University of Southern California Dissertations and Theses

Essays on the economics of infectious diseases

(USC Thesis Other)

Essays on the economics of infectious diseases

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYSONTHEECONOMICSOFINFECTIOUSDISEASES by EmmanuelFulgenceDrabo ADissertationPresentedtothe FACULTYOFTHEUSCGRADUATESCHOOL UNIVERSITYOFSOUTHERNCALIFORNIA InPartialFulllmentofthe RequirementsfortheDegree DOCTOROFPHILOSOPHY (HEALTHECONOMICS) August2016 Copyright 2016 EmmanuelFulgenceDrabo Dedication Thisdissertationisdedicatedtoallthosewhoseliveshavebeenandcontinuetobeafected by infectious diseases of all kinds, especially the so-called orphan diseases. I hope that in a foreseeablefuturesoundpublichealthpoliciesandinnovationofnewandefectivecuresand vaccines will not only alleviate their suferings, but also prevent others from continuing to suferfromtheseillnesses. ii Acknowledgments An axe does not cut down a tree by itself··· —BurkinaFasoproverb. Thisdissertationwouldnothavebeenpossiblewithouttheadvice,encouragementsandcon- stantsupportsofmanymentors,colleagues,friendsandfamilymembers.Iamrstandfore- mostindebtedtomyadvisor, Dr. NeerajSood, forhisintellectualgenerosity, patienceand permanentsupportthroughoutthisjourney.Theroleofhisguidanceandtimelyfeedbackin thedesign,research,andwritingofthisdissertationcouldnotbeoverstated. I am also extremely grateful to my dissertation committee members, Dr. Joel W. Hay, Dr. JasonN.Doctor,Dr.RafaeleVardavas,andDr.DariusLakdawalla,forforbeingasupportive team: Theirinsightfulcommentsandcritiqueoftheworkpresentedinthismanuscript, as wellastheirgenerosityinsharingwithmeresearchmaterialswereinvaluable. IamindebtedtomyDepartmentaswellastheLeonardSchaeferCenterforHealthPolicy for their reliable nancial support throughout the 4 years of my scholarly training at USC. IamparticularlygratefultoBaxterBiosciencefortheirnancialsupportthroughtheBaxter DoctoralFellowship. Ithankmyfellowclassmates,friends,andcolleagues,JustinMcGinnis,CynthiaGong,Laura Henkhaus,DarshanMehta,andthemanyotherswhosenamesarenotmentionedhere,for iii beingtrustedsoundingboardswithresearchideas,andfortheircompanionshipthoughthis experience. Lastbutnotleast, Iexpressmysinceregratitudetomyfamilyaswellasmypartner, Adina Cappell,fortheirsustainedsupport.Aspecialtreattomydog,Kalev,forstayingupwithme duringthosenumerouslatenights,asIstruggledtopreparethechaptersforthismanuscript. Icouldnotendtheseacknowledgmentswithoutthankingthevolunteerswhoparticipatedin thefocusgroupstudies,aswellastheMTurkworkerswhoparticipatedinsurveystudy. iv Contents Dedication ii Acknowledgments iii ListofTables ix ListofFigures xiii Abstract xvi 1 Introduction 1 1.1 Backgroundoninfectiousdiseases . . . . . . . . . . . . . . . . . . . . . 1 1.2 Burdenofinfectiousdiseasestoday . . . . . . . . . . . . . . . . . . . . . 4 1.3 Factorsinuencingtheemergenceandre-emergenceofEIDs . . . . . . . . 8 1.4 Objectivesofthisdissertation . . . . . . . . . . . . . . . . . . . . . . . . 10 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 2 NudginginfectiousdiseasevaccineuptakeintheUnitedStateswitha“no-fault”insur- anceagainsttheriskofvaccine-relatedsideefects:Adiscretechoiceexperiment 24 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.2 Theoreticalbasisforsmallincentivesandsmall-stakeinsuranceschemes . . 30 2.3 Methodsandstudydesign . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.3.1 Ethicsstatement . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.3.2 Studyparticipantsandrecruitment . . . . . . . . . . . . . . . . . 34 2.3.3 Experimentalprocedure . . . . . . . . . . . . . . . . . . . . . . 36 2.3.4 Discretechoiceexperiment(DCE) . . . . . . . . . . . . . . . . . 37 2.3.5 DevelopmentoftheDCEquestionnaire . . . . . . . . . . . . . . 38 2.3.6 DCEdesign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 2.3.7 Pilottest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 2.3.8 Samplesizecalculation . . . . . . . . . . . . . . . . . . . . . . . 44 2.4 Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 2.4.1 Willingnesstopayandwillingnesstoaccept . . . . . . . . . . . . 48 v 2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 2.5.1 Surveyparticipation . . . . . . . . . . . . . . . . . . . . . . . . 48 2.5.2 Characteristicsoftherespondents . . . . . . . . . . . . . . . . . 49 2.5.3 DCEresults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 2.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 3 Test-and-treatinLosAngeles: Amathematicalmodeloftheefectsoftest-and-treat forthepopulationofmenwhohavesexwithmeninLosAngelesCounty 73 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 3.2.1 Studydesign . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 3.2.2 Modelstructure . . . . . . . . . . . . . . . . . . . . . . . . . . 77 3.2.3 Keyinputparametersandranges . . . . . . . . . . . . . . . . . . 78 3.2.4 Surveillancedata . . . . . . . . . . . . . . . . . . . . . . . . . . 79 3.2.5 Modelcalibration . . . . . . . . . . . . . . . . . . . . . . . . . . 79 3.2.6 Test-and-treatmodel . . . . . . . . . . . . . . . . . . . . . . . . 80 3.2.7 Sensitivityanalysis . . . . . . . . . . . . . . . . . . . . . . . . . 83 3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 3.3.1 Sensitivityanalysis . . . . . . . . . . . . . . . . . . . . . . . . . 87 3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4 Acost-efectivenessanalysisofpre-exposureprophylaxisforthepreventionofHIV amongLosAngelesCountymenwhohavesexwithmen 95 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 4.2.1 Epidemiologicalmodelstructure . . . . . . . . . . . . . . . . . . 99 4.2.2 Economicmodelstructure . . . . . . . . . . . . . . . . . . . . . 102 4.3 Inputdata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 4.4.1 Ecientstrategies . . . . . . . . . . . . . . . . . . . . . . . . . . 111 4.4.2 Epidemiologicaloutcomes . . . . . . . . . . . . . . . . . . . . . 116 4.4.3 Sensitivityanalyses . . . . . . . . . . . . . . . . . . . . . . . . . 116 4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 5 Concludingremarks 140 5.1 Synthesisofthekeyndings . . . . . . . . . . . . . . . . . . . . . . . . . 140 5.2 Policyimplicationsandthewayforward . . . . . . . . . . . . . . . . . . 142 vi A Backgroundondecisiontheoryanddiscretechoiceexperiments, anddescriptionof theexperimentalprocedure 144 A.1 BriefoverviewofprospecttheoryandtheKőszegi-Rabinutilitytheory . . . 144 A.2 Methodsforelicitingpreferences . . . . . . . . . . . . . . . . . . . . . . 146 A.3 TheoreticalconsiderationsforconductingaDCE . . . . . . . . . . . . . . 147 A.3.1 ImportantstepsinvolvedinaDCE . . . . . . . . . . . . . . . . . 151 A.3.2 DesignconsiderationsinDCE . . . . . . . . . . . . . . . . . . . 151 A.4 Questionnairedevelopment . . . . . . . . . . . . . . . . . . . . . . . . . 156 A.4.1 Diseaseattributes . . . . . . . . . . . . . . . . . . . . . . . . . . 156 A.4.2 Vaccineattributesandattributelevels . . . . . . . . . . . . . . . . 159 A.4.3 Vaccineinjurycompensationattributelevels . . . . . . . . . . . . 161 A.5 Focusgroupinterviewguide . . . . . . . . . . . . . . . . . . . . . . . . 168 A.6 DCEsurveyinstrument . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 A.6.1 DCEdesign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 A.7 AdditionalDCEresults . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 B HIVModelingApproach:MathematicalDetails 207 B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 B.2 HIVepidemicsimulationmodelstructure . . . . . . . . . . . . . . . . . 208 B.2.1 Diseaseprogression . . . . . . . . . . . . . . . . . . . . . . . . . 210 B.2.2 DiseaseTransmission . . . . . . . . . . . . . . . . . . . . . . . . 212 B.2.3 Modeloutput. . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 B.3 Parameterranges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214 B.3.1 Perpartnershiptransmissibilities,β . . . . . . . . . . . . . . . . . 215 B.3.2 Populationparameters . . . . . . . . . . . . . . . . . . . . . . . 221 B.3.3 Diseaseprogressionparameters . . . . . . . . . . . . . . . . . . . 222 B.4 Samplingstyle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224 B.5 Initialconditionsandbaselinedata . . . . . . . . . . . . . . . . . . . . . 224 B.5.1 MSMpopulation . . . . . . . . . . . . . . . . . . . . . . . . . . 225 B.5.2 Inow-14yearolds . . . . . . . . . . . . . . . . . . . . . . . . . 226 B.5.3 AIDSprevalence . . . . . . . . . . . . . . . . . . . . . . . . . . 226 B.5.4 Non-AIDSHIVprevalence . . . . . . . . . . . . . . . . . . . . 230 B.6 Modelcalibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 B.6.1 Calibrationsensitivityanalysis . . . . . . . . . . . . . . . . . . . 238 B.7 Test-and-treatmodel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240 B.7.1 Test-and-treatmodelparametervalues . . . . . . . . . . . . . . . 242 B.7.2 Baselinetest-and-treatscenarioparameters . . . . . . . . . . . . . 243 B.8 Sensitivityanalysisoftheimpactoftest-and-treat . . . . . . . . . . . . . . 244 B.8.1 One-waysensitivityanalysis . . . . . . . . . . . . . . . . . . . . 244 B.8.2 Multi-waysensitivityanalysis . . . . . . . . . . . . . . . . . . . . 249 B.8.3 Sensitivitytovariationsintheinfectiousnessassumption . . . . . . 251 vii B.8.4 Calibrationvs.literaturebasedparameterestimates . . . . . . . . 252 B.9 Resultswithoutearlytreatment . . . . . . . . . . . . . . . . . . . . . . . 255 B.10 Robustnessanalysisunderresistance . . . . . . . . . . . . . . . . . . . . 257 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259 C HIVEpidemicModelingandCosts-efectivenessEstimationApproaches 265 C.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 C.2 HIVepidemicsimulationmodelstructure . . . . . . . . . . . . . . . . . 265 C.2.1 Systemofordinarydiferentialequations. . . . . . . . . . . . . . 271 C.2.2 Populationdata . . . . . . . . . . . . . . . . . . . . . . . . . . . 274 C.2.3 Age-weightedaveragelifeexpectancyfor15-to65-year-oldMSM. . 276 C.2.4 Modelinputparametervalues . . . . . . . . . . . . . . . . . . . 277 C.2.5 Modelcalibration . . . . . . . . . . . . . . . . . . . . . . . . . . 287 C.2.6 ProjectedHIVincidence . . . . . . . . . . . . . . . . . . . . . . 287 C.3 Economicmodelstructure . . . . . . . . . . . . . . . . . . . . . . . . . . 289 C.3.1 Calculationofcosts . . . . . . . . . . . . . . . . . . . . . . . . . 289 C.3.2 CalculationofQALYs . . . . . . . . . . . . . . . . . . . . . . . 294 C.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298 C.4.1 Thebasecaseanalysis . . . . . . . . . . . . . . . . . . . . . . . . 310 C.4.2 Resultsfromthefullanalysis . . . . . . . . . . . . . . . . . . . . 310 C.5 Sensitivityanalyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316 C.5.1 One-waysensitivityanalyses . . . . . . . . . . . . . . . . . . . . 316 C.5.2 Bootstrappingprobabilisticsensitivityanalysis . . . . . . . . . . . 328 C.5.3 Averagesocietalwillingnesstopaythresholds. . . . . . . . . . . . 331 C.5.4 Sensitivityonpricereductionfollowinggenericentry . . . . . . . 333 C.6 Systematicreviewoftheliterature . . . . . . . . . . . . . . . . . . . . . . 337 C.6.1 Inclusionandexclusioncriteria . . . . . . . . . . . . . . . . . . . 337 C.6.2 Searchstrategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 339 C.6.3 Publicationscreening . . . . . . . . . . . . . . . . . . . . . . . . 341 C.6.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343 viii ListofTables 2.1 DCEattributesandattributelevels. . . . . . . . . . . . . . . . . . . . . . 40 2.2 Characteristicsoftherespondents. . . . . . . . . . . . . . . . . . . . . . 50 2.3 Contributionsofthevaccineanddiseaseattributes. . . . . . . . . . . . . 57 2.4 Willingnesstopayestimates. . . . . . . . . . . . . . . . . . . . . . . . . 60 3.1 First-yearefectsofthetest-and-treatpolicy . . . . . . . . . . . . . . . . . 82 3.2 Resultswithoutearlytreatment. . . . . . . . . . . . . . . . . . . . . . . 86 4.1 Basecasemodelassumptions. . . . . . . . . . . . . . . . . . . . . . . . . 105 4.2 Summaryofkeymodelinputparameters. . . . . . . . . . . . . . . . . . 106 4.3 BenetsandcostsofthemostecientTest-and-TreatandPrEPstrategies. . 113 A.1 Diseasecharacteristics. . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 A.2 Alternative-invariantdiseaseattributesandattributelevels. . . . . . . . . . 159 A.3 Alternative-varyingvaccineattributesandattributelevels. . . . . . . . . . 160 A.4 Vaccineinjurycompensationamountsbyinjurytype. . . . . . . . . . . . . 165 A.5 Attributeandlevelsoftheattributeoftheinsuranceandsubsidyprograms. 167 A.6 Focusgroupexhibit1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 A.7 Focusgroupexhibit2–denitionsofthediseaseattributes . . . . . . . . 171 A.8 Focusgroupexhibit3–denitionsofthevaccineattributes . . . . . . . . . 172 ix A.9 Focusgroupexhibit4–denitionsofthecompensationprogramattributes 173 A.10 DCEdesign. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 A.11 Contributionsofthevaccineanddiseaseattributesbytreatmentstatus. . . 197 A.12 Contributionsofthevaccineanddiseaseattributesbygender. . . . . . . . 199 B.1 Rangesoftheαparametervaluesbydiseasestage.. . . . . . . . . . . . . . 216 B.2 Rangesofthescalingfactorparameter(η X ),bydiseasestage. . . . . . . . . 218 B.3 Unprotectedsexactsperpartnerperyear,ζ f X ,bystage . . . . . . . . . . . 220 B.4 Derivedperpartnershiptransmissibilities. . . . . . . . . . . . . . . . . . . 220 B.5 MDRtransmissibilitymultiplicativefactor. . . . . . . . . . . . . . . . . . 221 B.6 Parametersspecifyingcontactrange. . . . . . . . . . . . . . . . . . . . . . 221 B.7 Parametersspecifyingthepopulationofthemodel. . . . . . . . . . . . . . 222 B.8 Transitionrates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223 B.9 Initialpopulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226 B.10 TotalMSMAIDScasesinLAbyyear. . . . . . . . . . . . . . . . . . . . 228 B.11 TotalawareMSMHIVcasesinLAbyyear. . . . . . . . . . . . . . . . . . 232 B.12 Calibratedparametervaluesandnarrowedparameterranges. . . . . . . . . 237 B.13 Parametersforthebaselinetest-and-treatscenario. . . . . . . . . . . . . . 244 B.14 Percentreductioninnewinfectionsby2023. . . . . . . . . . . . . . . . . 246 B.15 PercentoftheHIV/AIDSpopulationwithMDRin2023. . . . . . . . . . 247 B.16 Comparisonofmodelresultsusingliterature-orcalibration-basedparameter values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 B.17 Resultswithoutearlytreatment. . . . . . . . . . . . . . . . . . . . . . . 256 C.1 Glossary–Denitionsofthemodelinputparameters. . . . . . . . . . . . 267 C.2 InitialpopulationofMSMinLosAngelescountyinyear2000 . . . . . . . 274 C.3 CountofLACmalepopulationbyyearandagegroup . . . . . . . . . . . 275 x C.4 AveragelifeexpectanciesforhealthyandHIV-positivemalepopulationsin theUnitedStates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276 C.5 ParametersspecifyinginowofnewMSMinthemodel. . . . . . . . . . . 278 C.6 Relationshipbetweenadherence toPrEP,PrEPecacy,andlevelsofprotec- tionofPrEP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279 C.7 AdherencetoPrEP:Estimatesbasedontenofovirconcentrationinbloodof nonseroconverters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280 C.8 Perpartnershiptransmissibilities . . . . . . . . . . . . . . . . . . . . . . 280 C.9 MDRtransmissibilitymultiplicativefactor . . . . . . . . . . . . . . . . . 280 C.10 ParameterSpecifyingContactRange . . . . . . . . . . . . . . . . . . . . 282 C.11 ReductionsinsexualinfectivityandriskybehaviorsowingtoART,PrEPand counseling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 C.12 Testsensitivityandspecicity. . . . . . . . . . . . . . . . . . . . . . . . 283 C.13 RateofHIVscreeninginthepopulation . . . . . . . . . . . . . . . . . . 284 C.14 ProbabilityofPrEPinitiation . . . . . . . . . . . . . . . . . . . . . . . . 286 C.15 DurationofsymptomaticHIVtreatedwithART. . . . . . . . . . . . . . 286 C.16 Costparameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291 C.17 USmalepopulationage-andsex-stratiedforeachcontinuoushealth-related QOLsummaryscore(EQ-5DUS)[32]. . . . . . . . . . . . . . . . . . . 294 C.18 Disease-stateQOLutilityweightsandotherefectivenessparameters. . . . 296 C.19 Completelistofallstrategiessimulatedintheanalysis. . . . . . . . . . . . 299 C.20 Basecaseanalysis–benetsandcostsofthetesting,test-and-treat,andPrEP strategies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311 C.21 CumulativeHIVinfectionsavertedduring20yearsintheLACMSMpop- ulationunderthestatusquoandrationaltest-and-treatandPrEPstrategies ontheecientfrontier. . . . . . . . . . . . . . . . . . . . . . . . . . . . 312 xi C.22 Mostcost-efectivetest-and-treatandPrEPstrategies(expanded). . . . . . 313 C.23 Inclusionandexclusioncriteriafortheliteraturereview. . . . . . . . . . . 338 C.24 Searchcodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340 C.25 Summarystatisticsofthedatabasesearches. . . . . . . . . . . . . . . . . . 342 xii ListofFigures 2.1 Expectedvaluesofthecompensationamountsunderthesubsidyandinsur- anceschemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 2.2 ApproachtothedevelopmentoftheDCEquestionnaire. . . . . . . . . . . 39 2.3 Unconditionalandconditionalmarginalefectsofinsuranceonvaccineuptake. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.1 Humanimmunodeciencyvirus(HIV)transmissionmodel. . . . . . . . . 78 3.2 Calibrationresults. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 3.3 Baselinetest-and-treatscenariocomparedtothestatusquo. . . . . . . . . . 84 4.1 SchematicsoftheHIVSimulationModelwithPrEP. . . . . . . . . . . . . 101 4.2 Ecientfrontierforresourceallocation. . . . . . . . . . . . . . . . . . . 112 A.1 Approachestopreferenceelicitation. . . . . . . . . . . . . . . . . . . . . 147 A.2 ExampleofaDCEchoiceproleusedinthesurvey. . . . . . . . . . . . . 192 A.3 Conditionalmarginalefectsofinsuranceonvaccineuptakebydiseasesever- ityandvaccinecharacteristics. . . . . . . . . . . . . . . . . . . . . . . . . 193 A.4 Conditionalmarginalefectsofinsuranceonvaccineuptakebyvaccineout- of-pocketcostandindividualcharacteristics. . . . . . . . . . . . . . . . . 194 xiii A.5 Conditionalmarginalefectsofinsuranceonvaccineuptakebyvaccinee- cacyandindividualcharacteristics. . . . . . . . . . . . . . . . . . . . . . 195 A.6 Conditionalmarginalefectsofinsuranceonvaccineuptakebyriskofvaccine sideefectsandindividualcharacteristics. . . . . . . . . . . . . . . . . . . 196 B.1 DivisionofMSMAIDScasesintocompartments. . . . . . . . . . . . . . 229 B.2 SSEsandLogSSEsagainsttheSSEranks . . . . . . . . . . . . . . . . . . 236 B.3 Sensitivitymapfortheparametercalibration. . . . . . . . . . . . . . . . . 239 B.4 Schematicsofthetest-and-treatscenariomodel. . . . . . . . . . . . . . . 241 B.5 Baselinetest-and-treatscenariocomparedtothestatusquo. . . . . . . . . . 248 B.6 Multi-waysensitivityanalysis–distributionsofthe%reductioninnewinfec- tions,deathsandnewAidscases. . . . . . . . . . . . . . . . . . . . . . . 250 B.7 Robustnessanalysisaroundtheinfectiousnessparameter. . . . . . . . . . . 252 B.8 Robustnessanalysis–baselinetest-and-treatscenariocomparedtothestatus quo. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258 C.1 Modelcalibrationresult:simulatedvs.surveillancedata. . . . . . . . . . . 288 C.2 ProjectedcumulativeHIVincidenceforthestatusquo,andbaselineTesting, Test-and-Treat,andPrEPscenarios. . . . . . . . . . . . . . . . . . . . . . 289 C.3 SensitivityoftheICERstovariationsintheepidemicparameters–TT(SQ +ImmediateEarlyART)strategy. . . . . . . . . . . . . . . . . . . . . . 317 C.4 SensitivityoftheICERstovariationsintheepidemicparameters–Enhanced TT(TT+Test6mo)strategy. . . . . . . . . . . . . . . . . . . . . . . . 318 C.5 SensitivityoftheICERstovariationsintheepidemicparameters–PrEP(TT +Test6mo+PrEP4y)strategy. . . . . . . . . . . . . . . . . . . . . . . 319 C.6 SensitivityoftheICERstovariationsintheepidemicparameters–Enhanced PrEP(PrEP+Test3mo+ImmediatePrEP)strategy. . . . . . . . . . . . . 320 xiv C.7 Sensitivity of the ICERs to variations in the cost parameters – TT (SQ + ImmediateEarlyART)strategy. . . . . . . . . . . . . . . . . . . . . . . 321 C.8 SensitivityoftheICERstovariationsinthecostparameters–EnhancedTT (TT+Test6mo)strategy. . . . . . . . . . . . . . . . . . . . . . . . . . 322 C.9 SensitivityoftheICERstovariationsinthecostparameters–PrEP(TT+ Test6mo+PrEP4y)strategy. . . . . . . . . . . . . . . . . . . . . . . . 323 C.10 Sensitivity of the ICERs to variations in the cost parameters – Enhanced PrEP(PrEP+Test3mo+ImmediatePrEP)strategy. . . . . . . . . . . . . 324 C.11 SensitivityoftheICERstovariationsintheefectivenessparameters–TT (SQ+ImmediateEarlyART)strategy. . . . . . . . . . . . . . . . . . . . 325 C.12 SensitivityoftheICERstovariationsintheefectivenessparameters–Enhanced TT(TT+Test6mo)strategy. . . . . . . . . . . . . . . . . . . . . . . . 326 C.13 SensitivityoftheICERstovariationsintheefectivenessparameters–PrEP (TT+Test6mo+PrEP4y)strategy. . . . . . . . . . . . . . . . . . . . 327 C.14 SensitivityoftheICERstovariationsintheefectivenessparameters–Enhanced PrEP(PrEP+Test3mo+ImmediatePrEP)strategy. . . . . . . . . . . . . 328 C.15 Sensitivityanalysis:Shareofcost-efectiveiterationsasafunctionofthesoci- etalaveragewillingnesstopay. . . . . . . . . . . . . . . . . . . . . . . . 332 C.16 Sensitivityanalysisrelativetopriorecientstrategy: Shareofcost-efective iterationsasafunctionofthesocietalaveragewillingnesstopay. . . . . . . 333 C.17 RobustnessofICERstochangesinthemodelinputparametervaluesatvar- iousgenericpricediscount(ICERsrelativetoStatusQuo). . . . . . . . . . 335 C.18 RobustnessofICERstochangesinthemodelinputparametervaluesatvar- iousgenericpricediscount(relativetopriorecientstrategy). . . . . . . . 336 C.19 Flowdiagramoftheliteraturescreening. . . . . . . . . . . . . . . . . . . 342 xv Abstract Infectious diseases are increasingly posing a threat to public health and economies in both developinganddevelopedcountries.TheUSisnotshieldedfromthesethreats,asevidenced by recent epidemic outbreaks of SARS, swine u, Ebola and Zika virus. Vaccination is a proven cost-efective strategy to mitigate these threats. However, both children and adult vaccinationratesintheUSremainbelowtheHealthyPeople2020targetlevels. Identifying strategies to increase vaccine uptake is therefore an important public health goal. In order tohelpshedsomelightonthisissue,weuseinsightsfrombehavioraleconomicstodevelop andconductedanexperimentaimedatdeterminingwhichofa“no-fault”insurancescheme against the risk of vaccine side efects vs a monetarily equivalent direct subsidy scheme (i.e. expectedvalueofinsurance)ismostefectiveatboostingvaccineuptake. Wepresentresults showingthata“no-fault”insuranceschemecanmodestlyincreasevaccineuptake,compared toasubsidyschemeofanequivalentmonetaryvalue. Thisresultleadsustoconcludethatit mightbepossibletoleverageonthecurrentvaccineinjurycompensationprogramintheUS toboostvaccineuptake. Next,wealsoexploittechniquesfromeconomicepidemiologytoinvestigatetheimpactsof diferentHIV/AIDSpreventstrategiesontheHIV/AIDSepidemicamongmenwhohavesex withmen(MSM)inLosAngelesCounty.First,wedevelopacompartmentalepidemiological modelofHIV/AIDStransmissiontoassesstheefectofthe“test-and-treat”policyonreducing HIVtransmission. Wealsoassesstheimpactofthispolicyontheprevalenceofmulti-drug xvi resistance. Second,weexpandthepolicysettoincludepre-exposureprophylaxis(PrEP)and develop an economic model to assess the trade-ofs between the costs and benets of these policies. This approach allows us to identify the ecient frontier for optimal allocation of resources. Collectively,thesestudiesoferpowerfultoolsforprioritizingandoptimallyallocatingresources inordertoefectivelycombatinfectiousdiseasesandaverttheirburden.Weconcludeouranal- ysisbyoutliningsomeareasforfutureresearch. xvii Chapter1 Introduction EmmanuelF.Drabo 1.1 Backgroundoninfectiousdiseases At the beginning of the 20 th century, infectious diseases were widely prevalent, and consti- tutedtheleadingcauseofmortalityinbothdevelopedanddevelopingnations.IntheUnited States (US), smallpox aicted 21,064 people in 1900 alone, killing 894 [13, 29]. The 1920 measlesepidemicoutbreakinfected469,924Americans,killing7,575. Adiphtheriaepidemic outbreak during the same year afected 147,991 people, claiming the lives of 13,170 patients. Twoyearslater, apertussisepidemicafected107,473people, killing5,099[13,73,74]. This high prevalence and death toll from infectious diseases in the 1900s was largely imputable tothelackofefectivetreatmentsandpreventivemeasuresagainstinfectiousdiseases. While the rst vaccine against smallpox existed long before the 1900 outbreak 1 , its adoption rate 1 EdwardJennerrstintroducedthesmallpoxvaccinein1796after20yearofexperimentationthatbegan in1776.However,theseissomecontroversywhetherJennerreceivedtheideaaboutthevaccinefromBenjamin Jetsy[34]. 1 remained signicantly below the critical vaccination threshold 2 , to prevent epidemic out- breaks [13, 39]. Vaccines were also available against other infectious diseases such as rabies, typhoid,choleraandplague,buttheiradoptionremainedverylow[13]. However,duringthecourseofthe 20 th century,theUSandtherestoftheworldmadesigni- cantstridesininfectiousdiseaseprevention,whichresultedinsubstantialdeclinesininfectious disease-relatedmortalityandmorbidity.Thesesignicantachievementsbecamelargelypossi- blethroughtheintroductionofmorevaccinesagainstinfectiousdiseases,aswellastheimple- mentationofefectivenationalvaccinerecommendationsandvaccinationprogramsthatsub- stantiallyincreasedtheadoptionofthesevaccinesbythepopulation.Between1798and1998, vaccineshavebeenintroducedorlicensedagainst21infectiousdiseasesintheUS[13,61],and becamewidelyrecommendedbyhealthocials[12,13]. Thewidespreadimplementationof nationalvaccinerecommendationspriorto1980translatedintoover92%declineininfections andover99%reductioninmortalityfromdiphtheria,mumps,pertussisandtetanus. These measuresalsohelpedtoeradicatetheendemictransmissionofpoliomyelitiswild-typeviruses andthemeaslesandrubellavirusesintheUS,aswellasthesmallpoxvirusworldwide[13,64]. Duringthatperiodvaccinationalsocontributedtoover80%reductioninmorbidityandmor- talityfromnumerousvaccine-preventableinfectiousdiseases(e.g.hepatitisA,acutehepatitis B,Haemophilusinfluenzaetypeb(Hib)invasivediseaseinchildren5yearsoryounger,vari- cella, measles), and helped to cut the morbidity and mortality from invasive Streptococcus pneumoniaeby34%and25%,respectively[13,64]. 2 Thecriticalvaccinationthresholdistheminimumvaccinationcoverageratethatneedstobeachievedin ordertopreventonecaseofinfectionfromspreadinginthepopulation. Whenthishappens, thepopulation issaidtohaveherd immunity. Thecriticalvaccinationthresholddeterminedatthepointatwhichtheefective reproductionnumber,R E ,fallsbelowone[13,50].ArelatednotiontoR E isthebasicreproductionnumber,R 0 , whichrepresentstheaveragenumberofnewinfectiousindividualsproduced,inatotallysusceptiblepopulation, bytheintroductionofaninfectiveindividual. 2 Despitethesubstantialadvancesduringthe 20 th century,infectiousdiseasescontinuetorank among the leading causes of morbidity and mortality worldwide, fueled by (i) the emer- gence of newly identied infectious diseases, (ii) the re-emergence of previously recognized diseasesinnew(traditionallyunafected)populationsorgeographicareas,oringreaterinten- sity(increasedprevalenceafterasignicantdeclineinincidence)ornewerforms(e.g.resistant orvirulentstrains),and(iii)thepersistenceofintractableinfectiousdiseases [30,44]. These diseasesarecollectivelyreferredtoasemerging andre-remerging infectionsdiseases(EIDs). While the sudden and unavoidable nature of the occurrence of EIDs has long been recog- nized, several outbreaks over the last two decades (e.g. HIV/AIDS, severe acute respiratory syndromecoronavirus[SARS-Cov],Middle-Eastrespiratorysyndromecoronavirus[MERS- Cov],methicillin-resistantStaphylococcusaureus[MRSA],H1N1,Zikavirus[ZIKV],Ebola virus)havecometoglobalhealthocialasasurpriseandexposedtheinadequacyofexisting preventiveandresponsemeasurestotheriskofsignicantpandemicsinbothdevelopedand developingcountries[52].Asaresult,intheir2014reportintheaftermathoftheebolacrisis, TrustforAmerica’sHealthandtheRobertWoodJohnsonFoundationnotedthat“theEbola outbreak has been a major wake-up call to the United States – highlighting serious gaps in thecountry’sabilitytomanageseverediseaseoutbreaksandcontaintheirspread”[63]. The pandemicofZIKVinfection 3 inSouthAmerica,CentralAmericaandtheCaribbeanisthreat- eningtoreachcontinentalUS(atthemomentofthiswriting,theinfectionhasclaimedalife inPuertoRico) [27,33],andthereisincreasingconcernthattheUSisunpreparedtocontain apotentialoutbreak[33]. 3 TheUSCentersforDiseaseControlandPrevention(CDC)andtheWorldHealthOrganization(WHO) considerZIKVtocauseneurologicaldisorders(includingmicrocephaly)innewborns[20]andGuillain-Barré Syndrome[84]. 3 1.2 Burdenofinfectiousdiseasestoday EIDsconstituteasignicantpublichealthchallengeduetotheirunexpectedillnessandmor- tality toll, as well as the reduction in quality of life they impose on individuals. Globally, infectiousdiseasescontinuetobetheleadingcauseofmortalityamongtheyoungerthan60 years[28].In2011,malariaaloneclaimed111,000livesglobally,andresultedinalossof55mil- liondisability-adjustedlifeyears(DALYs)[8,46,80].Worldwide,approximately35.0million (CI[33.2-37.2million])peoplewereinfectedwithHIVin2013;globalmortalityfromAIDS- relatedillnesswasestimatedat1.5millionduringthatyear[83],resultinginadiseaseburden of95millionDALYslost[8]. Tuberculosisafected9.6millionpeoplein2014,andresulted in1.5milliondeaths[82]. Theburdenofthediseasewasestimatedat42millionDALYslost in 2011 [8]. Since the beginning of the 2014-2016 Ebola epidemic outbreak, the number of conrmedebolacasessurpassed15,250,withmorethan11,316deaths[19]. Whilemuchoftheglobalburdenofinfectiousdiseasesisfeltbydevelopingandpoorcoun- tries,theyremainnonethelessaleadingcauseofmortalityandmorbidityintheUS,claiming approximately170,000liveseachyear[31,57].Inuenzaandpneumoniaranksamongthetop 10leadingcausesofdeathinthecountry,anddisproportionatelyafectsinfantsandchildren, theelderly,andimmunocompromisedindividuals;HIV/AIDSistheninthcauseofmortality amongthe25-44yearsold [31,53].Othervaccine-preventableinfectionssuchasviralhepatitis andinuenzaalsocontinuetobesignicantcausesofillnessanddeath[31]. Despite the availability of inuenza vaccine, seasonal u and its complications claim over 50, 000 American lives each year (55, 227 people died from inuenza and pneumonia in 2014) [53], resulting in an estimated 610,660 undiscounted life-years lost, 3.1 million hospi- talizeddays,and31.4millionoutpatientvisits[51].Similarly,despitesignicantpublichealth efortsandthelongevityafordedbyantiretroviraltherapy(ART)overthepastdecades,HIV 4 burdenstillremainshighintheUSandtheviruscontinuestoinfectapproximately500,000 americaneachyear[16]. Thesenewcasesareprimarilyconcentratedamongmenwhohave sexwithmen(MSM),AfricanAmericans,andtheadultpopulation(30-64yearsold),which accounted for 69%, 46% and 60% of new HIV infections in 2013 [18]. In 2012, HIV/AIDS claimedthelivesofapproximately13,712Americans[17]. Anapproximated25%ofindividu- alsinfectedwithHIVarealsoco-infectedwithhepatitisC(HCV)intheUS[77].Thepreva- lenceofHCVinfectionalsoremainshigh,withapproximately3.5millionpeopleinfectedin theUS[2,23],nearlyhalfofwhomignoretheirinfectionstatus[2,22,37]. Similarly,CDC estimatesthat700,000to1.4millionAmericanssuferfromchronichepatitisB(HBV)infec- tion,nearly65%ofwhomignoretheirinfectionstatus[15]. Besides their death tolls and impacts on quality of life, infectious diseases also constitute a threattoglobaleconomicactivityandproductivity,throughtheirdisruptiveefectsonindi- vidual productivity and consumption, travel, trade and commerce. We illustrate below the magnitudeofthisimpactthroughsomeexampleofspecicinfectiousdiseases. Forexample, the2003SARSoutbreakwasestimatedtohavecaused$30billionineconomiclosses(more than $3 million per case) globally [25]. The World Bank estimated that the two-year nan- cialcostofEbolacouldreach$32.6billion,andpushtheafectedcountriesintoadeepreces- sion[72]. IntheUS,EIDscosttheeconomyover$120billionayear[28].Eachyear,inuenzaepidemics inict$10.4(condenceinterval,CI[$4.1-$22.2])billionindirectmedicalcostsand$16.3(CI [$8.7-$31.0])billioninlostearningsfromillness[51]. Whenmeasuredintermsofstatistical valuesoflife,theeconomicburdenofthediseaseapproximates$87.1(CI[$47.2-$149.5])bil- lion[51].Arecentanalysisestimatedthediscountedaveragelifetimecosts(in2011USdollars) of HIV/AIDS infection to range between $253,222 (CI [$250,308 - $256,137]) for category I patients(CD4count≤ 200cells/μLatdiagnosisorentrytocare)and$402,238(CI[$399,571 5 -$404,904])forcategoryIVpatients(501−900CD4cellscount/μLatdiagnosisorentryto care)[26]. Withanindividuallifetimecostof$64,490,chronicHCVinfectioncoststheUS theUSanestimated$6.5(CI[$4.3-$8.4])billionin2011,andisexpectedtoreach$9.1(CI[$6.4- $13.3])billionby2024duetoanincreaseinlifeexpectancyandintheprevalenceofadvanced liverdisease[62]. Other infectious diseases constitute signicant economic burdens in the US. For example, the US Department of Agriculture (USDA) recently estimated the annual economic bur- denoffoodborneillnessescausedby15foodbornepathogenstobeoftheorderof$15.5bil- lion[36]. Thewestnilevirus-relatedhospitalizationcostsaloneexceed$56millionannually intheUS[5]. Collectively,thesedatasupporttheviewthatinfectiousdiseasesarenotdiseasesofthepastin theUS.Severalreasonsexplainthepersistenceoftherelativelyhighburdenofinfectiousin theUS.First, vaccineuptakeratesformanyvaccine-preventableinfectiousdiseasesremains below recommended target levels in the US (Healthy People 2020 targets) [7, 9, 14, 24, 35, 60,70].Itisestimatedthat2millionpre-schoolers,themajorityoftheadultpopulation,and approximately35%ofseniorsintheUSareincompliantwithrecommendedimmunizations. During the 2013-2014 u season, only 14 states achieved u vaccination coverage of at least 50%[63].AsofNovember5,2015,only39%ofAmericansreportedhavingreceiveauvaccine forthe2015-2016season[11]. Similarly,only35statesandWashington,D.C.mettheHealthy People 2020 immunization goal of 90% coverage for the recommended≥ 3 HBV vaccine dosesininfantsaged19-35months[35]. Several factors contribute to the low uptake of vaccines in the US. For certain consumers, theout-of-pocketcostofimmunization,inconvenientclinichours,longwaitingtimes,and distancetothepointofcareconstitutesignicantnancialandlogisticalbarrierstovaccine uptake.Second,vaccineuptakeislowbecauselossaversionprimesconsumerstowardsgrossly 6 overweightingtheriskofvaccine-relatedsideefectscomparedtotheriskandseverityofinfec- tion [41, 42], while underweighting the benets of vaccination. Vaccine uptake is also low becausevaccinationisaclassicalexampleofpublicgoodandhencecharacterizedbythefree- riderproblem(individualshaveanincentivetofree-rideonthevaccinationdecisionsofoth- ers). Thisoccursbecausethecostsofvaccinationarelargelyprivate(i.e. bornebytheindi- vidualconsumer),whereasitsbenetsarelargelysocial(theyspillovertotherestofsocietyin theformofherdimmunity),non-excludable(tosomeextent),andnon-rival.Finally,psycho- logicalfactorssuchaspersonalandreligiousbeliefsaboutvaccination,andpriorunpleasant experiencewithvaccinationinuenceprivatedemandforvaccination. AsecondfactorinuencingtheburdenofinfectiousdiseasesintheUSistheemergenceofre- emergenceofnewandpreviously-conqueredinfectiousdiseases. Theevidencedocumented todatesuggestsanincreaseinthenumberandfrequencyofEIDs.Forexample,aftercombin- ingdataon 12, 102infectiousdiseaseoutbreaksbetween1980and2013across219countries withecologicalcharacteristicsofthecausalpathogens, Smithetal.[67]examinedtemporal trendsinthetotalnumbersofdiseaseoutbreaksandconcludedthatthenumberanddiver- sityofinfectiousdiseasessignicantlyincreasedovertheirstudyperiod,withthemajorityof outbreaksbeingofbacterial, viral, zoonoticandvector-borneorigins. Someoftheincrease inthenumberandfrequencyofoccurrenceofEID-relatedoutbreakscanbeaccountedfor bytheimprovedtheabilityoverthepastdecadestodiagnoseanddetectpreviouslyunknown pathogensandtheillnessestheycause[10]. Acaseinpointisthediscoveryofthebacterium Helicobacter pylori as the causal pathogen for chronic gastric ulcers, which were previously thoughttobecausedbystressordiet[21]. Butotherfactorsbesidestheimprovedabilityto detect EID outbreaks also explain the rise in their number and frequency, and consist of a complexinteractionofhuman(e.g.humanbehaviors,evolvingnatureofhumanbiologyand genomics), microbial (e.g. microbial virulence factors; microbial adaptation), and environ- mental(e.g.)factorsovertime;webrieydiscusstheminwhatfollows. 