedtotreattheinfectedstockbystoppingfeeding,loadingwithprobiotics,andwater

exchangewithinthepond,resultingineithermoderate(28%,26of94:typeII)orlong-

terminfection(7%,7of94:typeIII).Thus,themodelproceededusingthefollowing

twosteps:(1)themodelassignedoneofthesethreetypesofbehaviourtoeachsite

(bothseed-producingandongrowingsites)whilekeepingallproportionsconstant(65

%of𝑁[7[<6 fortypeI,28%of𝑁[7[<6 fortypeII,and7%of𝑁[7[<6 fortypeIII),and(2)

theinfectionperiodwasgeneratedrandomlyforeachsiteunderthreeassumptions:a

1–7 day short-term infection for a group of farmers behaviour type I, an 8–30 day

moderate-terminfectionforagroupoffarmersbehaviourtypeII,anda31–120day

long-terminfectionforagroupoffarmersbehaviourtype III.

Themodelfixedtheexpectedperiodsforincubation,infectiousandfallowforeachsite

beforerunningasimulation.Thismeantthatifasitewasinfectedtwiceitbehavedthe

sameeachtime.

6.3.3.2Simulationapproach

Theepizooticsweremodelledusinga1-daytimestep.Assumingthataconnectionwith

a larger number of shrimp would carry more risk of pathogen transmission than

connectionswith fewer shrimp,we used the number of shrimpmoved,𝑐 (6.1) as a

weightfortheinfectiousnessofshrimp,𝜔(6.2):

(6.1)

(6.2)

where𝜇=1denotingtheinfectiousnessofshrimpactproportionaltotheirnumber.

Ateach1-daytimestep,avector𝑓representedaprobabilityofdestinationsites(𝑗)

becomingexposedviatheconnectionsof liveshrimpfromexposed𝐸and infected𝐼

sourcesites(𝑖),asin(6.3):

cjt =X

i

(Eit + Iit)hijt

!jt = cµjt


190

⌘

(6.3)

where𝛽6789=1impliesveryhighriskofinfectionviatheconnectionsofinfectedshrimp

since,inthepresentcircumstance,any𝜔 > 0causesinfection.

The 𝑓 included the stochastic nature of the epizootic process (sites successfully or

unsuccessful treated for AHPND). A vector 𝑄 of size 𝑁[7[<6 relied on a

𝐵𝑒𝑟𝑛𝑜𝑢𝑙𝑙𝑖distribution:0fornon-exposedand1forexposedwithaprobability=𝑓,asin

(6.4).

(6.4)

Additionally,aneffectoflocaltransmissionwasdependentonthenumberofinfections

(bothexposed𝐸andinfected𝐼sites)withincloseproximity,𝑧(6.5).Thecloseproximity

ofsiteswasbasedonthemodelledsub-districtareas.Themathshereworkedinparallel

totheabove.

(6.5)

The𝑧wasusedasaweightfortheinfectiousnessofsite(),asin(6.6):

(6.6)

where𝜖=1denotestheinfectiousnessofsites,actingproportionaltothenumberof

theirneighbours.

Avector𝑔representedtheprobabilityofsitesbecomingexposedvialocalspread,asin

(6.7):

(6.7)

where𝛽67;<6 =0.002perdayasthechanceofinfectionduetolocalnon-networkspread.

ThiswasestimatedfromthenumberofAHPNDcases(2012–2013)inThailand(vialocal

fjt

= 1� (1� �long

)!jt

Qjt ⇠ Bernoulli(fjt)

zjt =X

i

(Eit + Iit)bij


191

Ei + Si + Ii +Ri = 1

andlong-distancespread),basesontheresultsfromtheChapter3.Additionally,𝛽67;<6

=0.1perdayweremodelledtoevaluateanincreaseoflocalspreadeffect.

Althoughasusceptiblesitewasatriskofacquiringdiseasecausedbyinfectionvialocal

spread, the sitemightnotbe infectedasa resultofgood farmingmanagementand

control.Totakeaccountofthisinthemodel,𝑔wasincludedtocapturethestochastic

natureoftheepizooticprocesses.Anewvector𝑌ofsize𝑁[7[<6 wasgeneratedbythe

sameprocedureasusedforthevector𝑄,asin(6.8),but1denotingexposedreliedon

aprobability=𝑔.

(6.8)

Thenewstatevectorafterexposuretoinfectionwas:

(6.9)

where

Then,themodelranoverthetimeperiodoverwhichinfectionoccurred.Asdescribed

in the initialisation section, each site had a given length of time spent in its state

following each period for incubation, infection, and fallow. Hence, when a sitewas

exposed to infection, it was in the ‘𝐸’ state for the incubation period. After the

incubation period ended, the site entered the ‘𝐼’ state (𝐸 → 𝐼;𝐼> = 1, 𝐸> = 0). The

furtherchangesof‘𝐼’and‘𝑅’weredependentontheinfectiousperiod(for𝐼 → 𝑅;𝑅> =

1, 𝐼> = 0)andthefallowperiod(for𝑅 → 𝑆;𝑆> = 1, 𝑅> = 0),respectively.

6.3.3.3Finalresults

Atthefinalstageofthemodelsimulation,thenumberofinfectedsites‘𝐼’wascounted

andplottedeachdayoverthe13-monthnetworkdataperiod.Overall,themodelling

processcanbesummedupasshowninFigure6.3.

Ei,t+1 = Eit + (1� Eit)(1� (1�Qjt)(1� Yjt))


192

Figure 6.3 Design and implementation of an algorithm for an 𝑺𝑬𝑰𝑹𝑺 compartmental, individual-based epizootic model for shrimp disease in Thailand

Theepizooticsimulationmodelwascarriedout inthe𝑅ProgrammeEnvironment(R

foundation for statistical computing, 2015). The model was run 80 times for each

epizooticstarted.Intotal,1040resultswereobtained(13months×80simulations)per

parametersetstudied;thesewereusedtocomputethemeannumberofinfectedsites,

representinganestimatedepizooticsizeforAHPND.

Geographic distributions of AHPND prevalence at provincial level of Thailand were

shownintheresult.Thenumberofinfectedsites(𝐼)duringagivenperiodwasobtained

by themodelling of𝛽6789=1 and𝛽67;<6 =0.002. The prevalence of AHPND at each

provincewascalculatedastheproportions(%)ofsitesinfectedwithAHPND(6.10).

Proportions =Numberofinfectedsitesinaprovince

Totalnumberofinfectedsites ×100 (6.10)

The AHPND-infected site proportions were plotted in map form to represent the

geographic distributions of AHPND prevalence in Thailand. A map of Thailand as a

Initialisation

- Setupseedandmodelparameters,thensetinitialstatevectorof model

Mainloop

- Introducenetworksofdailyliveshrimpmovementsandlocalcontacts

- Calculatenewstatevectors,thencopythemallbacktothestatevectorsfornexttime step

Finalresults

- Countnumberofinfectedsiteseachday- Plotagraph


193

shapefilewasdownloadedfromhttp://www.diva-gis.org/gdata.Thismapwasmodified

using the rgdal package in the 𝑅 Programme Environment (Bivand et al., 2015; R

foundation for statistical computing, 2015), aswell as the tmap package (Tennekes,

2017).

To evaluate the accuracy of the disease prevalence prediction generated using the

𝑆𝐸𝐼𝑅𝑆model,themodelestimateswerecomparedagainsttherealpatternofAHPND

epizooticsinThailandatprovinciallevelreportedinJuly2013byFAO(2013).Thereal

AHPNDepizooticsinThairegionsandprovincesareshowninTable6.2.

Table 6.2 Real pattern of AHPND epizootics within shrimp farming sites of Thailand reported in July 2013 (FAO, 2013)

Region Diseasepresence Province

West NoAHPNDcase

Northeastern NoAHPNDcase

Central AHPNDcase Nakhonpathom

East AHPNDcase Chachoengsao,Chanthaburi,Rayong,andTrat

South

AHPNDcase Chumphon,Krabi,Nakhonsithammarat,Phuket,Songkhla,andSuratthani

Moreover,thereceiveroperatingcharacteristic(ROC)curvewasusedtoevaluatethe

model performance in predicting a binary outcome (Park et al., 2004). Here we

presentedthreetestsontheROCanalysis:

(1)Diseaseprevalenceofthe𝑆𝐸𝐼𝑅𝑆modelagainstdiseaseprevalenceinthefieldas

reportedbyFAO(2013),

(2)Diseaseprevalenceofasimplernullmodelwiththenumberofsitesagainstdisease

prevalenceinthefieldasreportedbyFAO(2013),and

(3)Diseaseprevalenceofasimplernullmodelwiththenumberofconnections(inter-

provincemovements‘connectionsin’andintra-provincemovements)againstdisease

prevalenceinfieldasreportedbyFAO(2013).


