Page 1
Predic'ngpropaga'onofdenguewithhumanmobility:
1
DanajaMaldeniya
Planningmee'ng:“Forecas'ngpropaga'onofdengue/zikainSriLanka
withMobileNetworkBigData”06May2016
ThisworkwascarriedoutwiththeaidofagrantfromtheInterna'onalDevelopmentResearchCentre,CanadaandtheDepartmentforInterna'onalDevelopmentUK..
APakistancasestudy
Page 2
Theroleofhumanmobilityinspreadingdengue
• Thedenguemosquitohasalifespanof2-4weeksandarangeless1km
• Dengueisspreadbeyondthenaturalrangeofthemosquitobythemovementofinfectedhosts
• AsaresultknowledgeofhumanmobilitypaWernscanshedlightonthepaWernofdenguepropaga'onandthelevelofdiseaseincidenceinaregion
2
Page 3
Outline• Background• Introduc'ontocasestudy• Data• Methodology
-Ento-epidemiologicalmodel-Predic'onofdengueimporta'onusingtravelpaWerns-Epidemicriskmaps
• Limita'ons3
Page 4
Outline
• Background• Introduc'ontocasestudy• Data• Methodology
-Ento-epidemiologicalmodel-Predic'onofdengueimporta'onusingtravelpaWerns-Epidemicriskmaps
• Limita'ons4
Page 5
Dengue in Pakistan
• Firstconfirmedcasein1994inKarachi• Priorto2008,majorityofcaseswereinKarachi
• Sincethendenguehasbeenspreadingtootherregions.(Lahoreepidemicin2011)
• PeakseasonoccursintheFall(October-November)
5
Page 6
Popula'onsandhumanmovementsinPakistan
6
Popula'ondensityandmobilephonedataavailabilitybytehsil
TravelintensitybetweenKarachi,Lahore,andMingoraregions
Page 7
Outline
• Background• Introduc'ontocasestudy• Data• Methodology
-Ento-epidemiologicalmodel-Predic'onofdengueimporta'onusingtravelpaWerns-Epidemicriskmaps
• Limita'ons7
Page 8
Amul'-disciplinaryresearcheffort
• Collabora'onbetweenanumberofresearchgroupsaffiliatedwith,– TelenorResearch– HowardT.HChanSchoolofPublicHealth– CenterofDiseaseControl– OxfordUniversityclinicalresearchunit– DepartmentofZoology,UniversityofPeshawar– etc.
• Leadauthor,AmyWeslowski,isaninfec'ousdiseaseepidemiologistwithpreviousexperienceonu'lizingmobilenetworkdataforepidemiologicalpurposes
8
Page 9
Research models spatial spread of dengue with human mobility across Pakistan as the primary
driver
9
Page 10
Themodelpredictsthe'mingofimporta'onofdenguefrom
endemicregions(Karachi)torestofthecountry
10
Page 11
Mapofepidemicriskbyevalua'ngclimatecondi'onsand
travelpaWerns
11
Page 12
Outline
• Background• Introduc'ontocasestudy• Data• Methodology
-Ento-epidemiologicalmodel-Predic'onofdengueimporta'onusingtravelpaWerns-Epidemicriskmaps
• Limita'ons12
Page 13
Data• MobilephoneCDRdata
– 7monthsofCDRdatafromTelenorPakistanfromJunetoDecemberof2013(~40millionSIMs)
– Onaverage28millionSIMsac'vedailywith15millionSIMsgenera'ngoutgoingcallswithloca'ondata
– Coverageof352of388tehsilsofsub-districtsinPakistan
• Denguedata– De-iden'fieddailydenguecountsaggregatedtothetehsilsorsub-
districts
13
Page 14
Data...