7 1.3 Factors inuencing the emergence and re-emergence of EIDs Anumberofhuman,microbialandenvironmentalfactorsaredeterminantintheemergence of new infectious disease as well as the re-emergence of previously well contained diseases. Amonghumandeterminants,poorcompliancewithvaccinationpolicies,weakenedsurveil- lancesystemsandthebreakdownorsystemicfailuresintheimplementationofpublichealth measuresinbothdevelopedanddevelopingcountriesduringthepastdecadesfacilitatedthe re-emergence of previously contained infectious diseases such as measles, mumps, pertussis andchickenpoxintheUSandinotherpartsoftheworld:yellowfeverhasrecentlyresurged inAngola,sickeningover1,200peopleandclaiming200lives[1,40,65]. Themisuseofthelimitedinventoryofefectiveantimicrobialsandpesticidesalsocreateseco- logicalandevolutionarypressuresthatcontributetothedevelopmentofresistantvectorsand pathogens(e.g. tuberculosis,malaria,nosocomial,food-borneinfections,HIV,andMERS) andresultinnewer(andoftenmorevirulent)formsofdiseases[10,40,65].Forexample,the prevalenceofantibiotic-resistantgonorrheahasrecentlysurgedintheUS,toreachoverathird ofallcases[63]. Migrations,travelinganddisplacementsofpopulationshavealsoincreasedtounprecedented levelssincethebeginningofthe 20 th century.Thesemassiveandfrequentmovementsofpop- ulationscontributetothespreadofnewinfectiouspathogens[40,65].Forexample,theSARS virusrapidlyspread(withinaweek)fromChinato17countriesthroughairtraveler,leading to the 2003 outbreak [38]. It has also been well documented that the development of the transportinfrastructureandtheassociatedmigrationsitfavoredhelpedacceleratethespread ofHIVinmainlandAfrica[71].Increasingurbanizationanddensicationoftheworldpop- ulation creates demographic pressures and precarious sanitary conditions that promote the 8 emergenceofinfectiousdiseasessuchastuberculosisandotherdiseasesoncelocalizedinrural areas[32,55]. Environmentalfactorsalsoplayanimportantroleintheemergenceofinfectiousdiseasesby directlyinuencingmicrobialevolution,orthroughtheirefectsontheecologyofinfectious diseasemicrobialhostsandvectors. Forexample,changesinclimaticconditionsarealtering thespatialandtemporaldistributionsoftemperaturesandrainfalls[38,43,78],andfavoring outbreaksofcertaininfectiousdiseases(e.g.hantaviruspulmonarysyndromeepidemic[38]). Similarly,theincreasingpaceofdeforestationincertainpartsoftheworldaswellasreforesta- tionefortsinotherpartsaresignicantlydisturbinglocalecosystemsandnaturalhabitatsof infectiousdiseasevectorsandtheirpredators,hencecreatinganimbalancethatfavorscertain vectorsforinfectiousdiseasespathogens(e.g.rodents,andcertaininsectssuchasmosquitoes) attheexpenseoftheirpredators,aswellasnewopportunitiesfortheinvasionandspreadof other opportunistic species. These new ecological pressures, which are often accompanied with competition for resources, are leading to more frequent contacts of infectious disease vectorsandpathogenswithhumans[40,58,65]. Intensivefarming, andincreasedproximityofhumanwithlivestockalsosignicantlyinu- encetheemergenceandre-emergenceofinfectiousdiseasesbyincreasinghumancontactswith newpathogensandfacilitatingtheirtransmissionbetweenspecies.Forexample,anumberof inuenzavirusespreviouslypresentonlyinanimals(e.gthetheinuenzastrainthatcausedthe 2009“swineu”pandemic)arenowtransmissibletohumansthroughdomesticatedanimals suchaspigsandducks, becausethelatteraremajor reservoirs ofinuenzaviruses, whilethe formerconstituteecientmixingvesselsfornewersubtypesofinuenzastrains(e.g.subtypes H1N1,H1N2andH3N2oftheinuenzaAvirus 4 )[38,66]. 4 The prevalent subtype in Europe (H1N2) is believed to have been introduced into pigs in 1979 through waterfowls,whilethosecirculatinginNorthAmerica(H1N1andH3N2)arebelievedtohavebeenintroduced intopigsafterthe1918SpanishinuenzapandemicthroughareassortmentbetweentheH1N1subtypeandthe 9 Finally,theexpansionofstandingwatersurfaces(e.g. dams,irrigationplants,reservoirs)cre- ates conditions for the proliferation of mosquitoes and many aquatic vectors for infectious diseases[56].Forexample,theconstructionsoflarge-scaledams(e.g.Gezira-Managildamin Sudan, AswandaminEgypt, MelkasadidaminEthiopiaandDanlingandHuangshidams inChina)werefollowedbytheemergenceandre-emergenceofdiseasessuchasschistosomi- asisandmalaria[48,59,85]. Similarly, oodinginricefarmswereassociatedwithJapanese encephalitis epidemics, while epidemics of the rift valley fever were linked to dam building andperiodsofheadyrainfalls[38]. Thisbriefoverviewsuggeststhatinfectiousdiseasesemergeandre-emergethroughavariety ofcomplexprocesses,someofwhichcanbeinuencedthroughbehavioralchanges(e.g.vac- cination). Inwhatfollows,wedescribesomeoftheseinfectiousdiseasepreventionstrategies andoutlinethegoalsofthedissertation. 1.4 Objectivesofthisdissertation The heavy burden EIDs calls for the implementation of efective disease prevention strate- gies. As the early 20 th experience suggests, improved hygiene and vaccination are the most efectivetoolsforpreventinginfectiousdiseasemorbidityandmortality,aswellastheirasso- ciatedsocial,economic,andhealthburden. Vaccinesarealsosomeofthemostcost-efective infectiousdiseasepreventiontechnologieswithaveryhighreturnoninvestment.Recentpro- jectionsfromtheGAVIAllianceandtheBill&MelindaGatesFoundationalsosuggestthat vaccinationcouldavert9.9milliondeathsgloballybetween2011and2020,andthatroutine andsupplementarymeaslesvaccinationcouldavertanother13.4milliondeaths[45].Itisalso H3N2orH1N1subtypes[38,75].Ageneticreassortmentorrecombinationisaprocessbywhichgenesegments from2ormorecellularunitscombinetoproduceanewcell. Thisprocessisdiferentfromthetypicalgenetic mutation(“antigenicdrift”)whicharerandomalterationsofthegeneticmaterial(e.g.copyingerrors,orchanges inducedbymutagens). 10 estimatedthatintheUSalone,routineimmunizationamongeachbirthcohortsavesapprox- imately33,000livesandpreventsnearly14millioninfections,resultinginnearly$9.9billion and $33.4 billion savings in direct and indirect healthcare costs, respectively [35]. Globally, immunizationavertsnearly2.5millionchildhooddeathseachyear[81]. Giventhesebenets,animportantpolicyquestionisthatofhowtomostefectivelyincrease privatedemandforvaccines. Sincecosts(out-of-pocketcostofvaccines,timecostofseeking vaccination)constitutesignicantbarrierstovaccinationformanyconsumers,cost-reducing strategies such schemes of direct subsidy of vaccination can help boost private demand for vaccines.Forthisreason,therecentNationalAdultImmunizationPlan(NAIP)setasobjec- tivetoreducenancialbarrierstotheuseofvaccinesrecommendedforadultsintheUS[54]. Evidencefromeldexperimentssuggestthatmodestnancialincentivesintheformofcash paymentorvoucherscanhelpachievethisobjective[4,47,79]. However,suchdemand-side subsidiescanbeveryexpensiveandmightonlyhaveamodestonvaccineuptake.Alternatively, nudgingmechanismssuchassmall-stakeinsuranceschemeshavebeenshowntomeaningfully inducedesiredhealthbehaviors[49]andcouldalsobeusedtohelpboostprivatedemandfor vaccines. However,itisunclearwhichofthesealternativeincentivestrategies(directsubsidy schemesvsinsuranceschemes)aremostefectiveforincreasingvaccineuptake.Inordertohelp shedsomelightonthisquestion,weinvestigateinChapter2theefectofa“no-fault”insur- anceschemevsamonetarilyequivalentdirectsubsidyscheme(expectedvalueoftheinsurance) onwillingnesstoutilizeinfectiousdiseasevaccinesintheUS.Weuseadiscretechoiceexperi- ment(DCE)approachtoinvestigatetheoptimalfeaturesofthesevaccineincentiveprograms, andstudyhowwillingnesstovaccinatevarieswiththecharacteristicsofthevaccineandthe disease,aswellasthecompensationlevel.Thisapproachallowsustogainabetterinsightinto thetypesofpoliciesthatcanbemostefectiveatboostingprivatedemandforvaccines,aswell asthetypesofvaccinesanddiseasesthatinnovatorsshouldtarget. 11 A second approach to infectious disease prevention is early diagnosis of infection followed by prompt treatment. This approach known as “treatment as prevention” is increasingly beingincorporatedintomanydisease-specicguidelines(especiallyforsexually-transmitted diseases[STIs]). Inordertoassesstheimpactsandefectivenessofthispreventionstrategy, weconductacasestudywithHIV/AIDS,adiseaseforwhichthe“treatmentasprevention” (“test-and-treat”) strategy has been particularly successful and is currently standard of care forHIV/AIDSintheUS[35,63],basedonemergingeldevidenceandmodelingpredictions supportingitsefectivenessinreducingHIVtransmission.WhilemostexistingHIVtransmis- sionmodelsagreeonthepreventiveefectivenessof“test-and-treat”atthepopulationlevel, onlyafewpredictthatitwillbesucienttoendtheHIV/AIDSepidemic[6,76],duetothe potentialbehavioralresponsestoART(“HIV-optimism”)andincreasesinbothacquiredand transmittedresistancetoART[68,69]. Wethereforeinvestigatetheepidemiologicalefects ofthe“test-and-treat”strategyinChapter3,usingarichmathematicalcompartmentalmodel ofHIVtransmissionthataccountsfortheefectsofdrugresistance. Ourmodelfocuseson theHIV/AIDSepidemicinLosAngelesCountyamongtheMSM,apopulationdispropor- tionatelyimpactedbytheepidemic. Thisapproachallowsustodescribeoptimalpoliciesfor combatingHIV/AIDSandotherinfectiousdiseaseswithsimilardynamics. Chemoprophylaxis (both pre- and post-exposure prophylaxis) can also be an efective approach to the prevention of infectious diseases. Post-exposure prophylaxis (PEP) treat- ments have been shown to prevent illness after exposure to the pathogen as well as reduce the risk of subsequent infection of other individuals. A number of PEP regimens are now approvedforuseandrecommendedagainstinfectionssuchasHAV,HBV,HCV,HIV,vari- cellazostervirus,inuenzavirus,invasivegroupAstreptococcalinfection,invasivemeningo- coccalinfection,pertussis,tetanus,tuberculosis,anthrax,diphtheria,measlesandplague[3]. Efective pre-exposure chemoprophylaxis treatments exist also for some infectious diseases: 12 forexample,chemoprophylaxisdrugsareroutinelyrecommendedfortravelersintomalaria- endemicareas.Recently,antiretroviraltherapy(ART)-basedpre-exposureprophylaxis(PrEP) wasintroducedtohelppreventHIVinfectioninindividualsathighriskofinfection.Thisrel- ativelynewHIVpreventionapproachwhichconsistsofprovidingreduceddosesofARTto uninfectedindividualswhoareatasubstantialriskofbecominginfectedwithHIV.Theclin- icalecacyofPrEPhasbeendemonstratedinnumerousrandomizedcontrolledtrialstudies; evidenceonitsreal-worldefectivenessisalsobeginningtoemerge. However,giventhisnew technologyisrelativelyexpensiveandcouldpotentiallyinduceanincreaseinriskybehaviors suchasareductionincondomuse,thetradeofsbetweenthecostsandbenetsofthispreven- tionapproachrelativetootheralternativestrategiessuchastestingaloneand“test-and-treat”, remainpoorlyunderstood. Further, becausePrEPisan imperfect vaccine(i.e. preventsbut doesnotconfercompleteimmunityagainstHIVinfection),itpresentssomeofthechallenges encounteredwithvaccination. Forexample,sincePrEPisbasedonARTregimens,itcauses severesideefectsthatcandeterhealthyat-riskindividualsfromadoptingit.Thesefactorshave profound implications for the efectiveness of the PrEP strategy. We tackle these questions inChapter4, byrstexpandingthesetofHIVpreventionpoliciestoincludevariationsof testing,“test-and-treat”andPrEP.Second,weextendtheepidemicmodeldevelopedinChap- ter3tocapturetheintroductionofPrEPandthesealternativepolicies,andthirdweusethis extended model in combination with an economic model to undertake a cost-efectiveness analysis of alternative HIV prevention strategies. Our integrated epidemiological and eco- nomicmodelsallowustodeterminetheecientfrontierforresourceallocationacrossalter- nativeHIVpreventionstrategies. WeconcludeouranalysisinChapter5withasummaryourndingsandcontributions,adis- cussionoftheimplicationsofourresults,andweofersomethoughtsforfutureanalyses. 13 References [1] Aizenman,N.A‘ForgottenDisease’IsSuddenlyCausingNewWorries.NationalPublic Radio,Mar252016. AccessedMar25,2016. [2] AmericanAssociationfortheStudyofLiverDiseasesandInfectiousDiseasesSocietyof America.HCVguidance:Recommendationsfortesting,managing,andtreatinghepati- tisC. Availableathttp://www.hcvguidelines.org/full-report-view,Mar2016. Accessed Mar26,2016. [3] Bader,M.S.andMcKinsey,D.S. Postexposureprophylaxisforcommoninfectiousdis- eases. American Family Physician,88(1):25–32,2013. [4] Banerjee, A.V.etal. ImprovingimmunisationcoverageinruralIndia: Calibrationlus- teredrandomisedcontrolledevaluationofimmunisationcampaignswithandwithout incentives. British Medical Journal,340:c2220,2010. [5] Barrett, A.D. Economic burden of West Nile virus in the United States. American Journal of Tropical Medicine and Hygiene,90(3):389–90,2014. [6] Bellan,S.E.etal.Extra-coupleHIVtransmissioninsub-SaharanAfrica:Amathematical modellingstudyofsurveydata. Lancet,381(9877):1561–9,2013. [7] Belluck,P.andRyzik,M. RobertDeNiroDefendsScreeningofAnti-VaccineFilmat TribecaFestival. The New York Times,Mar252016. AccessedMar25,2016. [8] Bhutta,Z.A.etal.Globalburden,distribution,andinterventionsforinfectiousdiseases ofpoverty. Infectious Diseases of Poverty,3(1):21,2014. 14 [9] Black,C.L.etal.InuenzaVaccinationCoverageAmongHealthCarePersonnel–United States,2014-15InuenzaSeason.MorbidityandMortalityWeeklyReport,64(36):993–9, Sep182015. [10] Caliendo,A.M.etal. Bettertests,bettercare: Improveddiagnosticsforinfectiousdis- eases. Clinical Infectious Diseases,57(Suppl3):S139–70,2013. [11] Center for Disease Control and Prevention. Fewer Than Half of Americans Report HavingGottenAFluVaccinethisSeason. Availableathttp://www.cdc.gov/u/news/ half-of-americans-received-u-vaccine.htm.,2015. AccessedMar18,2016. [12] CentersforDiseaseControlandPrevention. Recommendedchildhoodimmunization schedule–UnitedStates,1999. MorbidityandMortalityWeeklyReport,48(1):12,1999. [13] Centers for Disease Control and Prevention. Impact of vaccines universally recom- mended for children–United States, 1990-1998. Morbidity and Mortality Weekly Report,48(12):243,1999. [14] CentersforDiseaseControlandPrevention. FluVaccinationCoverage: UnitedStates, 2012-2013InuenzaSeason. Availableathttp://www.cdc.gov/u/pdf/uvaxview/vax- coverage-1213estimates.pdf,Sept252013. AccessedJan15,2016. [15] CentersforDiseaseControlandPrevention.2013-14State,Regional,andNationalVacci- nationReportII.Availableathttp://www.cdc.gov/u/uvaxview/reports/reporti1314/ reportii/index.htm,Sep232014. AccessedMar18,2016. [16] Centers for Disease Control and Prevention. Monitoring selected national HIV pre- vention and care objectives by using HIV surveillance data – United States and 6 US dependentareas–2012. HIV Surveillance Supplemental Report,19(3),2014. Accessed Mar23,2016. 15 [17] Centers for Disease Control and Prevention. HIV in the United States: At a Glance. Available at http://www.cdc.gov/hiv/statistics/overview/ataglance.html, Sep 15 2015. AccessedMar23,2016. [18] CentersforDiseaseControlandPrevention. DiagnosesofHIVInfectionintheUnited States and Dependent Areas, 2014. HIV Surveillance Report, 2014, 26, Nov 2015. AccessedMar23,2016. [19] CentersforDiseaseControlandPrevention. 2014EbolaOutbreakinWestAfrica-Case Counts. Availableathttp://www.cdc.gov/vhf/ebola/outbreaks/2014-west-africa/case- counts.html,March2016. AccessedMar23,2016. [20] CentersforDiseaseControlandPrevention.CDCConcludesZikaCausesMicrocephaly andOtherBirthDefects. Availableathttp://www.cdc.gov/media/releases/2016/s0413- zika-microcephaly.html,April132016. AccessedonApril13,2016. [21] ConsensusDevelopmentPanel. Helicobacter pyloriinpepticulcerdisease. NIH Con- sensus Development Conference Statement, 12(1):1–23, Jan 7-9 1994. Available at https: //consensus.nih.gov/1994/1994HelicobacterPyloriUlcer094html.htm. [22] Denniston, M.M. et al. Awareness of infection, knowledge of hepatitis C, and medi- cal follow-up among individuals testing positive for hepatitis C: National Health and NutritionExaminationSurvey2001-2008. Hepatology,55(6):1652–61,2012. [23] Denniston,M.M.etal.ChronichepatitisCvirusinfectionintheUnitedStates,National HealthandNutritionExaminationSurvey2003to2010. Annals of Internal Medicine, 160(5):293–300,2014. [24] Elam-Evans,L.D.etal. National,Regional,State,andSelectedLocalAreaVaccination CoverageAmongAdolescentsAged13-17Years–UnitedStates, 2013. Morbidity and Mortality Weekly Report,63(29):625–33,2014. 16 [25] Fan,E.X. SARS:Economicimpactsandimplications. 2003. [26] Farnham, P.G.etal. Updatesoflifetimecostsofcareandquality-of-lifeestimatesfor HIV-infected persons in the United States: late versus early diagnosis and entry into care. Journal of Acquired Immune Deficiency Syndromes,64(2):183–9,2013. [27] Fauci, A.S. and Morens, D.M. Zika Virus in the Americas – Yet Another Arbovirus Threat. New England Journal of Medicine,347(7):601–4,Feb2016. [28] Fauci,A.S.,Touchette,N.A.andFolkers,G.K. Emerginginfectiousdiseases:A10-year perspectivefromtheNationalInstituteofAllergyandInfectiousDiseases.International Journal of Risk & Safety in Medicine,17(3,4):157–67,2005. [29] Fenner,F.etal. Smallpoxanditseradication. Geneva: WHO,1987. [30] Fineberg,H.andWilson,M. Emerginginfectiousdiseases. IRGC Report,2010. [31] Fonkwo,P.N. Pricinginfectiousdisease. EMBO Reports,9(1S):S13–7,2008. [32] Gayer,M.etal. Conictandemerginginfectiousdiseases. EmergingInfectiousDiseases, 13(11):1625,2007. [33] Gostin, L.O. and Hodge, J.G. Is the United States Prepared for a Major Zika Virus Outbreak? Journal of the American Medical Association,April2016. Publishedonline April13,2016. [34] Hammarsten,J.,Tattersall,W.andHammarsten,J. Whodiscoveredsmallpoxvaccina- tion? Edward Jenner or Benjamin Jesty? Transactions of the American Clinical and Climatological Association,90:44,1979. [35] HealthyPeople.gov. Immunization and Infectious Diseases. Available at https://www.healthypeople.gov/2020/topics-objectives/topic/immunization-and- infectious-diseases,2014. AccessedMar18,2016. 17 [36] Hofmann, S., Maculloch, B.andBatz, M. Economicburdenofmajorfoodborneill- nessesacquiredintheUnitedStates. Technicalreport,May2015. [37] Holmberg, S.D. et al. Hepatitis C in the United States. New England Journal of Medicine,368(20):1859–61,2013. [38] Howard,C.R.andFletcher,N.F. Emergingvirusdiseases:Canweeverexpecttheunex- pected? Emerging Microbes & Infections,1(12):e46,2012. [39] Hoy,S.andDufy,J. The Sanitarians: A History of Public Health. Urbana,Illinois: UniversityofIllinoisPress,1990. [40] InstituteofMedicine. The impact of globalization on infectious disease emergence and control: ExploringtheConsequencesandOpportunities,WorkshopSummary-Forumon Microbial Threats. Washington,DC:TheNationalAcademiesPress,2006. [41] Kahneman, D. and Tversky, A. Prospect Theory - Analysis of Decision under Risk. Econometrica,47(2):263–91,1979. [42] Kőszegi, B. and Rabin, M. A model of reference-dependent preferences. Quarterly Journal of Economics,121(4):1133–65,2006. [43] Lambrechts,L.etal.Impactofdailytemperatureuctuationsondenguevirustransmis- sionbyAedesaegypti.ProceedingsoftheNationalAcademyofSciences,108(18):7460–5, 2011. [44] Lederberg,J.etal. Microbial Threats to Health: Emergence, Detection, and Response. Washington,DC:TheNationalAcademiesPress,2003. [45] Lee, L.A.etal. Theestimatedmortalityimpactofvaccinationsforecasttobeadmin- istered during 2011-2020 in 73 countries supported by the GAVI Alliance. Vaccine, 31: B61–72,2013. 18 [46] Lozano, R. et al. Global and regional mortality from 235 causes of death for 20 age groupsin1990and2010: AsystematicanalysisfortheGlobalBurdenofDiseaseStudy 2010. Lancet,381(9867):628,2013. [47] Mantzari,E.,Vogt,F.andMarteau,T.M. Financialincentivesforincreasinguptakeof HPVvaccinations: Arandomizedcontrolledtrial. Journal of Health Psychology,34(2): 160–71,Feb2015. [48] McManus,D.P.etal. SchistosomiasisinthePeople’sRepublicofChina:Theeraofthe ThreeGorgesDam. Clinical Microbiology Reviews,23(2):442–66,2010. [49] Meeker,D.etal.UseofInsuranceAgainstaSmallLossasanIncentiveStrategy.Decision Analysis,12(3):122–9,Sep2015. [50] Metcalf, C.J.E. et al. Seven challenges in modeling vaccine preventable diseases. Epi- demics,10:11–5,2015. [51] Molinari,N.A.M.etal. TheannualimpactofseasonalinuenzaintheUS:Measuring diseaseburdenandcosts. Vaccine,25(27):5086–96,2007. [52] Morens,D.M.andFauci,A.S. Emerginginfectiousdiseases: Threatstohumanhealth andglobalstability. PLoS Pathogens,9(7):e1003467,2013. [53] NationalCenterforHealthStatistics.Health,UniyedStState,2015: WithSpecialFeature on Racial and Ethnic Health Disparities. NumberDHHSPublicationNo.2016-1232. Hyattsville,MD,May2016. Availableat. [54] NationalVaccineAdvisoryCommittee. RecommendationsfromtheNationalVaccine Advisorycommittee:Standardsforadultimmunizationpractice.PublicHealthReports, 129(2):115–23,Mar–Apr2014. 19 [55] Neiderud,C.J. Howurbanizationafectstheepidemiologyofemerginginfectiousdis- eases. Infection Ecology & Epidemiology,5,2015. [56] Nelson, K. Infectious disease epidemiology, chapter Emerging and new infectious dis- eases,page337. Jones&BartlettPublishers,2013. [57] Noah,D.andFidas,G. Theglobalinfectiousdiseasethreatanditsimplicationsforthe UnitedStates. Technicalreport,2000. [58] Patz,J.A.etal. Unhealthylandscapes:Policyrecommendationsonlandusechangeand infectiousdiseaseemergence. Environmental Health Perspectives,pages1092–8,2004. [59] Petney, T. and Taraschewski, H. Water-borne parasitic diseases: Hydrology, regional developmentandcontrol. InFrimmel,F.,editor, Water Chemistry and Microbiology, volume3ofA Treatise on Water Science,pages303–66.Elsevier,2011. [60] Phadke,V.K.etal. AssociationBetweenVaccineRefusalandVaccine-PreventableDis- easesintheUnitedStates:AReviewofMeaslesandPertussis. Journal of the American Medical Association,315(11):1149–58,2016. [61] Plotkin,S.andOrenstein,W. Vaccines. Philadelphia,PA:WBSaundersCo.,3edition, 1999. [62] Razavi,H.etal.ChronichepatitisCvirus(HCV)diseaseburdenandcostintheUnited States. Hepatology,57(6):2164–70,2013. [63] RobertWoodsJohnsonFoundationandTrustforAmerica’sHealth. Outbreaks: Pro- tectingAmericansfrominfectiousdiseases. Technicalreport, RobertWoodsJohnson FoundationandTrustforAmerica’sHealth,2014. 20 [64] Roush, S.W. et al. Historical comparisons of morbidity and mortality for vaccine- preventablediseasesintheUnitedStates. JournaloftheAmericanMedicalAssociation, 298(18):2155–63,2007. [65] Saker, L.etal. Globalizationandinfectiousdiseases: Areviewofthelinkages, Social, EconomicandBehavioural(SEB)Research. TechnicalReport2,2004. [66] Smith,G.J.etal. Datingtheemergenceofpandemicinuenzaviruses. Proceedings of the National Academy of Sciences,106(28):11709–12,2009. [67] Smith, K.F. et al. Global rise in human infectious disease outbreaks. Journal of The Royal Society Interface,11(101),2014. [68] Sood,N.etal. Test-and-treatinLosAngeles: Amathematicalmodeloftheefectsof test-and-treatforthepopulationofmenwhohavesexwithmeninLosAngelesCounty. Clinical Infectious Diseases,56(12):1789–96,2013. [69] Sood,N.etal.ReplytoGonzalez-Sernaetal. ClinicalInfectiousDiseases,57(3):479–80, 2013. [70] Stokley, S. et al. Human Papillomavirus Vaccination Coverage Among Adolescents, 2007?2013, and Postlicensure Vaccine Safety Monitoring, 2006?2014 – United States. Morbidity and Mortality Weekly Report,63(29):620–4,2014. [71] Tatem, A.J.etal. SpatialaccessibilityandthespreadofHIV-1subtypesandrecombi- nants. AIDS,26(18):2351–60,2012. [72] The World Bank Group. The Economic Impact of the 2014 Ebola Epi- demic: Short and Medium Term Estimates for West Africa. Available at http://www.worldbank.org/en/region/afr/publication/the-economic-impact-of- 21 the-2014-ebola-epidemic-short-and-medium-term-estimates-for-west-africa, October 2014. AccessedMar23,2016. [73] US Department of Health, Education, and Welfare. Vital statistics – special report, nationalsummaries: Reportedincidenceofselectednotifiablediseases,UnitedStates,each divisionandstate,1920-50.Washington,DC:USDepartmentofHealth,Education,and Welfare,PublicHealthService,NationalOceofVitalStatistic,1953. [74] USDepartmentofHealth,Education,andWelfare. Vital statistics rates in the United States, 1940-1960. Washington,DC:USDepartmentofHealth,Education,andWelfare, PublicHealthService,NationalCenterforHealthStatistics,1968. [75] Vincent,A.L.etal. Swineinuenzaviruses: ANorthAmericanperspective. Advances in Virus Research,72:127–54,2008. [76] Walensky,R.P.etal.TestandtreatDC:ForecastingtheimpactofacomprehensiveHIV strategyinWashingtonDC. Clinical Infectious Diseases,51(4):392–400,2010. [77] Ward,J.ThehiddenepidemicofhepatitisCvirusinfectionintheUnitedStates:Occult transmissionandburdenofdisease. Topics in Antiviral Medicine,21(1):15–19,2012. [78] Wearing, H.J.andRohani, P. Ecologicalandimmunologicaldeterminantsofdengue epidemics. Proceedings of the National Academy of Sciences,103(31):11802–7,2006. [79] Wigham, S. et al. Parental nancial incentives for increasing preschool vaccination uptake:Systematicreview. Pediatrics,134(4):e1117–28,Oct2014. [80] WorldHealthOrganization. GlobalBurdenofDisease(GBD):DALYestimates,2000- 2012. Available at http://www.who.int/healthinfo/global_burden_disease/estimates/ en/index2.html,2011. AccessedMar18,2016. [81] WorldHealthOrganization. Globalvaccineactionplan2011–2020. 2013. 22 [82] WorldHealthOrganization. GlobalTuberculosisReport2015. 2015. [83] World Health Organization. Global Health Observatory (GHO) data: HIV/AIDS. Availableathttp://www.who.int/gho/hiv/en/,2016. AccessedMar23,2016. [84] World Health Organization. Zika virus microcephaly and Guillain-Barré syndrome: Situation report. Available at http://apps.who.int/iris/bitstream/10665/204718/1/ zikasitrep_31Mar2016_eng.pdf?ua=1,March2016. Accessedon15April,2016. [85] Yewhalaw,D.etal. Theefectofdamsandseasonsonmalariaincidenceandanopheles abundanceinEthiopia. BMC Infectious Diseases,13(1):1,2013. 23 Chapter2 Nudginginfectiousdiseasevaccineuptakein theUnitedStateswitha“no-fault” insuranceagainsttheriskofvaccine-related sideefects:Adiscretechoiceexperiment EmmanuelF.Drabo,NeerajSood,JoelW.Hay,JasonN.Doctor Abstract Importance: Both children and adult immunization rates for vaccine-preventable diseases remain below the Healthy People 2020 target levels in the US. Understanding the barriers tovaccineuptakeisessentialindevelopingefectivestrategiestoboostUSimmunizationrates againstinfectiousdiseases. Objectives: Directsubsidyschemescanserveaspowerfulmechanismsforincentivizingvac- cineuptakebyhelpingovercomethenancialbarrierstovaccination,butarecostlyandmight onlyyieldmodestoutcomes. Evidencefromeldexperimentssuggestthatsmall-stakeinsur- anceschemescaninducedesiredhealthbehaviors,includingvaccination. Further,lossaver- sioninprospecttheoryandtheKőszegi-Rabinutilitytheorypredictsthatvaccineuptakewith thistypeofinsuranceschemewillexceedthatunderanequivalentsubsidyscheme(i.e. the expected valueof the insurance), unlike expectedutility theory whichpredicts these mone- tarilyequivalentschemestoresultinsimilarefectsonvaccineuptake. Hence,itisunknown 24 whichofthesealternativestrategies(directsubsidyvsinsurance)ismostefectiveforincen- tivizingprivatedemandforinfectiousdiseasevaccines.Thisstudyaimstoshedsomelighton thisquestionanddeterminetheoptimaldesignofthisnancialincentiveprogram,aswellas thetypesofvaccinesthatinnovatorsshouldtarget. Design,SettingandParticipants:Weconductadiscreteexperiment(DCE)with 1, 257Ama- zon’sMTurksubjectsresidingintheUSandatleast 18yearsold.Theserespondentsareran- domizedwithequalprobabilityintotwocompensationprograms,namelyaninsurancearm (n = 627)andasubsidygroup(n = 630).Subjectsintheinsurancearmareoferedaninsur- anceagainsttheriskofseverevaccinesideefects.Thisinsurancecompensatesthemwiththe valueofstatisticallifeassociatedwiththevaccineinjury.Subjectsinthesubsidyarmareofered theexpectedvalueoftheinsuranceuponpurchaseofthevaccine.Participantsinbothgroups arethenpresentedasequenceof6hypotheticalvaccinationscenarios,eachconsistingofadis- easesituation,2vaccinealternatives,andano-vaccinationalternative,andaskedtochoosethe alternativetheymostpreferred. MainOutcomesandMeasures: Werstestimatethemarginalefectofinsurancerelativeto subsidy on vaccine uptake. Second, we estimate the marginal efects of vaccine attributes (administrationmode, ecacy, immunity, severityofsideefects, riskofsideefectsandthe out-of-pocketcostofvaccine),diseasecharacteristics(diseasetype,transmissionmode,infec- tionrisk,severityofdisease,riskofseveresymptoms),aswellasthecompensationamounton theprobabilityofvaccination. Third, weexaminepotentialheterogeneousefectsofinsur- anceonvaccination,withrespecttovaccine,disease,andindividualcharacteristics,aswellas thecompensationlevel. Finally, wecalculatethewillingnesstopayforselectedvaccineand diseaseattributes. 25 Results:Inthesampleofrespondentswhoprovidedconsistentresponsestothechoicetasks (i.e. selectedthesamealternativeforduplicatedchoicescenarios),vaccineuptakewas1.5per- centagepoints(0.015,CI[0.003,0.027])greateramongnon-femalerespondentsintheinsur- ancearmrelativetothesubsidyarm,buthadnosignicantefectamongfemalerespondents (-0.003[-0.014,0.008]). Non-femalesubjectsintheinsurancegroupwerealso1.1times(CI [0.976,1.207])aslikelyasfemalerespondentstoacceptvaccination.Thecorrespondinggure inthesubsidygroupwas0.918,CI[0.819,1.018]. Thecompensationamounthadnomean- ingfulefectonvaccineuptake.Theout-of-pocketcostofvaccine,vaccineecacyandimmu- nity,andtheriskandseverityofvaccinesideefectsconstitutedthemostinuentialvaccine attributes. Willingness to pay for a percentage point increase in vaccine ecacy was $23.28 [$21.33,$25.23]. Averagewillingnesstopaytoreduceinfectionriskby1in100,000was$0.22 [$0.19,$0.24].Theaveragewillingnesstoacceptfora1inamillionriskofseverevaccineside efectwas$17.11[$10.52,$23.69]. Conclusion and Relevance: Our results indicate that a “no-fault” insurance against vaccine sideefectshadamodestefectonvaccineuptakerelativetoamonetarilyequivalentsubsidy, suggestingthepossibilityofleveragingonthecurrentvaccineinjurycompensationprogram toboostUSvaccineuptake. Theresultsalsosuggestthattheinnovationofmoreecacious, saferandmoreprotectivevaccinescouldhelpsignicantlyboostdemandforvaccination. 26 2.1 Introduction Emerging and re-emerging infectious diseases such as Ebola, SARS, and MERS constitute a threat to public health and economies in both developed and under-developed nations. Immunizationisahighlycost-efectivepreventivetechnologyagainstthesediseases[61]. For someinfectiousdiseases,nosuchefectivepreventivetechnologyexistsduetoalackofinno- vation; for many others, however, relatively safe and ecacious vaccines exist, but uptake remainslow, despitethedemonstratedimportanceofhighervaccinationcoveragetoreduc- ingtheoccurrenceofvaccine-preventablediseases. Compliance with childhood-recommended vaccines remains relatively low in the US. For example,ofallchildhood-recommendedvaccinations,onlyTdapandMCVcoverageratehave met the US Healthy People 2020 goal coverage level of 80%. In 2013, uptake of Tdap and MCVvaccinesamongthe13-17yearswas86.0%and77.8%,respectively[21]. Coveragelevels forHPVvaccinationarealsodismallylow:Only57.3%and34.6%ofgirlsandboysaged13-17, respectively,hadinitiatedthethree-doseseriesasof2013,andonly37.6%and13.9%ofgirlsand boys,respectively,hadcompletedthe3-dosevaccinationseries[73]. Inaddition,HPVvacci- nationamonggirlshasonlyminimallyincreasedsince2013[73].Inuenzavaccineuptakealso remainsrelativelylowamongadolescents,withonly42.5%uptakerateduringthe2012-2013 season[10]. ThelowvaccinecoverageintheUSisnotpeculiartochildren;adultvaccinationcoveragefor most routinely recommended vaccines remain low and below the US Healthy People 2020 targetcoveragelevels[6]. Forexample,coverageforallrecommendedvaccinescontinuedto remainlowamongadultsbetween2012and2013,andcoverageratesmodestlyincreasedonly forTdap,herpeszoster,andHPV(2.9,4.1,and3.6percentagepointincreasesto17.2%,24.2% and5.9%amongthe≥ 19years,≥ 60years,and19-26yearsmales,respectively)[6]. 27 USchildhoodandadultvaccinationcoverageratesarelowbecauseconsumersfeartheymight experienceseverevaccine-relatedsideefects[21].Thisfearisevengreaterfornewvaccinesfor whichthesideefectscannotbeanticipateduntilthevaccineisoferedtothegeneralpopu- lation[13,67]. Only33%ofstateMedicaidprogramsincludeallrecommendedvaccinations foradultsasapreventivebenetforexistingpatientsandprohibitcostsharing[54]. Finally, incentives to free ride on the vaccination decisions of others, religious and personal beliefs aboutvaccinationcontributetoreduceduptakeforcertainvaccines. Thelowuptakeofbothchildhoodandadultvaccinesmakesfurtherreductionsininfectious diseasemortalityanddisabilityamajorchallengeforpublichealthocials.Identifyingefec- tiveinterventionsthatcanhelpincreasevaccinecoverageisthereforeanimportantpolicycon- cern.Tothatend,andbasedonevidencefromeldexperimentssuggestingthatmodestnan- cialincentivesintheformsofvouchersredeemableforcash,vaccinesubsidies,orconditional cashoferscanhelpboostvaccineuptake, somehavesuggestedencouragingvaccineuptake through the use of nancial incentives [4, 45, 51, 82]. The US Department of Health and HumanServices(HHS)NationalAdultImmunizationPlan(NAIP)endorsedthisviewby setting as one of its objectives the reduction of nancial barriers to the use of vaccines rou- tinelyrecommendedforadults[53]. However,demand-sidesubsidiescanbeveryexpensive andmightonlyhavemodestefectsonvaccineuptake. Furthermore,alternativetheoriesto expected utility theory (e.g., prospect theory and the Kőszegi-Rabin utility theory) predict thattheeliminationorreductionofriskthroughinsurancecanobtainasimilarorbetterout- comethanasubsidyofequivalentmonetaryvalue[37,39]. Thisbehaviorcanbeexplained bylossaversionandmightinturnexplainwhypeoplepurchaseinsurancepoliciesforsmall risks.Giventhatthefearofvaccine-relatedadverseeventsisanimportantdeterminantofthe demandforvaccines,suchinsuranceschemescouldbestructuredasano-faultinsuranceprod- uct. Theseresultsthereforepredictthatceterisparibus,vaccineuptakewillbegreaterunder the insurance scheme than under an equivalent subsidy scheme. The US National Vaccine 28 InjuryCompensationProgram(VICP)forchildren’svaccines(AppendixA,SectionA.4.3.1)is anexampleofsuchno-faultinsuranceschemeforvaccine-relatedsideefects.VICPhasexisted intheUSsince1988.Despitetheprogram,bothUSadultandchildhoodimmunizationrates continuetobelow,suggestingthattheefectsofano-faultinsurancemightbemodestrelative tothatofasubsidyprogram.ThereareseveralpotentialreasonswhyVICPinitscurrentstate mightbeaweakinstrumentforincentivizingimmunization: First,theVICPcompensation amountsmightbeincommensuratetotheriskofsideefects;second,theVICPclaimsling andprocessingtimesmightbetoocostlytoplaintifs,andthird,consumersmightbeunaware oftheexistenceoftheprogram. Inordertoshedsomelightonsomeoftheseissuesandinformthedevelopmentandimple- mentation of efective policies for increasing vaccine coverage and protecting public health, wedevelopanexperimentthathelpsusanswerthreeimportantquestions:(i)whichtypesof nancialincentivesaremostecientatencouragingvaccination?(ii)whatistheefectofthe incentive size on vaccine uptake?, and (iii) which vaccine and disease characteristics should innovators target? In order to answer the rst question, we compare vaccine uptake under ano-faultinsuranceschemethatcompensatesconsumerssomeawardwhentheyexperience certain vaccine-related side efects, relative to uptake under a subsidy scheme that pays the expectedvalueoftheinsuranceatvaccinepurchase. Thissetupalsoallowsustoconductan indirecttestofexpectedutilitytheoryvsalternativetheoriesofbehavior(prospecttheoryand K-Rutilitytheory). Toanswerthesecondquestion,weexaminevaccineuptakeatdiferent compensationlevelsandacrossdiferentdiseaseandvaccinetypes.Thisallowsustodetermine theoptimalcompensationdesignandassesshowitvarieswithvaccineanddiseasecharacteris- tics.Finally,toanswerthelastquestion,wedevelopadiscretechoiceexperimenttoassessthe contributionsofdiseaseandvaccinecharacteristicstoindividual’schoiceofvaccination. We alsoinvestigatethetradeofstheymakebetweenvariousdiseaseandvaccineattributes. 