194

Weuseddifferentcut-offvaluesofa“predictor”tomakethepredictorbinary inthe

ROCanalysis.Thepredictorvariedwiththepredictivediseaseprevalenceobtainedfrom

eachtestmodel.WiththeareaundertheROCcurve(AUC),themodelwasclassifiedto

beeitheraninformativemodel(AUC>0.5;thepredictiveresultofmodelisbetterthan

random)oranuninformativemodel (AUC=0.5; thepredictiveresultofmodel isnot

differentfromrandom)(AlonzoandPepe,2002;KumarandIndrayan,2011).TheROC

analysis was done in the 𝑅 Programme Environment (R foundation for statistical

computing,2015).

6.4 Results

6.4.1 SeasonalityofAHPNDepizooticdynamics

Themeannumberofsites infectedbyAHPNDdueto long-distanceand local spread

(default parameters𝛽6789was1 and𝛽67;<6 was 0.002) is shown in Figure 6.4. These

resultsareobtainedfromthesimulationsofthe𝑆𝐸𝐼𝑅𝑆models,andtheybasedonthe

realnetworkofliveshrimpmovementsofThailand(LSMN)overthe13-monthperiod

fromMarch2013toMarch2014.

The figure illustrates that inter-province movements are the major source of large

epizooticsaccordingtoourmodels.Wecanseethatthereisadifferenceinthenumber

ofinfectedsitesbetweentwoscenarios:A(bothintra-andinter-provincemovements),

andB(intra-provincemovementsalone).

WithscenarioA,wherethelargerepizooticsaregeneratedfrombothintra-andinter-

provincemovements,theresultsofthe𝑆𝐸𝐼𝑅𝑆modelsrevealthatAHPNDepizooticsin

Thailandaremorelikelytooccurduringthehotandrainyseasons(betweenApriland

August), and are less likely to happen throughout the rest of the year (mainly cool

season).With scenario B,meanwhile, where the epizootics are generated from the

intra-provincemovements alone, the𝑆𝐸𝐼𝑅𝑆 models gave amuch lower number of

infected sites and a shorter period for the risk ofAHPND (betweenApril and June),

comparedwiththescenarioA.


195

Themodelresultsforthetwoscenariossuggestthatcontrolstrategiesofrestrictions

on inter-provincemovements of live shrimp are obviously required tominimise the

spreadofAHPND.

Figure 6.4 Mean number of infected sites per seed for one-month epizootics. Infection occurs via long-distance and local spread (default parameters 𝜷𝒍𝒐𝒏𝒈 was 1 and 𝜷𝒍𝒐𝒄𝒂𝒍 was 0.002). The averaged values of 1 040 epizootics are shown, where 16 seeds were selected at random in each month (seed fraction was 0.02). The results were generated from two data sets: both intra- and inter-province movements (scenario A; solid line), and intra-province movements alone (scenario B; dash line).

6.4.2 Effectoflong-distanceandlocaltransmissiononAHPNDepizooticdynamics

Themodelalsoprovidedthechancetosimulatetrialbiosecuritymeasures,i.e.testing

for diseases in farmed shrimp before movements, and disease surveillance

programmes.ThemodelswithscenarioAuseddifferenttrialvaluesof𝛽6789=0,0.5and

1.Parameter𝛽67;<6 was0.002,andtheseedfractionwas0.02.

TheresultsareshowninFigure6.5.Ifthemovementsreliedonbiosecuritymeasures

(𝛽6789waslower),thenumberofinfectedsitesdecreasedmarkedlyandtheremaining

numberofinfectedsizeswasverysmall.Importantly,inthemodels,localspreadalone

couldsupporttheAHPNDepizooticsintheThaishrimpfarming(𝛽6789was0and𝛽67;<6

was0.002). This indicated that even if the Thai shrimp farming sector fully applied

0

5

10

15

20

25

30

Mar2013

Apr2013

May2013

Jun2013

Jul2013

Aug2013

Sep2013

Oct2013

Nov2013

Dec2013

Jan2014

Feb2014

Mar2014

Num

bero

finfectedsites()

Intra- andinter-provincemovements

Intra-provincemovementsalone


196

biosecurityonliveshrimpmovements,asmallnumberofseeds(16seeds)couldcause

anAHPNDepizooticvialocalspread.

Figure 6.5 Expected outcomes of the application of biosecurity measures on live shrimp movements in Thailand. With 𝜷𝒍𝒐𝒏𝒈 varied from 0 to 1, the lower 𝜷𝒍𝒐𝒏𝒈 gave smaller epizootics. The model set 𝜷𝒍𝒐𝒄𝒂𝒍 at 0.002, and the seed fraction at 0.02.

Toevaluatetheeffectofincreasedlocalspread,theepizooticswerecomparedwithtwo

valuesof𝛽67;<6 (0.002and0.01perday)inscenarioA.Thedefaultparameter𝛽6789was

settobeone.Figure6.6showsthatthegreaterlocalspreadresultsinahighernumber

ofinfectedsites.Interventionstoreducelocalspreadshouldthereforebeconsidered

aspartofcurrentdiseasesurveillanceandcontrolmeasures.

0

5

10

15

20

25

30

Mar2013

Apr2013

May2013

Jun2013

Jul2013

Aug2013

Sep2013

Oct2013

Nov2013

Dec2013

Jan2014

Feb2014

Mar2014

Num

bero

finfectedsites()

=0

=0.5

=1


197

Figure 6.6 Effects of larger local spread in Thai shrimp farming sectors. The higher local spread (𝜷𝒍𝒐𝒄𝒂𝒍 was 0.01) caused a higher number of infected sites, compared with setting 𝜷𝒍𝒐𝒄𝒂𝒍 at 0.002. The model set 𝜷𝒍𝒐𝒏𝒈 at 1, and the seed fraction at 0.02.

6.4.3 GeographicdistributionsofAHPNDprevalenceatprovinciallevelinThailand

Accordingtothemodel,thegeographicaldistributionsofAHPNDprevalenceonday153

(July 2013, 31) is presented in Figure 6.7a. Themodel shows that the southern and

eastern provinces of Thailandwere at greater risk of AHPND than other areas. The

resultsweresimilartotherealpatternofAHPNDepizooticsinThailandreportedinJuly

2013 as shown in Figure 6.7b (FAO, 2013 see Table 6.1). In respect to thewestern

provinces, however, the model clearly overestimated the number of AHPND, as in

realitynoneoccurredinthewestduringthestudiedperiod.

Num

bero

finfectedsites() =0.01

=0.002

0

5

10

15

20

25

30

Mar2013

Apr2013

May2013

Jun2013

Jul2013

Aug2013

Sep2013

Oct2013

Nov2013

Dec2013

Jan2014

Feb2014

Mar2014


198

(a) (b)

Figure 6.7 Geographic distributions of AHPND-infected provinces in Thailand. (a) The prevalence of AHPND between March 1st 2013 and July 31st 2013, are modelled using the 𝑺𝑬𝑰𝑹𝑺 models. The epizootic models started with 16 seeds in each month. Model parameters of 𝜷𝒍𝒐𝒏𝒈 and 𝜷𝒍𝒐𝒄𝒂𝒍 were 1 and 0.002, respectively. (b) The real AHPND presence in each province of Thailand as reported in July, 2013 (FAO, 2013).

6.4.4 Predictiveperformanceofthe𝑺𝑬𝑰𝑹𝑺models

Tofurthertestthemodelvalidity,theROCapproachwasusedtoshowtheAUCvalues

for threetestmodels: the𝑆𝐸𝐼𝑅𝑆model, thesimplernullmodelwith thenumberof

sites,andthesimplernullmodelwiththenumberofconnections.

Figure6.8showsthatalltestmodelshavegoodpredictiveperformancecomparedwith

anuninformativemodel(AUCof0.5).TheAUCvaluesofthe𝑆𝐸𝐼𝑅𝑆model,sitenumber

model,andconnectionnumbermodelwere0.88,0.87,and0.91,respectively.Although

theAUCofthe𝑆𝐸𝐼𝑅𝑆modelisclosetothatotheroftheothermodels,the𝑆𝐸𝐼𝑅𝑆model

isasystematicapproachandamoreflexibletoolforpredictingsitesatrisktoAHPND.

Iftheconnectionsbetweenshrimpfarmingsitehavetheconsistencyintermofnetwork

structure,the𝑆𝐸𝐼𝑅𝑆modeldoesnotrequirethereal-timeinformationformodelling.