• Popula'ondata– Uptodatepopula'ones'matesatthetehsillevelfromworldpop.co.uk
• Climatedata– Temperature&rela'vehumiditybasedontemperaturetakenfrom38weathersta'onsacrossPakistan
14
Page 15
Outline
• Background• Introduc'ontocasestudy• Data• Methodology
-Ento-epidemiologicalmodel-Predic'onofdengueimporta'onusingtravelpaWerns-Epidemicriskmaps
• Limita'ons15
Page 16
A clear distinction is made between endemic and naive
regions• Karachiandsorroundingregionsinthesouthofthe
countrywereiden'fiedasbeingendemicregions• Allotherregionsareassumedtobenaive.• Theanalysisinpar'cularfocusesonthenorthernregions
ofLahoreandMingorainparttoevaluatetheeffec'venessofthisassump'on
• Givena2011epidemic,Lahorewaslikelytohavebuiltupimmunityandalimiteddenguemosquitopopula'onby2013
• Mingorawaseffec'velyanaiveregionin2013
16
Page 17
Outline
• Background• Introduc'ontocasestudy• Data• Methodology
-Ento-epidemiologicalmodel-Predic'onofdengueimporta'onusingtravelpaWerns-Epidemicriskmaps
• Limita'ons17
Page 18
Anexis'ngento-epidemiologicalmodelwasusedtomodeldiseasedynamicsinlocalizedregions
• Themodelusestwosetsofrateordifferen'alequa'onstomodelhumananddenguemosquitopopula'ondynamics
• Humanpopula'ondynamics– Suscep'ble->Exposed->Infected->Recovered– Atthestarttheen'repopula'onisconsideredimmunologicallynaïve(i.e.suscep'ble)
– Theexistenceofmul'plestrainsofthediseaseisignored
– Re-infec'onisignored(followsfromignoringmul'plestrains)
18
Page 19
Anexis'ngento-epidemiologicalmodel…
• Denguemosquitopopula'ondynamics– Suscep'ble->Exposed->Infected– Elementsofthelife-cycleofthemosquitoisincorporated(Aqua'c->Adult)– Oviposi'onrate(egglayingrate)– carryingcapacityorsustainablemosquitopopula'onforaregion
limits
• Modelconstantssuchini'almosquitopopula'onsandhost-vectorincidenceratesarederivedfromtemperaturebasedformulae
• Otherconstantssuchasthesustainablemosquitopopula'on,bi'ngrateandrepor'ngrateswerees'matedusingsensi'vityanalysis
19
Page 20
Theuseofthemodelistwofold• ModelisfittotheendemicregionofKarachistar'ngatDay1,toes'matedailyinfectednumberofpeopleover'mebasedonreportedcases– Usedes'matelikelihoodofdengueimporta'onfromKarachitootherregionsonagivenday
• Modelisfit(inreverse)tonaïveregionstoes'matethe'meintroduc'onofdenguefromoutsidesolelywithreportedcases.– Usedtovalidatees'matesofdengueintroduc'onbasedonhumanmobilitypaWerns
20
Page 21
21
Weeklyreportedandes'mateddenguecasesinKarachi
Es'mateddateofintroduc'onofdenguetoMingoraandLahoreusingtheento-epidemiologicalmodelandreportedcases
Page 22
Theore'calmodel
22