29 This study hasseveral potential contributions. First, it could provide valuable information onconsumers’preferencesforvariousvaccineattributes,andcouldinforminnovatorsabout the need to develop varieties of vaccines that appeal most to consumers. Second, our nd- ingscouldinforminnovators,governmentsandotherpayersabouttherelativeeciencyof compensationandsubsidyschemesaimedatencouragingtheadoptionofnewandexisting vaccines. It would also inform public/private sector partnerships in the development and deliveryofthesetechnologies. Forexample,ndingsignicantdiferencesinthewillingness topaybetweenoralandinjectablevaccinesmaysuggestaroleforprivatesectorinbringing thesevarietiesinthemarket. Third,ourstudycouldhelpimproveourunderstandingofthe determinantsandheterogeneityofdemandfornewhealthtechnologies.Tothatend,itcould furtherinformmarketsegmentationstrategiesfortargetinginsuranceorsubsidiestofacilitate therapidadoptionofefectiveinfectiousdiseasevaccines. Finally,ourndingscouldinform thewelfareimplicationsofintroducingnewvaccinesbyforexamplesheddinglightonhow muchofthenetsurplusgainsaccruetoconsumersandproducers. 2.2 Theoreticalbasisforsmallincentivesandsmall-stakeinsur- anceschemes Standardeconomicanalysistypicallyassumesthatindividualsmakedecisionsfollowingthe tenetsofexpectedutilitytheory, meaningthatforadecisionoptionwithtwopossibleout- comesx 1 andx 2 withassociatedprobabilitiesofoccurrencepand (1−p)theagents’prefer- encesoverprospects [x 1 ,p;x 2 ]canberepresentedbytherelation: E{u(x 1 ,p;x 2 )} = V (p) =p·u(x 1 ) + (1−p)·u(x 2 ), (2.1) 30 whereu(x i )denotestheutilityoveroutcomex i .First,thisrelationimpliesthatu(x i )mustbe linearforsmall-stakesoutcomesovernalwealthstatesandpredictsanequivalencerelation- shipbetweeninsuranceanditsexpectedvalue[63]. Thetheoryofexpectedutilityovernal wealthstatesthereforepredictsthatpeoplewillbeapproximatelyrisk-neutralforarbitrarily small-stakesoutcomes 1 [3,63].Second,expectedutilitytheorypredictsthatthewayinwhich adecisionproblemisframedshouldnotafectindividuals’decisions. Animportantcharacteristicofthisexpectedutilityfunctionisitslinearityintheprobabili- ties,meaningthatV [α·p + (1−α)·p 0 ] = α·V (p) + (1−α)·V (p 0 ). Thelinearityof theexpectedutilityfunctionimpliesthattheagent’sindiferencecurvecanberepresentedby parallelstraightlines, becauseifV (p) = V (p 0 ), thenlinearityimpliesthatV (·)mustalso beconstantalongthelinejoiningptop 0 ,andthattheindiferencecurvesmustbeparallel.A secondimportantcharacteristicoftheexpectedutilityfunctionisthatitisamonotonicfunc- tion,apropertythatimpliesstochasticdominancepreference,meaningthatagentsaremade betterofbyincreasingtheprobabilityofhighpayofevents.Notethatthisisthecounterpart ofthenon-satiation(“moreisbetter”)implicationofthetraditionalutilityfunctions[44]. Thesepredictionsfromthetheoryarehoweverinconsistentwithmanyreal-worldbehaviors andhaveledtotheconclusionthatexpectedutilitytheoryovernalwealthstatesisill-suited formodelingcertainchoicebehaviors,becauseitisanormative,ratherthanadescriptiveorpre- dictivetheoryofchoices[37,38].Thetheory’srstprediction,whichderivesfromitsassump- tionthatpreferencesarelinearinprobabilitiesandthatriskattitudescomefromthecurva- ture of the utility-of-wealth function, is inconsistent with insurance purchase for inexpen- sivegoods[51,63,81]. Thesecondpredictionisalsoinconsistentwithndingsfromnumer- ousexperimentssinceTverskyandKahneman[77]’sstudywhichdemonstratedtheefectsof 1 Arrow[3]showsthatanexpectedutilitymaximizerwithadiferentiableutilityfunctionwillalwaysprefer asucientlysmall-stakeinanypositiveexpectedvaluebet,implyingthathe/shewillbearbitrariliyclosetorisk neutralwhenstakesarearbitrarilysmall. 31 framingonpeople’sdecisionsthroughitsefectonthereferencepointforevaluatingalterna- tivedecisions[22,39].Oneexplanationfortheimportanceofreferencepointislossaversion, whichisthetendencytoconsideralosstobemoreutility-reducingthanasimilargainwould increasetheirutility.Finally,thetheoryneitheraccommodatestheconsistentoverestimation ofindividuals’subjectiveprobabilitiesofrareeventsorverygoodorbadoutcomes,norcap- turestheirpropensitytooverweighttheoutcomesofsuchrareevents[24,41]. Inordertoaddressthesetheoreticallimitations,severalalternativenon-expectedutilitythe- oriesofdecision(e.g. prospecttheoryandtheKőszegi-Rabinutilitytheory)havebeenpro- posed.Prospecttheorypositsthatindividualsmakedecisionsbasedonthepotentialvalueof thelossesandgains–ratherthanthenaloutcome–ofthechoice,followingcertainheuris- tics[37,78]. Thetheorymakesthreekeyassumptionsaboutdecisions. First,itassumesthat individualsconsiderpossibleoutcomesrelativetoareferencepoint(usuallystatusquo[69]) ratherthantothenalstatus. Second,itassumesthatindividualsexhibitdiferentriskatti- tudes towards gains (outcomes above the reference point) and losses (outcomes below the referencepoint)inthattheycaremoreaboutpotentiallossesthanpotentialgains.Finally,the theory assumes that individuals overweight unlikely events, independently of their relative outcomes. TheKőszegi-Rabinutilitytheoryissimilartoprospecttheory,butpositsthatutilityisboth a function of consumption and an endogenous (with respect to the choice of the decision- maker)referencelevelforexpectationsoffutureoutcomes[39]. Prospecttheoryexplainsthedistortionsinprobabilitiesbyprobabilityweights, whicharise fromheuristics[76]andthepresenceofreferencepoints(impossibilityandcertainty)indeci- sionsinvolvingtworiskyprospects. TheKőszegi-Rabinutilitytheoryexplainstheoveresti- mationofprobabilitiesbythelawofsmallnumbers[16,64].Thesedistortionsreectindivid- uals’diminishingsensitivitytooutcomesthatarefartherawayfrombothreferencepointsand 32 explaininthecontextofvaccinationthegreaterfocusonvaccine-relatedadverseeventsthan ontheincreasedprobabilityofavoidinginfection: ononehand,refusalofvaccinationguar- anteesthattheindividualdoesnotexperiencevaccine-relatedsideefects,andindividualsare highlysensitivetothefactthatvaccinationmightslightlymovethemawayfromthisreference point. Ontheotherhand,vaccinerefusalexposestheindividualtothelikelihood(however smallitmightbe)ofinfectionandprematuredeath,andmightbeperceivedasamovefurther awayfrombothreferencepointshencemakingtheindividuallesssensitivetosuchchanges. Collectively,thesealternativedecisiontheoriestoexpectedutilitytheorypredictthatchoices insmall-stakegambleswilldeviatefromtheoptimalandrationalchoicepredictedbyexpected utility theory. In particular, they predict that individuals would prefer insurance to its expectedvalue: probabilityweightspredictthattheinsurancevaluewillexceeditsexpected value[18,62,81];loss-aversiononexpectedendowmentsalsopredictsapreferenceforinsur- anceoveritsexpectedvalue;andcertaintyefectspredictthattheeliminationofriskwillactas aparticularlypowerfulincentive.Thesepredictionsalsosuggestthatonewaytodesignefec- tiveincentiveschemesfordesirablebehaviorssuchasvaccinationmaybethroughinsurance, ratherthanthroughdirectsubsidies. 2.3 Methodsandstudydesign 2.3.1 Ethicsstatement ThisstudywasapprovedbytheUniversityofSouthernCaliforniaInstitutionalReviewBoard (IRB). 33 2.3.2 Studyparticipantsandrecruitment Participants for this study were recruited from Amazon’s Mechanical Turk (MTurk: http: //www.mturk.com). Since this study is investigating the US population’s valuation of vac- cines,weonlyrecruitUSresidentsintoourexperimentalsample,basedontheirownership ofaUSbankaccount,asdeterminedbyAmazon. Inordertoensuregoodresponserateand quality,andconsistentwithpriorstudies,weadditionallyrequiredthattheparticipantshave atleast90%taskapprovalratefortheirpreviousHumanIntelligenceTasks(HITs)[34,71]. Each participant in this study was asked to provide her/his informed consent. Participants who completed the survey tasks received $1.00 for approximately 8 minutes of their time. Thiscompensationrate($7.50perhour)exceededtheaverageMTurkHIThourlypaygrade ($1.40),themedianhourlywagefortasksperformedonMTurk($1.30),andthefederalhourly minimumwage($7.25)intheUS[31,59,79]. MTurkisincreasinglybeingusedbytheresearchcommunityinmanyelds,includingeco- nomics, psychology, and political science, for recruiting large samples of participants at rel- atively low costs for online surveys [47, 58]. The platform was created in 2005 initially as an online labor market that allows “requesters” to conveniently recruit large numbers of “workers”tocompletetasksthatarediculttoautomate[2,23,59].Inthismarket,workers haveaccesstoHITsthattheybrowsebytitle,keyword,reward,availability,andmanyother characteristics,andcompleteHITsthatinterestthem. Uponsuccessfulcompletionoftasks, therequestersthenpaytheworkers. OnMTurk, requesterscanalsodiscretionarilyrejecta worker’ssubmission,orassignthembonuses.Thisfeedbackmechanismensuresthatthetasks completedbyworkersareofrelativelyhighquality[47]. Studies of cross-sample investigations of the MTurk population demonstrate the diversity and representative nature of the MTurk population relative to in-person convenience sam- plesusedinavarietyofsocialsciencestudies[8,58,59,71,72]. Becauseofthesefeatures,the 34 MTurkplatformlendsitselfwelltoonlinesurveysandexperimentaldatacollection. How- ever,somehavedocumentedthattheMTurkpopulationmaynotbenaïve,alimitationthat canreduceefectsize[12,58]. Inordertomitigatethislimitation, wepreventsrespondents whoparticipatedinourpilotstudyfromparticipatingintheanalysissamplebyactivatingthe “PreventBallotBoxStung”featureintheQualtricssoftware. A second limitation of the MTurk sample is that it is not representative of the general US populationonsomecharacteristicssuchasage,gender,education,andincome.Forexample, inavalidationstudyoftheMTurksampleBerinskyetal.[5]comparedthecharacteristicsof MTurkrespondents(“turkers”)tothoseparticipatinginhigh-qualityin-personconvenience samples,Internetpanelsamples,andprobabilitysamplesusedintheCurrentPopulationSur- vey(CPS)andtheAmericanNationalElectionStudies(ANES).Theyfoundthatalthough MTurkrespondentsaregenerallymorerepresentativeoftheUSpopulationthanthoseinin- personconveniencesamples,theyarelessrepresentativeoftheUSpopulationthanthosein otherInternet-basedpanelsandnationalprobabilitysamples[5]. Inparticular, theyfound thatturkerswerepredominantlyWhite(84%)andfemale(60%),withanaverageof15years ofeducation. Turkerswerealsoyounger,withanaverageof32years,andmeanandmedian annualincomesof$55,000and$45,000respectively. Halfoftheturkersinthatsamplehad neverbeenmarriedandrentedahouse. ThemajorityoftherespondentsresidedintheMid- westandSouthoftheUS(22%,27%,30%and20%fortheNortheast,Midwest,South,and West,respectively)[5]. Huf and Tingley built on the Berinsky et al. [5] ndings to estimate joint distributions of severaldemographiccharacteristicsoftheMTurksample,usingdatafromofasimultaneous MTurkandCooperativeCongressionalElectionSurvey(CCES)survey,andhencecapturing similaritiesanddiferencesbetweenthetwosamplesatacommonpointintime. Consistent withtheCCESdistributions,theyfound75%ofturkerstobeWhite. However,theyfound 35 that the MTurk sample had 2-5% more Hispanics and Asians and, but 6% fewer African- AmericansthanCCES.TheyalsofoundtheMTurkparticipantstobeyoungerthanCCES, especiallyamongAsianmales[33]. HufandTingley[33]’sstudyalsoshowedthatthedistri- butionsoftheMTurkandCCESrespondentsaresimilarbyoccupationandrural-urbanloca- tion. Collectively,thesevalidationstudiesoftheMTurksamplesuggestthatMTurkremains areliablesamplingframeforvarioussocialsciencestudies[8]. 2.3.3 Experimentalprocedure Werandomizedparticipantswithequalprobability(1/2)intoaninsurance,orasubsidyarm. Respondentsintheinsurancearmareoferedvaccinevatapricep v aswellasano-faultinsur- ancethatpaysouttheamountY I (η v )iftheysuferandprovideproofoftheseverevaccine- related side efectsη v , which occurs with probabilityψ(η v ). Vaccinated individuals have a lower (but non-zero) risk ofρ d (1− φ v ) of becoming infected with the disease, whereρ d denotestheriskofinfectionwithoutvaccination,andφ v denotestheefectivenessofthevac- cine,sincethevaccinemightbeimperfect.Subjectsinthesubsidyarmareoferedvaccination ataprice (p v −Y S (η v )),whereY S (η v ) =ψ(η v )·Y I (η v )denotestheexpectedvalueofthe compensationamountoferedintheinsurancearm.Hence,thesubsidyschemeisanuncon- ditionaldiscountpaymentincentiveinthesensethatthesubjectreceivesitregardlessofthe side-efectoutcomesofvaccination. Noticethatinthisexperimentwedonotinsurevaccine failureduetolackofecacy,assuchaschemewouldonlyworkwellforhighlyefectivevac- cines, andcouldbeverydiculttoadminister. Thisexperimentaldesignyieldsatotalof2 experimentalgroupsacross2compensationschemes. WesummarizethedesigninFigure2.1. Respondentsineachexperimentalgrouparethenoferedchoiceprolesdevelopedusinga discretechoiceexperiment(DCE)design,asdescribedbelow. 36 Figure 2.1: Expected values of the compensation amounts under the subsidy and insurance schemes. Note:Respondentsareassignedtothesubsidy(S)orinsurance(I)armswithequalprobabilities1/2,andarepre- sentedchoiceprolesconstructedfromtheDCEexperiment.Theycanchoosetovaccinate(v)ornottovaccinate (v).ThosewhodecidetovaccinatecanreceivethecompensationamountsY S (η v ),orY I (η v )withprobabilities 1 andψ(η v ) under the subsidy, and insurance programs respectively, whereψ(η v ) denotes the probability of experiencingvaccineinjuryη v ,andY S (η v ) =ψ(η v )·Y I (η v )istheexpectedvalueofthecompensationforthe insuranceprogram. 2.3.4 Discretechoiceexperiment(DCE) ThisstudyfollowstheDCEmethodtoelicitconsumers’preferencesandwillingnesstopay forvariousvaccineattributes. DCEisamulti-attributeapproachwidelyusedinmanyelds, includingmarketing,transportation,andenvironmentalresourceeconomics,andpublicwel- fareanalysisthathasrecentlygainedtractioninhealthcare.Specically,DCEisaquantitative choice-based method for eliciting preferences that is theoretically rooted in random utility theory(RUT),abehavioraltheoryofchoice. ThetheorycanbetracedbacktoThurstone’s theoryofpairedcomparisonswhichwaslatergeneralizedbyMcFaddentomultiplecompar- isons[43,48–50,62,75].Thetheorypositsthatindividuals’choices(orpreferences)aregov- ernedbyalatentconstruct(i.e.“utility”),whichtheyseektomaximizebychoosingthealterna- tivethatyieldsthemthehighestutility.Assuch,thepredictionfromDCEarealsoconsistent 37 with economic demand theory [43]. In practice, DCE consists of presenting choice sets to individualsinturn,andaskingthemtostatetheirmostpreferredoption(alsoknownasalter- native)ineachchoiceset.EachDCEchoicesethastwoormorealternatives(e.g.hypothetical scenarios,goods,services),andeachalternativeisinturndescribedbyasetofattributesthat cancontainoneormorelevels[9,43,74]. DCEisamoreattractiveapproachthanotherstatedpreference(SP)techniques(e.g.ranking orratingofalternatives)becauseofitstheoreticalgroundingsinRUT.Contingentanalysis (CA)isanotherpreference-basedmethodrootedinconjointmeasurementtheory,butisill- suited for economic valuation because it cannot inform us about preference processes like DCEdoes[43]. DCEalsoaccommodatesreasonablystraightforwardtasksandavoidssome responsedicultiesencounteredwithcontingentvaluation(CV)approaches.Forexample,in DCE,taskscanbedesignedtomorecloselyresembleareal-worlddecision.Finally,andunlike inCV,theDCEapproachallowsforseparateestimationsoftheattributevalues,inaddition totheestimationofthetotalvalueofthegood. 2.3.5 DevelopmentoftheDCEquestionnaire In order to identify the most important vaccine and disease attributes that inuence vacci- nation decision, we rst surveyed the published literature. Using the information gathered from the literature review, we developed disease scenarios and identied important vaccine attributes. Second,weconsultedinfectiousdiseasecliniciansinordertounderstandthefac- torsthatinuencevaccinationdecision. Thesetwostepsallowedustoavoidtheoreticalmis- specicationofthescenariosintheDCEquestionnaire.Third,weconductedtwofocusgroup discussionsessionswithgraduatestudentsandinfectiousdiseasecliniciansfamiliarwithinfec- tious disease and vaccination issues, in order to determine which attributes to exclude or 38 includeinthestudy,andensurethatwecapturedallrelevantattributesandinformationonrel- evantaspectsofthediseasescenariosandvaccineattributesinourDCEdesignandintheques- tionnaire(AppendixA,SectionA.5forthefocusgroupinstrument). Therstfocusgroup discussionsessionconsistedof3graduatestudentsfromtheUniversityofSouthernCalifor- nia,andthesecondfocusgroupdiscussionsessionincluded3clinicians. Basedontheinputs fromthefocusgroupdiscussionsessions,werevisedtheDCEquestionnaire.Figure2.2sum- marizesourgeneralapproachtothedevelopmentoftheDCEquestionnaire. Basedonthis approach,weretainedforthisstudythedisease,vaccineandcompensationprogramattributes andattributelevelssummarizedinTable2.1. Figure2.2:ApproachtothedevelopmentoftheDCEquestionnaire. 39 Table2.1:DCEattributesandattributelevels. Attributes Levels References Disease Diseasetype,d a Acute;Chronic TableA.1 Transmission,γ d b Directcontact;Vector-borne;Airborne TableA.1 Riskofinfection,ρ d c 1outof1millionpeople TableA.1 30outof1millionpeople 20,000outof1millionpeople Severesymptoms,η d d Life-threateningcomplication TableA.1 Permanenthandicap Death Riskofseveresymptom,ψ(η d ) e 30outof100 TableA.1 2outof1000 1outof1million Vaccine Modeofadministration,α v f Needleinjection(1shotinthearm) TableA.3 Intranasal(sprayinsidethenose) Oralsolution(1liquiddropletinthemouth) Efectiveness,φ v g 70%;95%;99% TableA.3 Protectionduration,τ v h 1year;6years;Lifetime TableA.3 Severevaccinesideefect,η v i Life-threateningallergicreactions. TableA.3 Severeneurologicalinjurythatcanbecomepermanent(e.g.muscleparalysis). Death. Riskofseveresideefect,ψ(η v ) j 10outof1millionvaccinatedpeople TableA.3 4outof1millionvaccinatedpeople 1outof1millionvaccinatedpeople Continued... 40 Table2.1:DCEattributesandattributelevels. Attributes Levels References Out-of-pocketcost,p v k $0,$100;$700 TableA.3 Compensation Compensationamount,Y I (η v ) l Life-threateningallergicreaction:$30,00;$40,000;$60,000 Calculated:TableA.4 Severeneurologicalinjury:$1,000,000;$2,500,000;3,500,000 Deathfrominjury:$1,250,000;$5,000,000;$10,000,000 Notes: Thediseaseattributesaredenedtobealternative-invariant. Thevaccineandcompensationattributesaredenedtobealternative-varying,meaningthattheycan takedistinctvaluesacrossalternatives.Theirvaluesfortheopt-outoption(“Novaccination”)aredenotedby“n.a.”or“$0”.Foreachchoiceprole,respondentsareinformed thatbothvaccinesmightcausemildsideefectsthatdonotinterferewithnormalactivities,suchasmildinjectionsitesoreness,swelling,orrednessandmildfever,rash,or achiness[11]. AllriskprobabilitiesarepresentedtotherespondentinawaythatisconsistentwithCDCandothergovernmentagencies’reportingstyle. a Denedasthecourseanddurationofillness.Anacutediseaseisdenedasonethatdevelopsrapidlybutlastsashorttime(i.e.within3weeks);achronicdiseaseisdenedas onethatdevelopsmoreslowlyandhassymptomsthatarecontinualorrecurrentforlongperiods. b Denedasthewayinwhichpeoplebecomeinfectedwiththedisease. c Denedastheproportionofpopulationinfectedwiththenewdisease,i.e. havingdiseasesymptoms. d Denedastheseverityofthediseasesymptomsafterinfectionoccurs. e Denedastheproportionofinfectedpopulationthatsufersseveresymptoms,includingdeath. f Denedasthewayinwhichthesubjectwillreceivethevaccine. g Denedastheproportionofpeoplewhowillbeprotectedagainstthediseasewhenvaccinated. h Denedasthenumberofyearsduringwhichthevaccinewillprotectthesubjectagainstthedisease. i Denedassideefectsthatarelife-threateningorresultininpatient hospitalization,surgicalintervention,ordeath.Severesideefectsanddeathareonlyduetothesafetyofthevaccine,butnottoitsecacy,anddonotmakethevaccineless ecaciousagainstthedisease. j Denedastheproportionofvaccinatedpeoplewhosuferseverevaccinesideefects. k Denedastheamountpaidout-of-pocketbytherespondenttoreceivethevaccine.Forexample,foravaccinethatprovidesprotectionfor1yearandhasanout-of-pocket costof$700,thesubjectwouldpay$700eachyeartoreceivethevaccine. l Denedasthedollaramountofthecompensation;thisamountvarieswiththeinjuryseverity. Thelevelsofthisattributearealternative-invariant.FortheVICPprogram,theattributelevelsforeachinjuryaresettothesecondlevel. 41 2.3.6 DCEdesign OurDCEdesignconsistedof6vaccineattributes(eachwith3levels),5diseaseattributes(one with2levelsand4with3levelseach),andacompensationprogramattributewith3levels.If weweretorepresentallcombinationsoftheseattributelevelsinchoicesets,wewouldhave toconsider729(3 6 )hypotheticalvaccinealternativesfor162(2 1 · 3 4 )diseasescenariosand3 compensationalternatives,yielding 354, 294possiblevaccinationchoicealternatives. Given thatitisnotfeasibletopresentasingleindividualwithallthesescenarioalternatives(i.e.full factorialdesign),wegeneratedasubsetof36choicealternatives(i.e.fractionalfactorialdesign) usinganecientdesignthatmaximizesD-eciency,byassumingamodelwithzeropriorson the coecients [65] (see Appendix A, Section A.3.2.1). We constructed the ecient choice design using several macros (%MktRuns, %MktLab, %MktEx, %ChoicEf, and %MktEval) in SAS Studio 3.2 [70]. With this design, all main efects and some two-way interactions between attributes could be estimated. Given that presenting a single respondent with all 36 choice sets could result in lower response rate and/or lower response reliability [25], we reducedtheburdenonrespondentsbyusingablockeddesignanddividedthe36choicesets of the ecient fractional factorial design into 6 questionnaire versions containing 6 choice proleseach,inwhichweensuredsucientvariationintheattributelevelsbyndingblocks withnearattributelevelbalance[28,30,52]. Sinceindividualsarenotobligatedtoutilizeavaccineinreallifescenarios,weincludedan“opt- out”alternative(i.e. “Novaccination”)ineachchoiceprole. Priorstudiessuggestthatthe inclusionandlocationofthe“opt-out”optioninthechoiceproleresultsinsmalldiferences relativetotheforcedchoicemodel[80].However,somehavealsoraisedtheconcernthatthe inclusionofsuch“opt-out”optioninDCEtaskscouldinduceastatusquobias,whichcould inturnhaveprofoundimpactsonwelfaremeasures[1,26,68,69]. Forexample,Boxalletal. [7] recently showed that increasing the complexity of choice tasks also increased the choice 42 ofthestatusquooptionasaheuristicstrategy[7]. Despitetheseconcerns,weoptedforthe inclusionofthe“opt-out”optionasathirdalternativeineachchoiceprole,inordertomake ourchoicescenariosmorerealistic. Hence,eachchoiceproleinthisstudyconsistsofthree choicealternatives,namelytwounlabeledvaccinationalternatives(“VaccineA”and“Vaccine B”),andoneopt-outoption(“Novaccination”). Toavoidconfusinginterpretationsofthis option,allattributelevelsfortheopt-outoptionwerelabeledas“n.a.”or“$0”. Inordertoassesschoiceconsistencyforeachrespondent,werepeatedonequestionineach questionnaireversion.Wealsoincludedanattentioncheckchoiceprole,leadingtoatotalof 8questionsperrespondent[32]. Tominimizetheriskofbiasinducedbytheorderofques- tions, we randomized the questions in each questionnaire version such that the duplicated questionsdidnotfolloweachother. TheDCEsurveyinstrumentwithdetailsontheinformationpresentedtoeachrespondentis providedinappendixAppendixA,SectionA.6.First,wepresentedascreenwiththeinformed consent form. Upon consenting to participate in the study, the respondent was rst pre- sentedamixofsocio-demographicquestionsandquestionsabouttheirattitudestowardsvac- cination. Second,theywerepresentedanexampleofaDCEchoiceprole,beforereceiving thesequenceof8DCEchoiceproles. EachDCEchoiceproleconsistedofaninstruction promptfollowedbythetextofthediseasescenario. Thedescriptionofthediseasescenario diferedslightlybetweentheinsuranceandsubsidyarmstoreectthecompensationprogram. Thediseasescenariowasfollowedbyatableofthediseasecharacteristics,andanothertable ofthevaccineandcompensationattributelevelsforeachchoicealternative. Anexampleof a choice prole is presented in Figure A.2. Upon completion of the DCE choice proles, respondents who selected “No vaccination” on all 6 required DCE questions that allowed aresponse(i.e. theprolethatwasnotdesignedtotestattention)werepresented2follow- upquestionstodeterminetheminimumcompensationamounttheywerewillingtoaccept 43 inordertovaccinateaswellasthemaximumriskofdeathfromvaccinationtheywerewill- ingtotolerate. Third, respondentswerepresentedanothersequenceofsocio-demographic questionsandquestionsabouttheirattitudestowardsvaccination. Finally, theywereasked fortheirimpressionsandcommentsaboutthesurvey,aswellasdiseaseandvaccineattributes theymostconsideredintheirdecisions. 2.3.7 Pilottest Before elding the survey for data collection, we conducted two pilot studies of the ques- tionnaireamongasubgroup(n=90)oftheMTurkpopulation,inordertoensurethatthe wordingofthequestionswascorrectandwasunderstoodthesamewaybyallrespondents. Based on responses from these pilot studies, the follow-up questions for respondents who alwayschosetheopt-outoptionwereslightlymodiedtoincludetheseverityofthevaccine sideefect. Thepilottestofthequestionnairealsoprovidedussomepriorsaboutthevalue of the DCE coecients , which we used to estimate the minimum sample size required for detectingsignicantefectsizesontheDCEcoecients. 2.3.8 Samplesizecalculation Inareviewofhealth-relatedconjointanalysisstudies,Marshalletal.[46]reportedanaverage samplesizeof259participantsforDCEstudiespublishedbetween2005and2010.Nearly40% ofthesestudieshadsamplesizesrangingbetween100and300participants.deBekker-Grob et al. [15] reported that there were substantial variations in the sample sizes used in health- relatedDCEstudiespublishedin2012. Forexample, theyfoundthatamong32%ofthe69 studiesincludedintheirreviewhadsamplesizessmallerthan100respondents,while23%had samplesizesexceeding600respondents[15]. Untilrecently,andduetothelackofadequate statistical method for estimating sample sizes in DCEs, most researchers relied on rules of thumbtoestimatetheminimumrequiredsamplesizeinstudies[56].Forexample, Pearmain 44 etal.[60]suggestedthatsamplesizesover100respondentsaresucienttomodelpreference data.LancsarandLouviere[40]howeverreportedthattheir“[...]empiricalexperienceisthat onerarelyrequiresmorethan20respondentsperquestionnaireversiontoestimatereliable models,butundertakingsignicantposthocanalysistoidentifyandestimatecovariateefects invariablyrequireslargersamplesize”. JohnsonandOrme[36]suggestedthattheminimum samplesizerequiredforestimatingthemainefectsshouldbecalculatedasfollows: N > 500c/(t×a), (2.2) wheret denotes the number of choice tasks, a denotes the number of alternatives, andc represents the number of analysis cells. The parameterc is the largest number of levels of anyattributesinamainefectsanalysis;itisequaltothelargestproductoflevelsofanytwo attributeswhenconsideringalltwo-wayinteractions[15,36,55]. Generally,theserulesrecommendatleast300respondentsinDCEstudies,inadditiontocon- siderationsregardingthenumberofalternatives. Followingthisapproach,Determannetal. [17] estimated a required sample size of at least 500 respondents, and recruited 536 partici- pants. Inthisstudy, wefollowtheapproachindeBekker-Grobetal.[15]todeterminethe minimumsamplesizerequiredtotestspecichypothesesfortheDCEcoecients[15,57,66]. Wespecifytheminimumrequiredsamplesize(N)fortheestimatedcoecientsinourDCE experimenttosatisfythefollowingcondition: N >   (z 1−β +z 1−α ) s Σ γk δ   2 , (2.3) whereδ,α, and (1−β)denoterespectivelytheefectsize, signicancelevel, andstatistical powerlevel,and Σ γk denotesthek th diagonalelementoftheasymptoticvarianceoftheesti- matedparameters, anddependsonthestatisticalmodelusedintheestimationoftheDCE 45 coecients(e.g. multinomiallogit,mixedlogit,orgeneralizedmultinomiallogit),thepriors abouttheparameterestimates,andtheDCEdesign[15]. Inourstudy,thepriorsaboutthe parameterestimateswereobtainedfromthepilotstudy. Giventhatweaimedtoalsoassesstheefectsofinsurancevssubsidyonvaccineuptake,we conductedanadditionalpowercalculationforthetwogroupcomparison.Basedonourpriors abouttheefectsizeofinsurancefromthepilotdata(efectsizeof 0.003andstandarderror of 0.0353),weestimatedthattheminimumsamplesizerequiredtodetectasignicantefect at the 95% signicance level with 80% power was 1, 085 respondents. Based on the high surveycompletionrates(91.6%)reportedinpriorstudiesusingtheMTurksample,weopted torecruitatleast 1, 200participantsinthestudy[8,19,47]. 2.4 Analyses Our estimation approach follows the random utility maximization framework. The utility thatindividualnderivesfromalternativejisgivenby U nj = V nj + nj , (2.4) whereV nj = V (x nj ,z n )denotesthesystematiccomponentoftheutilityandisdenedto beafunctionofobservableattributesx nj ofthealternatives,andthecharacteristicsz n ofthe individual.V nj isusuallyassumedtobelinearoftheform V nj =x 0 nj β +z 0 n γ j , (2.5) 46 whereβandγ j denoterespectivelythecontributionsoftheattributesandthoseoftheindivid- ualcharacteristicstotheoverallutilityfromchoosingthealternative.Further,theprobability thatindividualnchoosesalternativejisgivenby P nj = P ( ni − nj <V nj −V ni ),∀j6=i. (2.6) Basedonthisframework,weassumedthateachindividual’sutilityfromvaccinationisafunc- tionofaconstantterm(intercept)reectinghis/herrelativepreferenceforvaccinationrelative tonovaccination,adummyvariableindicatingtheassignmenttotheinsurancevs. subsidy arm,aswellasavectorofdiseaseandvaccineattributes,thecompensationamount,andindi- vidualcharacteristics. Thesignsofthecoecientsindicatewhethertheassociatedattributes havepositiveornegativeefectsonutility,andthevaluesofthecoecientscapturetherelative importanceoftheparticularattributes. Inourestimationofthemodel,wetreateachchoice alternativeinthechoicesetasanobservation,andweestimateaconditionallogitregression model.Wealsoassumethatallattributeshadanindependentefect. Inordertodeterminewhethertheefectofinsurancevariedaccordingtocertainindividual, disease,orvaccinecharacteristicsorthecompensationamount,wealsoestimateinteraction models. Wealsoconductsubgroupanalysesinordertoassesswhetherrespondents’prefer- encesdiferbycertaincharacteristics. Inouranalysis,weexcludedatafromrespondentswhofailedtheattentioncheckquestions. Werstconductananalysiswiththefullsample,andthenwithsampleofrespondentswith consistentresponses. 47 2.4.1 Willingnesstopayandwillingnesstoaccept OneoftheobjectivesofDCEistohelpunderstandhowindividualsmaketradeofsbetween diferentattributes. Wethereforecalculatewillingnesstopaybyintegratingtheareabeneath theestimateddemandcurveasfollows: E (WTP k ) = − E (β k ) β p (2.7) whereβ p denotesthecoecientoftheprice(out-of-pocketcostofvaccine)attribute,which isgenerallyassumedtobeaxedparameter. 2.5 Results 2.5.1 Surveyparticipation WebeganeldingthesurveyonMarch15,2016andcompletedthedatacollectiononMarch 18,2016.Atotalof1,368respondentsprovidedaconsenttoparticipateinthissurveybut1,294 actuallycompletedandsubmittedthesurvey,yieldinganoverallresponserateof95%.Ofthis sampleofrespondentswhocompletedthesurvey,5respondents(0.39%)skippedatleast1of the 7 required DCE questions and were excluded from the analysis sample; 32 respondents (2.5%)failedtheattentioncheckquestionsandwerealsoexcludedfromtheanalysissample. Thisfailurerateisslightlybelowthe4.2%ratereportedinPaolaccietal.[59]andinDupuis et al. [19]. These exclusion criteria left us with a total of 1,257 respondents for the analysis. Thissamplerepresents92%ofthosewhoprovidedconsenttoparticipate,and97%ofthose who completed the survey. These completion rates are in line with 91.6% completion rates reportedinpriorstudies[8,19,47]. 48 2.5.2 Characteristicsoftherespondents The characteristics of the respondents are presented in Table 2.2. Our sample had more femalesthanmales(54.7%vs44.9%)andconsistedprimarilyofwhites(76.6%),followedby black(7.1%)andHispanics(3.6%). Roughly4.2%oftherespondentswerebi-racialormulti- racial.Thesamplewasalsorelativelyyoung,with70.7%ofrespondentsfallinginthe26to54 yearsoldagebracket. 18.4%oftherespondentswerebetween18and25yearsold. Asigni- cantshareofthesample(64.6%)hadreceivedsomecollegeeducation(24.0%)orcompleted theirbachelor’sdegree(40.6%). 12%oftherespondentshadcompletedagraduateorprofes- sionaldegree. 46.0%oftherespondentsweresingle(nevermarried)whereas39.9%ofthem weremarried.Themajorityoftherespondents(65.4%)hadnochildyoungerthan16yearsin theirhousehold.32.6%oftherespondentsreportedearninglessthan$50,000,butmorethan $25,000 during the previous tax season. 21.2% of the sample earned between $50,000 and $75,000whereas20.4%reportedearninglessthan$25,000. 62%oftherespondentsworked intheprivatesector. ThemajorityofthesampleresidedintheNortheast,Southeast,andin theWest.Theyalsolivedforthemostpartinasuburbanarea(52.5%). Mostrespondentshadaprivateinsurancecoverage(57.8%),butaconsiderablefraction(12.4%) stillremaineduninsured,while12.2%receivedcoveragethroughMedicaid,indicatingthesub- stantialdegreeofpovertyand/ordisabilityamongtherespondentsintheanalysissample.The respondentswererelativelyhealthy: 83.2%statedthattheywereingood,verygoodorexcel- lenthealth.9.1%oftherespondentsstatedhavingexperiencedorknowingsomeonewhohas experiencedseverevaccine-relatedsideefects.Intermsofhealthbehaviors,41.4%ofthesam- plestatedhavingreceivedauvaccineoverthepast12months,asharewhichissignicantly belowtheHealthyPeople2020targetlevelof80%[27].6.7%oftherespondentsbelievedthat vaccinescausedautism,andanother28.1%believednaturalproductstobebetteralternatives 49 tovaccinesandmodernmedicine. Themajorityofthesample(97.5%)knewnothingtolit- tleaboutVICP.Finally, thevastmajorityofrespondentsindicatedreceivingvaccine-related informationfromhealthocials(∼ 60%). Respondentsintheinsurancearmdidnotsignicantlydiferfromthoseinthesubsidyarm,as evidencedbytheinsignicantp-valuesonnearlyallcharacteristics.However,thetwogroups appearedtobeslightlyimbalancedwithrespecttoeducationlevelandtheemploymentsector oftherespondents.Forexample,morerespondentsintheinsurancearmhadreceivedagrad- uate, professional, or doctorate degree than those in the subsidy arm (15.5% vs 10.3%) while morerespondentshadachievedonlyabachelor’sdegreeinthesubsidyarmthanintheinsur- ancearm(44.6%vs36.5%). Withrespecttothesectorofemployment,thesubsidyarmhad morerespondentsworkinginthenon-protsectorthantheircounterpartsintheinsurance group(7.6%vs4.9%,respectively). Table2.2:Characteristicsoftherespondents. Variable Fullsample, n(%) Subsidy, n(%) Insurance, n(%) P-value a Samplesize 1,257 630 627 Gender 0.570 Female 688(54.7%) 345(54.8%) 343(54.7%) Male 564(44.9%) 282(44.8%) 282(45.0%) Other 4(0.3%) 3(0.5%) 1(0.2%) Don’tknow/Refusetotell 1(0.1%) 0(0.0%) 1(0.2%) Race 0.370 Black 89(7.1%) 51(8.1%) 38(6.1%) White 963(76.6%) 474(75.2%) 489(78.0%) Hispanic 45(3.6%) 24(3.8%) 21(3.3%) Multiracial 53(4.2%) 28(4.4%) 25(4.0%) Other 101(8.0%) 48(7.6%) 53(8.5%) Don’tknow/Refusetotell 6(0.5%) 5(0.8%) 1(0.2%) Age 0.880 18-25 231(18.4%) 113(17.9%) 118(18.8%) 26-34 460(36.6%) 225(35.7%) 235(37.5%) 35-54 429(34.1%) 217(34.4%) 212(33.8%) 55-64 107(8.5%) 58(9.2%) 49(7.8%) 65ormore 27(2.1%) 15(2.4%) 12(1.9%) Continued... 50 Table2.2:Characteristicsoftherespondents. Variable Fullsample, n(%) Subsidy, n(%) Insurance, n(%) P-value a Don’tknow/Refusetotell 3(0.2%) 2(0.3%) 1(0.2%) Education 0.067 Lessthanhighschool 10(0.8%) 5(0.8%) 5(0.8%) Highschoolgraduate 125(9.9%) 61(9.7%) 64(10.2%) Associate’sdegree 135(10.7%) 63(10.0%) 72(11.5%) Somecollege,nodegree 302(24.0%) 148(23.5%) 154(24.6%) Bachelor’sdegree 510(40.6%) 281(44.6%) 229(36.5%) Graduateorprofessionaldegree 151(12.0%) 62(9.8%) 89(14.2%) Doctoraldegree 11(0.9%) 3(0.5%) 8(1.3%) Don’tknow/Refusetotell 13(1.0%) 7(1.1%) 6(1.0%) Maritalstatus 0.470 Single 578(46.0%) 295(46.8%) 283(45.1%) Married 501(39.9%) 245(38.9%) 256(40.8%) Separated/Divorced 126(10.0%) 64(10.2%) 62(9.9%) Widowed 16(1.3%) 11(1.7%) 5(0.8%) Other 29(2.3%) 11(1.7%) 18(2.9%) Don’tknow/Refusetotell 7(0.6%) 4(0.6%) 3(0.5%) Children≤ 16years 0.970 0 822(65.4%) 412(65.4%) 410(65.4%) 1 209(16.6%) 103(16.3%) 106(16.9%) 2 132(10.5%) 65(10.3%) 67(10.7%) 3ormore 87(6.9%) 46(7.3%) 41(6.5%) Don’tknow/Refusetotell 7(0.6%) 4(0.6%) 3(0.5%) Householdincome 0.110 <$25,000 257(20.4%) 139(22.1%) 118(18.8%) $25,000-$49,999 410(32.6%) 194(30.8%) 216(34.4%) $50,000-$74,999 266(21.2%) 149(23.7%) 117(18.7%) $75,000-$99,999 164(13.0%) 79(12.5%) 85(13.6%) $100,000-$149,999 95(7.6%) 39(6.2%) 56(8.9%) $150,000ormore 34(2.7%) 16(2.5%) 18(2.9%) Don’tknow/Refusetotell 31(2.4%) 14(2.2%) 17(2.8%) Sectorofemployment 0.120 Public 172(13.7%) 76(12.1%) 96(15.3%) Private 787(62.6%) 401(63.7%) 386(61.6%) Non-prot 79(6.3%) 48(7.6%) 31(4.9%) Other 127(10.1%) 58(9.2%) 69(11.0%) Don’tknow/Refusetotell 92(7.3%) 47(7.5%) 45(7.2%) Regionofresidency 0.700 Northeast 270(21.5%) 130(20.6%) 140(22.3%) Southeast 324(25.8%) 172(27.3%) 152(24.2%) West 279(22.2%) 137(21.7%) 142(22.6%) Midwest 250(19.9%) 124(19.7%) 126(20.1%) Continued... 