Highprevalence

LowprevalenceNon-shrimpfarmingareas

Northeastern

Central

West

South

East

Non-shrimpfarmingareasNon-actualprevalence2013Actualprevalence2013

Northeastern

Central

West

South

East


199

Figure 6.8 ROC curves of three test models identifying the presence of AHPND in Thai shrimp farming sites. The ROC curves are plots of true positives, i.e. where the model gave positives for provinces identified as having AHPND (plotted on the y axis) versus false positives, i.e. where the model gave positives for provinces identified as having no AHPND (plotted on the x axis). The disease presence in the field reported by FAO (2013) was used to compare with the disease presence in the 𝑺𝑬𝑰𝑹𝑺 model, and for disease presence in the two simpler null models: prediction by number of sites, and number of connections.

6.5 Discussion

Thisstudyhasmodelledtheepizootiologicalpatternofanewbacterialdiseasenamed

acutehepatopancreaticnecrosisdisease(AHPND,alsoknownasEMS) inThaishrimp

farming, and has examined the effect of long-distance and local transmission on its

epizooticdynamics.Theepizooticestimationreliedonan𝑆𝐸𝐼𝑅𝑆(susceptible-exposed-

infected-remove-susceptible)model. Inaddition, theepizooticmodel simulationwas

controlledby the realnetworkof live shrimpmovementsofThailand (LSMN)overa

13-month period from March 2013 to March 2014. As a consequence, the model

systematically characterised the importance of two studied pathways for AHPND

transmission,therebyinformingthedesignofdiseasesurveillanceandcontrolforwhole

shrimpfarmingsitesoratacountry-widescale.

WhenliveshrimpmovefromalocationofhighAHPNDincidence,theycarrytheriskof

infection.Thus,patternsofliveshrimpmovementscanreflecttheseasonalityofAHPND

Falsepositiverate

True

positivera

te

Diseaseprevalenceof model

Diseaseprevalenceestimatedbysitenumber

Diseaseprevalenceestimatedbyconnectionnumber

0.0 0.2 0.8 1.00.4 0.6

0.0

0.2

0.8

1.0

0.4

0.6


200

epizootics in Thailand. Our model demonstrated that AHPND was likely to occur

betweenAprilandAugust(duringthehotandrainyseasons inThailand).Duringthis

period,thenumberofsiteconnectionstotallyincreasedbyabout50%,asshowninthe

Chapter4“Analysisofthenetworkstructureoftheliveshrimpmovementsrelevantto

AHPNDepizootic”,withapproximatelyhalfofthesehavingoriginallymovedfromthe

south. To implement successful control strategies, seed-producing sites in southern

areas should bemonitored closely for the AHPND pathogen, such as increasing the

samplingfrequencypriortotheoccurrenceofregularoutbreaks.

Intermsoflong-distancetransmissioneffects,inter-provincemovementsplayamajor

role inAHPNDepizootics,while intra-provincemovementsalonealsogeneratesmall

epizootic sizes.Althoughan importantbiosecuritypractice, i.e. testing forAHPND in

farmedshrimpbeforemovingtheshrimpfromsitetositehasbeenimplementedinthe

Thaishrimpfarmingsector(Uddin,2008;YamprayoonandSukhumparnich,2010),one

concern iswhether currentdiseasecontrolmeasuresarebasedon thesensitivityof

diagnostictechniquesanddelaysindiagnosis(i.e.thetimebetweensymptomonsetand

establishment of diagnosis) among infected sites (Ahmed et al., 2016; Lightner and

Redman, 1998; Liu et al., 2016; Saulnier et al., 2000a; Sithigorngul et al., 2007).

Additionally,previousworkhasindicatedthatvibriosiscanbefoundinhealthyshrimp,

which makes the infectious shrimp difficult to detect and treat at the initial stage

(Aguirre-Guzmánetal.,2004;Goarantetal.,2006;Gomez-Giletal.,1998;Vincentand

Lotz,2005).Fromthisevidence, it ispossible that the infectedshrimparestillbeing

accidentallymovedbetweensites.

Moreover, local transmission alone can cause AHPND epizootics according to our

modelling, although these epizootics are not large. Nevertheless, in this case if the

diseaseoccursinareaswithahighnumberofsites,orinasitewithhighconnections,

many infected sites may be observed across large areas. This local spread can be

generatedbycommonactivitieswithinThaishrimpfarmingsuchassharingofwater

bodies,aswellasbythecloseproximityofshrimpfarmingsites(Boonyawiwatetal.,

2016;Hazarikaetal.,2000).Waterdischargefrominfectedsitestocanals,riversorthe

sea,alsocauseslocalspreadincreases,ifthosesitesdonotconductappropriatewater

disinfection. Additionally, vibrios are often found growing in aquatic environments


201

(Kongkumnerd,2014),makingshrimpfarmingsitesvulnerabletoreceivingpathogens

vialocalspread.Thesepathwaysdrivinglocalspreadneedtobeeliminated.

Thediseasecontrolstrategiessuggestedbythisproposedmodelarebiosecurityonlive

shrimpmovementsandmonitoringofdiseasesinthenaturalenvironment.Themodel

isusefulfortestingcontrolstrategiesbychangingmodelparameters.First,applyinga

lowerchanceofinfectionviatheconnectionsofinfectedshrimp(𝛽6789)inthemodelling

resulted in smaller sizesofpotentialAHPNDepizootics.This illustrated thatapplying

biosecuritymeasurestoshrimpmovementscouldeffectivelycontrolAHPNDepizootic

dynamics.Anothertestwaschangingthechanceofinfectionduetolocalnon-network

spread (𝛽67;<6).The increased likelihoodof local spread led to largerepizootics.This

emphasised the importance of monitoring diseases in the natural environment of

shrimp farmsandalso theneed forwaterdisinfection for shrimp rearing. Theseare

commonpracticesofbiosecurity,but theymightbeneglected in recentThai shrimp

farming.

Theremovedsites(𝑅)inour𝑆𝐸𝐼𝑅𝑆modelmightbeinfected.Nevertheless,iftheyare

dryingout,forexample,andnotconnectedtowatersourcesorcarryingshrimp(called

sitefallowing),theyprobablydonotposeononwardriskofinfectionbecauseofalack

of host-pathogen-environment interactions (Rockett, 1999). In real practice, Thai

shrimpfarmersconductthefallowingprocessinvaryingperiodsfrom3to60days(data

obtained from the epizootiological survey in Chapter 3), although the Thailand

DepartmentofFisheriesrecommendsthatallsitesshouldbefallowedforbetweenone

andfourweeksafterharvestingshrimp(NationalBureauofAgriculturalCommodityand

FoodStandards,2014;ThailandDoF,2010).Thus,anappropriateperiodof fallowing

requiresfurthermodelling.

Amodelwillneveraddressexactlythereal-worldcomplexity.Itjustmeansthatasimple

solutiontoacomplexproblemmaywork.Forexample,inourmodels,theoccurrence

ofAHPNDwasoverestimatedinsomeareas.Nevertheless,themodelprovidedmore

accuratepredictionsofAHPNDoccurrencethanrandomguessmodels.Forcomparing

the performance of the three different test models, the value of the ROC curves

obtainedfromthe𝑆𝐸𝐼𝑅𝑆modelwasclosetothatofthetwosimplernullmodels(as


202

describedintheresultssection).The𝑆𝐸𝐼𝑅𝑆epizooticmodelwasheavilydrivenbythe

connectiondata(liveshrimpmovements),thustheobtainedresultwasnotunexpected.

Alsoofconnectionsverycloselywithsitecountthatwasnotunexpectedeithersince

the site counts are known to relate to low or high numbers of connections.

Nevertheless,notethatdatahereaggregatesatahighlevel(province),andaggregation

atasitelevelmaythereforeshowdifferentresults.

Oneofthelimitationsofthisresearchwasthatthemodelwasnotgreatlyconcerned

withtypesofshrimpfarmingsites,i.e.large,mediumandsmallscales.Differenttypes

ofsiteprovidevariousfarmingmanagementpractices(Erleretal.,2007;Lebeletal.,

2010).Themodeldid,however,considerthenumberofshrimp.Large-scalesitesthat

commonlyrelatetostockingahighnumberofshrimpshouldhavehighvulnerability

toinfectioncomparedwithsmallandmediumscales.Takingaccountofthiswouldbe

expectedtomakethemodelmoreaccurate.

Tosummarise,theAHPNDepizooticdynamicsweremodelledintheThaishrimpfarming

industryusingan𝑆𝐸𝐼𝑅𝑆compartmental,individual-basedepizooticmodel.Theresults

suggestedthatboth long-distanceand local transmission influencedthedynamicsof

AHPNDepizooticinThaishrimpfarming.Oftheeffortstocontroltheseepizooticsatthe

countryscale,biosecurityof liveshrimpmovementsand interventionstoreducethe

local spread (e.g.monitoringofpathogens innatural environments) are required. In

futuremodellingofAHPNDepizooticdynamics, control strategies in respect to local

spreadwillbeevaluated.


203

6.6 References

Aguirre-Guzmán,G.,MejiaRuíz,H.andAscencio,F.(2004)AreviewofextracellularvirulenceproductofVibriospeciesimportantindiseasesofcultivatedshrimp.AquacultureResearch,35(15),pp.1395-1404.