humanpopula'ondynamics mosquitopopula'ondynamicsN-Suscep'blehumanpopula'onEh-Exposedindividualsλv→h–vectortohumanincidencerate1/γh-meanincuba'onperiod(days)Ih-Infectedindividuals1/σh-meandura'onofinfec'onRh-Recoveredindividuals
A-earlystagemosquitopopula'onV-Adultmosquitopopula'onSv-Suscep'blemosquitoesEv-Incuba'ngmosquitoes1/γv-meanincuba'onperiod(days)λh→v–humantovectorincidencerateIv-InfectedmosquitoesεA-rateofprogressiontomaturityμA
V-mortalityrateofearlystagemosquitoesμV
V-mortalityrateofadultmosquitoesK-sustainablemaximummosquitopopula'on
Page 23
Es'ma'nghost-vectorincidenceratesandtransmission
probabili'es
23
Page 24
Es'ma'ngtemperaturedrivenconstantsofthedenguemosquito
lifecycle
24
Page 25
Outline
• Background• Introduc'ontocasestudy• Data• Methodology
-Ento-epidemiologicalmodel-Predic'onofdengueimporta'onusingtravelpaWerns-Epidemicriskmaps
• Limita'ons25
Page 26
CountrywidetravelpaWernsareextractedfromCDR
• Foreachday,eachsubscriberinthedataisassignedtothehis/hermostfrequentlyobservedmobilephonetower
• Travelises'matedeachdaybetweenmobilephonetowersbyconsideringasubscriber'sloca'onw.r.ttheirloca'onthepreviousday
• Movementsareaggregatedatthetehsillevelbasedontheoriginanddes'na'onandnormalizedbythenumberofac'vesubscribersintheorigintehsil(flux)
• Movementes'matesfortheperiod1stJanuaryto1stJuneweregeneratedbyassumingthesamemeannumberofnormalizednumberoftripsandaddingnoise
26
Page 27
Predic'ngimporta'onofdenguetonaiveregionsfromendemic
regionsthroughtravel• Approximately30%ofthesubscribersinKarachitraveledoutsidedaily(β)
• Themodelprovidesadailyes'mateofnumberofinfectedhostsinKarachi(mt)
• Naïvees'mateofinfectedtravelers:mtβ
• β isvariedbetween10%,20%and30%insimula'ons,toaccountforuncertaintyandlikelihoodofoveres'ma'ngtravelwithCDR27
Page 28
Predic'ngimporta'onofdenguetonaiveregions…
• Es'matedinfectedtravelersareappor'onedtodes'na'ontehsilsbasedonthepercentageofes'matedtraveltothosedes'na'ons
• Thestateofinfec'onofan‘infected’hostisaspectrumduetoviraldynamics.ThisaffectstransmissionlikelihoodwhenbiWen– Viraldynamicsarenotdealtwithdirectly– Insteadsimula'onsaredoneusingprobabili'estransmissionsampledfromaseriesofbinomialdistribu'onswithdifferentfixedprobabili'es(0.01to0.9)
28
Page 29
Predic'ngimporta'onofdenguetonaiveregions…
• 200simula'ondoneforeachcombina'onofthefrac'onofKarachipopula'ontravelingoutsideandthefixedtransmissionprobability
• Themostlikelydateoffirstcaseofimporteddengueatthedes'na'onsises'matedfromresultsofthesimula'ons
• Es'matespredominantlyaffectedbythenumberofpeopletravelingoutfromKarachiandisnotverysensi'vetotransmissionprobability 29
Page 30