51 Table2.2:Characteristicsoftherespondents. Variable Fullsample, n(%) Subsidy, n(%) Insurance, n(%) P-value a Southwest 124(9.9%) 60(9.5%) 64(10.2%) Other 2(0.2%) 1(0.2%) 1(0.2%) Don’tknow/Refusetotell 8(0.6%) 6(1.0%) 2(0.3%) Areaofresidency 0.340 Urban 327(26.0%) 162(25.7%) 165(26.3%) Suburban 660(52.5%) 333(52.9%) 327(52.2%) Rural 251(20.0%) 127(20.2%) 124(19.8%) Other 5(0.4%) 4(0.6%) 1(0.2%) Don’tknow/Refusetotell 14(1.1%) 4(0.6%) 10(1.6%) Healthstatus 0.290 Excellent/Verygood/Good 1,046(83.2%) 514(81.6%) 532(84.8%) Fair/Poor 206(16.4%) 113(17.9%) 93(14.8%) Don’tknow/Refusetotell 5(0.4%) 3(0.5%) 2(0.3%) Insurancetype 0.930 Commercial/privateinsurance 727(57.8%) 363(57.6%) 364(58.1%) Medicare 97(7.7%) 45(7.1%) 52(8.3%) Medicaid 153(12.2%) 76(12.1%) 77(12.3%) Militaryinsurance 30(2.4%) 17(2.7%) 13(2.1%) Uninsured 156(12.4%) 83(13.2%) 73(11.6%) Other 27(2.1%) 14(2.2%) 13(2.1%) Don’tknow/Refusetotell 67(5.3%) 32(5.1%) 35(5.6%) Receiveduvaccine 0.360 Yes 521(41.4%) 273(43.3%) 248(39.6%) No 722(57.4%) 351(55.7%) 371(59.2%) Don’tknow/Refusetotell 14(1.1%) 6(1.0%) 8(1.3%) Suferedvaccinesideefect 0.410 Yes 115(9.1%) 64(10.2%) 51(8.1%) No 1,008(80.2%) 497(78.9%) 511(81.5%) Don’tknow/Refusetotell 134(10.6%) 69(11.0%) 65(10.4%) Vaccinescauseautism 0.400 Agree/Stronglyagree 84(6.7%) 48(7.6%) 36(5.7%) Neutral/Don’tknow 333(26.5%) 162(25.7%) 171(27.3%) Disagree/Stronglydisagree 839(66.7%) 419(66.5%) 420(67.0%) Refusetotell 1(0.1%) 1(0.2%) 0(0.0%) Naturalproductssuperiortovaccine 0.270 Agree/Stronglyagree 274(21.8%) 131(20.8%) 143(22.8%) Neutral/Don’tknow 283(22.5%) 147(23.3%) 136(21.7%) Disagree/Stronglydisagree 697(55.4%) 349(55.4%) 348(55.5%) Refusetotell 3(0.2%) 3(0.5%) 0(0.0%) HeardofVICP 0.940 Alot 26(2.1%) 14(2.2%) 12(1.9%) Continued... 52 Table2.2:Characteristicsoftherespondents. Variable Fullsample, n(%) Subsidy, n(%) Insurance, n(%) P-value a Alittle 253(20.1%) 125(19.8%) 128(20.4%) Nothingatall/Don’tknow 973(77.4%) 489(77.6%) 484(77.2%) Refusetotell 5(0.4%) 2(0.3%) 3(0.5%) Sourceofvaccinationinformation 0.880 Electedleaders 31(2.5%) 16(2.5%) 15(2.4%) Familyandfriends 3(0.2%) 2(0.3%) 1(0.2%) Healthocials 748(59.5%) 370(58.7%) 378(60.3%) Media 119(9.5%) 57(9.0%) 62(9.9%) Others 356(28.3%) 185(29.4%) 171(27.3%) Religious 0.620 Very 214(17.0%) 116(18.4%) 98(15.6%) Somewhat 512(40.7%) 245(38.9%) 267(42.6%) Notatall 495(39.4%) 250(39.7%) 245(39.1%) Other 15(1.2%) 8(1.3%) 7(1.1%) Don’tknow/Refusetotell 21(1.7%) 11(1.7%) 10(1.6%) a Signicanceofthediferencebetweentheinsuranceandsubsidygroups(Pearson’schi-squaredtest). 2.5.3 DCEresults Themajorityofrespondentsindicatedthatthesurveywasclearandthattheyexperiencedno dicultyincompletingthesurvey. 2.5.3.1 Choiceconsistency We assessed the consistency of the respondents’ choices by examining the proportion of respondents who provided the same answers to the duplicated choice sets. Overall, 77.9% oftherespondentsprovidedaconsistentresponsetotheduplicatedquestions. Choicecon- sistencydidnotdiferbetweenthesubsidyandtheinsurancegroups(77.1%vs78.6%,respec- tively;χ 2 = 0.403,p = 0.526). 53 2.5.3.2 Vaccinationchoice No respondent refused vaccination on all 6 required DCE questions (excluding the dupli- catedconsistencycheckquestion). Inordertodetermineefectivepublichealthpoliciesfor increasingvaccineuptake,weexaminetheefectofinsuranceonwillingnesstovaccinate.Fig- ure2.3depictsthemarginalefectsofinsuranceontheprobabilityofvaccinationunderdif- ferentmodelspecicationsandindiferentsubsamples.Theunconditionalmarginalefectof insuranceontheprobabilityofvaccinationwaspositivebutstatisticallyinsignicantinthe fullsampleandinthesampleofconsistentrespondents(0.004;95%condenceinterval,CI [-0.004,0.011]vs0.006,CI[-0.002,0.014]). The conditional marginal efect of insurance on vaccine uptake was generally larger among non-femalerespondentscomparedtofemalerespondents.Inthefullsampleofrespondents, non-femalerespondentsintheinsurancearmwere1.1times(1.107,CI[1.004,1.210])aslikelyto vaccinateasanaveragefemalerespondentintheinsurancearm(0.109,CI[0.100,0.118]vs0.121, CI[0.111,0.131];χ 2 = 4.52,p = 0.034);thecorrespondingriskratioamongrespondentsin thesubsidyarmwas0.934(CI[0.846,1.022]). Comparedtosubsidy,themarginalefectsof insuranceontheprobabilitiesofvaccinationwasgreater(0.019,CI[0.004,0.034];χ 2 = 6.48, p = 0.011)amongnon-femalerespondentsthanfemalerespondents;themarginalefectsof insuranceonvaccineuptakeamongthetwogroupswere0.015(CI[0.004,0.026])and-0.005 (CI[-0.014,0.005]),respectively(seeFigure2.3). Amongconsistentrespondentsintheinsurancearm,non-femalerespondentswereagain1.1 times(1.091,CI[0.976,1.207])aslikelyasfemalerespondentstoacceptvaccination(0.117,CI [0.105,0.128]vs0.107,CI[0.097,0.117];χ 2 = 2.60,p = 0.107).Thecorrespondingriskfac- toramongrespondentsinthesubsidyarmwas0.918(CI[0.819,1.018]).Relativetothesubsidy arm,insuranceledtoa1.5percentagepointincreaseintheprobabilityofvaccinationamong 54 non-femalerespondents, buthadnosignicantefectamongfemalerespondents(0.015, CI [0.003,0.027]vs-0.003,CI[-0.014,0.008];χ 2 = 5.06,p = 0.024;seeFigure2.3). Figure2.3:Unconditionalandconditionalmarginalefectsofinsuranceonvaccineuptake. Note:Allmarginalefectsofinsurancerepresentdiscretechangesfromthebaselevel(subsidy),andareestimated fromconditionallogitregressionsoftheinsurancedummiesonvaccinationchoice,includinganinterceptterm, andcontrollingfortheefectsofvaccineattributes,compensationamount,diseaseattributes,aswellasindividual characteristics. The reference outcome is no vaccination. The horizontal lines represent the 95% condence intervalsaroundtheunconditionalandconditionalmarginalefects. Thesamplesizesfortheregressionswith thesampleandtheconsistentrespondentsare22,626and17,622observations,respectively. 2.5.3.3 Efectofthecompensationlevelonvaccineuptake. Inordertoanswerthequestionaboutthefeaturesofanoptimalvaccinationincentivepro- gram,weexaminedtheefectofthecompensationamountonvaccineuptakeandfoundthat neitherthecompensationamountnorcompensationlevelhadasignicantefectonvaccine uptakeinboththefullsampleandtherestrictedsampleofconsistentrespondents(TableA.11). 55 2.5.3.4 Contributionandrelativeimportanceofthevaccineanddiseaseattributes. Inordertoaddressthequestionaboutthetypesofvaccinesthatinnovatorsshouldtarget,we examinedthevaccineanddiseasecharacteristicsthatmostcontributedtovaccinechoice. All vaccineattributesinuencedvaccinationchoiceintheanticipateddirections:vaccinesadmin- isteredthroughtheintranasalroutewereleastpreferredcomparedtoinjectablevaccines;will- ingnesstovaccinatedidnotsignicantlydiferacrossorally-administeredandinjectablevac- cineinthefullsampleofrespondents,butwasslightlysignicantinthesampleofconsistent respondents. More ecacious vaccines and vaccines that conferred longer-term immunity, leastseveresideefects,lowerriskofseveresideefectsandlowerout-of-pocketcostweremore preferred(Table2.3). With respect to the inuence of disease characteristics, the results suggest greater vaccine uptakeforacutediseasescomparedtochronicdiseases. Respondentsalsoexpressedagreater willingnesstovaccinateagainstdiseasestransmittedthroughdirectcontactrelativetoairborne diseases.Vaccineuptakewasalsoassociatedwithgreaterriskofinfection,higherriskofexpe- riencingseverediseasesymptoms,andmoreseverediseasesymptoms. The contributions of most vaccine and disease attributes did not signicantly difer across the two treatment groups; only the efects of the risk of vaccine-related side efects and the out-of-pocketcostofvaccinediferedsignicantlybetweenthetwoexperimentalgroups:the magnitudeoftheefectoftheriskofsideefectswasgreaterintheinsurancearmcompared to the subsidy arm, while vaccine cost had a greater efect in the subsidy arm compared to theinsurancearm(TableA.11). Theformerresultcouldhaveamutingefectontheefectof insurance. 56 Table2.3:Contributionsofthevaccineanddiseaseattributes. Variables a FullSample ConsistentRespondents Estimate Marginalefect Estimate Marginalefect No-vaccinationconstant -5.4971*** – -5.8001*** – (0.1850) – (0.2121) – Insurance 0.0371 0.0037 0.0578 0.0056 (0.0375) (0.0037) (0.0430) (0.0042) Vaccineattributes Intranasaladministration b -0.2362*** -0.0225*** -0.2156*** -0.0203*** (0.0526) (0.0049) (0.0597) (0.0055) Oraladministration b -0.0616 -0.0063 -0.1066* -0.0104* (0.0499) (0.0051) (0.0569) (0.0055) Ecacy(%) 0.0477*** 0.0047*** 0.0480*** 0.0046*** (0.0011) (0.0001) (0.0013) (0.0001) 6yearsimmunity c 0.6863*** 0.0671*** 0.6583*** 0.0622*** (0.0520) (0.0059) (0.0593) (0.0064) Lifetimeimmunity c 0.9601*** 0.1047*** 0.9588*** 0.1025*** (0.0498) (0.0063) (0.0564) (0.0070) VSE:Neurologicaldisorder d -0.0886 -0.0092 -0.0134 -0.0014 (0.0563) (0.0059) (0.0641) (0.0066) VSE:Death d -0.3640*** -0.0341*** -0.3854*** -0.0346*** (0.0748) (0.0068) (0.0855) (0.0074) VSErisk(permillion) -0.0337*** -0.0033*** -0.0353*** -0.0034*** (0.0059) (0.0006) (0.0067) (0.0007) Out-of-pocketcost($) -0.0021*** -0.0002*** -0.0021*** -0.0002*** (0.0001) (0.0000) (0.0001) (0.0000) Compensationlevels Compensationamount($) -0.0000 -0.0000 0.0000 0.0000 (0.0000) (0.0000) (0.0000) (0.0000) Continued... 57 Table2.3:Contributionsofthevaccineanddiseaseattributes. Variables a FullSample ConsistentRespondents Estimate Marginalefect Estimate Marginalefect Diseaseattributes Acutedisease e 0.1725*** 0.0171*** 0.2085*** 0.0200*** (0.0412) (0.0040) (0.0470) (0.0045) Vector-bornedisease f -0.0055 -0.0006 0.0023 0.0003 (0.0516) (0.0061) (0.0583) (0.0067) Airbornedisease f -0.6200*** -0.0581*** -0.6170*** -0.0560*** (0.0497) (0.0050) (0.0569) (0.0055) Infectionrisk(per100,000) 0.0004*** 0.0000*** 0.0004*** 0.0000*** (0.0000) (0.0000) (0.0000) (0.0000) SEV:Handicap g 0.2033*** 0.0203*** 0.1935*** 0.0188*** (0.0483) (0.0048) (0.0550) (0.0053) SEV:Death g 0.1954*** 0.0195*** 0.2009*** 0.0195*** (0.0540) (0.0052) (0.0616) (0.0058) SEVrisk(permillion) 0.0000*** 0.0000*** 0.0000*** 0.0000*** (0.0000) (0.0000) (0.0000) (0.0000) Observations 22,626 17,622 AIC 17,567 13,558 BIC 17,912 13,893 Log-likelihood -8,740 -6,736 Pseudo-R 2 0.285 0.290 χ 2 6,973 5,505 P-value 0.000 0.000 Notes: Standarderrorsinparentheses;statisticalsignicance: ***p<0.01,**p<0.05,*p<0.1. Allmodelsusethesampleofconsistentrespondents,as denedinSection2.5.1,andcontrolforindividualcharacteristics.Thedependentvariableisthechoiceofvaccination.Abbreviations:VSE=vaccineside efect;SEV=severityofdisease. a Thereferencealternativeinallmodelsis“Novaccination”. b Referencelevel:administrationbyinjection. c Referencelevel:1yearimmunity. d Referencelevel:anaphylaxis(life-threateningallergicreaction). e Referencelevel:chronicdisease. f Referencelevel:transmissionbydirectcontact. g Referencelevel:severecomplicationrequiringhospitalization. 58 2.5.3.5 Heterogeneousefectsofinsuranceonvaccineuptake Wealsoinvestigatedotherheterogeneousefectsofinsuranceonvaccineuptakebyestimating models in which we interacted the insurance dummy with vaccine and disease attributes as wellasindividualcharacteristics.Theadditionoftheseinteractiontermsallowsustoassessthe diferentialefectofinsuranceonvaccineuptakewithrespecttotheseattributes. Wereport the results from these analyses in Appendix A: Section A.7. These results suggest that the conditionalmarginalefectsofinsurancedonotdifersignicantlyacrossanumberofindi- vidualcharacteristics(gender,race,age,education,income,areaofresidency,priorexperience with vaccine-related side efects, and belief that vaccines cause autism). For example, while theefectofinsuranceonvaccinationincreaseswiththeout-of-pocketcostsofvaccines,these efectsareinsignicantanddonotdiferbyindividualcharacteristics(FigureA.4).Theefects ofinsuranceonvaccinationdonotsignicantlydiferalongotherindividual,vaccineanddis- easecharacteristicsexplored(AppendixA:SectionA.7,FigureA.4-FigureA.3). 2.5.3.6 Willingness-to-payandwillingness-to-accept Weestimatedthewillingnesstopayforvariousvaccineanddiseaseattributes(Table2.4).The resultssuggestthattherespondentswereonaveragewillingtopayonaverageasmuchas$23 forapercentincreaseinvaccineecacy, andwerewillingtoacceptonaverageaslowas$17 averagetotakeonaoneinamillionriskofseverevaccinesideefect. Willingnesstopayto reduceinfectionriskwasgenerallyverylow(20centsonaverage),andtheaveragewillingness topayforinsurancewas$18(althoughinsignicant). Theseestimatesweresimilarbetween theinsuranceandsubsidygroups. 59 Table2.4:Willingnesstopayestimates. Attribute Toreceive vaccinationwith Willingness-to-pay(WTP)/Willingness-to-accept(WTA),$(CI) FullSample ConsistentRespondents Pooled Insurance Subsidy Pooled Insurance Subsidy Infectionrisk 1/100,000lessrisk 0.21 0.22 0.21 0.22 0.18 0.22 (0.19,0.24) (0.18,0.25) (0.18,0.23) (0.19,0.24) (0.13,0.24) (0.18,0.25) Vaccineecacy 1%moreecacy 23.15 26.22 20.54 23.28 20.27 20.64 (21.44,24.86) (23.30,29.15) (18.52,22.55) (21.33,25.23) (16.29,24.26) (18.30,22.97) Riskofseverevaccinesideefect 1/millionmorerisk -16.38 -26.49 -7.71 -17.11 -12.05 -6.56 (-22.14,-10.61) (-35.83,-17.15) (-14.93,-0.48) (-23.69,-10.52) (-26.35,2.24) (-14.93,1.82) Insurance Insurance 18.02 – – 28.02 – – (-17.69,53.74) – – (-12.88,68.90) – – Note:Theestimatesrepresenttherespondents’meanwillingness-to-payoftoreceivevaccinationwitheachattributeonthex-axis,andarecalculatedaccordingtoEquation2.7.Thehorizontalbarsdenotethe95%condenceintervalsaround theWTPestimates.AnegativeWTPindicatesthattheindividualismadeworseofbytheattribute;theabsolutevalueoftheWTPthereforerepresentstheindividual’swillingnesstoaccept(WTA),i.e.theminimumamountofmoneyhe/she willbewillingtoaccepttoutilizeavaccinewiththatattribute. 60 2.6 Discussion Thisstudyaimedtodeterminewhetherasubsidyprogramoranequivalentsmall-stakeno- fault insurance scheme against the risk of severe vaccine-related side efects was more efec- tive in incentivizing vaccine uptake. Based on the predictions from expected utility theory, prospect theory and the Kőszegi-Rabin utility theory, we hypothesized that vaccine uptake would be greater under an insurance scheme than under an equivalent subsidy scheme. In ordertotestthishypothesis,wedevelopedadiscretechoiceexperimentwithchoiceproles calibratedtothecharacteristicsofrealvaccinesandrealinfectiousdiseases.Thesechoicepro- leswerepresentedtorespondentsrandomizedwithequalprobabilityintoaninsuranceor subsidyarm.Theanalysisoftheresponsesfromthisexperimentyieldedseveralndings. First,wefoundthatinsuranceincreasedtheprobabilityofvaccinationby4and6percentage pointsinthefullsampleandinthesampleofrespondentswithconsistentchoiceresponses. We also found that insurance had a signicantly greater efect (1.5 percentage point) on the probability of vaccination among non-female respondents, but had no signicant efect on vaccine uptake among their female counterparts. These results suggest that while female respondents behaved according to expected utility theory, the vaccination choices of non- femalerespondentswerebestpredictedbyalternativebehavioraleconomicmodels.Theyfur- therimplythatfemalerespondentsweremorerationalandlessrisk-andloss-aversethantheir malecounterpartswithregardstovaccination. These results are however inconsistent with prior ndings from empirical investigations in laboratoryexperimentsandeldstudiessuggestingthatfemalesaregenerallymorerisk-averse thantheirmalecounterparts[14,20]. Otherstudiesfoundthattheseresultspersistedafter controllingforindividualcharacteristicssuchasage,education,familystatus,andwealth(see JianakoplosandBernasek[35]).Hersch[29]alsoreportedthatwomenmakesaferchoicesthan 61 meninanumberofriskyconsumerdecisions,includingsmoking,seatbeltuse,preventative dentalcare,regularbloodpressurechecks. Theinconsistencybetweenourndingandthepriorevidencecouldbeduetoseveralexpla- nations. First, thediferencemightbeexplainedbywomen?soverweightingofvaccineand diseaseattributesintheirvaccinationdecisionsrelativetothevalueofthecompensation. If thisisthecase,ourestimatesoftheefectofinsurancewillbebiaseddownward.Asecondrea- sonthatmightaccountforthediferentialefectacrossgendergroupsinthewrongdirection isrelatedtolossaversion:iffemalesaregenerallymorerisk-averseandloss-aversethanmales, thiscouldinducethemtooverweighvaccineattributes(e.g. severityandriskofvaccineside efects), andhenceviewvaccinationasatooriskyofagambletotake, regardlessofthesize oftheinsurancecompensation. Whiletheseconjecturesandhypothesesdonotdenitively explaintheresults,theyprovideleadsforfurtherinvestigationofthisresult. Asecondaimofthestudywastodeterminetheoptimalincentivestructurebyinvestigating theefectofthecompensationlevelonvaccination.Wefoundnosignicantandmeaningful efectofthecompensationamountonvaccineuptake. Athirdobjectiveofthestudywasto determinevaccinecharacteristicsmostvaluedbyconsumers,inordertohelpguideinnovators andpolicymakersonthetypesofvaccinestheyshoulddeveloportarget. Afterassessingthe contributions of various vaccine and disease characteristics to consumers’ vaccination deci- sions, we determined that vaccine price, safety prole (risk and severity of side efects), and ecacy (efectiveness and duration of immunity) had the largest efects on vaccine uptake. Theriskofbecominginfectedwiththediseaseandthemodeoftransmissionofthedisease werealsoimportantattributesthatinuencedrespondents’decisionstovaccinate.Thesend- ingsareconsistentwithourtheoreticalexpectationsaboutindividuals’preferencesforvaccine attributes,andsuggestthatinnovatorstoshouldinvestinthedevelopmentofvaccineswith thesecharacteristicsinordertoappealmoretoconsumerdemand. 62 Finally,weestimatedwillingness-to-paytobeforapercentincreaseinvaccineecacy($23.15 perpercentincreaseinecacy). Willingnesstopayforano-faultinsuranceagainstan“aver- age”riskofvaccinesideefectandseverityofsideefectwas$18,butvariedwidely. Theresultscollectivelysuggestthatitmaybefeasibletonudgevaccineuptakeinabudget- neutralmannerthroughano-faultinsurancescheme. Thendingsalsohavedirectimplica- tionsforthereductionofdrugresistance. Forexample,inthecaseofu,increasedvaccina- tioncouldhelpreducetheriskofinfectionwiththevirus, andhencereducetheutilization ofantibiotics.Giventhattheinappropriateutilizationofantibioticsinusymptommanage- mentisasignicantdriverofantibioticresistance,ano-faultinsuranceschemecouldtherefore helppreservetheantibioticstock. This study sufers from the typical criticism about stated preference methods and discrete choiceexperiments: therespondents’statedpreferencesmightnotaccuratelyrepresenttheir choicebehaviorsinareal-worldsituation[42].However,ourapproachisabetteralternative totherevealedpreferenceelicitationapproachwhichcannotisolatetheindependentefectof eachvaccineattribute,diseasescenario,andcompensationamountstotheoverallutility,since itwouldbeimpossibletoobserveindependentvariationsindisease,vaccineandcompensa- tionattributesinareal-worldcontext.Additionally,intheabsenceofamarketforvaccinesin thepipeline,surveymethodsaretheonlymeansforassessingprivatedemand. Thereisalso aconcernthatbyprovidingrespondentsinformationaboutthediseaseanddiseasescenario, wemighthaveprimedthemtothinkandactoutsideoftheirnormalbehaviorpatterns,and potentiallymuteanysmallefectthatinsurancecouldhavehadonuptakerates. Whilethis concernisvalid,oferingrespondentsthesamesetofinformationwasthemostreliableway tominimizetheriskofcofounding. Theexternalvalidityofourndingsarealsolimitedto internetuserswithsimilarcharacteristicsastheMTurkpopulation.Despitetheselimitations, thisstudyprovidesanewinsightsintoindividuals’preferencesforinfectiousdiseasevaccine 63 characteristics, aswellastheirresponsivenesstoano-faultinsuranceagainsttheriskofside efects. Futurestudiesshouldthereforeinvestigatetheefectivenessofthesenancialincen- tiveschemesonvaccineuptakeinareal-worldcontext. References [1] Adamowicz, W. et al. Stated preference approaches for measuring passive use values: Choiceexperimentsandcontingentvaluation. American Journal of Agricultural Eco- nomics,80(1):64–75,Feb1998. [2] Alter,A.L.,Oppenheimer,D.M.andZemla,J.C. MissingtheTreesfortheForest: A ConstrualLevelAccountoftheIllusionofExplanatoryDepth. Journal of Personality and Social Psychology,99(3):436–51,Sep2010. [3] Arrow,K.J. Essays in the theory of risk-bearing. MarkhamEconomicsSeries.Chicago, IL:MarkhamPub.Co.,1971. [4] Banerjee,A.V.etal. ImprovingimmunisationcoverageinruralIndia: Calibrationlus- teredrandomisedcontrolledevaluationofimmunisationcampaignswithandwithout incentives. British Medical Journal,340:c2220,2010. [5] Berinsky,A.J.,Huber,G.A.andLenz,G.S. Evaluatingonlinelabormarketsforexperi- mentalresearch:Amazon.com’smechanicalTurk. Political Analysis,20:351–68,2012. [6] Black, C.L. et al. Inuenza Vaccination Coverage Among Health Care Personnel– UnitedStates,2014-15InuenzaSeason. Morbidity and Mortality Weekly Report,64 (36):993–9,Sep182015. 64 [7] Boxall,P.,Adamowicz,W.L.andMoon,A. Complexityinchoiceexperiments:Choice ofthestatusquoalternativeandimplicationsforwelfaremeasurement.AustralianJour- nal of Agricultural and Resource Economics,53(4):503–19,Oct2009. [8] Buhrmester, M., Kwang, T. and Gosling, S.D. Amazon’s Mechanical Turk: A New SourceofInexpensive,YetHigh-Quality,Data? Perspectives on Psychological Science,6 (1):3–5,Jan2011. [9] Burgess, L.andStreet, D.J. Optimaldesignsforchoiceexperimentswithasymmetric attributes. Journal of Statistical Planning and Inference,134(1):288–301,2005. [10] CentersforDiseaseControlandPrevention. FluVaccinationCoverage: UnitedStates, 2012-2013InuenzaSeason. Availableathttp://www.cdc.gov/u/pdf/uvaxview/vax- coverage-1213estimates.pdf,Sept252013. AccessedJan15,2016. [11] CentersforDiseaseControlandPrevention. PossibleSide-efectsfromVaccines. Avail- able at http://www.cdc.gov/vaccines/vac-gen/side-efects.htm, Nov 5 2015. Accessed Feb12,2016. [12] Chandler,J.etal. UsingNonnaiveParticipantsCanReduceEfectSizes. Psychological Science,26(7):1131–9,Jul2015. [13] Chapman, G.B. and Coups, E.J. Predictors of inuenza vaccine acceptance among healthyadults. Preventive Medicine,29(4):249–62,1999. [14] Croson, R. and Gneezy, U. Gender diferences in preferences. Journal of Economic Literature,47(2):448–74,June2009. [15] deBekker-Grob,E.W.etal.SampleSizeRequirementsforDiscrete-ChoiceExperiments inHealthcare:APracticalGuide. Patient,8(5):373–84,Oct2015. 65 [16] DeBondt,W.F.M.andThaler,R.DoestheStockMarketOverreact? JournalofFinance, 40(3):793–805,Jul1985. [17] Determann,D.etal. AcceptanceofVaccinationsinPandemicOutbreaks: ADiscrete ChoiceExperiment. PloS One,9(7),Jul242014. [18] Diecidue,E.andWakker,P.P. OntheIntuitionofRank-DependentUtility. Journalof Risk and Uncertainty,23(3):281–98,Nov2001. [19] Dupuis,M.,Endicott-Popovsky,B.andCrossler,R.AnAnalysisoftheUseofAmazon’s MechanicalTurkforSurveyResearchintheCloud. InProceedings of the International Conference on Cloud Security Management: ICCSM 2013, page12.AcademicConfer- encesLimited,2013. [20] Eckel,C.C.andGrossman,P.J. Men,womenandriskaversion:Experimentalevidence. Handbook of experimental economics results,1:1061–73,2008. [21] Elam-Evans,L.D.etal. National,Regional,State,andSelectedLocalAreaVaccination CoverageAmongAdolescentsAged13-17Years–UnitedStates, 2013. Morbidity and Mortality Weekly Report,63(29):625–33,2014. [22] Ericson,K.M.M.andFuster,A. ExpectationsasEndowments:EvidenceonReference- DependentPreferencesfromExchangeandValuationExperiments. Quarterly Journal of Economics,126(4):1879–907,2011. [23] Eriksson,K.andSimpson,B. Emotionalreactionstolosingexplaingenderdiferences inenteringariskylottery. Judgment and Decision Making,5(3):159–63,Jun2010. [24] Fischhof,B.etal. Howsafeissafeenough? Apsychometricstudyofattitudestoward technologicalrisksandbenets. InSlovic,P.,editor,Risk Perception.Earthscan,2000. 66 [25] Hall, J. et al. What inuences participation in genetic carrier testing? Results from a discretechoiceexperiment. Health Economics,25(3):520–37,May2006. [26] Hartman, R.S., Doane, M.J.andWoo, C.K. Status-QuoBiasintheMeasurementof ValueofService. Resources and Energy,12(2):197–214,1990. [27] HealthyPeople.gov. Immunization and Infectious Diseases. Available at https://www.healthypeople.gov/2020/topics-objectives/topic/immunization-and- infectious-diseases,2014. AccessedMar18,2016. [28] Hensher,D.A.,Rose,J.M.andGreene,W.H. Appliedchoiceanalysis: Aprimer. Cam- bridge,MA:CambridgeUniversityPress,2005. [29] Hersch,J. Decisions: DiferencesbyGenderandRace. Managerial and Decision Eco- nomics,17:471–81,1996. [30] Hiligsmann,M.etal. Patients’preferencesforosteoporosisdrugtreatment: adiscrete- choiceexperiment. Arthritis Research & Therapy,16(1),2014. [31] Horton, J.J. and Chilton, L.B. The labor economics of paid crowdsourcing. In Pro- ceedingsfromEC’10: The11thACMConferenceonElectronicCommerce,pages209–18. NewYork,NY:ACM,2010. [32] Howard,K.andSalkeld,G. DoesAttributeFraminginDiscreteChoiceExperiments InuenceWillingnesstoPay? ResultsfromaDiscreteChoiceExperimentinScreening forColorectalCancer. Value in Health,12(2):354–63,Mar–Apr2009. [33] Huf,C.andTingley,D. “Whoarethesepeople?”Evaluatingthedemographiccharac- teristicsandpoliticalpreferencesofMTurksurveyrespondents. Research and Politics, (1):1–12,Jul–Sep2015. 67 [34] Jakobsson,M.andSiadati,H. Mobile Authentication: Problems and Solutions,chapter ImprovedVisualPreferenceAuthentication,pages57–72.SpringerBriefsinComputer Science.NewYork,NY:Springer,2013. [35] Jianakoplos,N.A.andBernasek,A.Arewomenmoreriskaverse? EconomicInquiry,36 (4):620–30,1998. [36] Johnson,R.andOrme,B.K. Getting the most from CBC. SawtoothSoftwareResearch PaperSeries.Sequim,2003. [37] Kahneman, D. and Tversky, A. Prospect Theory - Analysis of Decision under Risk. Econometrica,47(2):263–91,1979. [38] Kahneman,D.,Slovic,P.andTversky,A. Judgment under uncertainty: Heuristics and biases. NewYork,NY:CambridgeUniversityPress,1982. [39] Kőszegi, B. and Rabin, M. A model of reference-dependent preferences. Quarterly Journal of Economics,121(4):1133–65,2006. [40] Lancsar,E.andLouviere,J. Conductingdiscretechoiceexperimentstoinformhealth- caredecisionmaking:auser’sguide. Pharmacoeconomics,26(8):661–77,2008. [41] Lichtenstein,S.etal. Jugdedfrequencyoflethalevents. Journal of Experimental Psy- chology,4:551–78,1978. [42] Louviere,J.J.,Hensher,D.A.andSwait,J.D. Statedchoicemethods: Analysisandappli- cations. CambridgeUniversityPress,2000. [43] Louviere,J.J.,Flynn,T.N.andCarson,R.T.DiscreteChoiceExperimentsAreNotCon- jointAnalysis. Journal of Choice Modelling,3(3):57–72,2010. [44] Machina,M.J. ChoiceUnderUncertainty:ProblemsSolvedandUnsolved. Journal of Economic Perspectives,1(1):121–54,Summer1987. 68 [45] Mantzari,E.,Vogt,F.andMarteau,T.M. Financialincentivesforincreasinguptakeof HPVvaccinations:Arandomizedcontrolledtrial. Journal of Health Psychology,34(2): 160–71,Feb2015. [46] Marshall,D.etal. ConjointAnalysisApplicationsinHealth-HowareStudiesbeing DesignedandReported?: AnUpdateonCurrentPracticeinthePublishedLiterature between2005and2008. Patient,3(4):249–56,Dec12010. [47] Mason,W.andSuri,S. ConductingbehavioralresearchonAmazon’sMechanicalTurk. Behavior Research Methods,44(1):1–23,Mar2012. [48] McFadden,D. ConditionalLogitAnalysisofQualitativeChoiceBehavior. InZarem- bka,P.,editor,FrontiersinEconometrics,pages105–42.NewYork,NY:AcademicPress, 1974. [49] McFadden,D. TheChoiceTheoryApproachtoMarketResearch. Marketing Science, 5(4):275–97,1986. [50] McFadden, D. and Train, K. Mixed MNL models for discrete response. Journal of Applied Econometrics,15(5):447–70,2000. [51] Meeker,D.etal.UseofInsuranceAgainstaSmallLossasanIncentiveStrategy.Decision Analysis,12(3):122–9,Sep2015. [52] Mentzakis,E.,Stefanowska,P.andHurley,J. Adiscretechoiceexperimentinvestigat- ingpreferencesforfundingdrugsusedtotreatorphandiseases: Anexploratorystudy. Health Economics Policy and Law,6(3):405–33,Jul2011. [53] NationalVaccineAdvisoryCommittee. RecommendationsfromtheNationalVaccine Advisorycommittee:Standardsforadultimmunizationpractice.PublicHealthReports, 129(2):115–23,Mar–Apr2014. 69 [54] National Vaccine Program Oce. National Adult Immunization Plan Draft Report. Availableathttp://www.hhs.gov/nvpo/national_adult_immunization_plan_ draft.pdf,Feb52015. [55] Orme,B.K. Sawtooth Software Technical Paper,chapterSamplesizeissuesforconjoint analysisstudies. Sequim,1998. [56] Orme,B.K. Getting Started with Conjoint Analysis: Strategies for Product Design and Pricing Research. Madison,WI:ResearchPublishers,LLC,2edition,2006. [57] Orme, B.K. Sample size issues for conjoint analysis studies. In Getting Started with Conjoint Analysis: Strategies for Product Design and Pricing Research, pages 57–66. Madison,WI:ResearchPublishers,LLC,2edition,2010. [58] Paolacci, G. and Chandler, J. Inside the Turk: Understanding Mechanical Turk as a ParticipantPool. Current Directions in Psychological Science,23(3):184–8,Jun2014. [59] Paolacci,G.,Chandler,J.andIpeirotis,P.G.RunningexperimentsonAmazonMechan- icalTurk. Judgment and Decision Making,5(5):411–9,Aug2010. [60] Pearmain,D.etal. Statedpreferencetechniques: Aguidetopractice.SteerDaviesGleave andHagueConsultingGroup,2edition,1991. [61] Plotkin,S.A. Vaccines: Thefourthcentury. Clinical and Vaccine Immunology,16(12): 1709–19,Dec2009. [62] Quiggin,J. Atheoryofanticipatedutility. Journal of Economic Behavior & Organiza- tion,3(4):323–43,Dec1982. [63] Rabin,M. RiskAversionandExpected-utilityTheory:ACalibrationTheorem. Econo- metrica,68(5):1281–92,Sep2000. 70 [64] Rabin,M. InferencebyBelieversintheLawofSmallNumbers. Quarterly Journal of Economics,117(3):775–816,2002. [65] Reed Johnson, F. et al. Constructing experimental designs for discrete-choice experi- ments: report of the ISPOR Conjoint Analysis Experimental Design Good Research PracticesTaskForce. Value in Health,16(1):3–13,Jan–Feb2013. [66] Rose,J.M.andBliemer,M.C.J.Samplesizerequirementsforstatedchoiceexperiments. Transportation,40(5):1021–41,Sep2013. [67] Sadique, M.Z. et al. The Efect of Perceived Risks on the Demand for Vaccination: ResultsfromaDiscreteChoiceExperiment. Plos One,8(2),Feb82013. [68] Salkeld,G.,Ryan,M.andShort,L. Theveilofexperience:Doconsumerspreferwhat theyknowbest? Health Economics,9(3):267–70,Apr2000. [69] Samuelson,W.andZeckhauser,R.J.Statusquobiasindecisionmaking.JournalofRisk and Uncertainty,1:7–59,1988. [70] SASInstituteInc. SAS®Studio 3.2: Administrator’s Guide. Cary,NC:SASInstitute Inc.,2014. [71] Shapiro,D.N.,Chandler,J.andMueller,P.A. UsingMechanicalTurktoStudyClinical Populations. Clinical Psychological Science,1(2):213–20,Mar2013. [72] Sprouse,J. AvalidationofAmazonMechanicalTurkforthecollectionofacceptability judgmentsinlinguistictheory. Behavior Research Methods,43(1):155–67,Mar2011. [73] Stokley, S. et al. Human Papillomavirus Vaccination Coverage Among Adolescents, 2007âĂŞ2013, and Postlicensure Vaccine Safety Monitoring, 2006âĂŞ2014 – United States. Morbidity and Mortality Weekly Report,63(29):620–4,2014. 71 [74] Street, D.J., Burgess, L. and Louviere, J.J. Quick and easy choice sets: Constructing optimalandnearlyoptimalstatedchoiceexperiments.InternationalJournalofResearch in Marketing,22(4):459–70,Dec2005. [75] Thurstone,L.L. ALawofComparativeJudgment. PsychologyReview,34:273–86,1927. [76] Tversky, A. and Kahneman, D. Judgment under uncertainty: Heuristics and biases. Science,185(4157):1124–31,1974. [77] Tversky,A.andKahneman,D. Theframingofdecisionsandthepsychologyofchoice. Science,211(4481):453–8,1981. [78] Tversky,A.andKahneman,D.Advancesinprospecttheory:Cumulativerepresentation ofuncertainty. Journal of Risk and Uncertainty,5(4):297–323,1992. [79] USDepartmentofLabor. MinimumWageLawsintheStates–January1,2016. Avail- ableathttp://www.dol.gov/whd/minwage/america.htm,Jan2016. [80] Veldwijk, J.etal. Theefectofincludinganopt-outoptionindiscretechoiceexperi- ments. Plos One,9(11):e111805,2014. [81] Wakker,P.P. Prospect Theory: For Risk and Ambiguity. Cambridge,UK:Cambridge UniversityPress,2010. [82] Wigham, S. et al. Parental nancial incentives for increasing preschool vaccination uptake:Systematicreview. Pediatrics,134(4):e1117–28,Oct2014. 72 Chapter3 Test-and-treatinLosAngeles:A mathematicalmodeloftheefectsof test-and-treatforthepopulationofmen whohavesexwithmeninLosAngeles County Neeraj Sood, Zachary R. Wagner, Amber Jaycocks, Emmanuel F. Drabo,RaffaeleVardavas Abstract Background: Thereisevidencetosuggestthatantiretroviraltherapy(ART)andtestingfor humanimmunodeciencyvirus(HIV)reducetheprobabilityoftransmissionofHIV.This has led health ocials across the United States to take steps toward a test-and-treat policy. However,theextentofthebenetsgeneratedbytest-and-treatisdebatable,andtherearecon- cerns,suchasincreasedmultidrugresistance(MDR),thatremainunaddressed. Methods:WedevelopedadeterministicepidemiologicmodeltosimulatetheHIV/AIDSepi- demicformenwhohavesexwithmen(MSM)inLosAngelesCounty(LAC).Wecalibrated themodeltomatchtheHIVsurveillancedatafromLACacrossa10-yearperiod,startingin 73 2000. Wethenmodiedourmodeltosimulatethetest-and-treatpolicyandcomparedepi- demiologicoutcomesunderthetest-and-treatscenariotothestatusquoscenarioovertheyears 2012-2023.Outcomemeasuresincludednewinfections,deaths,newAIDScases,andMDR. Results:Relativetothestatusquo,thetest-and-treatmodelresultedina34%reductioninnew infections,19%reductionindeaths,and39%reductioninnewAIDScasesby2023.However, theseresultsarecounterbalancedbyaneardoublingoftheprevalenceofMDR(9.06%com- pared to 4.79%) in 2023. We also found that the efects of increasing testing and treatment werenotcomplementary. Conclusions: Althoughtest-and-treatgeneratessubstantialbenets,itwillnoteliminatethe epidemicforMSMinLAC.Moreover,thesebenetsarecounterbalancedbylargeincreases inMDR. Keywords: HIV/AIDS;antiretroviraltherapy;drugresistance;mathematicalmodel;testand treat. 74 3.1 Introduction Recentevidencesuggeststhatantiretroviraltherapy(ART)reducestheprobabilityofhuman immunodeciencyvirus(HIV)transmission,particularlyifinitiatedatearlystagesofthedis- ease[6,7,11]. EvidencealsosuggeststhattestingHIVpositivecausesdramaticreductionsin sexual activity levels, which also reduces HIV transmission probability [19, 20]. In light of thisevidence,healthocialsacrosstheUnitedStatesaretakingstepstowardscalinguptest- ingprogramsandrecommendingimmediateARTinitiationtoindividualstestingpositive, regardlessofCD4cellcount[9]. Somearguea“test-and-treat”policycouldleadtoelimina- tionoftheHIVepidemic[14,21,25]. Although reasons exist for optimism regarding the role of increased testing and treating in HIVpreventionagendas,someconcernsremainunaddressed.Oneimportantconcernisthat increasedARTusecreatesmoremultidrug-resistantstrains,whichmightlimitthebenetsof test-and-treat[13]. Treatmentresistancealreadyposesamajorpublichealthprobleminthe UnitedStates[22]. Therefore,determininghowtoimplementtest-and-treatpolicieswhere HIV prevention is maximized and multidrug resistance (MDR) prevalence is minimized is important. Previousmathematicalmodelsthatsimulatetheimpactofscalinguptest-and-treatpolicies show mixed results. Some nd dramatic benets [5, 14], whereas others nd only modest efects [18, 23, 26]. There is also evidence from a natural experiment suggesting that some modelresultsmaybeexaggerated[27]. Thediscrepancyinmodelndingsreectsthesensi- tivityoftest-and-treatoutcomestodiferingunderlyingassumptionsaboutHIVprevalence, proportion of undiagnosed cases, risky sexual behavior, and other population or location- specicparameters[10,12].Thishighlightsthecriticalneedtocalibratemathematicalmodels to mimic real-world HIV prevalence and incidence trends [8]. Additionally, none of these 75 articlesexplicitlyaddressespotentialefectsoftest-and-treatpoliciesonthespreadofMDR. However,anearlierstrandofliteratureexaminingexpandedARTuseaddressesthisissueand reportsmixedndings.SomendthatincreasingthepercentageofpeopleusingARTwould substantiallyreducetheHIVepidemic’sseverityeveninthepresenceofhighARTresistance levels [17, 25]. However, the authors note that emergence of highly transmissible resistant strainsofHIVcansignicantlyreducethebenetsofexpandeduseofART[25].Baggaley et al. [1] found that controlling sub-Saharan African HIV epidemics through treatment is inefective,asincreasingtheproportionontreatmentincreasestheemergenceandspreadof drugresistance[1].OthersndthatbenetsfromexpandedARTusearecounterbalancedby modestincreasesinriskysexualbehavior[1,2,16]. Ourstudycontributestothisliteraturebyusingamathematicalmodeltosimulateefectsof increasedtestingandearlyinitiationoftreatmentformenwhohavesexwithmen(MSM)in LosAngelesCounty(LAC).Nopreviousstudyhasfocusedontest-and-treatinLAC.Inthe UnitedStates,LAChasthelargestincidenceofHIV,andMSMaccountfor82%ofallpeo- plelivingwithHIV/AIDS(PLWHA)[3,15].WecalibrateourmodelusingHIVsurveillance data from LAC for the years 2000 to 2009. Following calibration, we manipulate parame- tersrelatingtotest-and-treattosimulatealternatetest-and-treatpolicyscenarios.Toassessthe individualandcomplementaryefectsoftestingandtreatingonepidemiologicaloutcomes, the intensity of both testing and treatment rates are varied. Finally, we also assess how the intensity of each mechanism afects the portion of the population with MDR. The results couldhelpinformpolicymakersonbestapproachesforthetest-and-treatpolicyandwhere tofocusfutureHIVpreventioneforts. 