Ahmed,R.,Rafiquzaman,S.,Hossain,M.T.,Lee,J.andKong,I.(2016)Species-specificdetectionofVibrioalginolyticusinshellfishandshrimpbyreal-timePCRusingthegroELgene.AquacultureInternational,24(1),pp.157-170.

Alonzo,T.A.andPepe,M.S.(2002)Distribution-freeROCanalysisusingbinaryregressiontechniques.Biostatistics(Oxford,England),3(3),pp.421-432.

Bénet,T.,Voirin,N.,Nicolle,M.,Picot,S.,Michallet,M.andVanhems,P.(2013)Estimationoftheincubationperiodofinvasiveaspergillosisbysurvivalmodelsinacutemyeloidleukemiapatients.MedicalMycology,51(2),pp.214-218.

Bivand,R.,Keitt,T.andRowlingson,B.(2015)rgdal:Bindingsforthegeospatialdataabstractionlibrary.

Boonyawiwat,V.,Patanasatienkul,T.,Kasornchandra,J.,Poolkhet,C.,Yaemkasem,S.,Hammell,L.andDavidson,J.(2016)Impactoffarmmanagementonexpressionofearlymortalitysyndrome/acutehepatopancreaticnecrosisdisease(EMS/AHPND)onpenaeidshrimpfarmsinThailand.JournalofFishDiseases,40(5),pp.649-659.


Chen,H.(2007)Useoflinear,Weibull,andlog-logisticfunctionstomodelpressureinactivationofsevenfoodbornepathogensinmilk.FoodMicrobiology,24(3),pp.197-204.

Chonsin,K.,Matsuda,S.,Theethakaew,C.,Kodama,T.,Junjhon,J.,Suzuki,Y.,Suthienkul,O.andIida,T.(2016)GeneticdiversityofVibrioparahaemolyticusstrainsisolatedfromfarmedPacificwhiteshrimpandambientpondwateraffectedbyacutehepatopancreaticnecrosisdiseaseoutbreakinThailand.FEMSMicrobiologyLetters,363(2),pp.1-8.

Dabu,I.M.,Lim,J.J.,Arabit,P.M.T.,Orense,SharlaineJoiAnnB,Tabardillo,J.A.,Corre,V.L.andManingas,M.B.B.(2015)ThefirstrecordofacutehepatopancreaticnecrosisdiseaseinthePhilippines.AquacultureResearch,pp.1-8.

Delignette-Muller,M.L.andDutang,C.(2015)fitdistrplus:AnRPackageforFittingDistributions.JournalofStatisticalSoftware,64(4),pp.1-34.



204

Durand,B.andMahul,O.(2000)Anextendedstate-transitionmodelforfoot-and-mouthdiseaseepidemicsinFrance.PreventiveVeterinaryMedicine,47(1–2),pp.121-139.


Erler,D.,Songsangjinda,P.,Keawtawee,T.andChaiyakam,K.(2007)Nitrogendynamicsinthesettlementpondsofasmall-scalerecirculatingshrimpfarm(Penaeusmonodon)inruralThailand.AquacultureInternational,15(1),pp.55-66.


Galappaththi,E.K.andBerkes,F.(2015)Canco-managementemergespontaneously?CollaborativemanagementinSriLankanshrimpaquaculture.MarinePolicy,60,pp.1-8.

Goarant,C.,Ansquer,D.,Herlin,J.,Domalain,D.,Imbert,F.andDeDecker,S.(2006)“SummerSyndrome”inLitopenaeusstylirostrisinNewCaledonia:Pathologyandepidemiologyoftheetiologicalagent,Vibrionigripulchritudo.Aquaculture,253(1–4),pp.105-113.

Gomez-Gil,B.,Tron-Mayén,L.,Roque,A.,Turnbull,J.F.,Inglis,V.andGuerra-Flores,A.L.(1998)SpeciesofVibrioisolatedfromhepatopancreas,haemolymphanddigestivetractofapopulationofhealthyjuvenilePenaeusvannamei.Aquaculture,163(1–2),pp.1-9.

Hazarika,M.,Samarakoon,L.,Honda,K.,Thanwa,J.,Pongthanapanich,T.,Boonsong,K.andLuang,K.(2000)MonitoringandimpactassessmentofshrimpfarmingintheeastcoastofThailandusingremotesensingandGIS.InternationalArchivesofPhotogrammetryandRemoteSensing,33(B7/2;PART7),pp.504-510.

Henryon,M.,Berg,P.,Olesen,N.J.,Kjær,T.E.,Slierendrecht,W.J.,Jokumsen,A.andLund,I.(2005)Selectivebreedingprovidesanapproachtoincreaseresistanceofrainbowtrout(Onchorhynchusmykiss)tothediseases,entericredmouthdisease,rainbowtroutfrysyndrome,andviralhaemorrhagicsepticaemia.Aquaculture,250(3–4),pp.621-636.

Huitric,M.,Folke,C.andKautsky,N.(2002)DevelopmentandgovernmentpoliciesoftheshrimpfarmingindustryinThailandinrelationtomangroveecosystems.EcologicalEconomics,40(3),pp.441-455.


205

Keeling,M.J.andEames,K.T.(2005)Networksandepidemicmodels.JournaloftheRoyalSocietyInterface,2(4),pp.295-307.


Kumar,R.andIndrayan,A.(2011)Receiveroperatingcharacteristic(ROC)curveformedicalresearchers.IndianPediatrics,48(4),pp.277-287.

Lai,C.C.,Ji,D.D.,Wu,F.T.,Mu,J.J.,Yang,J.R.,Jiang,D.D.,Lin,W.Y.,Chen,W.T.,Yen,M.Y.,Wu,H.S.andChen,T.H.(2016)EtiologyandriskfactorsofacutegastroenteritisinaTaipeiEmergencyDepartment:clinicalfeaturesforbacterialgastroenteritis.JournalofEpidemiology/JapanEpidemiologicalAssociation,26(4),pp.216-223.

Lebel,L.,Mungkung,R.,Gheewala,S.H.andLebel,P.(2010)Innovationcycles,nichesandsustainabilityintheshrimpaquacultureindustryinThailand.EnvironmentalScience&Policy,13(4),pp.291-302.

Leclerc,M.,Doré,T.,Gilligan,C.,Lucas,P.,Filipe,J.andNishiura,H.(2014)Estimatingthedelaybetweenhostinfectionanddisease(incubationperiod)andassessingits.PloSOne,9(1),e86568.Available:http://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0086568&type=printable[Accessed:1May2016].

Lightner,D.V.,Redman,R.M.,Pantoja,C.R.,Noble,B.I.andTran,L.(2012)EarlymortalitysyndromeaffectsshrimpinAsia.GlobalAquacultureAdvocateMagazine,40Available:https://pdf.gaalliance.org/pdf/GAA-Lightner-Jan12.pdf[Accessed:31March2013].

Lightner,D.V.andRedman,R.M.(1998)Shrimpdiseasesandcurrentdiagnosticmethods.Aquaculture,164(1–4),pp.201-220.

Liu,X.,Guan,Y.,Cheng,S.,Huang,Y.,Yan,Q.,Zhang,J.,Huang,G.,Zheng,J.andLiu,T.(2016)DevelopmentofahighlysensitivelateralimmunochromatographicassayforrapiddetectionofVibrioparahaemolyticus.JournalofMicrobiologicalMethods,131,pp.78-84.

Muniesa,A.,Perez-Enriquez,R.,Cabanillas-Ramos,J.,Magallón-Barajas,F.J.,Chávez-Sánchez,C.,Esparza-Leal,H.andBlas,I.(2015)IdentifyingriskfactorsassociatedwithwhitespotdiseaseoutbreaksofshrimpsintheGulfofCalifornia(Mexico)throughexpertopinionandsurveys.ReviewsinAquaculture,pp.1-9.



206


Nagai,F.,Mektrairat,N.andFunatsu,T.(2008)LocalgovernmentinThailand:analysisofthelocaladministrativeorganizationsurvey.Chiba,Japan:InstituteofDevelopingEconomies.


Nunan,L.,Lightner,D.V.,Pantoja,C.andGomez-Jimenez,S.(2014)Detectionofacutehepatopancreaticnecrosisdisease(AHPND)inMexico.DiseasesofAquaticOrganisms,111(1),pp.81-86.


Park,S.H.,Goo,J.M.andJo,C.H.(2004)Receiveroperatingcharacteristic(ROC)curve:practicalreviewforradiologists.KoreanJournalofRadiology,5(1),pp.11-18.

Patil,A.A.,Annachhatre,A.P.andTripathi,N.K.(2002)Comparisonofconventionalandgeo-spatialEIA:Ashrimpfarmingcasestudy.EnvironmentalImpactAssessmentReview,22(4),pp.361-375.