Es'matesdemonstrateaccuracyofmodelbutalsotheeffectofpriorepidemicsandimmunity
30
Page 31
Outline
• Background• Introduc'ontocasestudy• Data• Methodology
-Ento-epidemiologicalmodel-Predic'onofdengueimporta'onusingtravelpaWerns-Epidemicriskmaps
• Limita'ons31
Page 32
AmapofdengueepidemicriskwasdevelopedforPakistanby
combiningclima'csuitabilityandmobilitypaWerns
𝑟𝑖𝑠𝑘↓𝑒𝑝𝑖𝑑𝑒𝑚𝑖𝑐(𝑋) = ∑𝑡=1↑𝑁▒𝑍↓𝑋 ( 𝑇↓𝑡 ) 𝑌↓𝑋,𝑡 Where,𝑍↓𝑋 −𝑉𝑒𝑐𝑡𝑜𝑟 𝑠𝑢𝑖𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑎𝑡 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑋
𝑇↓𝑡 −𝑡𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒 𝑎𝑡 𝑋 𝑎𝑡 𝑡𝑖𝑚𝑒 𝑡𝑌↓𝑋,𝑡 −𝑡ℎ𝑒 𝑛𝑢𝑚𝑏𝑒𝑟 𝑖𝑛𝑓𝑒𝑐𝑡𝑒𝑑 𝑡𝑟𝑎𝑣𝑒𝑙𝑒𝑟𝑠 𝑎𝑟𝑟𝑖𝑣𝑖𝑛𝑔 𝑎𝑡 𝑋 𝑎𝑡 𝑡𝑖𝑚𝑒 𝑡
32
Page 33
Environmentsuitabilityfordenguemosquitoesisdrivenby
temperature 𝑍↓𝑋 (𝑇)= exp(−𝜇↓𝑉↑𝑉 (𝑇)𝛾↓𝑉↑𝑉 (𝑇))/𝜇↓𝑉↑𝑉 (𝑇)↑2 Where,𝑇 −𝑡𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒𝑍↓𝑋 (𝑇) −𝑆𝑢𝑖𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑎𝑡 𝑋
𝜇↓𝑉↑𝑉 − 𝑖𝑛𝑠𝑡𝑎𝑛𝑡𝑎𝑛𝑒𝑜𝑢𝑠 𝑑𝑒𝑎𝑡ℎ 𝑟𝑎𝑡𝑒 𝑜𝑓 𝑎𝑑𝑢𝑙𝑡 𝑓𝑒𝑚𝑎𝑙𝑒 𝑚𝑜𝑠𝑞𝑢𝑖𝑡𝑜𝑒𝑠𝛾↓𝑉↑𝑉 −𝑖𝑛𝑐𝑢𝑏𝑎𝑡𝑖𝑜𝑛 𝑝𝑒𝑟𝑖𝑜𝑑 𝑜𝑓 𝑑𝑒𝑛𝑔𝑢𝑒• Suitabilityisarela'vequan'typropor'onaltothevectoralcapacity.• Itisdifficulttoes'matevectoralcapacityduetolackofdataonfactorssuchas
humanbloodfeedingrateetc.
33
Page 34
Epidemicriskisdominatedbyimporta'onthroughtravel
34
Page 35
Outline
• Background• Introduc'ontocasestudy• Data• Methodology
-Ento-epidemiologicalmodel-Es'ma'nghumanmobility-Predic'onofdengueimporta'onusingtravelpaWerns-Epidemicriskmaps
• Limita'ons35
Page 36
Researchhighlightspoten'alvalueofhumanmobility
es'mates,butisnotwithoutissues
• Thereisadisconnectbetweentheepidemiologicalmodelandthepredic'onofdengueintroduc'onthroughtravel
• Ignoresexistenceofdiseaseserotypes(strains)• Ignoreshistoryofdengueintheregionand
immunologicalcovariates• Validityandalignmentoftemperaturebasedexpressions
forentomologicalproper'esfordengueinalocalcontext• Mul'pleparameterses'matedthroughbruteforcesearch
(travelerpercentage,mosquitocarryingcapacity,bi'ngrateetc.)
36
Page 37
Issues…
• Rela'velylowspa'alsamplingofweatherdata(38weathersta'ons,Pakistanisabout14'mesSriLankainarea)
• Differencesbetweenairandwatertemperatures(alsomicroclimates?)
37
Page 38
Placementofweatherdatausedintheresearch
38
Page 39
Keydataandparametersthatmaydeterminethesuccessofsimilar
workinSriLanka• Unitofspa'alanalysis-(MoHregion,basesta'oncoverage
region,DSD?)• Accuratespa'o-temporaldataonreporteddenguecases
(daily?)• Repor'ngrate(~100%inSL?)• Regionaldenguepopula'ons• Regionalserotypeprevalence• Regionalimmunologicaldata/historicalcaseserotypes• Entomologicalparametersacceptablefortheregion• Spa'otemporalinforma'ononpreven'vemeasures?
39