76 3.2 Methods 3.2.1 Studydesign TomodeltheLACHIV/AIDSepidemic,wecalibrateadeterministicepidemiologicalmodel to match LAC’s HIV surveillance data across a 10-year period (2000-2009). We obtain the set of model parameters that best reproduce the observed HIV/AIDS prevalence trends for MSM.Wethenmodifyourmodeltoinvestigateefectsofthetest-and-treatpolicywhereby HIV testing rates are increased and newly diagnosed HIV-positive individuals are immedi- atelytreatmenteligible.WefocusontheMSMpopulationbecausetheyrepresent> 82%of PLWHAinLAC,andtheirbiologicalandbehavioralcharacteristicsarelikelytodiferfrom thoseofheterosexuals(seeAppendixBformoredetails). 3.2.2 Modelstructure We construct a compartmental HIV transmission model. Our model structure follows the generalapproachusedinpreviousHIVtransmissionmodels,andinparticularthatusedby Smithetal.[22].Ourmodelconsidersthedynamicsintheeraoftripletherapyandtherefore considersdynamicsofacquiredandtransmittedMDR. Figure 3.1 displays our compartmental HIV model, whereby individuals in our population progressoverthediferentstagesoftheinfection. Understatusquo, treatmenteligibilityis basedonstandardCentersforDiseaseControlandPreventionguidelines: advanceddisease and/orCD4cellcount< 350cells/L,which,althoughlowerthancurrenttreatmentguide- lines,wastheguidelineformostofthecalibrationyears.Themodelisintegratednumerically andtracksthedynamicsofthepopulationineachcompartmentasitchangesovertime. 77 Figure3.1:Humanimmunodeciencyvirus(HIV)transmissionmodel. Note:Individualsinourmodelaredividedinto8keyHIVinfectionstatuses:susceptibleorHIVnegative(S), infectedintheprimarystageofinfection(P), infectedandunawareofbeinginfected(I), diagnosedinfected but not treatment-eligible (J), infected and treatment-eligible (E), treated but no progression to AIDS (T), progressiontoAIDSbutnottreated(A),andprogressiontoAIDSandtreated(TA). Wedistinguishbetween infectedindividualswithdrug-sensitiveHIVstrain(subscripts)fromthoseinfectedwiththemultidrug-resistant HIVstrain(subscriptr). 3.2.3 Keyinputparametersandranges Ourmodelincludes34parameters,forwhichweprovideuncertaintyrangesbaseduponlit- eraturereview,expertopinion,andoursubjectiverangeassessment(AppendixB:SectionB.3, Table B.1 - Table B.4). Key parameters for this study include the rate at which treatment- eligible individuals initiate and adhere to treatment (σ), the rate at which unaware HIV- positive individuals get tested for HIV (ω), the rate at which MDR is acquired for people on treatment (r), and the probability of HIV transmission per sexual partnership (β). To accountfordiferentinfectiousnesslevelsandriskysexualactivityatdiferentdiseasestages, 78 theβparameterisuniqueforeachmodelcompartment.AkeymodelassumptionisthatHIV- positiveindividuals,awareoftheirserostatus,signicantlyreduceriskybehavior;weaccount forthisbyensuringthesamplingrangeofβ forstageJ is56%-76%lowerthanthesampling rangeofβforstageI [19]. 3.2.4 Surveillancedata Thesurveillancedataweuse,asareferencetocalibrateourmodel,weretakenfromLACsemi- annualsurveillancereportsfrom2008to2010.Thesurveillancedataprovideannualnumbers ofPLWHAbytransmissiongroupfortheyears2000-2009.ThesereportsdiferentiateHIV- positiveindividualsfromthosewithAIDS,andwedistributethesestagesaccordinglyinour model(AppendixB:SectionB.8). 3.2.5 Modelcalibration Inthissection,webrieysummarizehowwecalibratethemodel;moredetailsonthecalibra- tionareprovidedinAppendixB,SectionB.6. Tocalibrateourmodel,weuseaLatinhyper- cubesampling(LHS)methodandmatchthemodeloutputtoLACsurveillancedatafrom 2000 to 2009. We rst run 500,000 LHS simulations based on parameter ranges from the priorliterature. Next,wedropsimulationsthatdonotmatchpriorestimatesofthepropor- tion of PLHWA who are unaware of their HIV status [4], and MDR prevalence [22, 24]. Finally, we select the top 1,000 simulations that best match estimates of total numbers of PLHWAinLACsurveillancedatafortheyears2000to2009.Weusethesesimulationstocre- atenarrowerparameterrangesandrunanadditional 500, 000LHSsimulations. Asbefore, we drop simulations that do not match prior estimates of the proportion of PLHWA who areunawareoftheirHIVstatusandMDRprevalence. Inaddition, weonlyretainsimula- tionswherepredictedAIDScasesandnon-AIDSHIV-awarecasesareboth< 5%diferent 79 fromthe2009surveillancedata. Finally,oftheremainingsimulations,wechoosethesimu- lationthatbestmatchesthetotalnumberofPLHWAinLACsurveillancedatafortheyears 2000-2009.Welabelthisasthe“bestrun”andusetheparametersproducedbythisrunasthe main model parameters. Figure 3.2 compares results from the “best run” simulation to the surveillancedata. Figure3.2:Calibrationresults. 2000 2002 2004 2006 2008 10 15 20 25 30 35 Y ear Number of Living Cases ('000s) 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 HIV/AIDS Aware AIDS Non−AIDS Aware Surveillance Data Best Run Note:Asshown,thecalibratedmodelcanaccuratelypredicttrendsinnumberofAIDSandnon-AIDShuman immunodeciency virus-aware cases in Los Angeles County for the years 2000-2009. Abbreviation: HIV = humanimmunodeciencyvirus. 3.2.6 Test-and-treatmodel To simulate the test-and-treat policy, we extend the HIV model described above by incor- porating2additionalcompartments,TJ s andTJ r ,representingthenondrugresistantand 80 drug-resistant infected individuals who initiate treatment early prior to CD4 counts falling beloweligibilityguidelines(AppendixB:FigureB.4).TheJ toTJ path,σ TJ ,representsthe earlytreatmentinitiationrateforinfectedindividualswithaCD4cellcount> 350cells/L. WeassumecompartmentTJ’sriskoftransmissionis96%lowerthantheriskofcompartment J’stransmission[6]. Testingratesandratesatwhichpeoplebegintreatmentarekeyfortest-and-treatpolicyout- comes. Wetestarangeofparametercombinationstoassesshowmoreaggressivetestingand moreaggressivetreatmentindividuallyandjointlyimpactHIV-relatedoutcomes. Ourbaselinetest-and-treatsimulationassumesthefollowing:(1)Theexpectedtimetotreat- mentinitiationforthosewithCD4cellcountsabovethethresholdof 350cells/Listhesame asforthosewithCD4countbelowthisthreshold;(2)theaveragedurationforundiagnosed PLWHAtogettestedchangesfrom4.4yearsto1year.Inadditiontoourbaselinescenarios, weprovidearangeofotherscenariosthatcombinediferenttestingandtreatmentintensity levels. InLAC,achievableincreasesintestingandtreatmentlevelsareunclear. Toestimatefeasibil- ity,wecalculatenewdiagnosesandnewpeopleinitiatingtreatmentaftertherstyear,which serves as a proxy for efort and resources put toward the policy. Table 1 presents new HIV diagnosesandnewpeoplebeginningtreatmentduringtherstyearofthetest-and-treatpol- icywithalternativescenariosoftestingandtreatmentintensity. Ourbaselinescenario(test every1yearandtreatmentinitiation2.5yearsafterdiagnosis)resultsina115%increaseinnew diagnosesanda36%increaseinnewpeopleinitiatingtreatmentduringtherstyearcompared tothestatusquo.Themostaggressivestrategywesimulateassumesannualtestingandtreat- mentinitiation6monthsafterbecomingeligible. Thisresultsinincreasesof114%and156%, respectively. Theleastaggressivestrategywesimulate,whichassumestestingevery4.4years 81 andtreatmentinitiation2.5yearsafterbecomingeligible(simplyachangeinthetreatmenteli- gibilityguidelines),stillresultsinan18%increaseinnewpeopleinitiatingtreatmentintherst yearcomparedtothestatusquo.Thisisexplainedbythenewtreatmentguidelinesthatallow people to become eligible immediately after diagnosis rather than waiting until they have a CD4cellcount< 350cells/L. Table3.1:First-yearefectsofthetest-and-treatpolicy Scenario NewHIVDiagnosis (%ChangeFromStatusQuo) NewPeopleInitiatingTreatment (%ChangeFromStatusQuo) Statusquo 4,437 5,696 Treatment2.5y a Testevery1y b 9,537(115%) 7,724(36%) Testevery2y 7,576(71%) 7,317(28%) Testevery3y 6,689(51%) 7,154(26%) Testevery4.4y a 4,433(0%) 6,732(18%) Treatment1y Testevery1y 9,514(114%) 11,248(97%) Testevery2y 7,562(70%) 10,343(82%) Testevery3y 6,678(51%) 9,973(75%) Testevery4.4y a 4,427(0%) 9,071(59%) Treatment6mo Testevery1y 9,485(114%) 14,562(156%) Testevery2y 7,544(70%) 13,175(131%) Testevery3y 6,663(50%) 12,596(121%) Testevery4.4y a 4,419(0%) 11,198(97%) Note: First-year efects of the test-and-treat policy on HIV diagnoses and treatment initiation with diferent levelsofHIVtestingandtreatmentintensitycomparedtotheStatusQuo. a Currentdurationbasedonmodelcalibration. b Baselinescenario. 82 3.2.7 Sensitivityanalysis We rst conduct a one-way sensitivity analysis to determine how the model results are impactedbyeachparameterindividually. Inparticular,wereestimatethemodelbyvarying eachparameterby10%aboveandbelowthebaselinevalues,whileholdingallotherparameters xed. Weusedthesenewparametervaluestoestimateefectsoftest-and-treatonnewinfec- tionsandMDR.Thisallowsustounderstandwhichparametersmostinuenceourresults. Wealsoconductamultivariatesensitivityanalysisbyresampling 100, 000parametersetsfrom theparameterrangeabove. Weuseestimatesfromthisanalysistoreportthe95%condence intervalforourbaselinetest-and-treatresults(AppendixB:SectionB.8). 3.3 Results Figure 3.3 displays results under the baseline test-and-treat scenario compared to the status quo.Thebaselinetest-and-treatscenariorepresentsa115%increaseinthenumberofrst-year diagnosedHIVcasesanda36%increaseinthenumberofPLWHAtreatedintherstyear (Table3.1).Figure3.3showsthatthepercentageofthepopulationthatisunawarereducesover a10-yearperiodfrom20%in2013toapproximately5%in2023,andthepercentageofPLWHA onARTincreasesfrom62%in2013toapproximately76%in2023. Overthese10years,the dramaticchangesintestingandtreatmentleadtolargereductionsinnewinfections(34%;95% condenceinterval[CI],30%-37%),deaths(19%;95%CI,17%-21%),andnewAIDScases(39%; 95%CI,36%-41%).However,contrastingthesebenets,theMDRprevalencenearlydoubles to9%by2023(95%CI,7.8%-10.2%). 83 Figure3.3:Baselinetest-and-treatscenariocomparedtothestatusquo. 84 Table3.2presentsepidemiologicoutcomesforallsimulatedscenarios. Ourbaselinescenario resultsinasubstantialreductioninallepidemiologicmeasures.Themostaggressivestrategy, testingannuallyandtreatmentinitiationafter6months,increasesnewdiagnosesby114%and treatmentby156%intherstyear(Table3.1). Thisstrategyalsocreatesa47%reductionin newinfections,a28%reductionindeaths,anda64%reductioninnewAIDScasesby2023. However,underthisscenario,theportionofPLWHAwithMDRnearlytriplesto13.7%in 2023. Theleastaggressivestrategywesimulated,testingevery4.4yearsandtreatmentinitia- tionafter2.5years— essentiallyjustachangeintreatmenteligibilityguidelines— stillyields a6%reductioninnewinfections,a6%reductionindeaths,andan11%reductioninnewAIDS cases,withonlya1.3%increaseinMDR. 85 Table3.2:Resultswithoutearlytreatment. Scenario CumulativeFrom2013to2023 %ofTotalHIV/AIDSPopulationin2023 NewInfections (%Reduction) Deaths (%Reduction) NewAIDScases (%Reduction) % MDR % Unaware Status Quo 54,067 42,083 48,907 4.79% 20.0% Treatment2.5y a Testevery1y b 35,795(34%) 34,127(19%) 29,824(39%) 9.06 4.50% Testevery2y 39,707(27%) 36,079(14%) 34,131(30%) 8.19% 7.37% Testevery3y 42,249(22%) 37,417(11%) 37,190(24%) 7.64% 9.45% Testevery4.4y a 50,676(6%) 39,748(6%) 43,350(11%) 6.07% 18.51% Treatment1y Testevery1y 31,251(42%) 30,237(28%) 22,152(55%) 11.90% 3.67% Testevery2y 35,271(35%) 31,890(24%) 25,709(47%) 10.77% 6.14% Testevery3y 37,935(30%) 33,015(22%) 28,246(42%) 10.05% 7.97% Testevery4.4y a 47,226(13%) 35,610(15%) 34,763(29%) 7.85% 16.57% Treatment6mo Testevery1y 28,489(47%) 27,807(34%) 17,642(64%) 13.70% 3.20% Testevery2y 32,373(40%) 29,172(31%) 20,517(58%) 12.51% 5.39% Testevery3y 34,957(35%) 30,084(29%) 22,536(54%) 11.75% 7.02% Testevery4.4y a 44,589(18%) 32,633(22%) 28,737(41%) 9.22% 15.18% Abbreviation:HIV=humanimmunodeciencyvirus;MDR=multipledrugresistance. a Currentdurationbasedonmodelcalibration; b Baselinescenario. 86 Table3.2alsoshowsthe2mechanisms,testingandtreating,arenotcomplementary. Wesee roughly equal benets from increasing testing rates regardless of treatment initiation rates andviceversa.Forexample,undertheleastaggressivetreatmentstrategy(averagedurationto treatmentof2.5years),thereisanadditional28%reductioninnewinfections,whereasunder themostaggressivetreatmentstrategy(averagedurationtotreatmentof6months),thereisan additional29%reduction. Similarly,achangefromtheleastaggressivetreatmentstrategyto themostaggressivetreatmentstrategyaddsroughlythesamebenetunderalllevelsoftesting intensity. Wealsoexaminedscenarioswithoutearlytreatmentguidelinesimplemented(ie,onlyincreas- ing testing rates). This allows analysis of how much test-and-treat benet is derived from early treatment and how much comes from increasing testing (Appendix B: Sec- tion B.9, Table B.17). We nd that without any change to treatment guidelines and only increasingtesting,wegetabouthalftheepidemiologicbenets,whileMDRremainsstable. 3.3.1 Sensitivityanalysis Overall,ourresultsareratherstable,withthelargestimpactofany10%changeinourbaseline parametersresultinginonlya1.82percentagepointchangeinnewinfections. Wendnew infections are most sensitive to the parameters that represent sexual behavior (ie, the trans- missibilityparametersandnumberofpartners). ForMDR,wendthatourresultsaremostsensitivetotherateofresistanceparameter(r). Similartonewinfections,theresultsforMDRwereratherstable;thelargestimpactofany10% changeinourbaselineparametersresultsina0.45percentagepointchangeinMDRpreva- lence.Acompletedescriptionandresultsofthemultivariatesensitivityanalysisusedforcon- denceintervalsisfoundinAppendixB:SectionB.8. 87 Wealsorananadditionalsimulationusingourbestparameterestimatesfromtheclinicallit- eraturewithoutanycalibration.Underthisscenario,wendpercentagereductionsindeaths and new AIDS cases due to test-and-treat to be underestimated by 6.45 percentage points and 14.32 percentage points, respectively, whereas percentage reductions in new infections remainrelativelystable.Absolutereductionsinalloutcomesaresignicantlyunderestimated (AppendixB,SectionB.8.4). Also, because there are no data on the relative reduction in infectiousness due to ART for MSM — data are for discordant couples — we test the sensitivity of our assumption of a 96%reductionininfectiousnessfromARTusingtheupperandlowerendsoftheCohenet al95%condenceinterval,99%and73%reductions,respectively[6].Wealsoconductathird simulationassuminga50%reductionforadditionalcontrast.Wendthatassumptionsabout thisparameterhaveonlyamodestinuenceonourresults(AppendixB:SectionB.8.3). 3.4 Discussion WecalibratedourmodeltoaccuratelypredicttrendsinHIV/AIDSprevalenceinLACfrom 2000to2009. Weusedthismodeltoevaluatetheefectsofthetest-and-treatpolicyinLAC. The results show that the test-and-treat policy can generate substantial reductions in new infections, death, and new AIDS cases. However, contrary to Granich et al. [14], even the mostaggressivetest-and-testpolicydoesnoteliminatetheHIVepidemic[14].Thesendings areconsistentwiththoseofLongetal.[18], Walenskyetal.[26],andSorensenetal.[23],who ndonlymodestpolicyefectswhenmodelingtheepidemicintheUS[18,23,26]. We also nd that the epidemiologic benets of test-and-treat are counterbalanced by large MDRincreases.ItisuncleartowhatextentincreasesinMDRwillafectthecourseoftheHIV epidemic. ConsistentwithVelasco-Hernandezetal.[25]andLimaetal.[17],weshowthat evenwhenMDRincreases3-fold,increasingtreatmentratesstillbringsepidemiologicbenets 88 in the short run [17, 25]. Even when we simulate the model far into the future and MDR prevalenceapproaches23%,thepolicyisstillbenecial(AppendixB:SectionB.10).However, in the long run it is possible new forms of MDR evolve with higher resistance intensity or greater transmissibility, which could exacerbate MDR consequences. Conversely, it is also possible that MDR could be mitigated by the invention of new variants of drugs that are lesssusceptibletoresistance. WithsuchuncertaintyaboutthecourseofMDR,itisunclear whetherbenetsfromincreasingtest-and-treatoutweighcostsofincreasingMDR.Ifpolicy makersarerisk-averse,amorecautiousapproachtoHIVpreventionmightinvolvemorefocus ontestingwithoutchangingtreatmentguidelinestocoverearly-stageHIV.Wendthatby simply increasing testing rates, we derive roughly half the benets of the full test-and-treat policywithnoMDRincrease.Theresultsalsosuggestthatbenetsofincreasingtestingrates andtreatmentratesoccurindependently,suggestingthatimplementationneednotoccurin unison. Ourmodelhasitslimitations.First,mathematicalmodelsareonlyasgoodastheavailabledata usedfortheparametersandcalibration.Werigorouslycalibratedourmodeltoveryaccurately matchLACsurveillancedataandweshowthatifwehadnotcalibratedandinsteadsimply chosenparametersbasedonclinicalliterature,wewouldhaveunderestimatedthebenetsof test-and-treat. However,manyofourparametersarediculttomeasure,andHIVsurveil- lancedatausedforcalibrationitselfmighthavemeasurementerror.Second,ourmodelisnot stratiedbyrisk,ethnicity,orage.Futuremodelsshouldattempttoachievemoregranularity. Third,ourmodeldoesnotconsidertheuseofpreexposureprophylaxis.RecentUSFoodand DrugAdministrationapprovalofTruvadaforpreexposureprophylaxismayplayanimpor- tantroleinfutureHIVincidenceanddrugresistance. Fourth,oursimulationsdonotfully exploretheimplicationsofbehavioramongpeoplewhoinitiatetreatmentearly. Thepossi- bility exists that early treatment will result in increased risky behavior in HIV-positive and 89 susceptible populations because transmission is less likely. Another possibility is that peo- ple who initiate early-stage treatment are less likely to fully adhere to treatment; this could increase the acquired MDR rate, although the precise relationship between adherence and MDR is unknown. In summary, behavioral changes may mitigate the benets of test-and- treat,andfutureresearchshouldexploreimplicationsofthesebehavioralchanges. Fifth,we omitcostsfromourmodel,whichprecludesusfromidentifyingtheoptimalcombinationof testingandtreating. Futureresearchshouldincludeacost-efectivenessanalysisofalternate test-and-treat policy options to identify the optimal policy. Understanding budgetary and scaling-upconstraintsofalternate preventionprogramswouldprovideinsightonwhere to focusfuturepreventioneforts. Notes FinancialSupport: ThisworkwassupportedbytheEuniceKennedyShriverNationalInsti- tuteofChildHealthandHumanDevelopment(grantnumberR01HD054877).Thefunding sourcehadnoroleinthedesignorconductofthisstudy. AuthorContributions:Allauthors,andespeciallyR.V.andE.D.,contributedtothedesign ofthedeterministicepidemiologicalmodel.N.S.,Z.W.,andR.V.contributedtothewriting of the manuscript. Z. W., A. J., and E. D. contributed to the literature review and genera- tionofparameterranges.A.J.calculatedtheinitialconditionsandthemodelbaseline.Z.W. andN.S.createdallmanuscripttables. E.D., Z.W., andA.J.createdalltablesandgures inAppendixB.Allauthorscontributedtothestudydesign,modelcalibration,andwriting ofAppendixB. PotentialConictsofInterest:Allauthors;Noreportedconicts. 90 AllauthorshavesubmittedtheICMJEFormforDisclosureofPotentialConictsofInter- est. Conictsthattheeditorsconsiderrelevanttothecontentofthemanuscripthavebeen disclosed. References [1] Baggaley,R.F.,Garnett,G.P.andFerguson,N.M.Modellingtheimpactofantiretroviral useinresource-poorsettings. PLoS Medicine,3(4):e124,Apr2006. [2] Blower, S.M., Gershengorn, H.B. and Grant, R.M. A tale of two futures: HIV and antiretroviraltherapyinSanFrancisco. Science,287(5453):650–654,Jan2000. [3] CentersforDiseaseControlandPrevention.CDCFactSheet.NewestimatesofUSHIV prevalence,2006. Availableathttp://www.cdc.gov/hiv/topics/surveillance/,Oct2008. [4] CentersforDiseaseControlandPrevention. DiagnosesofHIVInfectionintheUnited StatesandDependentAreas,2009.Availableathttp://www.cdc.gov/hiv/pdf/statistics_ 2009_hiv_surveillance_report_vol_21.pdf,Feb2011. [5] Charlebois,E.D.etal. Theefectofexpandedantiretroviraltreatmentstrategiesonthe HIVepidemicamongmenwhohavesexwithmeninSanFrancisco. Clinical Infectious Diseases,52(8):1046–9,Apr2011. [6] Cohen,M.S.etal. PreventionofHIV-1infectionwithearlyantiretroviraltherapy. New England Journal of Medicine,365(6):493–505,Aug2011. [7] DelRomero,J.etal. Combinedantiretroviraltreatmentandheterosexualtransmission ofHIV-1: crosssectionalandprospectivecohortstudy. British Medical Journal,340: c2205,2010. 91 [8] Delva, W.etal. HIVtreatmentasprevention: PrinciplesofgoodHIVepidemiology modellingforpublichealthdecision-makinginallmodesofpreventionandevaluation. PLoS Medicine,9(7):e1001239,2012. [9] DepartmentofHealthandHumanServices. GuidelinesfortheUseofAntiretroviral AgentsinHIV-1-InfectedAdultsandAdolescents.Availableathttp://aidsinfo.nih.gov/ contentles/lvguidelines/adultandadolescentgl.pdf,2011. Accessed10August2012. [10] Dodd,P.J.,Garnett,G.P.andHallett,T.B. ExaminingthepromiseofHIVelimination by‘testandtreat’inhyperendemicsettings. AIDS,24(5):729–35,Mar2010. [11] Donnell,D.etal.HeterosexualHIV-1transmissionafterinitiationofantiretroviralther- apy:aprospectivecohortanalysis. Lancet,375(9731):2092–8,Jun2010. [12] Eaton,J.W.etal. HIVtreatmentasprevention: Systematiccomparisonofmathemati- calmodelsofthepotentialimpactofantiretroviraltherapyonHIVincidenceinSouth Africa. PLoS Medicine,9(7):e1001245,2012. [13] Garnett, G.P.andBaggaley, R.F. TreatingourwayoutoftheHIVpandemic: Could we,wouldwe,shouldwe? Lancet,373(9657):9–11,Jan2009. [14] Granich, R.M. et al. Universal voluntary HIV testing with immediate antiretrovi- raltherapyasastrategyforeliminationofHIVtransmission: Amathematicalmodel. Lancet,373(9657):48–57,Jan2009. [15] LADHSP. Programbrief: HIVpreventionthroughcareandtreatment. Availableat http://ph.lacounty.gov/aids/,2012. [16] Law,M.G.etal. ModellingtheefectofcombinationantiretroviraltreatmentsonHIV incidence. AIDS,15(10):1287–94,Jul2001. 92 [17] Lima,V.D.etal. Expandedaccesstohighlyactiveantiretroviraltherapy: Apotentially powerfulstrategytocurbthegrowthoftheHIVepidemic.JournalofInfectiousDiseases, 198(1):59–67,Jul2008. [18] Long, E.F., Brandeau, M.L. and Owens, D.K. The cost-efectiveness and population outcomesofexpandedHIVscreeningandantiretroviraltreatmentintheUnitedStates. Annals of Internal Medicine,153(12):778–89,Dec2010. [19] Marks,G.etal.Meta-analysisofhigh-risksexualbehaviorinpersonsawareandunaware theyareinfectedwithHIVintheUnitedStates:ImplicationsforHIVpreventionpro- grams. Journal of Acquired Immune Deficiency Syndromes,39(4):446–53,Aug2005. [20] Marks,G.,Crepaz,N.andJanssen,R.S. EstimatingsexualtransmissionofHIVfrom personsawareandunawarethattheyareinfectedwiththevirusintheUSA. AIDS,20 (10):1447–50,Jun2006. [21] Montaner,J.S.G.etal. Thecaseforexpandingaccesstohighlyactiveantiretroviralther- apytocurbthegrowthoftheHIVepidemic. Lancet,368(9534):531–6,Aug2006. [22] Smith, R.J. et al. Evolutionary dynamics of complex networks of HIV drug-resistant strains:ThecaseofSanFrancisco. Science,327(5966):697–701,Feb2010. [23] Sorensen,S.W.etal.Amathematicalmodelofcomprehensivetest-and-treatservicesand HIVincidenceamongmenwhohavesexwithmenintheUnitedStates. PLoS One,7 (2):e29098,2012. [24] vandeVijver,D.A.M.C.,Wensing,A.M.J.andBoucher,C.A.B. Theepidemiologyof transmissionofdrugresistantHIV-1.InHIVSequenceCompendium2007,pages17–36, 2007. 93 [25] Velasco-Hernandez,J.X.,Gershengorn,H.B.andBlower,S.M.Couldwidespreaduseof combinationantiretroviraltherapyeradicateHIVepidemics? LancetInfectiousDiseases, 2(8):487–93,Aug2002. [26] Walensky,R.P.etal.TestandtreatDC:ForecastingtheimpactofacomprehensiveHIV strategyinWashingtonDC. Clinical Infectious Diseases,51(4):392–400,Aug2010. [27] Wilson, D.P. HIV treatment as prevention: natural experiments highlight limits of antiretroviraltreatmentasHIVprevention. PLoS Medicine,9(7):e1001231,2012. 94 Chapter4 Acost-efectivenessanalysisofpre-exposure prophylaxisforthepreventionofHIV amongLosAngelesCountymenwhohave sexwithmen EmmanuelF.Drabo,JoelW.Hay,RaffaeleVardavas,ZacharyR. Wagner,NeerajSood Abstract Importance: Substantialgapsremaininunderstandingthetrade-ofsbetweenthecostsand benetsofchoosingalternativeHIVpreventionstrategies,includingmaintainingtheStatus Quo, expandedHIVscreening, expandedHIVscreeningcombinedwithearlyinitiationof treatmentwithantiretroviraltherapy(ART),andusingpre-exposureprophylaxis(PrEP). Objective:Toassessthepotentialtrade-ofsbetweenthecostsandbenetsofchoosingalter- nativeHIVpreventionstrategies,includingtheStatusQuo(currentHIVtestingwithART initiation at CD4 ≤ 500), Testing (expanded HIV testing alone with ART initiation at CD4≤ 500),Test-and-Treat(expandedHIVtestingcombinedwithimmediatetreatment), andPrEP(initiationofPrEPbyuninfectedindividuals)strategies. Design,SettingandParticipants: Wedevelopamathematicalepidemiologicalmodeltosim- ulateHIVincidenceamongmenresidinginLosAngelesCounty,CA,aged15-65year,who 95 havesexwithmen.Wecombinetheseincidencedatawithaneconomicmodeltoestimatethe costandefectivenessofvariousHIVpreventionstrategiesusingasocietalperspectiveanda lifetimehorizon. MainOutcomesandMeasures: OurintegratedmodelsallowustoestimatenewHIVinfec- tions averted, discounted costs, quality-adjusted life years (QALYs), and incremental cost- efectivenessratios(ICERs)ofvariousHIVpreventionstrategies. Results: PrEP and Test-and-Treat yield the largest reductions in HIV incidence. At a US willingness-to-paythresholdof$150,000/QALYsavedandrelativetoStatusQuo,Test-and- Treat and PrEP are highly cost-efective ($19,302/QALY and $27,863/QALY, respectively). StatusQuoandtwelveTest-and-TreatandPrEPstrategiesdeterminethefrontierforecient decisions. More aggressive strategies are more costly, but more efective, albeit with dimin- ishingreturns.Themostaggressivestrategy(EnhancedPrEPstrategywithHIVtestingevery 3 months and immediate ART and PrEP start) remains highly cost-efective relative to Sta- tusQuo($37,181/QALY).Thesendingsarerobusttoperturbationsoftheparametervalues withinreasonable bounds. Therelativeefectiveness ofPrEPissensitive toPrEPand ART adherenceandinitiationrates,aswellastheprobabilitiesofHIVtransmission,andtherates ofsexualpartnermixing. ConclusionsandRelevance:PrEPandTest-and-Treatofercost-efectivealternativestoStatus Quo.ThesuccessofthesestrategiesdependsonARTandPrEPadherenceandinitiationrates. ThelackofevidenceonadherencebehaviorstowardPrEP,therefore,warrantsfurtherstudies. Keywords:Cost-efectiveness,HIV,PrEP,Pre-exposureprophylaxis,Test-and-Treat,Testing. 96 4.1 Introduction The human immunodeciency virus (HIV) infects approximately 50,000 individuals each yearintheUnitedStates.In2012,nearly13%ofthe1.2millionHIV-positiveAmericanswere unawareoftheirserostatus[57]. Althoughthenumberofnewinfectionshasremainedrel- atively stable for the past decade [69], HIV prevalence has increased, in part owing to the longevity aforded by antiretroviral therapy (ART). However, disparities remain in the dis- easeburden,withmenwhohavesexwithmen(MSM),AfricanAmericansandthenonelderly adultpopulation(30-64yearsold)beingthemostaictedgroups,andaccountingrespectively for69%,46%and60%ofnewHIVinfectionsin2013[18,20,21,145]. Early detection of HIV infection followed by prompt treatment initiation and counseling, mayminimizeriskysexualbehaviorsandavertsecondaryHIVinfections[56,91,159]. Thus, theCentersforDiseaseControlandPrevention(CDC)reviseditsguidelinesin2006bycalling forroutineHIVscreeninginallhealthcaresettingsforpatientsaged13to64years[20,120].In 2013,theUSPreventiveServicesTaskForcealsorecommendedHIVscreeningforallpregnant women,adolescentsyoungerthan15years,andadultsolderthan65yearswhoareatincreased riskofHIVinfectionaswellasforalladolescentsandadultsaged15to65years,regardlessof risk[104]. Evidencefromnationalandinternationalstudiessuggeststhisapproachmaybe cost-efective[15,22,26,62,65,70,86,87,89,94,107,115]. Before2010,treatmentguidelinesrecommendedARTinitiationonlyinknownHIV-positive individuals with CD4 ≤ 350cells/L, or those with certain HIV-related comorbidi- ties[19,144]. Howeverin2010,themoreaggressivetest-and-treatpolicywasrecommended andbecamewidelyadopted. Test-and-Treatcallsforroutinetestingandprompttreatment start for all diagnosed cases, regardless of CD4 count, viral load, or the presence of AIDS- deningillness[33,48,49,51,56,91,98,138,150].Althoughthisapproachcouldhelpidentify 97 morecasesandavertnewinfections,itmayresultinincreasedresistancetoARTandcouldbe nanciallyburdensome[6,7,35,129,130]. Pre-exposureprophylaxis(PrEP)ofersanotherviablestrategyforpreventingHIVinfections, particularlyinhigh-risksubgroups[68].Withthisstrategy,uninfectedsusceptibleindividuals athighriskofcontractingHIVreceivedailydosesofacombinationcocktailofemtricitabine (FTC)withtenofovirdisoproxilfumarate(TDF)[42],theonlyPrEPregimenapprovedby theFoodandDrugAdministrationandcurrentlyendorsedbyWorldHealthOrganization (WHO)andUShealthocials[9,98,99,147].Recently,LosAngelesCounty(LAC)ocials also voted to roll out PrEP to its residents with high HIV-exposure risk, following similar programsinSanFranciscoandNewYorkState[126]. Despite strong evidence from randomized controlled trials (RCTs) on the preventive e- cacy of PrEP, limited evidence supports its cost-efectiveness relative to other strategies [3, 32, 42, 51, 105, 136]. Cost-efectiveness studies of PrEP for the US MSM population ofer mixedresultsandhavelimitations: MostonlycomparedscenariosimplementingPrEPwith scenarios without PrEP, rather than comparing competing strategies (e.g. Test-and-Treat, Testing) [31, 47, 71, 76, 112, 123]. Others did not capture secondary infections [112] or resis- tance to ART and PrEP [31, 76]. Finally, some did not capture the efect of ART-related adverse events [31, 71], behavioral changes that may be induced by PrEP (e.g. decreased condom use) [76] or incomplete adherence to PrEP, which could undermine the ecacy of PrEP [91, 93, 148] and contribute to spreading the epidemic and developing drug resis- tance[35,83,90,129]. This study assesses the potential trade-ofs between choosing the Status Quo (testing with treatment initiation at CD4 ≤ 500cells/L), Testing (expanded HIV testing), Test-and- Treat(expandedHIVtestingcombinedwithimmediatetreatment),andPrEP(initiationof 98 PrEP)strategiesamongthe15-to65-year-oldMSMinLAC,usingasocietalperspective.Sim- ilartoJuusolaetal.[71], weuseacompartmentalHIVtransmissionmodel. Ourmodeling approachaddressesthelimitationsofJuusolaetal.[71]andotherpriorworkbycomparing PrEPtocompetingpreventionstrategies,andbyaccountingforsecondaryinfections,changes indrugresistance,andincompleteadherence. 4.2 Methods 4.2.1 Epidemiologicalmodelstructure WeextendthemodelpresentedinChapter3(seealsoSoodetal.[129]),whichreproducedthe dynamicsoftheHIVepidemicinLACfrom2000to2010,andsimulatedtheefectofTesting andTest-and-TreatintheLACMSMpopulationbeyond2010. Webeginthesimulationin year2000with176,971MSMinLAC.Weestimatethat17.0%ofthemenareHIV-positive,and thatnearly25.0%oftheinfectedareunawareoftheirinfectionstatus(AppendixC:TableC.2). We use a one-year time step for each iteration of the model. In each simulation year, new susceptiblemenenterthemodelthroughaginganddiscoveryofsexualorientation,andexit the model through death. Once in the model, they can transition between health states, which comprise the uninfected (S, SJ, SPrEP), primary (P k , PPrEP k , PJ k , TPJ k ), asymptomatic (I k ,IPrEP k ,J k ,TJ k ), symptomatic (E k ,EJ k ,T k ), and AIDS (A k ,AJ k , TA k )stagesofHIVinfection,wherethesubscriptk denotesthedrug-sensitive(s)ordrug- resistant(r)strata(Figure4.1). Inthesespecications,theS,P,I,E andAcompartments denote the populations of individuals unaware of their serostatus whithe theSJ, PJ, J, EJ andAJ compartmentsdenoteindividualsawareoftheirserostatusthroughtesting.The TPJ,TJ,T andTA compartments denote individuals treated with ART in the primary, asymptomatic, symptomaticandAIDSstages, respectively. Finally, theSPrEP,PPrEP 99 andIPrEP compartments denote individuals using PrEP in the uninfected, primary and asymptomatic stages. In the model, individuals are ofered HIV testing at specied rates, based on current programs or more aggressive strategies that increase the screening rate. Starting in year 2013, those who test negative can initiate PrEP at specied rates of uptake, adherence, and ecacy in preventing HIV infection. We estimate that 12% of uninfected MSM and 24% of undiagnosed HIV+ MSM are eligible for PrEP (as determined by high- riskbehaviors)[27,46,52,78,81,82,101,114,125],andweassumeinourbasecaseanalysisthat 10%ofuninfectedMSMand25%oftheMSM(uninfectedandundiagnosedHIV+)adopt PrEP[50,53]. Similarly, thosewhotestpositiveandbecomeawareoftheirinfectionstatus areoferedARTatspeciedratesofuptake,adherence,andecacyinpreventingdiseasepro- gressionandtransmission. 100 Figure4.1:SchematicsoftheHIVSimulationModelwithPrEP. Note: Individualsenterthemodelthroughbirthatanannualrateπ. Onceinthemodel, theycantransition betweenhealthstates,whichcomprisetheuninfected(S,SJ,SPrEP),primary(P k ,PPrEP k ,PJ k ,TPJ k ), asymptomatic(I k ,IPrEP k ,J k ,TJ k ),symptomatic(E k ,EJ k ,T k ),andAIDS(A k ,AJ k ,TA k )stagesofHIV infection, where the subscriptk denotes the drug-sensitive (s) or drug-resistant (r) strata. S,P,I,E andA denotethepopulationsofindividualsunawareoftheirserostatus;SJ,PJ,J,EJ andAJ denoteindividuals awareoftheirserostatusthroughtesting. TPJ,TJ,T, andTAdenoteindividualstreatedwithARTinthe primary,asymptomatic,symptomaticandAIDSstages,respectively.Finally,SPrEP,PPrEP,andIPrEP denoteindividualsusingPrEPintheuninfected,primaryandsymptomaticstages. Atanystageinthemodel, individualscanexitthemodelthroughdeathatanaturaldeathrateμ,orfromHIV/AIDScomplications(γ A k , γ AJ k orγ TA k ). Estimatesanddenitionsoftheparametersanddatasourcesenteringthemodelareprovided in Table 4.2 and in Table C.1, and are discussed in Section 4.2. The directed arrows denote the transition of men between health states; the associated transition rates are denoted by the adjacent symbols to the arrows, andaredenedinTable4.2.Abbreviations:HIV=humanimmunodeciencyvirus;AIDS=acquiredimmune deciencysyndrome;PrEP=pre-exposureprophylaxis;ART=antiretroviraltherapy. AdetaileddescriptionofthismodelanditscalibrationtotheLACHIVepidemicisalsopro- videdinChapter3andinAppendixB,SectionB.2(seealsoSoodetal.[129]).Anattractivefea- tureofthispopulation-levelmodelisitsabilitytocapturesecondaryinfectionsandpredictthe population-levelefectsofvariousHIVinterventionstrategies.Inthisstudy,weusethemodel tosimulatetheHIVepidemicintheLACMSMpopulationunder623alternativestrategies 101 consistingofvariantsandcombinationsoftheTesting,Test-and-Treat,andPrEPstrategies, wherebytheintensityoftesting,ARTcoverageandPrEPuptakearechanged(AppendixC: TableC.19). 