Rizzo,A.,Pedalino,B.andPorfiri,M.(2016)AnetworkmodelforEbolaspreading.JournalofTheoreticalBiology,394,pp.212-222.

Rockett,I.R.(1999)Populationandhealth:anintroductiontoepidemiology.PopulationBulletin,54(4),pp.1.

Rvachev,L.A.andLongini,I.M.(1985)Amathematicalmodelfortheglobalspreadofinfluenza.MathematicalBiosciences,75(1),pp.3-22.

Saulnier,D.,Avarre,J.,LeMoullac,G.,Ansquer,D.,Levy,P.andVonau,V.(2000)RapidandsensitivePCRdetectionofVibriopenaeicida,theputativeetiologicalagentofSyndrome93inNewCaledonia.DiseasesofAquaticOrganisms,49(2),pp.109-115.


207

Sithigorngul,P.,Rukpratanporn,S.,Pecharaburanin,N.,Suksawat,P.,Longyant,S.,Chaivisuthangkura,P.andSithigorngul,W.(2007)AsimpleandrapidimmunochromatographicteststripfordetectionofpathogenicisolatesofVibrioharveyi.JournalofMicrobiologicalMethods,71(3),pp.256-264.

Smith,P.T.(1999)ApplicationofGIStotheThaishrimpfarmsurvey.In:P.T.Smith,ed.TowardsSustainableShrimpCultureinThailandandtheRegion.Songkhla,Thailand,28thOctoberto1stNovember1996.Canberra,Australia:AustralianCentreforInternationalAgriculturalResearch,pp.113-130.Available:http://aciar.gov.au/files/node/2196/pr90_pdf_11112.pdf[Accessed:17November2014].

Taylor,N.G.,Norman,R.A.,Way,K.andPeeler,E.J.(2011)Modellingthekoiherpesvirus(KHV)epidemichighlightstheimportanceofactivesurveillancewithinanationalcontrolpolicy.JournalofAppliedEcology,48(2),pp.348-355.

Tennekes,M.(2017)tmap:Thematicmaps.

ThailandDoF(2010)Thaigoodaquaculturepractices(GAP)certificationstandardsformarineshrimpfarms.Available:http://www.fisheries.go.th/thacert/images/pdf/SF_GAPcriteria.pdf[Accessed:1May2015].

Tojinbara,K.,Sugiura,K.,Yamada,A.,Kakitani,I.,Kwan,N.C.L.andSugiura,K.(2016)Estimatingtheprobabilitydistributionoftheincubationperiodforrabiesusingdatafromthe1948–1954rabiesepidemicinTokyo.PreventiveVeterinaryMedicine,123,pp.102-105.

Tran,L.,Nunan,L.,Redman,R.M.,Lightner,D.V.andFitzsimmons,K.(2013a)EMS/AHPNS:infectiousdiseasecausedbybacteria.GlobalAquacultureAdvocate.Available:http://apfa.com.au/wp-content/uploads/2017/03/GAA-Tran-July13.pdf[Accessed:6May2016].

Tran,L.,Nunan,L.,Redman,R.M.,Mohney,L.L.,Pantoja,C.R.,Fitzsimmons,K.andLightner,D.V.(2013b)Determinationoftheinfectiousnatureoftheagentofacutehepatopancreaticnecrosissyndromeaffectingpenaeidshrimp.DiseasesofAquaticOrganisms,105(1),pp.45-55.

Uddin,M.T.(2008)ValueChainsandStandardsinShrimpExport.

Ushijima,B.(2014)PoolingKnowledge:Devisingnext-generationtechnologiestodetectdiseasesaffectingaquaculture.Japan:PublicRelationsOfficeoftheGovernmentofJapan.Available:http://dwl.gov-online.go.jp/video/cao/dl/public_html/gov/pdf/hlj/20141101/24-25.pdf[Accessed:6May2016].


208

Vandergeest,P.,Flaherty,M.andMiller,P.(1999)ApoliticalecologyofshrimpaquacultureinThailand.RuralSociology,64(4),pp.573-596.

Vincent,A.G.andLotz,J.M.(2005)Timecourseofnecrotizinghepatopancreatitis(NHP)inexperimentallyinfectedLitopenaeusvannameiandquantificationofNHP-bacteriumusingreal-timePCR.DiseasesofAquaticOrganisms,67(1-2),pp.163-169.

Werkman,M.,Green,D.M.,Murray,A.G.andTurnbull,J.F.(2011)TheeffectivenessoffallowingstrategiesindiseasecontrolinsalmonaquacultureassessedwithanSISmodel.PreventiveVeterinaryMedicine,98(1),pp.64-73.

Yamprayoon,J.andSukhumparnich,K.(2010)Thaiaquaculture:achievingqualityandsafetythroughmanagementandsustainability.JournaloftheWorldAquacultureSociety,41(2),pp.274-280.

Zorriehzahra,M.andBanaederakhshan,R.(2015)Earlymortalitysyndrome(EMS)asnewemergingthreatinshrimpindustry.Adv.Anim.Vet.Sci,3(2s),pp.64-72.


209

Chapter 7 - General discussion

7.1 Summary

Thisfinalchaptersummarisesthemainconclusionsoftheresearch.Inthelastpartof

thischapter,wesuggestsomedirectionsforfurtherwork.

The main objective of this research was to apply epizootiological tools to study of

AHPND and other infectious disease spread in the Thai shrimp farming industry.

Chapter 3 (the first research chapter), we evaluated the risk factors for AHPND

occurrenceatsitelevelusingacross-sectionalstudydesign.TheAHPND-infectedcases

wereidentifiedfromadecisiontreethatwasdevelopedbasingongrosssignsofAHPND

andnon-laboratoryconfirmation.Theresultsfromunivariateanalysisandconditional

logisticregressionindicatedthatnopondharrowingincreasedtheriskofAHPNDand

the use of earthen ponds for shrimp rearing decreased the risk of the disease

occurrence.

Chapter4utilisedthedataoftheliveshrimpmovementnetwork(LSMN)toexamine

thespreadofinfectiousdiseasesinshrimpfarmingsites.Theresultsemphasisedthat

theLSMNhadahighnumberofinter-provincemovements,whichincreasedthechance

ofdiseaseepizootics acrossprovinceand regionbordersof Thailand.Moreover, the

LSMNrepresentedascale-freenetworkwithsmall-worldproperties.Thissuggeststhat

regulatorsanddecisionmakersneedtoposemoreattentionineffortstodesigndisease

surveillanceandcontrolstrategies.

Consequently, in Chapter 5 disease-control algorithms were developed using four

targetedapproachesandanon-targetingapproach.Theseaimedtofindoptimalcontrol

strategies to reduce the potential epizootic size in the LSMN. The results from the

studied network demonstrate that targeted strategies, i.e. biosecurity measures in

respecttotargetedliveshrimpmovements,supportthesuccessofdiseasesurveillance

andcontrolinThaishrimpfarming,andimprovethecost-effectivenessofsurveillance

resources.


210

Chapter6modelledthespreadofAHPNDfromsitetosite,basedontwotransmission

routesforthedisease(long-distanceandlocaltransmission).An𝑆𝐸𝐼𝑅𝑆compartmental,

individual-basedepizooticmodelshowedthatAHPNDwaslikelytooccurduringthehot

andrainyseasonsofThailand(betweenAprilandAugust).Duringthisoutbreakperiod,

themovementsweremainlysourcedfromtheareasinthesouth.Althoughthenumber

ofintra-provincemovementswassmall(23%oftotalconnections),theyalsoaffected

theestimatedepizooticsize.Fordiseasepreventionandcontrol,themodelsindicated

thatlowerlong-distancetransmissionrates(𝛽6789 <1,and𝛽67;<6 was0.002)strongly

reduced thenumberof infectedsites. Local spreadalone,however, couldstill cause

epizootics(𝛽6789was0,and𝛽67;<6 was0.002).GiventhesetwomajorroutesforAHPND

transmission, two measures are suggested in this research. First is an increase in

biosecurityonliveshrimpmovements(effectivetestingofdiseasesinfarmedshrimp

beforemovements,andtargeteddiseasesurveillanceandcontrolofdiseasespreadin

the live shrimp movement network). The second is interventions to reduce local

transmissionsuchasmonitoringofdiseasesinnaturalenvironments.