4.2.2 Economicmodelstructure Weusetheestimatesofnewinfectionsandtheannualpopulationsineachcompartmentas inputs for our economic model. The economic model estimates the total discounted costs and QALYs of the various interventions assessed in this study, and the incremental cost- efectivenessratio(ICER)ofthe623alternativestrategiessimulatedwiththeepidemicmodel (TableC.19). WeadoptaUSsocietalperspectiveandmeasurethecostandefectivenessout- comesindiscountedsocietallifetimecostsandQALYs[45]. Starting in year 2013, and for each policy scenario, we use the epidemic model to simulate theannualnumberofMSMineachhealthstateofthemodeloveralifetimehorizon. Next, wemultiplytheestimatedpopulationineachhealthstateandeachyearbytheirassociated annualcoststoobtainthetotallifetimecosts.Similarly,wecomputethetotalsocietallifetime QALYsbymultiplyingthepopulationineachhealthstateandeachyearbythehealthstate- specic QALY estimates. Third, we calculate both the discounted total lifetime costs and QALYsforeachstrategy,usinga3%annualdiscountrate[45,64,86].Finally,foreachpairof interventionscompared,wecomputetheincrementaldiscountedsocietalcostsandQALYs andcalculatethecorrespondingICER.WealsocalculatesequentialICERs[34]andtracethe cost-efectivenessfrontier,whichrepresentsthesetofalldominantandcost-efectivestrategies orstrategiesthatachievethemaximumefectivenessforagivenvalueofsocietalcosts[43,44]. We implement both epidemiological and economic models in the R programming pack- age[142]. 102 4.3 Inputdata WederivetheinitialLACMSMandmaledemographicandHIVprevalencedatafromChap- ter3(seealso Soodetal.[129]),theLACannualHIVsurveillancereports,andtheRAND CaliforniaPopulationandDemographicsdatabase[135].Ourmethodologyforestimatingthe compartment-specicpopulationsisdescribedinAppendixB(SectionB.3.2)andAppendixC (Section C.2.2). Average age group specic life expectancy estimates for the uninfected and HIV-infected male populations are derived from the CDC life tables [2, 63] (Appendix C: SectionC.2.2,TableC.3andTableC.4).Weobtaintheepidemicmodelparameters(e.g.HIV transmissionrates,parametersassociatedwithHIVnaturalhistory,HIVtestingrates,ART andPrEPinitiationanddiscontinuationrates,ARTandPrEPecacy)fromChapter3(see alsoSoodetal.[129])andthepublishedliterature, followingasystematicreviewoftheevi- dencefromlargeRCTstudiesofHIVtreatmentandprevention[3,51,91,93,136,148](see AppendixC:SectionC.2.4.3-C.2.4.9,TableC.6-C.16,andSectionC.6). We also derive the health state specic costs and efectiveness parameters for the economic model from the published literature following a systematic review approach [31, 47, 58, 70, 71, 76, 86, 87, 112, 155], and from government fee schedules, which reect the best approx- imation to the societal drug cost [24, 146]. Using these estimates, we calculate the annual costs for the health states to include goods and services involved in the delivery of medical care,suchasphysicianvisits,drugs(ARTandPrEPregimens),managementofopportunistic infections, tests for HIV (eg, enzyme immunoassay [EIA]; enzyme-linked immunosorbent assay[ELISA];rapidHIVtest;conrmatorytestingusingnucleicacidamplicationtesting [NAAT]), STIs, serum BUN and creatinine levels; CD4 count and viral load monitoring; andpretestandposttestcounseling,andlinkagetocare.Weconvertallhealthcarepricesinto 2013USdollars, assuming9%peryearincreaseinhealthcareprices[45]. Weexcludedirect nonmedicalcosts,i.e.,thoseincurredbeyondthehealthcaresetting(eg,transportationcosts, 103 otherout-of-pocketexpenses,resourcesfromotheragencies)becausetheyaresmallandlikely similaracrosstheassessedalternatives[45]. Indirectcostsassociatedwithinformalcaregiver supportandunpaidhelpbyfamilyandfriendsarecalculatedusingestimatedaveragehome healthcarecosts[100],aswellasestimatesofAIDSpatients’homecareutilization[36,37], weightedbythenationalaveragehourlycompensationratesofhomehealthandpersonalcare aides(2010US$11.63/hour)[17].Weexcludechangesincaregivers’qualityoflifeduetotheir caregivingactivities,becausethedenominatorshouldexclusivelyincludehealth-related(not care-related)qualityoflife[116]. Otherindirectcostsrelatedtothevalueoftheindividual’s forgone(orgained)productivityattributabletotheillness-relatedmorbidityandmortality, aremeasuredinutilityandcapturedbytheQALYestimates[64]. WecalculatethehealthstateQALYsbymultiplyingthehealth-state-adjustedaveragehealth- related quality of life score (HRQOL) with the number of MSM in that health state [71] (AppendixC:TableC.17-C.18andEquationC.81).TheQALYandHRQOLscoreestimates areobtainedfromthepublishedliterature.Alldemographic,costs,andefectivenessparame- tersaresummarizedinTable4.2,anddescribedingreaterdetailinAppendixC:SectionC.2.1 -C.2.2. 104 Table4.1:Basecasemodelassumptions. Parameter StatusQuo Testing Test-and-Treat PrEP HIVtestingrate(Frequency) 0.227 (Every4.4y) 1.000 (Annually) 0.500 (Every2y) 0.500 (Every2y) Bloodureanitrogenconcentration, serumcreatininelevelsandSTI testingfrequency - - - Every3mo ARTinitiationrateat CD4≤ 500cells/L(frequency) 0.404 (Every2.5y) 0.404 (Every2.5y) 0.404 (Every2.5y) 0.404 (Every2.5y) EarlyARTinitiationrate - - Immediate Immediate PrEPinitiationrate - - - 0.250 (Every4y) ARTandPrEPadherencerates 0.282 0.282 0.282 0.282 ARTandPrEPdiscontinuationrates 0.116 0.1161 0.116 0.116 Reductioninriskysexualbehavior owingtotestingandcounseling - - - 0.200 ReductionsexualinfectivityowingtoART 0.900 0.900 0.900 0.900 ReductionsexualinfectivityowingtoPrEP - - - 0.920 ReductioninriskofinfectionowingtoPrEP - - - 0.440 105 Table4.2:Summaryofkeymodelinputparameters. Parameters Value Range Reference EpidemicParameters Demographic Parameters π:Annualinowofsusceptibleindividuals 3,597 3,143-3,825 [129,143] μ:Naturalrateofdeath 4.0000 0.0003-0.0004 [129,143] HIV Transmission Parameters C mix :Sexualmixingrate 4.5046 2.2798-8.4753 [74,75,109,127–129,140,156] λ s = λ sJ : Transmissionratefortheawareandunawaresusceptiblepopu- lationsnotreceivingPrEP(drugsensitive) Varies - [129] λ r =λ rJ : Transmissionratefortheawareandunawaresusceptiblepopu- lationsnotreceivingPrEP(drugresistant) Varies - [129] λ sPrEP : Transmission rate for the susceptible populations treated with PrEP(drugsensitive) Varies - [4,71] λ rPrEP : Transmission rate for the susceptible populations treated with PrEP(drugresistant) Varies - [4,71] Disease Progression ρ = ρ PrEP : ProgressionratefromtheprimarytotheasymptomaticHIV stage 11.0136 6.7713-22.1852 [127,129] ξ s = ξ r : Progression rate from asymptomatic to untreated symptomatic HIV(unawareHIV+individuals) a 0.32 0.2501-0.8302 [77,84,102,127,129] ν s = ν r : Progression rate from asymptomatic to untreated symptomatic HIV(awareHIV+individuals) 0.32 0.000-0.8302 [77,84,102,127,129] θ: RateofdiseaseprogressionfromthetreatedasymptomaticHIVstageto thetreatedsymptomaticHIVstage 0.1949 0.1158-1.7500 [59,71,86,88,95,129] γ EJs = γ Es : Progression rate to AIDS in treatment-eligible individuals (drug-sensitivestrata) 0.6658 0.2674-0.6693 [103,129,153] γ EJr = γ Er : Progression rate to AIDS in treatment-eligible individuals (drug-resistantstrata) 1.3080 0.5314-1.3386 [129] γ Ts :ProgressionratetoAIDSinART-treatedindividuals 0.0777 0.0468-0.0875 [127,129] γ Tr :ProgressionratetoAIDSinART-treatedindividuals 0.1947 0.1045-0.5370 [5,10,129] Continued... 106 Table4.2:Summaryofkeymodelinputparameters. Parameters Value Range Reference HIV/AIDS-related Mortality γ AJs =γ As :UntreatedindividualswithAIDS(drugsensitive) 0.5427 0.5093-7.8389 [103,129] γ AJr =γ Ar :UntreatedindividualswithAIDS(drugresistant) 1.7016 1.1098-19.8800 [129] γ TAs :ART-treatedindividualswithAIDS(drugsensitive) 0.1187 0.0795-0.4120 [5,10,124,127,129] γ TAr :ART-treatedindividualswithAIDS(drugresistant) 0.4891 0.1643-0.8296 [129] Screening and Counseling 1/ψ:Averagedurationofidenticationforsusceptibleindividuals 1 0.5000-3.0000 [71] ω S :Rateofidenticationforsusceptibleindividuals Varies - [127,129,145] ω SPrEP :RateofHIVtestinginsusceptibleindividualsdiscontinuingPrEP Varies - [127,129,145] ω P :Rateofserostatusidenticationfornon-PrEPusersintheprimarydis- easestage Varies - [127,129,145] ω PPrEP :RateofserostatusidenticationforPrEPusersintheprimarydis- easestage Varies - [127,129,145] ω I :Rateofserostatusidenticationfornon-PrEPusersintheasymptomatic diseasestage Varies - [127,129,145] ω IPrEP : Rate of serostatus identication for PrEP users in the asymp- tomaticdiseasestage Varies - [127,129,145] ω E :Rateofserostatusidenticationforindividualsinthesymptomaticdis- easestage Varies - [127,129,145] ω A :RateofserostatusidenticationforindividualsintheAIDSstage Varies - [129,132,145] τ C :Reductioninriskysexualbehaviorowingtotestingandcounseling 0.2 0.0000-0.5000 [71,88,92,119,121,131] ART and PrEP σ:ARTinitiationrateinHIV+individualswithoutAIDS 0.4040 0.3353-6.8907 [129] σ SPrEP :PrEPinitiationrateinthesusceptiblepopulation 0.0987 0.0888-0.1098 [46,78,82,101] σ PPrEP :PrEPinitiationratebyunawareHIV+individualsintheprimary stageofinfection Varies - [46,78,82,101] σ IPrEP : PrEPinitiationratebyunawareHIV+individualsintheasymp- tomaticstageofinfection Varies - [46,78,82,101] σ TPJ =σ TJ : ARTinitiationratebyidentied(aware)HIV+individuals intheasymptomaticstageofinfection 365 0-365 Assumed Continued... 107 Table4.2:Summaryofkeymodelinputparameters. Parameters Value Range Reference σ A :ARTinitiationrateinindividualswithAIDS 10.8637 0.6766-20.6688 [129] g =g SPrEP =g PPrEP =g IPrEP =g TPJ =g TJ :ARTdiscontinua- tionrateinHIV+individualswithoutAIDS 0.1160 0.0234-0.1576 [127,129] g A :ARTdiscontinuationrateinindividualswithAIDS 0.0314 0.0024-0.0774 [129] τ ART :ReductioninsexualinfectivityowingtoART 0.9 0.5000-0.9900 [1,25,26,71,72,86,88,96,117,121,134, 152,157] τ PrEP :ReductioninsexualinfectivityowingtoPrEP 0.92 0.5000-0.9900 [4,71] Resistance r =r TJ :RateofacquiredMDR 0.0278 0.0061-0.0535 [127,129,149] r PrEP :RateofacquiredMDRinPrEPusers 0 0.0000-0.0161 [51,80] h r :MDRtransmissibilitymultiplicativefactor 0.1232 0.1000-0.1756 [127,129] q:Rateofmutationfromacquiredresistanttothedrugsensitivestrain 0.0043 0.0006-0.0307 [14,129] CostParameters Annual HIV-related Health Care Costs ($) AcuteHIV c 30 10-500 [30,70,71,122,151] AsymptomaticHIV-untreated c 4,130 3,000-6,000 [8,13,71,88] SymptomaticHIV-untreated c 6,934 5,000-9,000 [8,13,71,88] SymptomaticHIV-treatedwithART,excludingARTcosts c 6,181 5,000-7,000 [8,13,71,88] AIDSuntreated cd 21,863 15,000-25,000 [13,36,37,41,66,71,85,88,100,146] AIDS-treatedwithART,excludesARTcosts cd 9,950 6,000-17,000 [8, 36, 37, 41, 71, 85, 88, 100, 121, 124, 146] Annualcostofantiretroviraltherapy(ART) c 15,000 13,520-17,109 [66,88,121,124,146] Cost of PrEP (cost per test, refill, or visit; $) PrEP(tenofovir,TDF/emtricitabine,FTC):30-dsupply c 776 672-925 [112,137,146] STItesting:costpertest c 54 25-75 [23,71] Bloodureanitrogenconcentrationandserumcreatinineleveltesting: cost pertest c 23 10-40 [23,71] Physicianvisit:costpervisit c 100 10-200 [23,71] Cost of HIV Testing (cost per test; $) Continued... 108 Table4.2:Summaryofkeymodelinputparameters. Parameters Value Range Reference Costofinitialtest:3rd/4thgenerationtest(EIA/ELISA;CPT:86703,G0432, G0433,87389)orrapidHIVtest(CPT:G0345) b 19 9-45 [24] CostofconrmatorytestingorHIVRNAtest(NAATtestforHIVRNA; CPT:87535) b 48 16-158 [24] CostofCD4cellcountmonitoring(CPT:86359,86360,86361) b 52 10-87 [24] CostofHIVgenotypetest(CPT87901,87906) 177 54-239 [24] Cost of Counseling (cost per visit; $) Pretestcounseling c 13 0-100 [38,71,111] PosttestcounselingforHIVnegativeindividuals c 7 0-50 [38,71,111] Posttestlinkage/counselingforHIV-positiveindividuals c 14 0-100 [38,71,111] Other Costs and Cost-related Parameters CostofHIVdiagnosis($) c 500 125-1,200 [23,71] Annualcostdiscountrate 0.03 0.0000-0.0500 [29,70,71,86] 2013to2010inationfactor 1.0900 - [16] EfectivenessParameters Disease State QOL Utility Weights Uninfected(noPrEP) 1 - [40,70,71,86,88] Uninfected(PrEP) 1 0.9000-1.0000 [51,71,82] AcuteHIV,unidentied 0.92 0.7300-0.9700 [12,30,70,71,73,106,108,122,141,151] AcuteHIV,identied 0.86 0.6800-0.9100 [12,30,61,70,71,73,106,108,121,122, 141,151] AcuteHIV,treatedwithART 0.88 0.6800-0.9400 [54,76] AsymptomaticHIV,unidentied 0.91 0.8500-0.9500 [61,70,71,121] AsymptomaticHIV,identied(Year1) 0.84 0.8400-0.9500 [29,70,71,86,121] AsymptomaticHIV,identied(Year2+) 0.89 0.8500-0.9500 [29,70,71,86,121] AsymptomaticHIV,treatedwithART 0.91 0.8500-0.9500 [29,54,70,71,86] SymptomaticHIV,unidentied 0.8 0.7000-0.8000 [60,70,71,86,121,133] SymptomaticHIV,identied 0.72 0.7000-0.8000 [61,70,71,86] SymptomaticHIV,treatedwithART 0.83 0.7800-1.0000 [29,60,70,71,86,88,121] AIDS,unidentied 0.72 0.2400-0.8000 [29,71,87,88,121] Continued... 109 Table4.2:Summaryofkeymodelinputparameters. Parameters Value Range Reference AIDS,identied 0.72 0.6000-0.7500 [60,71,87,88,133] AIDS-treatedwithART 0.82 0.8200-0.8700 [60,71,87,88] Other Eectiveness Parameters QOLdecrementfactorforfalse-positiveresult 0.12 0.0000-0.4800 [61,71,110,121] QOLdecrementfactorowingtoresistancetoARTorPrEP 0 0.0000-0.0100 Assumed AnnualQOLdiscountrate 0.03 0.0000-0.0500 [29,70,71,86] Note:DiseasestateQOLutilityweightsareage-unadjusted.Valuesareindicatedbythemention“varies”whenevertheparameterestimatevarieswithtimeorthescenario; detailsonparametercalculationsareprovidedinSectionC.2.4. a Estimatebasedonthe2011treatmentguidelines.Priorto2011,theestimateusedis0.1700(range,0.1450-0.1956); b Year2010USdollars; c Year2013USdollars; d Inclusiveofinformalsupportcosts. 110 4.4 Results 4.4.1 Ecientstrategies Our results suggest that the Status Quo and 12 of the 623 strategies assessed determine the ecient cost-efectiveness frontier. These rational decisions consist of variants of the Test- and-Treat, andPrEPstrategies(Table4.3andFigure4.2), andexcludeallTestingstrategies, whichareallextendedlydominated(haveahigherICERthanthenextmoreefectivealter- natestrategy)orstronglydominated(havehighercostsandlowerefectivenessthanthealter- nate strategy). As Table 4.3 and Figure 4.2 indicate, the least costly ecient decision rela- tivetoStatusQuoistheTest-and-Treatstrategy(SQ+ImmediateEarlyART;strategy2on Figure4.2)whichcosts$19,302perQALYgained,andishighlycost-efectiveatthecurrent US willingness-to-pay threshold of $150,000 per QALY gained (based on the WHO’s soci- etal willingness to pay estimate) [158]. Test-and-Treat extendedly dominates strategies that combineStatusQuowithlessaggressiveearlytreatmentorHIVtesting(e.g.earlyARTstart everymonth,HIVtestingevery4yearscombinedwithearlyARTstartevery6months;see Table4.3).Test-and-Treatisfollowedby4Test-and-TreatstrategiesenhancedwithHIVtest- ingevery3years,2years,1year,and6months(strategies3-6onFigure4.2). Relativetothe precedingrationalstrategyonthefrontier,andrelativetoStatusQuo,all4EnhancedTest- and-Treatstrategiesarehighlycost-efective.Forexample,relativetoTest-and-Treat,theTest- and-TreatstrategyenhancedwithHIVtestingevery3years(strategy3onFigure4.2)would cost$20,451perQALYgained. Similarly,themostaggressiveenhancedTest-and-Treatstrat- egyonthefrontier(TT+Test6mo;strategy6onFigure4.2)wouldcost$25,654perQALY gained relative to Test-and-Treat, and $38,492 per QALY gained, relative to Test-and-Treat enhancedwithannualtesting(TT+Test1y;strategy5onFigure4.2). 111 Figure4.2:Ecientfrontierforresourceallocation. 3.0 3.1 3.2 3.3 3.4 3.5 ● ● ● ● ● ● ● 3.0 3.1 3.2 3.3 3.4 3.5 45 50 55 60 65 19 20 24 31 38 63 85 105 139 146 189 235 1 2 3 4 5 6 7 8 9 10 11 12 13 Discounted Effectiveness (QAL Ys, million) Discounted Lifetime Costs (2013 US$, billion) ICER relative to prior strategy on the efficient frontier ($ '000/QAL Y gained) ● ● 1. Status Quo, SQ (Test 4.4 y + ART 2.5 y at CD4 ≤ 500) 2. Test−and−treat, TT (SQ + Immediate Early ART at CD4>500) 3. Enhanced TT (TT + Test 3 y) 4. Enhanced TT (TT + Test 2 y) 5. Enhanced TT (TT + Test 1 y) 6. Enhanced TT (TT + Test 6 mo) 7. PrEP (TT + Test 6 mo + PrEP 4 y) 8. Enhanced PrEP (PrEP + PrEP 3 y) 9. Enhanced PrEP (PrEP + PrEP 2 y) 10. Enhanced PrEP (PrEP + PrEP 1.2 y) 11. Enhanced PrEP (PrEP + Test 3 mo + PrEP 2 y) 12. Enhanced PrEP (PrEP + Test 3 mo + PrEP 1.2 y) 13. Enhanced PrEP (PrEP + Test 3 mo + Immediate PrEP) Dominated Testing Dominated Test−and−treat Dominated PrEP Efficient frontier Note:Theecientfrontier,indicatedbythesolidblackline,denotesstrategiesthatyieldthehighestvalue(low- estcostperQALYsgained)foradenedlevelofsocietalwillingnesstopay. Itcanbeusedtodeterminehow muchhealthbenetsareobtainablefromtheresourcesusedbyaspecicclinicalinterventionandunderagiven budget constraint. Points on the ecient frontier (strategies 1-13) are cost-efective; the grey points to the left oftheecientfrontierindicatestronglyandextendedlydominatedalternativestrategies,whicharevariantsor combinationsofthetesting,test-and-treat,andPrEPstrategies,wherebythefrequenciesoftesting,ARTcov- erageand/orPrEPuptakearevaried(thesestrategiesarelistedinAppendixC,TableC.19). Positivegradients (e.g. betweenpoints3and4)reecttheICERsofeachstrategyonthefrontierrelativethepriorstrategyonthe frontier (i.e. additional costs for increased health benets), and are captured by the values on the right-hand sidey-axis(e.g. $24,394perQALYgainedbetweenpoints3and4). Abbreviations: y=year;mo=month;SQ =StatusQuo;ART=antiretroviraltherapy;TT=test-and-treat;PrEP=pre-exposureprophylaxis;QALY= quality-adjustedlifeyear. 112 Table4.3:BenetsandcostsofthemostecientTest-and-TreatandPrEPstrategies. RationalDecision (OnEcientFrontier) Discounted a Inc.Values b Extendedly DominatedStrategies c Cost, $B QALYs, M Costs, $B QALYs ICER, $/QALY StatusQuo(SQ) d 43.58 2.96 - - - - Test-and-treat,TT (SQ+Imm.EarlyART) d 45.18 3.05 1.6 82,915 19,302 SQ+Test4y SQ+EarlyART1mo SQ+T4y+EarlyART6mo E.TT (TT+Test3y) 48.97 3.23 3.79 185,522 20,451 E.TT(TT+Test4y) E.TT(TT+Test2y) TT+T4y+PrEP3y E.TT (TT+Test2y) 50.85 3.31 1.88 76,882 24,394 SQ+T2y+EarlyART3mo SQ+T2y+EarlyART1mo E.TT (TT+T1y) 53.32 3.39 2.47 79,527 31,036 SQ+T1y+EarlyART3mo SQ+T1mo+EarlyART3mo TT+T3y+PrEP2y E.TT (TT+Test6mo) 55.22 3.44 1.90 49,415 38,492 SQ+Test6mo+EarlyART1mo TT+T2y+PrEP2y PrEP (TT+T6mo+PrEP4y) 58.03 3.48 2.81 44,457 63,269 SQ+T1y+EarlyART1mo+PrEP3y TT+T1y+PrEP2y E.PrEP (PrEP+PrEP3y) 58.55 3.49 0.52 6,111 85,117 SQ+T3mo+EarlyART1mo+PrEP3y E.PrEP (PrEP+PrEP2y) 59.36 3.5 0.81 7,707 104,788 SQ+T6mo+EarlyART1mo+PrEP2y E.PrEP (PrEP+PrEP1.2y) 60.33 3.5 0.97 6,945 139,346 E.PrEP(PrEP+T3mo+PrEP4y) E.PrEP(PrEP+T3mo+PrEP3y) Continued... 113 Table4.3:BenetsandcostsofthemostecientTest-and-TreatandPrEPstrategies. RationalDecision (OnEcientFrontier) Discounted a Inc.Values b Extendedly DominatedStrategies c Cost, $B QALYs, M Costs, $B QALYs ICER, $/QALY E.PrEP (PrEP+T3mo+PrEP2y) 61.01 3.51 0.68 4,676 145,956 E.PrEP(PrEP+PrEP1.2y) E.PrEP (PrEP+T3mo+PrEP1.2y) 61.93 3.51 0.92 4,892 188,714 E.PrEP(PrEP+PrEP6mo) E.PrEP (PrEP+T3mo+Imm.PrEP) 64.38 3.52 2.4500 10,429 234,726 E.PrEP(PrEP+Imm.PrEP) Abbreviation: Inc. =incremental;B=billion;M=million;ICER=incrementalcostefectivenessratio;Imm. =immediate;T=testing;E=enhanced;y=year;mo= month;SQ=statusquo;ART=antiretroviraltherapy;TT=test-and-treat;PrEP=pre-exposureprophylaxis;QALY=quality-adjustedlifeyear. a Discountedat3%annualdiscountrate;costspresentedin2013US$. b Relativetothepriorrationaldecisionontheecientfrontier;costspresentedin2013US$. c Onlyincludesselectedextendedlydominatedstrategiesduetospacing.ThecompletelistoftheextendedlydominatedstrategiesisprovidedinAppendixC,TableC.22. d Test4.4y+ART2.5yatCD4≤ 500. e EarlyARTdenedasARTstartatCD4> 500. 114 ThemostaggressiveandoptimalEnhancedTest-and-Treatstrategy(strategy6onFigure4.2) is followed by the PrEP strategy, which combines Test-and-Treat with HIV testing every 6 monthsandPrEPstartevery4years(TT+Test6mo+PrEP4y;strategy7onFigure4.2). PrEPwouldcost$27,863;$29,492;and$63,269perQALYgainedrelativetoStatusQuo,Test- and-Treat,andTest-and-TreatenhancedwithHIVtestingevery6months,respectively.PrEP isfollowedby3PrEPstrategiesenhancedwithPrEPstartevery3,2,and1.2years(strategies 8-10onFigure4.2), aswellas3PrEPstrategiesenhancedwithHIVtestingevery3months, andPrEPstartevery2years,1.2years,andimmediately(strategies11-13onFigure4.2).Relative tothepriorrationalPrEPstrategiesonthefrontier,4enhancedPrEPstrategies(strategies8-11 on Figure 4.2) are highly cost efective and would cost $85,117; $104,788; $139,346; $145,956 per QALY gained, respectively (Table 4.3). The two remaining enhanced PrEP strategies (strategies12-13onFigure4.2)arecost-inefectiverelativetotheprecedingrationalenhanced PrEP strategy on the frontier ($188,714 and $234,726 per QALY gained, respectively). All enhanced PrEP strategies are highly cost efective relative to Status Quo ($28,529; $29,633; $31,045;$32,033;$33,429and$37,181perQALYgained,respectively;seeTableC.22)andTest- and-Treat($30,261;$31,538;$33,178;$34,321;$35,942,$40,292perQALYgained,respectively). RelativetoPrEP,allenhancedPrEPstrategies(exceptforstrategy13whichcosts$155,770per QALYgained)arealsocostefective($85,117;$96,088;$110,557;$117,064;$128,622perQALY gained,respectively). Collectively, theseresultssuggestthatthemostaggressivestrategiesaremoreexpensive, but alsomoreefective,albeitwithdiminishingreturns,asindicatedbythecurvatureofthefron- tier (Figure 4.2). However, relative to Status Quo, even the most aggressive intervention (EnhancedPrEPwithHIVtestingevery3monthsandimmediatePrEPstart)remainshighly 115 cost-efective. TheresultsalsosupportthehypothesisthatPrEPandTest-and-Treatarecost- efectivealternativestoStatusQuoinHIVprevention.Thislikelyowestothepreventiveben- etsofPrEPandthesurvivalgainsfromearlyknowledgeofserostatusthroughearlydetection andpromptARTinitiation. 4.4.2 Epidemiologicaloutcomes Intermsofepidemiologicaloutcomes,oursimulationindicatesthattheStatusQuoapproach wouldyieldacumulativeHIVincidenceof99,874cases.Relativetothisgure,Test-and-Treat andPrEPwouldrespectivelyavert4,332(4.3%)and58,881(59.0%)newinfections(TableC.21). ThemostaggressiveEnhancedTest-and-Treatstrategy(Test-and-Treatenhancedwithannual HIVtesting; strategy6onFigure4.2)wouldavert47,759(47.8%),43,427(45.5%),and5,711 (9.9%)infections,relativetoStatusQuo,Test-and-Treat,andTest-and-Treatenhancedwith annualHIVtesting(strategy5onFigure4.2),respectively. PrEPwouldavert58,881(59.0%), 54,549(57.1%),and11,123(21.3%)infectionsrelativetoStatusQuo,Test-and-Treat,andTest- and-TreatenhancedwithHIVtestingevery6months(strategy6onFigure4.2),respectively. ThemostaggressiveEnhancedPrEPstrategy(PrEPenhancedwithimmediatePrEPstartand HIV testing every 3 months; strategy 13 on Figure 4.2) would avert 77,301 (77.4%), 42,969 (76.4%),29,543(56.7%),18,420(44.9%),and7,911(26.0%)infectionsrelativetoStatusQuo, Test-and-Treat,Test-and-TreatenhancedwithHIVtestingevery6months(strategy6onFig- ure4.2),PrEP,andPrEPenhancedwithimmediatePrEPstartandHIVtestingevery3months (strategy12onFigure4.2),respectively. 4.4.3 Sensitivityanalyses Weconductseveralsensitivityanalysestoassesstheefectofuncertaintyinourmodelparame- tervaluesontheICERs.First,weconductaseriesof1-waysensitivityanalysesbyvaryingeach 116 parametervalueoneatatimewithintheuncertaintyrangesoftheparameters.Second,wecon- ductaprobabilisticsensitivityanalysisbysimultaneouslyvaryingallparametervalueswithin theiruncertaintyranges.Wesamplepolicyandefectiveness(QOL)parametersaccordingtoa programevaluationandreviewtechnique(PERT)distribution,whichisavariantofthebeta distributionandiswellsuitedformodelingexpertestimatesandpolicyvariables[129,154]. Cost parameters are sampled following a lognormal distribution to account for the skewed and fat-tail nature of cost data [139]. Other parameters are sampled following a normal or uniformdistribution.Weindependentlysampledallparametersthroughouttheanalysis.We nd that all ICERs remain robust to perturbations of the epidemic, cost, and efectiveness parametervalues(FigureC.3-FigureC.14). Thesexualmixingandtransmissionparameters, theintensityoftesting,theratesofadherencetothetreatmentregimen,andinitiationofPrEP andARTaresignicantmodulatorsoftheICERs.Thesendingsagreewithpriorstudieson the benets of adherence to ART in reducing viral load, averting CD4 cells depletion, and reducingtheriskofresistancetoART[113]. Third, we assess the efect of ART and PrEP price reductions on the ICERs. Several stud- ies showed that the expiration of a brand-name drug patent increases market competition, therebyreducingprices[79]. Thesepricereductionsrangetypicallybetween20%and70% ofthebrand-nameproductprice[118]. Oursensitivityanalysisinthisrangeofpricereduc- tion suggests that all cost-efectiveness proles improve with generic entry (Figure C.17 and FigureC.18).The95%simulationintervalsarerelativelynarrow,suggestingrobustnessofthe ICERs. 117 4.5 Discussion Our study suggests that PrEP and Test-and-Treat constitute cost-efective HIV prevention alternativestotheStatusQuo,andthatrelativetoStatusQuo,themostecientPrEPstrate- gies could cost $27,863 to $37,181 per QALY gained, whereas the Test-and-Treat strategies couldcost$19,302to$24,544perQALYgained. Theseresultsareconsistentwiththe Desai etal.[31]ndingthatPrEPfortheNewYorkCityhigh-riskMSMwouldcost$32,000/QALY. Theydifer, however, fromthe Juusolaetal.[71]estimatesof$50,000/QALYinhigh-risk MSMand$172,091/QALYinthegeneralMSMpopulationat20%PrEPcoverage.Theyalso diferfrom KoppenhaverandSorensen[76]estimatesof$353,739and$570,273perQALY gainedamonghighlyadherentMSM(i.e. thosetaking>90%ofPrEPdoses, asdetermined by pill counts) and in the overall population, respectively, at universal PrEP coverage. The discrepancywith Juusolaetal.[71]and KoppenhaverandSorensen[76]likelyowestodif- ferencesinmodelingassumptionsandepidemictrendsinoursettings.Forexample,whereas Juusola et al. [71] studied the entire US MSM population, in which the HIV prevalence is relatively low (12.3%), our study is restricted to the LAC MSM population, in which HIV prevalenceismuchhigher(17.0%in2010). WhiletheinitialHIVprevalenceinourpopula- tion (17.0%) is similar to that in Koppenhaver and Sorensen [76] (17.5%), the two studies difer importantly in their assumptions about PrEP coverage: Koppenhaver and Sorensen [76] assumed universal PrEP coverage among susceptible MSM, whereas we assumed that only10%oftheMSMwouldinitiatePrEP. OurresultssupportpriorndingsthatPrEPcanbecost-efectiveinhighlyconcentratedepi- demicsettingsevenwhenarichersetofalternateHIVpoliciesareevaluated[47]. However, theoptimalstrategydependsonthecostssocietyiswillingtoincurforHIVprevention.With constrained budgets Test-and-Treat is the optimal policy and with less constrained budgets 118 Test-and-Treat combined with PrEP is the optimal policy. Overall these results help poli- cymakersandpublichealthocialschoosetheoptimalHIVpreventionstrategygiventheir budgetconstraints.TheresultsalsosupporttherecentLAC,WHOandUSocials’endorse- ment of PrEP, as well as New York Governor Cuomo’s call for a statewide adoption of the strategy[9,98,99,126].However,ourresultsalsosuggestthateventhemostaggressivecost- efectiveHIVpreventionstrategyisunlikelytoeliminatetheHIVepidemic. Thesuccessofthesestrategiesdependsontheuptakeofandadherencetotreatment.Thelack ofevidenceonbehavioralresponsestoPrEP,therefore,warrantsfurtherstudies.Tothatend, theALERTstudyofadherencetoPrEPatseveralsouthernCaliforniaclinicsshouldgenerate valuableresults[97]. AmonthlyorquarterlyinjectablePrEP(GSK-744)inaPhaseIIclini- caltrial[55]couldalsohelpimproveadherenceifsuccessful,andtherebyrevolutionizeHIV prevention. Thisstudyhasseverallimitations.First,ourmodelassumesaproportionalsexualmixing,but thisassumptionmightbeunrealistic[11,67].Ourrobustnessanalysismitigatesthislimitation byaccountingfortheefectsofvariationsinthemixingratesontheICERestimates.Second, ourmodelaccountsforneitherMSMinjection-drugusersnorthosewithfemalepartners.An explicitaccountingfortheseindividualsmightafectthesexualmixingratesandtransmission parameters,althoughwithmarginalefectonourICERs.Third,theHRQOLestimatesinthe studyweredevelopedusingthewidelyusedEQ-5Dinstrument. BecauseQOLestimatesare sensitivetotheinstrumentused,ourICERsmaybeafected[28,39]. However,thesensitiv- ityanalysismitigatesthisthreatbecausetheestimatesremainedrobusttovariationsinQOL weights. ThisanalysisdemonstratesthatPrEPcombinedwithearlyARTinitiationwouldbeacost- efectivestrategyinpreventingandminimizingtheriskofHIVinfectionintheLACMSM population. Our results are robust to perturbations in the model input parameter values. 119 PooradherencetoPrEPandARTremainsamajorthreattothisapproach.Therefore,under- standingandimprovingbehavioralresponsestobothPrEPandARTmayhelpdramatically improvehealthoutcomesandreducecosts. References [1] Abbas,U.L.,Anderson,R.M.andMellors,J.W. Potentialimpactofantiretroviralther- apyonHIV-1transmissionandAIDSmortalityinresource-limitedsettings. Journalof Acquired Immune Deficiency Syndromes,41(5):632–41,Apr152006. [2] Arias,E.,Curtin,L.R.andWei,R. USdecenniallifetablesfor1999âĂŞ2001,United Stateslifetables. National Vital Statistics Reports,2008. [3] Baeten,J.M.etal. AntiretroviralprophylaxisforHIVpreventioninheterosexualmen andwomen. New England Journal of Medicine,367(5):399–410,Aug022012. [4] Baeten,J.M.etal. PreexposureprophylaxisforHIVprevention: wherehavewebeen andwherearewegoing? JournalofAcquiredImmuneDeficiencySyndromes,63(Suppl 2):S122–9,Jul2013. [5] Baggaley, R.F., Ferguson, N.M. and Garnett, G.P. The epidemiological impact of antiretroviralusepredictedbymathematicalmodels: Areview. Emerging Themes in Epidemiology,2:9,Sep102005. [6] Bangsberg,D.R.etal. HighlevelsofadherencedonotpreventaccumulationofHIV drugresistancemutations. AIDS,17(13):1925,Sep052003. [7] Bangsberg,D.R.etal.ModelingtheHIVproteaseinhibitoradherence-resistancecurve byuseofempiricallyderivedestimates. Journal of Infectious Diseases, 190(1):162–165, Jul012004. 120 [8] Barnett, P.G.etal. Determinantsofthecostofhealthservicesusedbyveteranswith HIV. Medical Care,49(9):848–56,Sep2011. [9] Barro,J. PilltoPreventH.I.V.GetsaProminentBacker: AndrewCuomo. The New York Times,Jul32014. [10] Blower, S.M., Gershengorn, H.B. and Grant, R.M. A tale of two futures: HIV and antiretroviraltherapyinSanFrancisco. Science,287(5453):650–4,Jan282000. [11] Boily,M.C.andAnderson,R.M.Sexualcontactpatternsbetweenmenandwomenand thespreadofHIV-1inurbancentresinAfrica. MathematicalMedicineandBiology,8 (4):221–247,1991. [12] Bollinger,R.C.etal. RiskfactorsandclinicalpresentationofacuteprimaryHIVinfec- tion in India. Journal of the American Medical Association, 278(23):2085–9, Dec 17 1997. [13] Bozzette, S.A. et al. Expenditures for the care of HIV-infected patients in the era of highlyactiveantiretroviraltherapy. NewEnglandJournalofMedicine,344(11):817–23, Mar152001. [14] Brenner,B.etal. Persistenceofmultidrug-resistantHIV-1inprimaryinfectionleading tosuperinfection. AIDS,18(12):1653–60,Aug202004. [15] Broder,M.S.etal. Costefectivenessofatazanavir-ritonavirversuslopinavir-ritonavir in treatment-naÃŕve human immunodeciency virus-infected patients in the United States. Journal of Medical Economics,14(2):167–178,Apr242011. [16] BureauofLaborStatistics. ConsumerPriceIndex-AllUrbanConsumers,USMedical Care. Availableathttp://data.bls.gov/cgi-bin/surveymost?cu,Jun52013. 121 [17] BureauofLaborStatistics. NationalCompensationSurvey(NCS). Availableathttp: //www.bls.gov/ncs/data.htm,Apr282015. [18] Centers for Disease Control and Prevention. HIV in the United States: At a glance. Avaliable at http://www.cdc.gov/hiv/resources/factsheets/PDF/HIV_at_a_ glance.pdf,2012. AccessedJun05,2015. [19] Centers for Disease Control and Prevention. Using viral load data to monitor HIV burden and treatment outcomes in the United States. Available at http://www.cdc. gov/hiv/topics/surveillance/resources/factsheets/viral_load.htm,2012. [20] CentersforDiseaseControlandPrevention. Today’sHIV/AIDSEpidemic. Available at http://www.cdc.gov/nchhstp/newsroom/docs/HIVFactSheets/TodaysEpidemic- 508.pdf,Dec2013. AccessedJun05,2013. [21] CentersforDiseaseControlandPrevention.DiagnosesofHIVInfectionintheUnited StatesandDependentAreas,2013. HIVSurveillanceReport,25,Feb2015. AccessedJan 22,2016. [22] Chou,R.etal. ScreeningforHIV:Systematicreviewtoupdatethe2005USPreventive Services Task Force recommendation. Annals of Internal Medicine, 157(10):706–18, Nov202012. [23] CMS. Medicare Information for Providers, Partners and Health Care Professionals. Availableathttp://www.cms.gov/home/medicare.asp,2010. [24] CMS.2013ClinicalDiagnosticLaboratoryFeeSchedule.Availableathttp://www.cms. gov/apps/ama/license.asp?le=/ClinicalLabFeeSched/downloads/13CLAB.ZIP,Jun5 2013. 122 [25] Cohen,M.S.etal. Narrativereview:Antiretroviraltherapytopreventthesexualtrans- missionofHIV-1. Annals of Internal Medicine,146(8):591–601,Apr172007. [26] Cohen,M.S.etal.PreventionofHIV-1infectionwithearlyantiretroviraltherapy.New England Journal of Medicine,365(6):493–505,Aug112011. [27] Cohen,S.E.etal. ImplementationofPrEPinSTDClinicsandaCommunityHealth Center:HighUptakeandDrugLevelsamongMSMintheDemoProject.CROI,Mar 3-6;Boston,MA. 2014. [28] Coons,S.J.etal. Acomparativereviewofgenericquality-of-lifeinstruments. Pharma- coeconomics,17(1):13–35,Jan2000. [29] Cragin,L.etal. Cost-efectivenessofafourth-generationcombinationimmunoassay forhumanimmunodeciencyvirus(HIV)antibodyandp24antigenforthedetection ofHIVinfectionsintheUnitedStates. HIV Clinical Trials,13(1):11–22,Jan–Feb2012. [30] Daar, E.S.etal. DiagnosisofprimaryHIV-1infection.LosAngelesCountyPrimary HIVInfectionRecruitmentNetwork. Annals of Internal Medicine,134(1):25–9,Jan2 2001. [31] Desai, K. et al. Modeling the impact of HIV chemoprophylaxis strategies among menwhohavesexwithmenintheUnitedStates: HIVinfectionspreventedandcost- efectiveness. AIDS,22(14):1829–39,Sep122008. [32] Desai, M. Pre-exposureprophylaxis: increasingevidenceforefectivenessbutshould the perspective return to test and treat? Sexually Transmitted Infections, 88(6):475– 476,Oct012012. [33] Dodd,P.J.,Garnett,G.P.andHallett,T.B. ExaminingthepromiseofHIVelimination by“testandtreat”inhyperendemicsettings. AIDS,4(5):729–35,Mar132010. 123 [34] Drummond, M. et al. Methods for the Economic Evaluation of Health Care Pro- grammes. Oxford,UK:OxfordUniversityPress,2005. [35] El-Khatib,Z.,DeLong,A.K.andKatzenstein,D.Drugresistancepatternsandvirusre- suppressionamongHIV-1subtypeCinfectedpatientsreceivingnon-nucleosidereverse transcriptaseinhibitorsinSouthAfrica. Journal of AIDS & Clinical Research,2(117), Feb182011. [36] Ettner,S.L.andWeissman,J. UtilizationofformalandinformalhomecarebyAIDS patients in Boston: A comparison of intravenous drug users and homosexual males. Medical Care,32(5):459–70,May1994. [37] Ettner,S.L.etal. Triply-diagnosedpatientsintheHIV/AIDSTreatmentAdherence, HealthOutcomesandCostStudy:PI-treatedatternsofhomecareuse. AIDS Care,20 (10):1177–89,Nov2008. [38] Farnham,P.G.etal. ComparingthecostsofHIVscreeningstrategiesandtechnologies inhealth-caresettings. Public Health Reports,123(Suppl3):51–62,Nov–Dec2008. [39] Fayers,P.andMachin,D. Qualityoflife: Theassessment,analysisandinterpretationof patient-reported outcomes. JohnWiley&Sons,2edition,Apr272007. [40] Fryback,D.G.etal.TheBeaverDamHealthOutcomesStudy:Initialcatalogofhealth- statequalityfactors. Medical Decision Making,13(2):89–102,Apr–Jun1993. [41] Gebo, K.A.etal. CostsofHIVmedicalcareintheeraofhighlyactiveantiretroviral therapy. AIDS,13(8):963–9,May281999. 124 [42] Gilead Sciences and United States Food and Drug Administration Antiviral Drugs AdvisoryCommittee. Truvada®(emtricitabine/tenofovir disoproxil fumarate) for pre- exposure prophylaxis of HIV-1: Antiviral Drugs Advisory Committee Meeting Briefing Document. FosterCity,CA:GileadSciences,Inc.,2012. [43] Glick, H.A. Economic evaluation in clinical trials. HandbooksinHealthEconomic EvaluationSeries.Oxford,UK:OxfordUniversityPress,2007. [44] Glick,H.A. Cost-EfectivenessAnalysis: Averageandincrementalratios;choicecrite- ria;strategiesforidentifyingthecostefectivetherapy. Availableathttp://www.uphs. upenn.edu/dgimhsr/documents/1b.introcea.fda13.pdf,2013. AccessedJun30,2014. [45] Gold,M.R.etal. Cost-eectiveness in health and medicine: Report of the panel on cost- eectiveness in health and medicine. NewYork:OxfordUniversityPress,1996. [46] Golub,S.A.etal.Fromecacytoefectiveness:FacilitatorsandbarrierstoPrEPaccept- ability and motivations for adherence among MSM and transgender women in New YorkCity. AIDS Patient Care and STDs,27(4):248–54,Apr2013. [47] Gomez, G.B. et al. The cost and impact of scaling up pre-exposure prophylaxis for HIV prevention: A systematic review of cost-efectiveness modelling studies. PLoS Medicine,Mar122013. [48] Gonzalez-Serna,A.andLima,V.D. “Test-and-treat”strategyforcontrolofHIVand AIDS can lead to a decrease, not an increase, of multidrug-resistant viruses. Clinical Infectious Diseases, 57(3):479–480, Apr262013. URLhttp://cid.oxfordjournals.org/ content/57/3/478.short. [49] Granich, R.M. et al. Universal voluntary HIV testing with immediate antiretrovi- raltherapyasastrategyforeliminationofHIVtransmission: Amathematicalmodel. Lancet,373(9657):48–53,Jan2009. 125 [50] Grant,R.M.andSmith,D.K.IntegratingAntiretroviralStrategiesforHumanImmun- odeciencyVirusPrevention:Post-andPre-ExposureProphylaxisandEarlyTreatment. Open Forum Infectious Diseases,2(4):ofv126,Dec2015. [51] Grant, R.M. et al. Preexposure chemoprophylaxis for HIV prevention in men who havesexwithmen. New England Journal of Medicine,363(27):2587–99,Dec302010. [52] Grant,R.M.etal. Uptakeofpre-exposureprophylaxis,sexualpractices,andHIVinci- denceinmenandtransgenderwomenwhohavesexwithmen:Acohortstudy. Lancet Infectious Diseases,14(9):820–9,Sep2014. [53] Grant,R.M.etal.Scale-upofpre-exposureprophylaxisinSanFranciscotoimpactHIV incidence. InCROI, Feb 23-26, Seattle, W A,2015. [54] Grijsen,M.etal. TemporaryantiretroviraltreatmentduringprimaryHIV-1infection hasapositiveimpactonhealth-relatedqualityoflife:datafromthePrimo-SHMcohort study. HIV Medicine,13(10):630–5,Nov2012. [55] Gulick,R. NewARVdrugs. BMC Infectious Diseases,14(Suppl2):S2,May232014. [56] Günthard,H.F.etal.AntiretroviraltreatmentofadultHIVinfection:2014recommen- dations of the International Antiviral Society–USA Panel. Journal of the American Medical Association,312(4):410–425,2014. [57] Hall, H.I.etal. PrevalenceofDiagnosedandUndiagnosedHIVInfection–United States, 2008-2012. Morbidity and Mortality Weekly Report, 64(24):657–62, Jun 26 2015. [58] Hanmer, J., Lawrence, W.F. and Anderson, J.P. Report of nationally representative valuesforthenoninstitutionalizedUSadultpopulationfor7health-relatedquality-of- lifescores. Medical Decision Making,26(4):391–400,Jul–Aug2006. 126 [59] Hogg,R.etal. Lifeexpectancyofindividualsoncombinationantiretroviraltherapyin high-incomecountries:Acollaborativeanalysisof14cohortstudies. Lancet,372(9635): 293–9,Jul262008. [60] Holtgrave, D.R. and Pinkerton, S.D. Updates of cost of illness and quality of life estimates for use in economic evaluations of HIV prevention programs. Journal of Acquired Immune Deficiency Syndromes,16(1):54–62,Sep11997. [61] Honiden,S.etal. TheefectofdiagnosiswithHIVinfectiononhealth-relatedquality ofLife. Quality of Life Research,15(1):69–82,Feb2006. [62] Hornberger,J.,Holodniy,M.andRobertus,K.Asystematicreviewofcost-utilityanal- yses in HIV/AIDS: Implications for public policy. Medical Decision Making, 27(6): 789–821,Nov–Dec2007. [63] Hoyert, D.L. and Xu, J. Deaths: PI-treatedreliminary data for 2011. National Vital Statistics Reports,61(6):1–52,Oct102012. [64] Hsu,H.andWalensky,R.P. Cost-efectivenessanalysisandHIVscreening:Theemer- gencymedicineperspective. Annals of Emergency Medicine, 51(Suppl1):S145–50, Jul 2011. [65] Hutchinson,A.andSansom,S. Cost-efectivenessofmorefrequentHIVscreeningof menwhohavesexwithmenintheUnitedStates. In19th International AIDS Confer- ence. Washington, DC; Jul 22–27. Abstract THAB0302,2012. [66] Hutchinson,A.B.etal.TheeconomicburdenofHIVintheUnitedStatesintheeraof highlyactiveantiretroviraltherapy:Evidenceofcontinuingracialandethnicdiferences. Journal of Acquired Immune Deficiency Syndromes,43(4):451–7,Dec12006. 127 [67] Hutchinson,R. Sexualpartnerselectionvariesbyraceamongmenwhohavesexwith men. Perspectives on Sexual and Reproductive Health,42(3):214–214,Aug182010. [68] Jin,J. MedicationstoPreventHIVInfection. Journal of the American Medical Asso- ciation,312(4):452–452,2014. [69] Johnson,A.S.etal. TrendsindiagnosesofHIVinfectionintheUnitedStates,2002- 2011. Journal of the American Medical Association,312(4):432–4,Jul23–302014. [70] Juusola,J.L.,Brandeau,M.L.andLong,E.F. Thecost-efectivenessofsymptom-based testingandroutinescreeningforacuteHIVinfectioninmenwhohavesexwithmenin theUnitedStates. AIDS,25(14):1779–87,Sep102011. [71] Juusola,J.L.etal.Thecost-efectivenessofpreexposureprophylaxisforHIVprevention intheUnitedStatesinmenwhohavesexwithmen. Annals of Internal Medicine,156 (8):541–50,Apr172012. [72] Kamb,M.L.etal. Ecacyofrisk-reductioncounselingtopreventhumanimmunode- ciencyvirusandsexuallytransmitteddiseases: Arandomizedcontrolledtrial.Project RESPECTStudyGroup.JournaloftheAmericanMedicalAssociation,280(13):1161–7, Oct71998. [73] Khazeni,N.etal. Efectivenessandcost-efectivenessofvaccinationagainstpandemic inuenza(H1N1)2009. Annals of Internal Medicine,151(12):829–39,Dec152009. [74] Koblin, B.A.etal. High-riskbehaviorsamongmenwhohavesexwithmenin6US cities: BaselinedatafromtheEXPLOREStudy. American Journal of Public Health, 93(6):926–32,Jun2003. [75] Koblin,B.A.etal. RiskfactorsforHIVinfectionamongmenwhohavesexwithmen. AIDS,20(5):731–9,Mar212006. 128 [76] Koppenhaver,R.T.andSorensen,S.W. Thecost-efectivenessofpre-exposureprophy- laxisinmenwhohavesexwithmenintheUnitedStates:Anepidemicmodel. Journal of Acquired Immune Deficiency Syndromes,58(2):e51–2,Oct12011. [77] Korenromp,E.L.etal.ClinicalprognosticvalueofRNAviralloadandCD4cellcounts duringuntreatedHIV-1infection–aquantitativereview. PLoS One,4(6):e5950,2009. [78] Leonardi,M.,Lee,E.andTan,D.H.Awarenessof,usageofandwillingnesstouseHIV pre-exposureprophylaxisamongmenindowntownToronto, Canada. International Journal of STD & AIDS,22(12):738–41,Dec2011. [79] Lichtenberg,F.R.andDuos,G. Timerelease:TheefectofpatentexpirationonU.S. drugprices,marketing,andutilizationbythepublic. MedicalProgressReport,(11),Oct 2009. [80] Liegler,T.etal. HIV-1drugresistanceintheiPrExpreexposureprophylaxistrial. Jour- nal of Infectious Diseases,210(8):1217–27,Oct152014. [81] Liu,A.etal.Earlyexperiencesimplementingpre-exposureprophylaxis(PrEP)forHIV preventioninSanFrancisco. PLoS Medicine,11(3):e1001613,Mar2014. [82] Liu,A.Y.etal. LimitedknowledgeanduseofHIVpost-andpre-exposureprophylaxis amonggayandbisexualmen. Journal of Acquired Immune Deficiency Syndromes,47 (2):241–7,Feb12008. [83] Liu,H.etal. Repeatedmeasuresanalysesofdosetimingofantiretroviralmedication anditsrelationshiptoHIVvirologicoutcomes. StatisticsinMedicine,26(5):991–1007, Feb282007. [84] Lodi,S.etal. Timefromhumanimmunodeciencyvirusseroconversiontoreaching CD4+ cell count thresholds <200, <350, and <500 Cells/mm(3): assessment of need 129 following changes in treatment guidelines. Clinical Infectious Diseases, 53(8):817–25, Oct2011. [85] London,A.S.etal.UseofunpaidandpaidhomecareservicesamongpeoplewithHIV infectionintheUSA. AIDS Care,13(1):99–121,Feb2001. [86] Long, E.F. and Brandeau, M.L. The cost-efectiveness and population outcomes of expandedHIVscreeningandantiretroviraltreatmentintheUnitedStates. Annals of Internal Medicine,153(12):778–89,Dec212010. [87] Long,E.F.andOwens,D.K.Thecost-efectivenessofamodestlyefectiveHIVvaccine intheUnitedStates. Vaccine,29(36):6113–24,Aug182011. [88] Long,E.F.,Brandeau,M.L.andOwens,D.K. Potentialpopulationhealthoutcomes andexpendituresofHIVvaccinationstrategiesintheUnitedStates. Vaccine, 27(39): 5402–10,Aug272009. [89] Lucas,A.andArmbruster,B. Thecost-efectivenessofexpandedHIVscreeninginthe UnitedStates. AIDS,27(5):795–801,Mar132013. [90] Maggiolo, F.etal. EfectofadherencetoHAARTonvirologicoutcomeandonthe selection of resistance-conferring mutations in NNRTI- or PI-treated patients. HIV Clinical Trials,8(5):282–92,Sep–Oct2007. [91] Marazzo,J.M.etal. Pre-exposureprophylaxisforHIVinwomen:Dailyoraltenofovir, oraltenofovir/emtricitabineorvaginaltenofovirgelintheVOICEstudy(MTN003). In20thConferenceonRetrovirusesandOpportunisticInfections;Marc3-6,Atlanta,GA, 2013. 130 [92] Marks,G.etal.Meta-analysisofhigh-risksexualbehaviorinpersonsawareandunaware theyareinfectedwithHIVintheUnitedStates:ImplicationsforHIVpreventionpro- grams. Journal of Acquired Immune Deficiency Syndromes,39(4):446–53,Aug12005. [93] Mascolini,M. PooradherencemayexplainFEM-PrEPfailuretondprotectionfrom HIV with Truvada. In 19th Conference on Retroviruses and Opportunistic Infections. Mar 5-8; Seattle, W A,2012. [94] Mauskopf, J. et al. Cost efectiveness of darunavir/ritonavir in highly treatment- experienced, HIV-1-infectedadultsintheUSA. Pharmacoeconomics, 28(Suppl1):83– 105,Dec2010. [95] May, M. et al. Prognosis of HIV-1-infected patients up to 5 years after initiation of HAART: collaborative analysis of prospective studies. AIDS, 21(9):1185–97, May 31 2007. [96] McCormick,A.W.andWalensky,R.P.Theefectofantiretroviraltherapyonsecondary transmissionofHIVamongmenwhohavesexwithmen. Clinical Infectious Diseases, 44(8):1115–1122,Apr152007. [97] McEnery,R.PreparingforPrEP:Weknowitworks.Buthowdoyougetpeopletotake it? IAVI Report,17(3):9–14,Fall2013. [98] McNeilJr.,D.G. WHOIssuesGuidelinesforEarlierHIVTreatment. The New York Times,Jun302013. [99] McNeilJr.,D.G. AdvocatingPill,USSignalsShifttoPreventAIDS. The New York Times,May152014. [100] Meara,E.,White,C.andCutler,D.M. Trendsinmedicalspendingbyage,1963-2000. Health Aairs,23(4):176–83,Jul–Aug2004. 131 [101] Mimiaga,M.J.etal.Preexposureantiretroviralprophylaxisattitudesinhigh-riskBoston area men who report having sex with men: limited knowledge and experience but potentialforincreasedutilizationaftereducation. Journal of Acquired Immune Defi- ciency Syndromes,50(1):77–83,Jan12009. [102] Minga,A.K.etal.CD4celleligibilitythresholds:ananalysisofthetimetoantiretroviral treatmentinHIV-1seroconverters. AIDS,25(6):819–23,Mar272011. [103] Morgan,D.etal. HIV-1infectioninruralAfrica: Isthereadiferenceinmediantime to AIDS and survival compared with that in industrialized countries? AIDS, 16(4): 597–603,Mar82002. [104] Moyer, V.A. Screening for HIV: US preventive services task force recommendation statement. Annals of Internal Medicine,159(1):51–60,2013. [105] Mugo, N.R. et al. Pregnancy incidence and outcomes among women receiving pre- exposureprophylaxisforHIVprevention: arandomizedclinicaltrial. Journal of the American Medical Association,312(4):362–371,2014. [106] Neuner, J.M. et al. Diagnosis and management of adults with pharyngitis: A cost- efectivenessanalysis. Annals of Internal Medicine,139(2):113–22,Jul152003. [107] Nosyk,B.andMontaner,J.TheevolvinglandscapeoftheeconomicsofHIVtreatment andprevention. PLoS Medicine,9(2):e1001174,Feb142012. [108] O’Brien, B.J. et al. Oseltamivir for treatment of inuenza in healthy adults: pooled trial evidence and cost-efectiveness model for Canada. Value in Health, 6(2):116–25, Mar–Apr2003. 132 [109] Osmond,D.H.etal.ChangesinprevalenceofHIVinfectionandsexualriskbehaviorin menwhohavesexwithmeninSanFrancisco:1997–2002. American Journal of Public Health,97(9):1677–83,Sep2007. [110] Paltiel, A.D. et al. Expanded screening for HIV in the United States–an analysis of cost-efectiveness. New England Journal of Medicine,352(6):586–95,Feb102005. [111] Paltiel,A.D.etal. ExpandedHIVscreeningintheUnitedStates:efectonclinicalout- comes, HIV transmission, and costs. Annals of Internal Medicine, 145(11):797–806, Dec52006. [112] Paltiel, A.D., Freedberg, K.A. and Scott, C.A. HIV preexposure prophylaxis in the United States: Impact on lifetime infection risk, clinical outcomes, and cost- efectiveness. Clinical Infectious Diseases,48(6):806–15,Mar152009. [113] Parienti,J.J.,Ragland,K.andLucht,F.Averageadherencetoboostedproteaseinhibitor therapy,ratherthanthepatternofmisseddoses,asapredictorofHIVRNAreplication. Clinical Infectious Diseases,50(8):1192–7,Apr152010. [114] Paz-Bailey,G.etal. HIVTestingandRiskBehaviorsAmongGay,Bisexual,andOther Men Who Have Sex with Men – United States. Morbidity and Mortality Weekly Report,62(47):958–962,2013. [115] Prabhu,V.S.,Farnham,P.G.andHutchinson,A.B.Cost-efectivenessofHIVscreening in STD clinics, emergency departments, and inpatient units: a model-based analysis. PloS One,6(5):e19936,May202011. [116] Prachakul,W.andGrant,J.S.InformalcaregiversofpersonswithHIV/AIDS:Areview andanalysis. JournaloftheAssociationofNursesinAIDSCare,14(3):55–71,May–Jun 2003. 133 [117] Quinn, T.C. et al. Viral load and heterosexual transmission of human immunode- ciencyvirustype1.RakaiProjectStudyGroup. NewEnglandJournalofMedicine,342 (13):921–9,Mar302000. [118] Richardson,E.etal. HealthPolicyBrief:Biosimilars. Health Aairs,pages1–5,Oct10 2013. [119] Rose,C.D.etal.HIVinterventionforprovidersstudy:arandomizedcontrolledtrialof aclinician-deliveredHIVrisk-reductioninterventionforHIV-positivepeople. Journal of Acquired Immune Deficiency Syndromes,55(5):572–81,Dec152010. [120] Rothman, R.E. and Merchant, R.C. Update on emerging infections from the Cen- tersforDiseaseControlandPrevention.RevisedrecommendationsforHIVtestingof adults,adolescents,andpregnantwomeninhealth-caresettings. Annals of Emergency Medicine,49(5):575–9,May2007. [121] Sanders,G.D.etal. Cost-efectivenessofscreeningforHIVintheeraofhighlyactive antiretroviraltherapy. New England Journal of Medicine,352(6):570–85,Feb102005. [122] Schacker,T.etal.ClinicalandepidemiologicfeaturesofprimaryHIVinfection.Annals of Internal Medicine,125(4):257–64,Aug151996. [123] Schackman,B.R.andEggman,A.A. Cost-efectivenessofpre-exposureprophylaxisfor HIV:Areview. Current Opinions in HIV and AIDS,7(6):587–92,Nov72012. [124] Schackman, B.R. et al. The lifetime cost of current human immunodeciency virus careintheUnitedStates. Medical Care,44(11):990–7,Nov2006. [125] Schrager,S.M.etal. Humanimmunodeciencyvirustestingandriskbehaviorsamong youngmenwhohavesexwithmeninLosAngelesCounty.AmericanJournalofHealth Promotion,25(4):244–7,Mar–Apr2011. 134 [126] Sewell, A. LA County planning to distribute controversial HIV-prevention drug. Available at http://www.latimes.com/local/lanow/la-me-ln-county-hiv-truvada- 20150609-story.html,Jun192015. AccessedDec15,2015. [127] Smith,R.J.etal. EvolutionarydynamicsofcomplexnetworksofHIVdrug-resistant strains:thecaseofSanFrancisco. Science,327(5966):697–701,Feb52010. [128] Smith,R.J.etal. SupplementaryMaterialtoEvolutionarydynamicsofcomplexnet- worksofHIVdrug-resistantstrains:thecaseofSanFrancisco. Science,327(5966):697– 701,Feb52010. [129] Sood,N.etal. Test-and-treatinLosAngeles: Amathematicalmodeloftheefectsof test-and-treatforthepopulationofmenwhohavesexwithmeninLosAngelesCounty. Clinical Infectious Diseases,56(12):1789–96,Mar132013. [130] Sood,N.etal. MathematicalDetailsS1: HIVModelApproach;SupplementaryData to“Test-and-TreatinLosAngeles: AMathematicalModeloftheEfectsofTest-and- TreatforthePopulationofMenWhoHaveSexWithMeninLosAngelesCounty”. Clinical Infectious Diseases,56(12):1789–1796,2013. [131] Steward, W.T. et al. Behavior change following diagnosis with acute/early HIV infection-amovetoserosortingwithotherHIV-infectedindividuals.TheNIMHMul- tisiteAcuteHIVInfectionStudy: III. AIDS and Behavior,13(6):1054–60,Dec2009. ISSN1573-3254(Electronic)1090-7165(Linking). [132] Sykes,B.L.andPiquero,A.R.StructuringandRe-CreatingInequality:HealthTesting Policies,Race,andtheCriminalJusticeSystem. Annals of the American Academy of Political and Social Science,623(1):214–227,2009. [133] Tengs,T.O.andLin,T.H. Ameta-analysisofutilityestimatesforHIV/AIDS. Medical Decision Making,22(6):475–81,Nov–Dec2002. 135 [134] The National Institute of Mental Health (NIMH) Multisite HIV Prevention Trial Group. TheNIMHMultisiteHIVPreventionTrial:ReducingHIVsexualriskbehav- ior. Science,280(5371):1889–94,Jun191998. [135] The Rand Corporation. Rand California Population and Demographic Statistics. Availableathttp://ca.rand.org/stats/popdemo/popraceage.html,June52013. [136] Thigpen,M.C.,Kebaabetswe,P.M.andSmith,D.K.Dailyoralantiretroviraluseforthe preventionofHIVinfectioninheterosexuallyactiveyoungadultsinBotswana:Results from the TDF2 study. In 6th IAS Conference on HIV Pathogenesis, Treatment and Prevention. Jul 17-20; Rome, Italy,2011. [137] Thompson,M.A.etal. AntiretroviraltreatmentofadultHIVinfection: 2010recom- mendationsoftheInternationalAIDSSocietyâĂŞUSApanel.JournaloftheAmerican Medical Association,304(3),Jul212010. [138] Thompson,M.A.etal. AntiretroviraltreatmentofadultHIVinfection: 2012recom- mendationsoftheInternationalAntiviralSocietyâĂŞUSApanel.JournaloftheAmer- ican Medical Association,308(4):387–402,Jul252012. [139] Thompson, S.G. and Nixon, R.M. How sensitive are cost-efectiveness analyses to choiceofparametricdistributions? MedicalDecisionMaking,25(4):416–423,Jul–Aug 2005. [140] Tieu,H.V.etal. SexualpartneringandHIVriskamongblackmenwhohavesexwith men:NewYorkCity. Journal of Urban Health,87(1):113–21,Jan2010. [141] Turner, D.A. et al. The cost-efectiveness of inuenza vaccination of healthy adults 50-64yearsofage. Vaccine,24(7):1035–43,Feb132006. 136 [142] Urbanek,S.etal. R:Alanguageandenvironmentforstatisticalcomputing.RFoun- dationforStatisticalComputing,Vienna,Austria.2013,2014. [143] US Census Bureau. S0101. Age and Sex: 2006-2008 American Community Survey 3-YearEstimates. Availableathttp://factnder.census.gov/servlet/STTable?_bm=y&- geo_id=01000US&-qr_name=ACS_2008_3YR_G00_S0101&-ds_name=ACS_ 2008_3YR_G00_,2008. [144] US Department of Health & Human Services. AIDS.gov. Available at https://www.aids.gov/hiv-aids-basics/just-diagnosed-with-hiv-aids/understand- your-test-results/cd4-count/index.html,2010. AccessedJune05,2013. [145] US Department of Health and Human Services. Panel on Antiretroviral Guidelines for Adults and Adolescents. Guidelines for the use of antiretroviral agents in HIV- 1-infected adults and adolescents. Available at http://aidsinfo.nih.gov/contentles/ lvguidelines/adultandadolescentgl.pdf,Aug072013. [146] USDepartmentofVeteransAfairs.FSS,BIG4andNationalContractPrices.Available athttp://www.pbm.va.gov/PharmaceuticalPrices.asp,Jun52013. [147] US Food and Drug Administration. FDA approves rst drug for reducing the risk of sexually acquired HIV infection. Available at http://www.fda.gov/NewsEvents/ Newsroom/PressAnnouncements/ucm312210.htm,Jul172012. [148] Van Damme, L. et al. The FEM-PrEP trial of emtricitabine/tenofovir disoproxil fumarate (Truvada) among African women. In 19th Conference on Retroviruses and Opportunistic Infections; Mar 5–8; Seattle, W A,2012. [149] vandeVijver,D.A.,Wensing,A.M.andBoucher,C.A.TheEpidemiologyofTransmis- sionofDrugResistantHIV-1. InLeitner,T.etal,editors,HIVSequenceCompendium 137 2006/2007,pages17–36.LosAlamos,NM:TheoreticalBiologyandBiophysicsGroup, LosAlamosNationalLaboratory,2007. [150] Velasco-Hernandez,J.X.andGershengorn,H.B.Couldwidespreaduseofcombination antiretroviraltherapyeradicateHIVepidemics? Lancet Infectious Diseases,2(8):487– 93,Aug2002. [151] Vergis, E.N.andMellors, J.W. NaturalhistoryofHIV-1infection. Infectious Disease Clinics of North America,14(4):809–25,Dec2000. [152] Vernazza,P.etal. HIVtransmissionunderhighlyactiveantiretroviraltherapy. Lancet, 372(9652):1806–7,Nov222008. [153] Veugelers,P.J.etal. DeterminantsofHIVdiseaseprogressionamonghomosexualmen registeredintheTricontinentalSeroconverterStudy. American Journal of Epidemiol- ogy,140(8):747–58,Oct151994. [154] Vose,D.Quantitativeriskanalysis: AguidetoMonteCarlosimulationmodelling.John Wiley&Sons,3edition,Sep2009. [155] Walensky,R.P.etal. Thecost-efectivenessofpre-exposureprophylaxisforHIVinfec- tioninSouthAfricanwomen. Clinical Infectious Diseases,54(10):1504–13,May2012. [156] Wilson,D.P.ModellingbasedonAustralianHIVnoticationsdatasuggestshomosex- ualagemixingisprimarilyassortative. Journal of Acquired Immune Deficiency Syn- dromes,51(3):356–60,Jul12009. [157] Wilson,D.P.etal.RelationbetweenHIVviralloadandinfectiousness:Amodel-based analysis. Lancet,372(9635):314–20,Jul262008. [158] WorldHealthOrganization. Thewordhealthreport2002:Reducingrisks,promoting healthylife. Technicalreport,WorldHealthOrganization,2002. 138 [159] Zakher,B.,Blazina,I.andChou,R. AssociationbetweenknowledgeofHIV-positive statusoruseofantiretroviraltherapyandhigh-risktransmissionbehaviors:Systematic review. AIDS Care,26(4):514–521,May032014. 139 Chapter5 Concludingremarks EmmanuelF.Drabo 5.1 Synthesisofthekeyndings Thisdissertationsetouttoinvestigateefectivepoliciesforincreasingvaccineuptakeandassess theefectivenessaswellasthetradeofbetweenthecostsandbenetsofalternativeHIV/AIDS preventionstrategies.Weinvestigatedthesequestionsinthreeessays.Intherstessay(Chap- ter2),weconductedadiscretechoiceexperimenttoanswertwoimportantquestions. First, weinvestigatedwhichnancialincentivestrategiesaremostefectiveatencouragingvaccina- tionbycontrastingtheefectsofasubsidyschemeagainstanequivalent‘no-fault”insurance scheme.Thisexperimentalsetupalsoallowedustoconductanindirecttestofexpectedutil- itytheoryvsalternativebehavioraleconomicstheories(i.e. prospecttheoryandtheKőszegi- Rabinutilitytheory).Wefoundevidenceofapositiveefectofano-faultinsuranceonvaccine uptakeamongnon-femalesubjects.Thisresultsuggeststhatano-faultinsuranceschemecan serveasapowerfulincentiveforincreasingvaccineuptake.Second,weexaminedtheoptimal featuresofthesevaccineincentiveprogramsbystudyinghowtheirefectsvarywiththesizeof thecompensation.Wefoundthatthecompensationamounthadnodiferentialefectonvac- cineuptakebetweentheinsuranceandsubsidygroups. Finally,weinvestigatedvaccineand diseasecharacteristicsthatinnovatorsshouldtarget,byexploringthecontributionsofvaccine and disease attributes to vaccination decisions. Our results suggested that more ecacious 140 (ecacy, immunity), safer (less severe and low risk of severe side efects), and less expensive vaccinesarepreferredbyconsumers.Wealsofoundthatvaccineuptakewasgreaterformore severediseases,anddiseasewithgreaterriskofinfection. Collectively,thesendingssuggest thata“no-fault”insuranceschemethatinsuresagainsttheriskofseverevaccineadverseevents canserveaspowerfulincentivesforincreasingthedemandforinfectiousdiseasevaccines. The second and third essays described optimal policies for combating and preventing HIV/AIDS. In Chapter 3, we studied the efects of the “test-and-treat” policy on the HIV/AIDS epidemicamong MSMin LosAngeles County. Wefound that “test-and-treat” could lead to a 39% reduction in the cumulative incidence of AIDS over a 20-year period, suggestingsubstantialpreventivebenetsofthe“test-and-treat”strategyintheLosAngeles CountyMSMpopulation.However,the“test-and-treat”strategyalonewouldnoteliminate theepidemic. Thesendingsarelargelyconsistentwithpriorandtheemergingevidenceon theepidemiologicalbenetsofthe“test-and-treat”strategy.Wealsofoundthatthepreventive benetsof“test-and-treat”wouldbeofsetbyanapproximatelytwo-foldsincreaseinMDR prevalence (from 9.1% to 4.8%) over the 20-years period of the program. Finally, we found thatincreasingHIVtestingandtreatmentwithARTwerenotcomplementary.InChapter4, weextendedtheuniverseofpoliciestoincludevariationsof“test-and-treat”and“PrEP”,and conducted a cost-efectiveness analysis of these policies. Our results suggest that PrEP and “test-and-treat” strategies are cost-efective HIV prevention alternatives compared to a sta- tusquostrategyofHIVtestingevery4.4yearscombinedwithARTinitiationevery2.5year amongtreatment-eligiblesubjects(i.e. atCD4cellcounts≤ 500cells/μL). Inparticular,we nd that at the current US average willingness-to-pay threshold of $150,000/QALY saved, test-and-treatandPrEPwouldcostrespectively$19,302and$27,863perQALYgained. We alsofoundthatmoreaggressivestrategiesweremoreexpensivebutmoreefective,although with diminishing returns. These results remained generally robust to perturbations of the 141 parametervalueswithintheiruncertaintyranges.Withrespecttoepidemiologicaloutcomes, thePrEPandtest-and-treatstrategiesalsoresultedinthelargestHIVinfectionsaverted. 5.2 Policyimplicationsandthewayforward TheexperimentinChapter2suggeststhatmodeststakenancialincentivescansignicantly increasewillingnesstovaccinate.ThissuggeststhatitmaybepossibletoleverageontheVICP toboostvaccineuptake. Infactas,outbaselinedatasuggests,manyAmericansareunaware oftheexistenceoftheVICP.Greaterawarenessaboutthisno-faultinsuranceprogramcanbe apowerfultoolforincreasingUSvaccinationrates.However,giventhatourexperimentonly demonstratedtheefectofinsuranceamongnon-femalerespondents,futurestudiesshould furtherinvestigatetheissue.Specically,theexperimentshouldberepeatedwithamorerep- resentativesample,forbroaderapplicability. ThendingsinChapter3suggestthatwhile“test-and-treat”isaveryefectiveHIVprevention strategy,thisstrategycannotaloneeliminateHIV.Hence,itwouldbemisguidedtosolelyrely onthepromisesof“test-and-treat”inthecombatagainstHIVtransmission. Whilesystem- aticallyinitiatingARTininfectedindividualsmayhelpreduceHIVtransmission,ourstudy suggeststhatthesebenetscouldbeovershadowedbywidespreadMDR,ifnothingisdone toimproveadherencetoandcompliancewiththetreatmentregimen,orinnovatenewerefec- tivelinesoftherapy.Thissuggeststhatthe“test-and-treat”policyshouldalsobeaccompanied withcounselingprogramsaswellasotherstrategiesaimedatincreasingadherencetoantiretro- viraltreatment.Similarly,theinnovationoflonger-actingagentssuchasweeklyoralregimens orquarterlyinjectableARTregimenscouldsignicantlyhelppreservethebenetsofART. Similarly,theinnovationofARTregimenwithbettersafetyprole(fewersideefects)canalso helpreducetherateofnon-adherencetoART. 142 The results from Chapter 4 suggest that combining pre-exposure prophylactic treatments with the “test-and-treat” strategy is a cost-efective strategy for preventing HIV among the MSMpopulationinhighlyconcentratedHIVepidemicsettingssuchasthatofLAC. The results presented in Chapter 3 and Chapter 4 are only applicable to the MSM popula- tionandtoepidemicsettingssimilartothatinLosAngelesCounty. However, withminor adjustmentstoaccountforcontext-specicepidemicfeatures(e.g. baselineHIVprevalence andincidence),themodelscanbeusedtomakepredictionsaboutothersettings.Inaddition, themodelsweredevelopedtoonlyaccountforhomosexualtransmissionamongmen.How- ever, HIV transmission occurs through many other routes such as infected needles among injection-drugusers(IDUs),andtheheterosexualroute.Inordertobetterinformtheoverall nationaldynamicsoftheHIVtransmission,futureworkshouldalsoaccountforthesealter- nativemodesofinfection. ManymodelsofHIVtransmission,includingourwork,donotaccountfortheadaptiverisk behaviors in response to changes in the epidemic trends, the availability of new prevention or treatment technologies (e.g. highly efective and safe ARVs, condoms, PrEP), as well as policy changes. However, there is substantial evidence suggesting that these variations can haveprofoundimpactsontheoveralldynamicofHIVtransmission,andfailuretoproperly accountforthemcanleadtogrossoverestimationsofdiseaseprevalenceintheabsenceofthe changes,andanunderestimationofdiseaseprevalenceunderthechanges. 143 AppendixA Backgroundondecisiontheoryanddiscrete choiceexperiments,anddescriptionofthe experimentalprocedure A.1 Brief overview of prospect theory and the Kőszegi-Rabin utilitytheory Inprospecttheory,decisionprocessesfollowtwostages:Intherststage,knownastheediting stage,outcomesofpotentialchoicesarerankedfollowingcertainheuristicsinwhichindividu- alsdecidewhichoutcomestheyconsiderequivalent,setareferencepoint,andconsiderlesser outcomesaslossesandgreateronesasgains.Inthesecondstage,knownastheevaluationstage, probabilityweightsareappliedtotheutilitiesofpotentialoutcomesandalternativeyielding thehighestvaluationarechosen.Formally,theoverallexpectedutilitytotheindividualofthe outcomesx =hx 1 ,··· ,x n iandtheirassociatedobjectiveprobabilitiesp =hp 1 ,··· ,p n iis denedasfollows: V (x,p) = n X i=1 w(p i )u(x i ), (A.1) whereu(·)istheutilityfunctionthatassignsavaluetoeachoutcomex i ,andw(·)denotes theprobabilityweightingfunction. Thevaluefunction,whichpassesthroughthereference point,isS-shapedandasymmetrical,hencecapturingthenotionoflossaversion. 144 However, this original formulation of prospect theory violated rst-order stochastic domi- nance,alimitationthatwasaddressedbycumulativeprospecttheorythroughaprobability weightingfunctionderivedfromrank-dependentexpectedutilitytheory. Inthisrevisedthe- ory,weightingisappliedtothecumulativeprobabilitydistributionfunction,ratherthanto the probabilities of individual outcomes as done in prospect theory. Cumulative prospect theoryalsodifersfromtheoriginalprospecttheoryinthatitassumesthatindividualsover- weight extreme, and rare events, but underweight “average” events. Efectively, the theory modiesexpectedutilitytheoryby(i)replacingthenalwealthwithpayofsrelativetoaref- erencepoint,(ii)replacingtheutilityfunctionwithanoverallvaluefunctionthatdependson therelativepayof,and(iii)replacingthecumulativeprobabilitieswithweightedcumulative probabilitiesaspresentedintheequationbelow: V (x,p) = Z 0 −∞ v(x) x dx [w(F (x))] dx + Z ∞ 0 v(x) d dx [−w(1−F (x))] dx, (A.2) whereF (x) = R 0 −∞ representsthecumulativeprobabilitydistributionfunctionoftheout- comesx,v(·)denotesthevaluefunctionwithatypicalformsimilartothatinFigure??,and w(·)representstheweightingfunctionforthecumulativeprobabilitydistributionfunction. IntheKőszegi-Rabinutilitytheory,theagent’sutilityfunctionisdenedasfollows: U(F|G) = x u(c|r)dG(r)dF (c), (A.3) where F (·) and G(·) denote respectively the stochastic outcomes and reference points, u(c|r) = m(c) + n(c|r) denotes the overall utility function withm(c) = X k m k (c k ) representingtheconsumptionutilityandn(c|r) = X k n k (c k |r k )the“gain-lossutility”. In thisspecication,n k (c k |r k ) = μ (m k (c k )−m k (c k ))capturesthesensationofgainorloss 145 duetodeparturefromthereferencepoint,andμ(·)satisesthepropertiesofthevaluefunc- tioninKahnemanandTversky[25]. IntheKőszegi-Rabinutilitytheory,thereferencepoint isendogenouslydeterminedbytheindividual’srationalexpectationsheldintherecentpast aboutoutcomes,ratherthanbythestatusquoasisassumedbyprospecttheory. Hence,the theorypredictsloss-aversiononexpected,ratherthanonrealizedendowments. A.2 Methodsforelicitingpreferences Two broad approaches are used in the economic valuation of non-market goods: Revealed preference (RP) and stated preference (SP) approaches. RP is a utility-based approach in whichdataaboutareobtainedbyobservingindividualbehaviorinrealmarkets. However, itisnotalwayspossibletouseRPtovaluegoodsandservices,becausemanygoodsandser- vices are not traded on a market (e.g. new products under development, products not yet commerciallyavailable,illicitgoodsandservices,someintangiblegoods). Furthermore,itis sometimesnotdesirabletousetheRPapproachtovaluegoodsandservices,becauseitmaybe impossibletovaryeachproductattributeindependently(e.g. single-product/homogeneous productswherethereisnovariation)inordertoisolatetheindependentefectofeachprod- uctattributetotheoverallutility. RPdataisalsooftentarnishedwiththepresenceofother confounding efects that are beyond the control of the analyst, making the isolation of the efectachallengingtask.SPapproachesareusefulforstudyingmarketswhereitisimpossible toobtainexogenousrevealedpreferencedataontheactualchoicesmadebyindividuals, by allowingtheexperimentertocontrolthestimulithatgeneratethedata. TherearemanySP approaches, including contingent valuation approaches and multi-attribute valuation tech- niques(FigureA.1). 146 FigureA.1:Approachestopreferenceelicitation. A.3 TheoreticalconsiderationsforconductingaDCE DCEisrootedinRUTandassumesthatanindividualn’sperceivedutilityU nj fromachoice alternativej isalatentvariableconsistingofasystematiccomponentV nj ,whichdependson theobservableattributesofthealternative,x nj ,aswellastheindividual’sobservablecharac- teristics,z n ,thatinuenceherchoicebehavior,andarandomcomponent nj ,whichcaptures unobservableandtheunmeasuredattributesofthealternativesinthechoiceset,theindivid- ual’sunobservableandunmeasuredcharacteristicsthatinuenceherchoicebehavior,aswell asmeasurementerrors[28,39].Mathematically,thiscanbewrittenas U nj = V nj + nj =V (x nj ,z n ) + nj . (A.4) 147 This relation is often referred to as the a“basic axiom of RUT” [30]. The functionV (·) is known as the representative utility and is in practice assumed to be a linear function of its arguments V (x nj ,z n ) = x 0 nj β +z 0 nj γ j , (A.5) whereβandγ j areunknownsetsofweightparametersthatestablishtherelativecontribution ofeachattributetotheutilityassociatedwithalternativej[39]. Therandomcomponent nj isoftendescribedbyajointprobabilitydistributionwithdensity f ( n )andacumulativedistributionfunctionF ( n ), where 0 n =h n1 ,··· , nJ i, withJ dened as the total number of alternatives in the alternative setJ n . Since nj can never be observed and is by denition stochastic, one can never predict a specic individual’s choice behavioronaspecicchoiceevent. However,usingthedensityfunction,predictionscanbe made about the individual’s choice patterns over many choice events. The probabilityP nj thatindividualnwillchoosealternativejfromthechoicesetJ n isthereforegivenby P nj = P U nj > max{U ni } ∀i6=j∈Jn , (A.6) where max{U ni }denotestheutilityderivedfromthebestalternative(diferentfromalter- nativej)inJ n [39].Thisexpressionisalsoacumulativedistribution,ofthediferenceofthe randomerrorterms,asexpressedbelow P nj = P ( ni − nj >V nj −V ni ∀j6=i∈J n ) = Z I{ ni − nj >V nj −V ni }f ( n )d n , (A.7) whereI(·) is an indicator function taking the value of 1 or 0 when its argument is true or false[39]. 148 Fromthisexpression,itcanbenoticedthatthechoiceprobabilitywillbesensitivetothedis- tributionalassumptionsabouttherandomcomponents,sincetheprobabilitydependsonthe densityfunctionoftheerrorterms. Indeed,entirefamiliesofdiscretechoicemodelscanbe obtainedbymakingdiferentassumptionsabouttheerrorterms.Thedistributionalassump- tionabouttherandomcomponentalsohasimplicationsforthedicultyofestimatingthe integral. Forexample, thelogitspecicationhasaclosed-formsolution, whereastheprobit specicationdoesnot.Undertheassumptionthattheerrortermsareindependentandiden- ticallydistributed(i.e. IID,andwritten nj ∼ IID)typeIextremevalue,thechoiceproba- bilitytakestheform P ni = exp{σ n V ni } J X j=1 exp{σ n V nj } , (A.8) whereσ n isascalingparametertypicallynormalizedto1.Themultinomial(conditional)logit modelassumesthatallattributesindependentlyinuencetherespondents’preferences,and thatrespondentshavethesamepreferences,orthatthesepreferencesdependontheirobserv- ablecharacteristics. Theseassumptionsleadtotheindependencefromirrelevantalternatives (IIA)propertywhichimpliesthatremovingsomeunchosenalternativesfromthealternative sethasnoimpactontheselectionofthechosenalternative.Theestimationofthemultinomial logitmodelisconductedbymaximizingthefollowinglog-likelihoodfunction: LL = N X n=1 ln   J Y j=1 P d nj nj   , (A.9) whered nj isanindicatorvariabletakingthevalue1whenalternativejischosenbytheindivid- ual,and0otherwise. IntheRUTframework,theestimatedparameterβ canbeinterpreted asthemarginalutilityofthealternative. 149 Clearly, these assumptions the multinomial logit model relies on strong assumptions. For example,itsrelianceontheIIApropertyhasbeenchallenged,becauseitcanyieldunrealistic predictions. In order to overcome this limitation and account for the unobservable prefer- ence heterogeneity of the respondents, the mixed logit model was developed. This model accommodatespreferenceheterogeneitybyallowingoneormoreparameterstoberandomly distributed.Theprobabilityofchoosingalternativeisgivenby P ni = Z exp n x 0 ni β o P J j=1 exp n x 0 nj β o f(β|θ)dβ, (A.10) wheref(β|θ)denotesthedensityfunctionoftheβ parameters. Itisthroughthisvariation inthecoecientsthatpreferenceheterogeneityiscapturedinthemixedlogitmodel,andthe strong IIA assumption is relaxed. The mixed logit model also accommodates sequences of choicesbythesameindividual,andtheprobabilityofagivensequenceofchoicesiscalculated as S n = Z T Y t=1 J Y j=1     exp n x 0 ni β o P J j=1 exp n x 0 nj β o     d njt f(β|θ)dβ, (A.11) whered njt isanindicatorvariabletakingthevalue1whenalternativejischosenbytheindivid- ualninchoicesituationt,and0otherwise.Themeansandvariancesofthemixedlogitmodel parametersareestimatedbymaximizingthethefollowingsimulatedlog-likelihoodfunction SLL = N X n=1 ln          1 K K X k=1 T Y t=1 J Y j=1     exp n x 0 ni β (k) n o P J j=1 exp n x 0 nj β (k) n o     d njt          , (A.12) whereβ (k) n denotesthek th drawfromthedistributionofβforindividualn. Inouranalysisofthedata,weonlyestimateconditionallogitmodels. 