7.2 Generaldiscussion

7.2.1 Diseasecaseconfirmation

Chapter3developedanAHPNDdecisiontreefortheidentificationofAHPND-infected

cases,basedonthediseaseatthepointbeinganunknownetiology.Thiseffectivetool

wasabletonarrowthegapperiodfordiseasepreventionandcontrols.Thetoolalso

couldalsohelpthedecisionoffarmerstodealwiththeinfectionatsitelevel,providing

evidencetosupporttheearlystoppingofanypathwayfortransmittingpathogens,such

as stoppingwaterdischarge tonatural sources.Theoutcomeshould lead to smaller

sizedepidemics,asproposedinFergusonetal.(2001).Amatterofgreatconcernisthat

waitingforlaboratoryresultscausedthedelayedoutbreakreporting,andconsequently

the late implementation of mitigation measures, as happened with a waterborne

outbreakofGiardiasis(agastrointestinaldisease) intheNorwegianpopulationin2004

(Nygård et al., 2006). Hence, until there is a rapid diagnosticmethod for infectious

diseases, the decision tree will remain useful to the study epidemiology of these

diseases.


211

The AHPND decision tree can reflect on current disease surveillance and control

programmes in Thailand. Shrimp samples obtained from both active and passive

surveillancesareconfirmedbythePCRtesting(adiagnosistoolforAHPND).Thistestis

quitecostly(approximately25USDpersample)andtime-consuming(aroundtwodays)

(CoastalAquaticAnimalHealthResearchInstitute,2016b).Thus,thedecisiontreecan

beusedasaprimaryscreen todetectcasesofAHPND.Theuseof thedecision tree

techniqueasascreeningtoolfordiseaseshasalreadybeenshowntobeeffectiveinthe

epidemiologicalinvestigationofrheumatoidarthritisinalargedatabaseofaroundtwo

millionpatients(Zhouetal.,2016).Foranotherpurpose,thedecisiontreegivesdecision

makers a tool to choose optimised control measures for foot-and-mouth disease

transmittedbetweenlivestockanimals(Tomassenetal.,2002).

7.2.2 Shrimpfarmingdatausedforepizoology

7.2.2.1Datarecordedonshrimpfarmingmanagementandpractices

DuringtheepizootiologicalsurveyinChapter3,akeysuccessforthedatacollectionwas

thatmany interviewees hadwritten their daily farming practices in their own diary

booksorelectronic files.Therecordingfollowstherequirementofgoodaquaculture

practice (GAP) standards for Thai shrimp farming (National Bureau of Agricultural

CommodityandFoodStandards,2014),withmostshrimpfarmingsitesinThailandnow

being certified by the GAP standard (Fisheries Commodity Standard System and

TraceabilityDivision,2017).Thus,theintervieweescouldprovideaccuratedatatous,

leading to robustepizootiological studies.The recordingof farmingactivities ismost

helpfulinepizoologyifsuchrecordingiscollectedforpurposesofdiseasesurveillance

andcontrols(Birkheadetal.,2015;Wendtetal.,2015).

7.2.2.2Liveshrimpmovementnetworkdata

Itshouldbeemphasisedthattheliveshrimpmovementnetworkdataprovidesrobust

epizootiological information.Modellingthesedataasanetworkprovidedapowerful

toolfortheunderstandingofdiseasespreadbetweensites(Chapter4).Additionally,its

modellingaidedthedesignofdiseasesurveillanceandcontrolstrategiesthatmightbe

applied given limitations of surveillance resources (Chapter 5). Nonetheless, the


212

networkmaychangeitsstructureoverthelongtermbecausepatternsofconnections

change(Greenetal.,2012;Stoddardetal.,2013).Implementationsofcontrolstrategies

such as vaccination (McReynolds et al., 2014), and case isolation and quarantine

(Lipsitchetal.,2003;Pandeyetal.,2014)alsoinfluencethestructureofnetworks.Thus,

furtherresearchisneededtodeterminetheconsistencyofthisnetworkstructure,in

whichsuchhistoricaldataoftheliveshrimpmovementnetworkprovideabenchmark.

ThenetworkdatathatweobtainedfromtheThailandDepartmentofFisheriescanbe

consideredtobehighlyaccuratefortworeasons.First,thedatawerebasedonofficial

reports;second,suchinformationwerekeptincomputeriseddatabase.Nonetheless,

thereweresomeminorfailuresthatshouldbementioned.Errorswerefoundinrespect

tositeaddresseswithinthevillagedetails. Ifwereliedonthelocationofsitesatthe

lowest level of the village, therefore, these errorswould affect the reliability of our

results.Hence,computerdataareonlyasgoodasthatwhichareentered.

7.2.3 Modellingdiseaseepizooticdynamics

Thenetworkmodelling inChapters4and5 considers thepotential routeofdisease

transmission(liveshrimpmovements)butinrealitymostshrimpdiseasescantransmit

fromsitetositeviaotherpathways,suchashydrologicalconnectivityandthephysical

proximity of sites (see Chapter 1). In addition, the non-random mixing and

heterogeneityofconnectionsaffectdiseasetransmission.Thus,wemodelledthelive

shrimpmovementsasaweightednetwork,andtooktheeffectof localnon-network

spread and the number of shrimp into account while modelling AHPND epizootic

dynamics(Chapter6).

Thecapacityoffarmerstomanagetheirsitesafterinfectionalsoplaysanimportantrole

indiseaseepizootics.Wuetal.(2001)proposedthatthemortalityofwhitespotdisease-

challengedshrimpthatwerestockedathighdensitybecamecriticalifmoribundshrimp

were not immediately removed as to minimise the cannibalism and waterborne

transmission.Theevidenceindicatesthatmodelsfordiseasespreadsbetweenanimal

farmsitesshouldconsiderthefarmers’behaviourindiseasepreventionandcontrolas

anexampleinChapter6.


213

7.3 Futurework

7.3.1 Controlstrategiesforlocalnon-networkspread

To control disease spread via live shrimpmovements, Thailand applied a process of

testing farmed shrimp for diseases before their movements (National Bureau of

Agricultural Commodity and Food Standards, 2015). Additionally, Thai regulators

routinelymonitorfordiseasesinfarmedshrimpthroughactiveandpassivesurveillance

programmes (Coastal Aquatic Animal Health Research Institute, 2016a). Here, a

targeted disease surveillance approach was developed in Chapter 5 using network

modellingapproacheswiththeaimofimprovingthesesurveillanceprogrammes.While

the effect of the local non-network spread on the AHPND epizootic dynamics was

demonstratedinChapter6,thebenefitofthetargetedstrategiesforlocalnon-network

spreadwasnotclearinpractice.

Commonly,Vibriospp.arefoundtobefreelivinginaquaticenvironments(Thompson

etal.,2004).Theirlifecyclesoftenrelatetolocalnon-networkspreadofdiseases(i.e.

hydrologicaltransmission)forfarmedshrimp(Apanakapanetal.,2016;Heenatigalaand

Fernando,2016;Naganathanetal.,2014;Puietal.,2014).Importantly,theabundance

ofVibriospp.innaturalwatersourcescanbepromotedbycohabitationwithprotozoans

(Laskowski-ArceandOrth,2008).Thus,targetingofnetworkspreadmaynotsufficient.

For this case, hydrodynamicmodelsmayneed tobe a part of thedesignof control

strategies.Parallelexamplesforthisincludesalmonlice,whichareabundantonboth

wildandfarmedfish(Torrissenetal.,2013),andforwhichmanyhydrodynamicmodels

havealreadybeendeveloped(GillibrandandWillis,2007;MurrayandGillibrand,2006).

7.3.2 Coinfectionepizooticmodels

DuringAHPNDoutbreaks,farmedshrimpinAsiafaceanotherinfectiousdiseasenamed

hepatopancreaticmicrosporidiosis (HPM)causedbyamicrosporidianEnterocytozoon

hepatopenaei (NACA, 2015; Thitamadee et al., 2016). HPM does not lead to mass

mortalities, but it is associatedwith slow growth in affected stocks (Salachan et al.,

2017).Nevertheless,recently,Arangurenetal.(2017)foundthatHPM-infectedshrimp

haveahighriskofAHPNDinfection.Theevidenceshowsthatshrimpareinfectedby


214

bothHPMandAHPND,leadingtomorecomplexdynamicsofAHPNDepizooticsthan

ourmodelsinChapter6.Additionalparameters,suchasaprobabilityofHMPinfection,

thereforeneedtobeaddedtothemodelsdevelopedinthisresearch.

7.3.3 Geographicalinformationsystems(GIS)forshrimpfarmingsites

Withapproachesofcomputerisationandgeographical informationsystems(GIS),the

description of a disease epidemic in terms of real spatial results can be undertaken

(Cromley,2003;Tamietal.,2016).ThisisbecauseGIScanworkonareas,whilemany

networkmodelssuchasourstreatsitesessentiallyaspointsources.Forhumandisease

models,GIS has been used to examine transmission of tuberculosis (Moonan et al.,

2004), chikungunya fever (Rodriguez-Morales et al., 2016), and dengue disease

(Palaniyandietal.,2014),forexample.GIShasalsorecentlybeenusedinthemodelling

of livestock diseases such as bovine tuberculosis (Ribeiro‐Lima et al., 2016). For

aquatic animals, GIS is typically applied to identify suitable areas for aquatic animal

farming(Giapetal.,2005;Salametal.,2003;Salametal.,2005),neverthelessGIShas

beenlesswidelyusedinshrimpandfishepizoology(Bayotetal.,2008).

ForThaishrimpfarming,theGISapproachcouldallowusmoreaccuracyinsitelocation

to determine a change in the distance of movements caused by any rewiring.

Additionally, the geographic distribution of AHPND, and the spread of other shrimp

diseasesinThailand,couldbedrawnatsitelevelinsteadofprovincelevelasshownin

Chapter6.

7.4 Conclusions

- Theuseofepidemiologicalandepizootiologicaltoolsdemonstratedfourmain

epizootiologicalfindingsforAHPNDspreadinThaishrimpfarming.

- A cross-sectional study revealed a significant association between AHPND

transmission at site level and farming management practices, i.e. types of

ongrowing ponds and harrowing. No pond harrowing increased the risk of

AHPNDonset,whileearthenpondsdecreasedtheriskofAHPNDoccurrencein

ongrowingsites.


215

- Moreover, case confirmation with the AHPND decision tree allowed rapid

investigationsoftheriskfactorsfortheAHPNDtransmission,giventhatatthe

timeofdatacollection,AHPNDwasadiseasewithanunknownetiology.This

case-identification AHPND decision tree can be adapted when there is

recurrenceofAHPND,orthegrosssignsofdiseasepathologycanbechanged

whenthereisincidenceofothernewdiseases.

- Inordertopreventandcontrolinfectiousdiseaseoutbreaksatthecountryscale,

networkmodellingisappliedhereforthefirsttimetotheshrimpfarmingsector.

ThisillustratedthepropertiesoftheliveshrimpmovementnetworkinThailand

(LSMN)inaidingandhinderingthetransmissionofdiseases.Mostimportantly,

our research into network modelling emphasised that the random disease

surveillanceandcontrolcouldnoteffectivelyminimisediseasespreadsitetosite

via long-distance transmission, compared with the optimal targeted disease

surveillanceandcontrol.

- Accordingtoour𝑆𝐸𝐼𝑅𝑆compartmental,individual-basedepizooticmodel,the

inter-provincemovementswerethemainsourceofAHPNDspreadintheThai

shrimpfarming.The𝑆𝐸𝐼𝑅𝑆modellingstronglydemonstratedthattheincrease

ofbiosecuritywithintheliveshrimpmovementnetworkcouldleadtosmaller

potentialAHPNDepizootics.Additionally,itsuggestedthattherewasaneedfor

diseasesurveillanceandcontrolinnaturalsourcesorotherpathwaysrelativeto

localnon-networkspread.

- Ournetworkepizooticmodelmayapplytootherdiseasesandindustries.The

modelcanbeappliedtootherdiseasesbyusingdifferentparametervalues,or

addingotherpathwaysfordiseasetransmission.Themodelcanbeappliedto

other industries by including other types of sites (e.g. fish, mussel and crab

farming, or processing plants) and their connections if the data is available.

Importantly, the model is useful to evaluate current regulations or desired

strategies to prevent and control disease outbreaks such as biosecurity and

surveillancestrategies.


216

The epidemiological and epizootiological tools used in this research provide unique

contributionstothestudyofshrimpepizoology.Theresultsgivemoreunderstandingof

site-to-sitetransmissionofAHPNDandotherinfectiousdiseasesinfarmedshrimp.Thai

regulatorsanddecisionmakerscanusetheminimprovementofdiseasesurveillance

andcontrol.


217

7.5 References

Apanakapan,J.,Chuchird,N.,Tongsri,P.,Taemmesub,T.,Taparhaudee,W.andSritunyalucksana,K.(2016)Effectofspore-formingbacteriaonVibriospp.,survivalrateandwaterqualitiesinthelarvalrearingofthePacificwhiteshrimp(Litopenaeusvannamei).KasetsartUniversityAnnualConference.KasetsartUniversity,2ndto5th

February2016,pp.648-655.Available:https://www.cabdirect.org/cabdirect/FullTextPDF/2016/20163376046.pdf[Accessed:29April2017].

Aranguren,L.F.,Han,J.E.andTang,K.F.(2017)Enterocytozoonhepatopenaei(EHP)isariskfactorforacutehepatopancreaticnecrosisdisease(AHPND)andseptichepatopancreaticnecrosis(SHPN)inthePacificwhiteshrimpPenaeusvannamei.Aquaculture,471,pp.37-42.

Bayot,B.,Sonnenholzner,S.,Ochoa,X.,Guerrerro,J.,Vera,T.,Calderón,J.,deBlas,I.,delPilarCornejo-Grunauer,M.,Stern,S.andOllevier,F.(2008)Anonlineoperationalalertsystemfortheearlydetectionofshrimpepidemicsattheregionallevelbasedonreal-timeproduction.Aquaculture,277(3),pp.164-173.

Birkhead,G.S.,Klompas,M.andShah,N.R.(2015)Usesofelectronichealthrecordsforpublichealthsurveillancetoadvancepublichealth.AnnualReviewofPublicHealth,36,pp.345-359.

CoastalAquaticAnimalHealthResearchInstitute(2016a)Diseasesurveillanceandcontrolprojectsinaquaticanimals.Available:http://www.aquathai.org/wed/projects/[Accessed:23May2017].

CoastalAquaticAnimalHealthResearchInstitute(2016b)Proceduresforaquaticanimalhealthservices.Available:http://www.aquathai.org/wed/home/service/[Accessed:22May2017].

Cromley,E.K.(2003)GISanddisease.AnnualReviewofPublicHealth,24(1),pp.7-24.

Ferguson,N.M.,Donnelly,C.A.andAnderson,R.M.(2001)TransmissionintensityandimpactofcontrolpoliciesonthefootandmouthepidemicinGreatBritain.Nature,413(6855),pp.542-548.

FisheriesCommodityStandardSystemandTraceabilityDivision(2017)ListcertifiedfarmandoperatorforGAPCoCandorganiconfisheriesandaquaculturestandards.Available:http://www.fisheries.go.th/thacert[Accessed:29May2017].

Giap,D.H.,Yi,Y.andYakupitiyage,A.(2005)GISforlandevaluationforshrimpfarminginHaiphongofVietnam.Ocean&CoastalManagement,48(1),pp.51-63.


218

Gillibrand,P.A.andWillis,K.J.(2007)Dispersalofsealouselarvaefromsalmonfarms:modellingtheinfluenceofenvironmentalconditionsandlarvalbehaviour.AquaticBiology,1(1),pp.63-75.


Heenatigala,P.andFernando,M.(2016)OccurrenceofbacteriaspeciesresponsibleforvibriosisinshrimppondculturesystemsinSriLankaandassessmentofthesuitablecontrolmeasures.SriLankaJournalofAquaticSciences,21(1),pp.1-17.

Laskowski-Arce,M.A.andOrth,K.(2008)AcanthamoebacastellaniipromotesthesurvivalofVibrioparahaemolyticus.AppliedandEnvironmentalMicrobiology,74(23),pp.7183-7188.

Lipsitch,M.,Cohen,T.,Cooper,B.,Robins,J.M.,Ma,S.,James,L.,Gopalakrishna,G.,Chew,S.K.,Tan,C.C.,Samore,M.H.,Fisman,D.andMurray,M.(2003)Transmissiondynamicsandcontrolofsevereacuterespiratorysyndrome.Science(NewYork,N.Y.),300(5627),pp.1966-1970.

McReynolds,S.W.,Sanderson,M.W.,Reeves,A.andHill,A.E.(2014)ModelingtheimpactofvaccinationcontrolstrategiesonafootandmouthdiseaseoutbreakinthecentralUnitedStates.PreventiveVeterinaryMedicine,117(3),pp.487-504.

Moonan,P.K.,Bayona,M.,Quitugua,T.N.,Oppong,J.,Dunbar,D.,Jost,K.C.,Burgess,G.,Singh,K.P.andWeis,S.E.(2004)UsingGIStechnologytoidentifyareasoftuberculosistransmissionandincidence.InternationalJournalofHealthGeographics,3(1),pp.23.

Murray,A.andGillibrand,P.(2006)ModellingsalmonlicedispersalinLochTorridon,Scotland.MarinePollutionBulletin,53(1),pp.128-135.

NACA(2015)DiseasesofCrustaceans:hepatopancreaticmicrosporidiosiscausedbyEnterocytozoonhepatopenaei(EHP).Available:http://enaca.org/publications/health/disease-cards/ehp-disease-card-2015.pdf[Accessed:10March2017].

Naganathan,S.,Muruganandam,M.andVenkatesan,V.(2014)EcologicalandquorumsensingpotentialofVibriosp.fromprawnfarmsofTamilNaducoastalregions.WorldJournalofPharmacyandPharmaceuticalSciences,3(3),pp.1102-1112.



219

NationalBureauofAgriculturalCommodityandFoodStandards(2015)GuidanceontheapplicationofThaiagriculturalstandardTAS7401(G)-2015.Available:http://www.acfs.go.th/standard/download/GAP_MARINE-SHRIMP-FARM.pdf[Accessed:1April2017].

Nygård,K.,Schimmer,B.,Søbstad,Ø,Walde,A.,Tveit,I.,Langeland,N.,Hausken,T.andAavitsland,P.(2006)Alargecommunityoutbreakofwaterbornegiardiasis-delayeddetectioninanon-endemicurbanarea.BMCPublicHealth,6(1),pp.141.

Palaniyandi,M.,Anand,P.andManiyosai,R.(2014)GISbasedcommunitysurveyandsystematicgridsamplingfordengueepidemicsurveillance,control,andmanagement:acasestudyofPondicherryMunicipality.Int.JofMosquitoResearch,1(4),pp.72-78.

Pandey,A.,Atkins,K.E.,Medlock,J.,Wenzel,N.,Townsend,J.P.,Childs,J.E.,Nyenswah,T.G.,Ndeffo-Mbah,M.L.andGalvani,A.P.(2014)StrategiesforcontainingEbolainwestAfrica.Science(NewYork,N.Y.),346(6212),pp.991-995.

Pui,C.,Bilung,L.,Mohd,N.,Zainal,N.,Vincent,M.andApun,K.(2014)RiskofacquiringVibrioparahaemolyticusinwaterandshrimpfromanaquaculturefarm.KuroshioSci,8(1),pp.59-62.

Ribeiro-Lima,J.,Carstensen,M.,Cornicelli,L.,Forester,J.andWells,S.(2016)Patternsofcattlefarmvisitationbywhite-taileddeerinrelationtoriskofdiseasetransmissioninapreviouslyinfectedareawithbovinetuberculosisinMinnesota,USA.TransboundaryandEmergingDiseases,pp.1-11.

Rodriguez-Morales,A.J.,Bedoya-Arias,J.E.,Ramírez-Jaramillo,V.,Montoya-Arias,C.P.,Guerrero-Matituy,E.A.andCárdenas-Giraldo,E.V.(2016)Usinggeographicinformationsystem(GIS)tomappingandassesschangesintransmissionpatternsofchikungunyafeverinmunicipalitiesoftheCoffee-TriangleregionofColombiaduring2014–2015outbreak:Implicationsfortraveladvice.TravelMedicineandInfectiousDisease,14(1),pp.62-65.

Salachan,P.V.,Jaroenlak,P.,Thitamadee,S.,Itsathitphaisarn,O.andSritunyalucksana,K.(2017)Laboratorycohabitationchallengemodelforshrimphepatopancreaticmicrosporidiosis(HPM)causedbyEnterocytozoonhepatopenaei(EHP).BMCVeterinaryResearch,13(1),pp.9.

Salam,M.A.,Ross,L.G.andBeveridge,C.M.(2003)AcomparisonofdevelopmentopportunitiesforcrabandshrimpaquacultureinsouthwesternBangladesh,usingGISmodelling.Aquaculture,220(1),pp.477-494.

Salam,M.A.,Khatun,N.A.andAli,M.M.(2005)CarpfarmingpotentialinBarhattaUpazilla,Bangladesh:aGISmethodologicalperspective.Aquaculture,245(1),pp.75-87.


220

Stoddard,S.T.,Forshey,B.M.,Morrison,A.C.,Paz-Soldan,V.A.,Vazquez-Prokopec,G.M.,Astete,H.,Reiner,R.C.,Jr,Vilcarromero,S.,Elder,J.P.,Halsey,E.S.,Kochel,T.J.,Kitron,U.andScott,T.W.(2013)House-to-househumanmovementdrivesdenguevirustransmission.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,110(3),pp.994-999.

Tami,A.,Grillet,M.E.andGrobusch,M.P.(2016)Applyinggeographicalinformationsystems(GIS)toarboviraldiseasesurveillanceandcontrol:Apowerfultool.TravelMedicineandInfectiousDisease,14(1),pp.9-10.


Thompson,F.L.,Iida,T.andSwings,J.(2004)Biodiversityofvibrios.MicrobiologyandMolecularBiologyReviews,68(3),pp.403-431.

Tomassen,F.H.M.,deKoeijer,A.,Mourits,M.C.M.,Dekker,A.,Bouma,A.andHuirne,R.B.M.(2002)Adecision-treetooptimisecontrolmeasuresduringtheearlystageofafoot-and-mouthdiseaseepidemic.PreventiveVeterinaryMedicine,54(4),pp.301-324.

Torrissen,O.,Jones,S.,Asche,F.,Guttormsen,A.,Skilbrei,O.T.,Nilsen,F.,Horsberg,T.E.andJackson,D.(2013)Salmonlice–impactonwildsalmonidsandsalmonaquaculture.JournalofFishDiseases,36(3),pp.171-194.

Wendt,A.,Kreienbrock,L.andCampe,A.(2015)Zoonoticdiseasesurveillance–inventoryofsystemsintegratinghumanandanimaldiseaseinformation.ZoonosesandPublicHealth,62(1),pp.61-74.

Wu,J.L.,Namikoshi,A.,Nishizawa,T.,Mushiak,K.,Teruya,K.andMuroga,K.(2001)EffectsofshrimpdensityontransmissionofpenaeidacuteviremiainPenaeusjaponicusbycannibalismandthewaterborneroute.DiseasesofAquaticOrganisms,47(2),pp.129-135.

Zhou,S.,Fernandez-Gutierrez,F.,Kennedy,J.,Cooksey,R.,Atkinson,M.,Denaxas,S.,Siebert,S.,Dixon,W.G.,O’Neill,T.W.andChoy,E.(2016)Definingdiseasephenotypesinprimarycareelectronichealthrecordsbyamachinelearningapproach:acasestudyinidentifyingrheumatoidarthritis.PloSOne,11(5),pp.e0154515.

AppendixAShrimpdiseasepictures

221

Appendix A: Shrimp disease pictures

Picture cards showing gross signs of AHPND were shown alongside those showing clinical signs of other high-prevalence shrimp diseases while collecting data for AHPND cases and non-cases.

AppendixAShrimpdiseasepictures

222

Appendix A: Shrimp disease pictures

Picture cards showing gross signs of AHPND were shown alongside those showing clinical signs of other high-prevalence shrimp diseases while collecting data for AHPND cases and non-cases.

AppendixBQuestionsfortelephonesurvey

223

Appendix B: Questions used in brief telephone survey

(Notethattheanswersbaseonshrimpfarmingfromyear2011tocurrent)

Questions:

1. Does your farm face any disease problem? If ‘yes’ please tellme the name of

disease?

2. IfthefarmisinfectedbyAHPND,WhendoyouseetheAHPND-typelossesforthe

firsttime?Whatarethesignsoftheshrimpthatareclinicallyill(Thaispeakerscall

กุง้ป่วย/Kûng-P�wy/)includingtheageofAHPND-diseasedshrimp?

3. Howmany ponds affected with AHPND?What are themortality rates in each

affectedpond?

4. Pleasegivemethedetailsaboutgeneralmanagementpractices

- Featuresofongrowingponds

- Ponddryingduration

- Pondharrowingbeforestocking

5. Whatmeasuresdoyoutaketomitigatediseasetransmissionorrecurrence?

AppendixCNationalprovincialcentres

224

Appendix C: National provincial centres and abbreviation

Provincecode Nationalprovincialcentre Abbreviation

11 Samutprakan SPK18 Chainat CNT73 Nakhonpathom NPT72 Suphanburi SPB61 Uthaithani UTI74 Samutsakhon SKN13 Pathumthani PTE14 Phranakhonsiayutthaya AYA26 Nakhonnayok NYK75 Samutsongkhram SKM10 Bangkok BKK60 Nakhonsawan NSN12 Nonthaburi NBI23 Trat TRT24 Chachoengsao CCO21 Rayong RYG20 Chonburi CBI22 Chanthaburi CTI25 Prachinburi PRI30 Nakhonratchasima NMA77 Prachuapkhirikhan PKN70 Ratchaburi RBR71 Kanchanaburi KRI76 Phetchaburi PBI90 Songkhla SKA94 Pattani PTN80 Nakhonsithammarat NRT84 Suratthani SNI86 Chumphon CPN82 Phangnga PNA81 Krabi KBI83 Phuket PKT96 Narathiwat NWT91 Satun STN85 Ranong RNG92 Trang TRG93 Phatthalung PLG

Date post:	26-Aug-2020
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times

Epizoological Tools for Acute Hepatopancreatic Necrosis ... version of... · at country level,...

Documents