150 A.3.1 ImportantstepsinvolvedinaDCE ADCEstudyinvolvesseveralimportantsteps.First,attributesmustbeestablishedandlevels mustbeassignedtoeachattribute.Thisistypicallydonebycombininginformationfromthe literature,expertopinions,andinputsfromfocusgroupdiscussions. Basedontheseresults, thesetofattributearerevised. Second,choicesetsmustbeconstructedusingreliableexper- imentaldesignapproaches. ThirdaDCEquestionnairemustbegeneratedandpilot-tested among a small sample of respondents, in order to ensure that respondents understand the choicetasksandareansweringtheminameaningfulway.Basedonresultsfromthepilottest, thequestionnaireisrevised.Thepilottestisalsoanimportantstepforobtainingpriorsabout thecoecients,inordertocalculatetheminimalsamplesizerequiredforthestudy. Fourth, the revised questionnaire is administered to the study sample and nally the DCE data are analyzed[24]. A.3.2 DesignconsiderationsinDCE Inthedesignphase,experimentaldesignmethodsareusedtoecientlycombineattributes andtheirlevelsintochoicesets.Adesignthatincludesallpossiblecombinationsofthelevels ofallattributesorfactorsisknownasafull-factorial design. Toillustratethisidea,consider situationwherewehave5factorssuchthat2factorseachconsistof7levelsandtheremaining 3factorseachconsistof3levels.Itisconvenienttowritethissituationas 7 2 ·3 3 .Thenthereare 1, 323possiblecombinationsofthelevelsoftheseattributes.Inthistypeofdesign,allmain- efects,two-wayinteraction,andhigher-orderinteractionsareuncorrelated,andcanbeesti- mated.However,itwouldbeburdensomeandpracticallyimpossibleforasubjecttoconsider allthese 1, 323possiblecombinations.Hence,itiscommonpracticetousefractional-factorial designs,whichconsistofasubsetofthepossiblecombinationsproducedbythefull-factorial 151 design. However,thisreductioninthenumberofrunsresultsinaliasing,whichisthedelib- erateconfoundingofsometwo-wayandhigher-orderinteractionefects,sothattheyareno longerestimableandareuncorrelated. Whenallestimableefectsofafractional-factorialdesignareuncorrelated,thedesigniscalled anorthogonaldesignororthogonalarray.Strictlyspeaking,theterm“orthogonalarray”refers todesignsthatarebothbalancedandorthogonal,andhenceoptimal.Abalanceddesignisone inwhicheachleveloccurswiththesameprobabilitywithineachfactor.Thisimpliesthatthe interceptwillbeorthogonaltoeachefect,sothatbalanceimpliesorthogonality. However, theterm“orthogonalarray”isalsolooselyusedtorefertodesignsthatareorthogonal,butnot balanced,hencepossiblynon-optimal. Orthogonaldesignsaredistinguishedbytheirresolution(r),whichhelpsdeterminetheefects andpossibleinteractionsthatcanbeestimated. Fordesignswithoddresolution(ier≡ 1 mod 2), we can estimate independent efects of order up toe = (r− 1)/2, but at least someoftheefectsoforderewillbealiasedwithinteractionsoforder (e + 1). Fordesigns withevenresolution(ier≡ 0 mod 2),wecanestimateallefectsofordere = (r− 2)/2, independently from each other and from all interactions of order (e + 1). These relations suggestthathigherresolutionsalsorequirelargerdesigns. ResolutionIIIorthogonaldesigns arethemostfrequentlyuseddesigns,especiallyinmarketing. A.3.2.1 Constructionofoptimaldesigns Theprimarygoalinchoicedesignsistodeneagroupofchoicesetsthatminimizesthe“size” of the covariance matrix of the parameters estimated from that design. The “size” of this matrixisrelatedtothenotionofthe efficiencyofthedesignmatrixX N D ×p . Hence, ane- cientdesignisonethatminimizesthe“size”ofthevariance-covariancematrixoftheparam- eter vectorβ. In a least squares analysis, the variance-covariance matrix is proportional to 152 (X 0 X) −1 ,whichistheinverseoftheinformationmatrix (X 0 X)ofthedesignmatrix. The sizeofvariance-covariancematrixisgivenbytheeigenvaluesof (X 0 X) −1 . Therearethreecommonlyusedmeasuredofeciency.Therst,A-efficiency,isbasedonthe A-error measure,whichisthearithmeticmeanoftheeigenvalues,orequivalentlythearith- meticmeanofthetraceofthevariance-covariancematrix,trace n (X 0 X) −1 o /p. Asecond measure, D-efficiency, isbasedonthe D-error, whichisthegeometricmeanoftheeigenval- ues(ie|(X 0 X) −1 | 1/p ).Thethirdcommonlyusedmeasure,G-efficiency,isbasedonameasure knownasσ M ,whichrepresentsthemaximumstandarderrorforpredictionoverthecandidate set.Itisconvenienttoscalethesemeasuresofeciencytoarangeof0to100,asfollows A eff = 100· h N D ·trace n (X 0 X) −1 o /p i −1 (A.13) D eff = 100· h N D ·|(X 0 X) −1 | 1/p i −1 (A.14) G eff = 100·σ −1 M q p/N D (A.15) Eachoftheseeciencymeasuresassessesthegoodnessofthedesignrelativetoahypothetical orthogonal design, which may not exist in practice. Hence, they have no value as absolute measuresofdesigneciency. Rather,theyareonlyusefulforcomparingtwodesignsforthe samesituation.Asaresult,ecienciesthatarenotnear100maystillbesatisfactorydesigns. Giventhatallthesemeasuresareconvexfunctionsoftheeigenvaluesof (X 0 X) −1 ,theywill usuallybehighlycorrelated. Hence, ifabalancedandorthogonaldesignexists, it willhave optimaleciencybyallthreemeasures. Conversely, themoreecientadesignisbyanyof thesethreemeasures, themoreitwilltendtosatisfythebalanceandorthogonalityproper- ties[23]. 153 A.3.2.2 Propertiesofanoptimaldesign Fourfundamentalprinciples(orthogonality, levelbalance, minimaloverlap, andutilitybal- ance)jointlyensurethatadesignhasanoptimalD-eciency[23].Theorthogonalityprinciple requiresthelevelsofeachattributetovaryindependentlyofoneanother.Adesignisorthog- onalifthesub-matrixof (X 0 X) −1 thatresultsfromtheexclusionoftherowofcolumnfor theinterceptisdiagonal. Level balancerequiresthelevelsofeachattributetooccurwiththe sameprobability.Adesignisbalancedwhenallof-diagonalelementsintheinterceptrowand columnarezero. Balanceandorthogonalityarerelatedinthesensethatnon-orthogonality increasesthevarianceoftheparameterestimates. Adesignisbothorthogonalandbalanced when (X 0 X) −1 is diagonal. Minimal overlap occurs when the alternatives in each choice sethavenon-overlappingattributelevels. Finally,utility balancerequiresthattheutilitiesof alternativeswithineachchoicesetbethesame.Thismeansthatthedesignwillbecomemore ecientastheexpectedprobabilitieswithinachoicesetamongalternativesapproaches 1/J n . Inpractice, itisoftenimpossibletosatisfyallthesecriteriaatthesametime; hencecurrent designapproachesattempttoapproximatelysatisfythemthroughcomputeralgorithms. Themajorityofmethodsavailabletobuildchoicedesignsrelyonlinearmodels.Intheselinear models,optimaldesignsareconstructedbychoosingthevectorofattributesandtheirlevels X in order to minimizevar (β|X) = (X 0 X) −1 σ 2 , whereσ 2 is the variance of the error termsfromthelinearmodel. Recently,however,Klausetal.[27]proposedanapproachfor identifyingecientdesignsbasedonnon-linearmodels.Theirapproachreliesonthevariance ofthevectorofparameterestimates’ă ˆ βinthemultinomiallogitmodel: var ˆ β =     ∂ 2 L(β) ∂β 2     −1 =   N X k−1 N{Δ(β)− Γ(β)}   −1 (A.16) 154 where L(β) = N Y k−1 exp n P m j−1 f j x 0 j β o P m j−1 exp n x 0 j β o N , (A.17) Δ(β) = P m j−1 exp n x 0 j β o x 0 j x j P m j−1 exp n x 0 j β o , (A.18) Γ(β) = P m j−1 exp n x 0 j β o x j P m j−1 exp n x 0 j β o x j 0 P m j−1 exp n x 0 j β o 2 , (A.19) andm,nandN denoterespectivelythenumberofbrands,choicesets,andindividuals[27]. This approach adapts Cook and Nachtsheim [12]’s modication of the Fedorov algorithm, which was applied in the context of linear model designs [18]. Conceptually, the approach consistsofrstconstructingacandidatesetofpotentialalternatives.Second,arandomchoice setofthesepotentialalternativesisselectedtobethestartingdesign.Third,thestartingdesign isthenmodiedbythealgorithmthroughanexchangeofitsalternativeswiththecandidate alternatives. Fourth,thealgorithmidentiesthebestexchangefortherstalternativeinthe startingdesign,ifsuchanexchangeexists. Itisworthwhilenotingthatinbothlinearandnon-linearmodels,theoptimaldesigndepends onβ,theparametervectorwearetryingtoestimateintherstplacewiththroughtheexper- iment. Sinceβ is unknown, it is typical to make assumptions about its possible values. In empiricalapplications,β issettothevector 0,orreplacedbypriorsabouttheparameterval- uesinthedesignphase.Thesepriorscanbeobtainedfromtheliteratureorfrompilotstudies. 155 A.4 Questionnairedevelopment A.4.1 Diseaseattributes Priorstudiessuggestthatthediseasecharacteristicsthatmostinuenceprivatevaccination decisionaretheriskofinfection,theseverityofthevaccine-preventabledisease,andtheriskof deathfromthevaccine-preventabledisease[11,15,29,35].Inordertocapturetheseconsidera- tions,webeganwithaninitiallistofinfectiousdiseases(TableA.1)andconstructed5disease attributesandattributelevels, takingintoaccountinputsfromthefocusgroupdiscussions (Table A.2). These disease attributes and attribute levels were used to construct unlabeled diseasescenariosforeachchoiceprole. 156 TableA.1:Diseasecharacteristics. Disease Type Transmission Symptoms InfectionRisk CFR a Severity Reference Ebola Acute Person-to-person (direct contact, dropletspread) Highfever;severeheadacheandmusclepain; weakness; frequent diarrhea, vomiting, and severestomachpain;frequentbleeding. 10in100million 25% High [7,10,16] SARS-Cov Acute Person-to-person (dropletspread) High fever (temperature greater than 100.4 ◦ F [> 38.0 ◦ C]); severe headache and bodyaches;pneumonia;drycough;diarrhea. 10in100million 15% High [3,36,47] MERS-Cov Acute Person-to-person (dropletspread) Fever,cough,shortnessofbreath. 1in100million 35% High [5] HIV Chronic Person-to-person (directcontact,i.e.sex, or vehicle-borne, i.e. blood) Fever;chills;rash;nightsweats;muscleaches; sore throat; fatigue; swollen lymph nodes; mouth ulcers; rapid weight loss; extreme and unexplained tiredness; diarrhea lasting morethanaweek;pneumonia;memoryloss, depression,andneurologicdisorders. 4in100million 3% Medium [2,21,42] Tuberculosis Chronic Person-to-person (airborne) Bad cough lasting 3 weeks or longer; pain in the chest; coughing up blood or spu- tum;weaknessorfatigue;weightloss;lossof appetite;chills;fever;sweatingatnight. 30%in1million 0.1% Low [8,17,19,38] Inuenza Acute Person-to-person (direct contact; droplet spread; vehicle-borne; air- borne) Fever or feeling feverish/chills; cough; sore throat; runnyorstufynose; muscleorbody aches;headaches;fatigue(tiredness). 20,000in1million 4% Medium [1,4] Pertussis Acute Person-to-person (dropletspread) Runnynose; low-gradefever(generallymin- imal throughout the course of the disease); mild,occasionalcough;apnea. 100in1million 10% Medium [1,6] Continued... 157 TableA.1:Diseasecharacteristics. Disease Type Transmission Symptoms InfectionRisk CFR a Severity Reference Measles Acute Person-to-person (airborne) Highfever; cough; runnynose(coryza); red, wateryeyes(conjunctivitis). 1in1million 0.2% Medium [6,10] Mumps Acute Person-to-person (airborne, direct con- tact,dropletspread) Fever; Headache; Muscle aches; Tiredness; Lossofappetite; Swollenandtendersalivary glands under the ears on one or both sides (parotitis). 30in10million 0.0001% Low [1,9] Rubella Acute Person-to-person (dropletspread) Lowfever;rash;swollenglands;achingjoints; birth defects in pregnant women (deafness, cataracts, heart defects, mental retardation, liverandspleendamage). 5in10million 0.05% Low [1,9,10,48] a CFR=casefatalityrate,iethenumberoffractionofpeoplewithseverediseasesymptomswhodie. 158 TableA.2:Alternative-invariantdiseaseattributesandattributelevels. Attributes Levels References Diseasetype,d a Acute TableA.1 Chronic Transmission,γ g b Directcontact TableA.1 Vector-borne Airborne Riskofinfection,ρ d c 1outof1millionpeople TableA.1 30outof1millionpeople 20,000outof1millionpeople Severesymptoms,η d d Life-threateningcomplication TableA.1 Permanenthandicap Death Riskofseveresymptom,ψ(η d ) e 30outof100 TableA.1 2outof1,000 1outof1million Notes: Respondentsareinformedthatcommonsymptomsofthediseaseincludefever(morethan100 ◦ F),headache,muscle pain,andtiredness. Thediseaseattributesaredenedtobealternative-invariant,meaningthattheytakethesamevalueacross allalternativesinanygivenchoiceset.TheriskprobabilitiesarepresentedinawaythatisconsistentwithCDCandothergov- ernmentagencies’reportingstyle. a Denedasthecourseanddurationofillness.Anacutediseaseisdenedasonethatdevelopsrapidlybutlastsashorttime(ie within3weeks);achronicdiseaseisdenedasonethatdevelopsmoreslowlyandhassymptomsthatarecontinualorrecurrent forlongperiods. b Denedasthewayinwhichpeoplebecomeinfectedwiththedisease. c Denedastheproportionofpopulationinfectedwiththenewdisease,iehavingdiseasesymptoms. d Denedastheseverityofthediseasesymptomsafterinfectionoccurs. e Denedastheproportionofinfectedpopulationthatsufersseveresymptoms,includingdeath. A.4.2 Vaccineattributesandattributelevels The empirical literature on vaccination choice suggests that the out-of-pocket cost, ecacy (iethereductionintheriskofinfectionduetovaccination), durationofprotectionagainst thedisease, andsafety(i.e. severityofvaccine-relatedsideefects, riskofoccurrenceofsuch adverseevents)ofthevaccineareimportantvaccineattributesthatarelikelytoafectvaccine utilization[11,15,29,35].Weusedtheseresultsasthebasisforthedevelopmentofthevaccine attributes. The levels of each vaccine attribute were derived from the published literature, andwerecalibratedtothoseoflicensedvaccinesormostadvancedcandidatevaccinesinthe 159 pipeline, in order to make the choice proles more realistic and hence increase the external validityofourndings.Thelistofvaccineslicensedforimmunizationanddistributioninthe USareobtainedfromvaccines.gov,aUSDepartmentofHealthandHumanServices(HHS) portal[43].WederivedthevaccinepricerangesusingdatafromtheCDCvaccinepricelists[5]. Thenallistofattributesandattributelevelswasdeterminedafterinputsfromexpertand focusgroupinterviewsandarepresentedinTableA.3. TableA.3:Alternative-varyingvaccineattributesandattributelevels. Attributes Levels References Modeofadministration,α v a Oralsolution(1liquiddropletinthemouth) Focusgroup Needleinjection(1shotinthearm) Intranasal(sprayinsidethenose) Efectiveness,φ v b 70%;95%;100% [26,31] Protectionduration,τ v c 1year;6years;Lifetime Focusgroup Severevaccinesideefect,η v d Life-threateningallergicreactions. [6] Severeneurologicalinjurythatcanbecomeperma- nent(e.g.muscleparalysis). Death. Riskofseveresideefect,ψ(η v ) e 10outof1millionvaccinatedpeople [6] 4outof1millionvaccinatedpeople 1outof1millionvaccinatedpeople Out-of-pocketcost,p v f $0;$100;$700 [5] Notes:Theseattributesaredenedtobealternative-varying,meaningthattheycantakedistinctvaluesacrossalternatives.Theirvalues fortheopt-outoption(“Novaccination”)aredenotedby“n.a.” or“$0”. Foreachchoiceprole,respondentsareinformedthatboth vaccinesmightcausemildsideefectsthatdonotinterferewithnormalactivities,suchasmildinjectionsitesoreness,swelling,orred- nessandmildfever,rash,orachiness[6]. Theriskprobabilitiesforthevaccinesideefectsandthevaccineecacylevelsarepresented inawayconsistentwithCDCandothergovernmentagencies’reportingstyle. a Denedasthewayinwhichthesubjectwillreceivethevaccine. b Denedastheproportionofpeoplewhowillbeprotectedagainstthediseasewhenvaccinated. c Denedasthenumberofyearsduringwhichthevaccinewillprotectthesubjectagainstthedisease. d Denedassideefectsthatarelife-threateningorresultininpatienthospitalization,surgicalintervention,ordeath.Severesideefects anddeathareonlyduetothesafetyofthevaccine,butnottoitsecacy,anddonotmakethevaccinelessecaciousagainstthedisease. e Denedastheproportionofvaccinatedpeoplewhosuferseverevaccinesideefects. f Denedastheamountpaidout-of-pocketbytherespondenttoreceivethevaccine.Forexample,foravaccinethatprovidesprotection for1yearandhasanout-of-pocketcostof$700,thesubjectwouldpay$700eachyeartoreceivethevaccine. 160 A.4.3 Vaccineinjurycompensationattributelevels A.4.3.1 BackgroundontheUSNationalVaccineInjuryCompensationProgram(VICP) In the US, vaccine product liability and injury malpractice lawsuits for all vaccines recom- mended by the Advisory Committee on Immunization Practices (ACIP) are administered through the National Vaccine Injury Compensation Program (VICP). VICP is a no-fault compensationsystemforindividualsfoundtohavebeeninjuredbyACIP-recommendedvac- cines. Althoughtheprogramwasdesignedtoonlycoverchildhood-recommendedvaccines, italsocoversmanyadultvaccines.Theprogramwasestablishedin1986throughtheNational ChildhoodVaccineInjuryAct(whichbecameefectiveinOctober1998)inresponsetovac- cineshortagescausedbythehostilelegalenvironmentfacedbyvaccinemanufacturers[46]. The Act aimed to ensure an adequate supply of vaccines, stabilize vaccine costs, and estab- lishandmaintainanaccessibleandecientsettingforprovidingcompensationtoconsumers injuredbycertainchildhoodvaccines.Throughano-faultcompensationprogram,thedesign- ersoftheprogramaimedtocompensateplaintifsofvaccineinjury,whilerelievingmanufac- turersfromthelegalburdenofvaccineinjuryclaimsbyrequiringcasestorstbeexaminedby therestrictedcivillitigationprocessoftheVaccineCourt[46]. Thelegislationalsocalledfor thereportingofvaccine-relatedadverseevents,thecreationofvaccine-informationmaterials detailing the benets and risk of each vaccine, the conduct of Institute of Medicine studies ofpotentialvaccine-relatedinjuries,andtheresearchanddevelopmentofnewandsafervac- cines[13]. TherstinstanceforadjudicatingallclaimsagainsttheseACIP-recommendedvaccineman- ufacturersistheCourtofClaims(VaccineCourt),whichsitswithoutajury.Thismeansthat beforeanyvaccineclaimcanbelitigatedinStateorFederalcivilcourts,theymustrstbeheard bytheVaccineCourt.Therearetwoexceptionstothisrule:Therstexceptionconcernsthe precedentsetbytheHoldervs. AbbottLaboratoriesInc.,444F.3d383caseinwhichtheUS 161 FifthCircuitCourtofAppealsruledthatplaintifssuingthreemanufacturersofthiomersal couldbypasstheVaccineCourtandlitigateineitherStateorFederalcourt, usingthetradi- tionaltortsystem[14]. Thesecondexceptionconcernsvaccinesafectedbythe2002Home- landSecurityAct,whichstipulatesthatifasmallpoxvaccineweretobewidelyadministered byfederalauthoritiesinresponsetoaterroristthreatorotherbiologicalwarfareattack,per- sonsadministeringorproducingthevaccinewouldbedeemedfederalemployeesandwould besubjecttotheFederalTortClaimsAct,therebyrequiringtheplaintiftosuetheUSGov- ernmentintheUSDistrictCourts,andendorsetheburdenofthemuchhigherstandardof provingthedefendant’snegligence[32]. VICPcompensationsincludesmonetarydamagesforpainandsufering,pastandfuturemed- icalexpenses,pastandfuturelostwages,andreasonableattorneys’feesandcosts. Compen- sations for pain and sufering and death are both capped to $250,000, but there is no limit ofcompensationformedicalexpensesandlostwages[41].Theprogramtypicallypaysattor- neys’ĂŹfeesforinjuryclaimsledona“goodfaithandreasonablebasis”regardlessofenti- tlementoutcome[41].Compensationsaretypicallymadeasaninitiallumpsumandannuity whichpayslifetimestreamofbenets.PaymentsforclaimsaremadefromtheVaccineInjury CompensationTrustFundwhichisfundedthroughanexcisetaxof75centsperdoseofvac- cinesold.Sinceitsinceptionin1988,theprogramhasawardedmorethan$2.8billionintotal compensation. Inordertobeeligibleforcompensation,theplaintifmustleaclaimwithin36monthsfrom dateofoccurrenceoftherstsymptomsormanifestationofonsetorofsignicantaggravation ofinjury. Inthecaseofdeathmustbeledtheclaimwithin24monthsfromdateofdeath, butnomorethan48monthsafterdateofoccurrenceofrstsymptomormanifestationof onsetorofsignicantaggravationofinjuryfromwhichdeathresulted. 162 Inordertofacilitatetheclaimsprocess,VICPpublishedanInjuryTablewhichlistsinjuries presumedtohavebeencausedbythevaccinestheprogramcovers.Toreceivecompensation, the plaintif must demonstrate either (i) a proof of a Vaccine Injury Table condition, (ii) a proofofcausation,or(iii)aproofofsignicantaggravation.Thestandardofproofisprepon- deranceofevidence(morelikelythannot),andthetheproofsaretypicallymedicalrecords. Sometimes, the program also compensates individuals for injuries not on the Injury Table; however,inthosecases,causationisnotpresumed. HHSprovidesrecordsofcompensationamountspaidforspecicvaccineinjuryclaimsunder VICP during the 1989-2015 period in the court decisions for each claim [40]. During that period,atotalof 16, 014petitionswereledand 4, 112oftheseclaimswereawardedcom- pensation. Thelawallowsamaximumcompensationamountof$250,000incaseofdeath; compensationforpainandsuferingislimitedto$250,000perinjury. However,thereisno limitofcompensationformedicalexpensesandlostwages.Dependingontheseverityofthe injury,themedicalexpensesandlostwagescouldbeasmuchas$1million. Usingthecourt decisionrecords,wedenedthecurrentVICPcompensationamountsforeachvaccineinjury asthemaximumpaymentsforthatinjuryunderthecurrentVICP.Thesevalueswerechosen asthelowerleveloftheinsurancecompensationscheme.Themediumandhighlevelsofthe insurancecompensationschemeweredeterminedbasedonthevalueofstatisticallife(VSL) in the US. The US Department of Transportation recommends using $9.4 million (range: $5.2-$13.0million)asanestimateoftheVSLforanalysespreparedin2015. Thisestimatecan beupdatedforadiferentperiodusingthefollowingequation: VSL t+k = VSL t ·     CPI t+k CPI t     ·     Y t+k Y t     (A.20) 163 whereVSL y ,CPI t andY t denoterespectivelythevalueofstatisticallife,theconsumerprice indexforallurbanconsumerscurrentseries(CPI-U),andtherealincomeinyear.Forthepur- posesofthisanalysis,weusethe2015estimates.GiventhatVICPcompensationamountsvary accordingtotheinjurytype(TableA.4),wemodeledtheinsurancecompensationamounts tovarywiththeseverityoftheinjury.Tothatend,weusedtheAbbreviatedInjuryScale(AIS) todeterminetherelativedisutilityfactors(i.e.fractionoffatality)associatedwitheachinjury andcalculateareasonablecompensationamountassociatedwitheachinjury(TableA.4).For example,thevalueofa“severe”injury(AIS4)wouldbeestimatedbymultiplyingtheFraction ofVSLforasevereinjury(0.266)bytheVSL($9.4million),yieldingthevalueof$2.5million. 164 TableA.4:Vaccineinjurycompensationamountsbyinjurytype. AIS Level Severity GeneralPrognosis[22] Fraction ofVSL[44] VaccineInjury,η v Compensationamount VICP Insurance,Y I (η v ) AIS1 Minor Canbetreatedandreleased[45] 0.003 Life-threatening allergic reac- tions within 4 hours of receiv- ingthevaccine. $30,000 $30,000 $40,000 $60,000 AIS2 Moderate Requires follow-up and weeks tomonthstoheal[45] 0.047 – – – AIS3 Serious Requires substantial follow- up; some minor disability likely[45] 0.105 – – – AIS4 Severe Hospitalization; substantial temporary and moderate long-termdisability[45] 0.266 Severe neurological injury (e.g. muscleparalysis)within2to28 daysofreceivingthevaccine. $1,000,000 $1,000,000 $2,500,000 $3,500,000 AIS5 Critical Extended hospitalization; signicant long-term disabil- ity[45] 0.593 – – – AIS6 Unsurvivable Usually fatal (though not invariably)[20,33,34,37] 1 Death within 48 months after the date of occurrence of the rstsymptomormanifestation ofonsetorofsignicantaggra- vationoftheinjuryfromwhich deathresulted. $1,250,000 $1,250,000 $5,000,000 $10,000,000 Notes:Thevalueofa“severe”injury(AIS4)wouldforexamplebeestimatedbymultiplyingtheFractionofVSLforasevereinjury(0.266)bytheVSL($9.4million),yieldingthevalueof$2.5million.All compensationamountsonlyincludemonetarydamageamountsforpainandsufering,pastandfuturemedicalexpenses,andpastandfuturelostwages. Attorneys’feesandcostsareexcludedfromthese gures. 165 Wedenedtheequivalentsubsidyamountastheexpectedvalueoftheinsurance,andcalcu- lateditbymultiplyingtheinsuranceamountbytheriskoftheinjury,asintheEquationA.21. Y S (η v ) = ψ(η v )·Y I (η v ) (A.21) OtherfactorsthatmoderatetheattractivenessoftheVICPincludethelengthoftheclaimpro- cessingtimeandthelikelihoodofreceivingcompensation.AccordingtotheUSGovernment AccountabilityOce(GAO)nearly24%ofpetitionsledsince2009wereresolvedwithin2 years(11%withinayear,and13%between1and2years);16%ofclaimswereresolvedbetween 2and5years;51%ofclaimstookmorethan5yearstoberesolved[41].TheUSDepartmentof HealthandHumanServices(HHS)reportedthat29%ofpetitionsledbetween10/01/1988 and02/03/2016wereresolvedwithacompensation. Whiletheseareallimportantfactorsto consider,weonlyincludeinourdesignthetimetoclaimcompensationasanattributewith multiple levels, and inform respondents that about 30% of claims are approved under the insurancescheme. Itisimportanttonotvarytheapprovalprobability,becauseitisendoge- nousinthesensethatitdependshighlyonthemeritsofclaim,aswellasthediscretionofthe rulingjudgeinthecase.WesummarizedinTableA.5theattributesandattributelevelsofthe variouscompensationprogramsconsideredforthisstudy.Afterthefocusgroupdiscussions, onlythecompensationamountswereretainedasattributestobeincludedinthedesign. 166 TableA.5:Attributeandlevelsoftheattributeoftheinsuranceandsubsidyprograms. Attributes Levels References Compensationamount,Y I (η v ) a Life-threatening allergic reaction TableA.4 $30,000;$40,000;$60,000 Severe neurological injury $1,000,000;$2,500,000;$3,500,000 Death from injury $1,250,000;$5,000,000;$10,000,000 Claimprocessingtime,τ C b Lessthan1year [41] 1to2years Morethan2years Note:Respondentsareinformedthat29%ofallclaimsledundertheinsuranceprogramreceivedacompensation. a Denedasthedollaramountofthecompensation;thisamountvarieswiththeinjuryseverity.Thelevelsofthisattributearealternative- invariant.FortheVICPprogram,theattributelevelsforeachinjuryaresettothesecondlevel. b Denedasthetimeittakesfromthedateoftheclaimlingtotheresolutionoftheclaim. 167 A.5 Focusgroupinterviewguide A1.Couldwegoaroundthetablesothateveryonecanintroducehimself/herself?(Pleasetell usyourname,whereyouarefrom,andwhatyoudoforliving) A2.Didyoureceiveaushotduringthelast12months?Didyoureceiveothervaccines?What motivatedyourdecisiontovaccinateornottovaccinate? A3. Haveyouorsomeoneyouknoweverfallenillwithavaccine-preventableinfectiousdis- ease?Ifso,whatwastheexperiencelike(trytothinkofbothgoodandbadexperiences)?How didthatexperienceafectthewayyouthinkaboutinfectiousdiseasesandvaccination? A4. Whatarethemostimportantdiseasecharacteristicsthatyouconsiderinyourattitudes towardsinfectiousdisease? A5. PleaseconsiderthelistofdiseasesinTableA.6,focusingrstonthesymptoms. Which one would consider to be the most severe? Which one would you consider to be the least severe?Whatifweweretofocusonthecasefatalityrate?Whatifweweretoconsiderallthe informationinthetable? A6.Now,let’sfocusonTableA.7.Whichdiseaseattributeswouldyouaddtoorremovefrom thistable?Whichlevelswouldyouchange?Dotheseseemreasonabletoyou? A7. For you, what are the most important characteristics of a vaccine? What are the most importantfactorsthatyouconsiderwhendecidingwhetherornottovaccinate? A8. Areyouworriedatallaboutpotentialvaccinesideefects? Whatdoyouconsidertobe aseverevaccinesideefect? Haveyouorsomeoneclosetoyoueverexperiencedvaccineside efects?Howdidthatexperienceafectthewayyouthinkaboutvaccines? 168 A9.Now,let’sfocusonTableA.8.Whichvaccineattributeswouldyouaddtoorremovefrom thistable?Whichlevelswouldyouchange?Dotheseseemreasonabletoyou? A10.Whatdoyouthinkoftheideaofgettingarebateonvaccineprice?Wouldthisinuence yourdecisiontovaccinate? Underwhatconditions? Whataboutaninsuranceagainstsevere vaccinesideefects? A11. PleasetakealookatTableA.9.. Whichcompensationattributeswouldyouaddtoor removefromthistable?Whichlevelswouldyouchange?Dotheseseemreasonabletoyou? A12.PleasetakeatimetoreadthroughModuleB.Couldyoutellusyourgeneralimpressions aboutthechoicetask?Isanythinguncleartoyou?Whichpartswereconfusingtoyou?What isunclearaboutthem?Howwouldyoureformulatethem? A13. What are your thoughts about the presentation of risk in the choice prole? Are the diferentrangesreasonabletoyou? A14. Let’ssummarizewhatwediscussed;didIgeteverythingrightabouttoday’sdiscussion? Doyouhaveanyquestionsaboutwhatwediscussed?Doyouhaveanysuggestionsforfurther interviews,orideasforimprovements? Thankyouforyouractiveandhonestparticipationinthediscussion.Yourfeedbacksarevery usefultous. Wewillrevisethesurveyquestionsbasedonyourinput. Afterthisstep,wewill administerthesurveytoasampleofrespondents.Theiranswerswillhelpusbetterunderstand thefactorsthatafectdemandforinfectiousdiseasevaccines.Theywillalsohelpusunderstand policiesthataremoreefectiveatincreasingprivatedemandforvaccines. 169 TableA.6:Focusgroupexhibit1 Disease Type TransmissionMode SevereSymptoms InfectionRisk CFR(%) a Ebola Acute Person-to-person (directcontact,dropletspread) Highfever;severeheadacheandmusclepain;weakness;frequentdiarrhea,vomiting,and severestomachpain;frequentbleeding. 10in100million 25 SARS/COV Acute Person-to-person (dropletspread) High fever (temperature greater than 100.4 ◦ F [>38.0 ◦ C]); severe headache and body aches;pneumonia;drycough;diarrhea. 10in100million 15 MERS/COV Acute Person-to-person (dropletspread) Fever,cough,shortnessofbreath. 1in100million 35 HIV Chronic Person-to-person (direct contact, i.e. sex, or vehicle- borne,i.e.blood) Fever;chills;rash;nightsweats;muscleaches;sorethroat;fatigue;swollenlymphnodes; mouthulcers;rapidweightloss;extremeandunexplainedtiredness;diarrhealastingmore thanaweek;pneumonia;memoryloss,depression,andneurologicdisorders. 4in100million 3 Tuberculosis Chronic Person-to-person (airborne) Bad cough lasting 3 weeks or longer; pain in the chest; coughing up blood or sputum; weaknessorfatigue;weightloss;lossofappetite;chills;fever;sweatingatnight. 30in1million 0.1 Inuenza Acute Person-to-person (direct contact; droplet spread; vehicle-borne;airborne) Feverorfeelingfeverish/chills;cough;sorethroat;runnyorstufynose;muscleorbody aches;headaches;fatigue(tiredness). 20000in1million 4 Pertussis Acute Person-to-person (dropletspread) Runnynose; low-gradefever(generallyminimalthroughoutthecourseofthedisease); mild,occasionalcough;apnea. 100in1million 10 Measles Acute Person-to-person (airborne) Highfever;cough;runnynose(coryza);red,wateryeyes(conjunctivitis). 1in1million 0.2 Mumps Acute Person-to-person (airborne, direct contact, droplet spread) Fever;Headache;Muscleaches;Tiredness;Lossofappetite;Swollenandtendersalivary glandsundertheearsononeorbothsides(parotitis). 30in10million 0.0001 Rubella Acute Person-to-person (dropletspread) Lowfever;rash;swollenglands;achingjoints;birthdefectsinpregnantwomen(deafness, cataracts,heartdefects,mentalretardation,liverandspleendamage). 5in10million 0.05 a CFR=casefatalityrate,i.e.thenumberoffractionofpeoplewithseverediseasesymptomswhodie.tablenotes 170 TableA.7:Focusgroupexhibit2–denitionsofthediseaseattributes AttributesandLevels Denitions Courseandduration Howquicklythediseasedevelopsandlasts. Acute Thediseasedevelopsrapidlybutlastsashorttime(i.e.within3weeks). Chronic Thediseasedevelopsmoreslowlyandhassymptomsthatarecontinualorrecurrentforlong. Riskofinfection Theproportionofpopulationinfectedwiththenewdisease,i.e.havingdiseasesymptoms. 1outof100millionpeople 100outof1millionpeople 2,000outof1millionpeople Severesymptoms Theseverityofthediseasesymptomsafterinfectionoccurs. Life-threateningcomplication Seriouscomplicationrequiringhospitalization(e.g.severebleedingandbruising,pneumonia,meningitis,swelling ofthebrain,neurologicdisorder). Permanenthandicap Blindness,deafness,paralysis,mentalretardation,birthdefects. Death Deathasaresultofthedisease. Riskofseveresymptom Theproportionofinfectedpopulationthatsufersseveresymptoms,includingdeath. 30outof100infectedpeople 2outof1,000infectedpeople 1outof1millioninfectedpeople 171 TableA.8:Focusgroupexhibit3–denitionsofthevaccineattributes AttributesandLevels Denitions Ecacy:70%;95%;99% Theproportionofpeoplewhowillbeprotectedagainstthediseasewhenvaccinated. Protectionduration:1year;6years;Lifetime Thenumberofyearsduringwhichthevaccinewillprotectyouagainstthedisease. Severevaccinesideefect Sideefectsthatarelife-threateningorresultininpatienthospitalization,surgicalintervention,ordeath.Severesideefects anddeathareonlyduetothesafetyofthevaccine, butnottoitsecacy, anddonotmakethevaccinelessecacious againstthedisease. Life-threateningallergicreactions Severeallergicreactions(anaphylaxis)followingreceptionofthevaccine. Severeneurologicaldisorder Severeneurologicalinjurythatcanbecomepermanent(e.g.Guillain-BarréSyndrome,muscleparalysis). Death Deatheventassociatedwithreceptionofthevaccine. Riskofseverevaccinesideefect Theproportionofvaccinatedpeoplewhosuferseverevaccinesideefects. 10outof1millionvaccinatedpeople 4outof1millionvaccinatedpeople 1outof1millionvaccinatedpeople Out-of-pocketcost:$0;$100;$700 Theamountofmoneyyouwillpayoutofyourownpockettoreceivethevaccine.Forexample,foravaccinethatprovides protectionfor1yearandhasacostof$700,youwillpay$700eachyearyoureceivereceivethatvaccine. 172 TableA.9:Focusgroupexhibit4–denitionsofthecompensationprogramattributes AttributesandLevels Denitions Compensationamount Thedollaramountofthecompensation;thisamountvarieswiththeinjuryseverity. Life-threatening allergic reactions $30,000;$40,000;$60,000 Severe neurological disorder $1,000,000;$2,500,000;$3,500,000 Death $1,250,000;$5,000,000;$10,000,000 173 ModuleB Inthispartofthesurveyweaskyoutomake7choicesbetween3vaccinationalternatives (“Novaccination”,“VaccineA”and“VaccineB”).Ineachchoice,weaskyoutoconsiderthe followingscenario: ImaginethatamassiveoutbreakofnewinfectiousdiseaseisabouttooccurintheUS.There iscurrentlynodrugthatcuresthisdisease; youcanonlytakecertaindrugstomanagethe diseasesymptoms. Healthocialshavedeterminedthatvaccinationisthemostefectivepreventivemeasure againstthisdisease. Twonewvaccines(VaccineAandVaccineB)havebeendevelopedand canbothreduce(butnoteliminateentirely)yourriskofinfection. Thevaccinesareavail- ableforeveryoneintheUSatanymajorlocaldrugstoreordoctor’soce.Bothvaccinescan causemildsideefectssuchasmildinjectionsiteswellingandmildfever,butareotherwise generallyverysafe.Bothvaccinesarecoveredbyaninsurancethatpaysamonetarycompen- sationifcertainseverevaccineinjuriesoccur. Approximately30%ofallclaimsledunder theinsuranceprogramreceiveacompensation. Thetwonewvaccinesaresimilarinallcharacteristicsanddiferonlyinthefeaturesdescribed below. Foreachchoice,youwilldecidewhichalternativeyoupreferbasedonthecharacteristicsin TableA.Foreachchoice,youwilldecidewhichalternativeyoupreferbasedonthecharac- teristicsinTableAbelow. Foreachchoice,acopyof TableAwillbeavailableasapop-up windowbyclickingonthehyperlinkedtext“here”. 174 ModuleB(continued) PLEASECONSIDERALLTHEINFORMATIONBELOWBEFOREMAKINGACHOICE. Pleaseindicatebelowwhetheryouprefer“Novaccination”,“VaccineA”,or“VaccineB”. Thescenarioisavailablehereforyourreference. Thetableofcharacteristics(TableA)isavailablehereforyourreference. Whatdoyouchooseforyourself?PleasecheckonlyoneoftheradiobuttonsunderTableC(i.e.“No vaccination”or“VaccineA”or“VaccineB”)toindicateyourchoice.THEREARENORIGHTOR WRONGANSWERS. TableB Attributes Levels Diseasetype Acute. Riskofinfection 1outof100millionpeople. Severesymptoms Life-threateningcomplications. Riskofseveresymptoms 30outof100infectedpeople. TableC Attributes Novaccination VaccineA VaccineB Vaccineefectiveness n.a.Thismeans1outof 100,000peoplegetinfected 95%.Thismeans95outof100 vaccinatedpeopleareprotected againstthedisease 70%.Thismeans70outof100 vaccinatedpeopleareprotected againstthedisease Duration of protection against thedisease n.a. 1year 3years Severevaccinesideefects n.a. Life-threateningallergic reactions. Severeneurologicdisorder. Risk of severe vaccine side efects n.a. 1outof1millionvaccinated people. 10outof1millionvaccinated people. Compensation amount in case ofthisseverevaccinesideefect n.a. $75,000 $150,000 Out-of-pocket cost for a com- pletevaccination $0 $100 $1,000 Whatdoyouchooseforyourself?(Pleasecheckoneboxonly) # Novaccination # VaccineA # VaccineB A.6 DCEsurveyinstrument [MESSAGETOAPPEARONTHEMTURKPAGE] 175 Description Youarebeinginvitedtoparticipateinastudyinvestigatingdecision-makingaboutvaccines forinfectiousdiseases.Youwillearn$1.00forcompletingasurveythattakesapproximately8 minutestocomplete.Wewillpresentyouaseriesof8hypotheticalvaccinationscenariosand askyoutomakeadecisionaboutwhetherornotyouwouldvaccinate. Wewillalsoaskyou somebasicquestionsaboutyoursocio-demographicsituationandissuesrelatedtovaccina- tion. Thedatawearecollectingisbeingusedforascienticresearch;pleasedonottakethissurvey ifyouarenotwillingtocommit8minutesofyourtimetoit. PleaseleavethisMTurkwindowopenasyoucompletethesurvey. Attheendofsurvey,you willbegivenasurveycompletioncodetopasteintotheboxcalled“Provideyourcompletion codehere”,inordertoreceiveyour$1.00pay. PleaseclickontheSurveylinkbelowtostartthesurvey. Thankyouagainforhelpinguswiththisresearch. Surveylink:[INCLUDESURVEYLINK] Provide your completion code here: [BLANK TO BE FILLED WITH COMPLETION CODE] 176 [CONSENTFORMFOLLOWS] [GENERATE ASSIGNMENT TOKEN TO ASSIGN TO INSURANCE OR SUBSIDY GROUPSANDBLOCKS] MODULEA:Experiencewithvaccination C1.Whatisyourgender?(Pleaseselectone)1=Female 2=Male 3=Other(Pleasespecifybelow) 4=Don’tknow/Notsure 5=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] C2.Whatisyourage?(Pleaseselectone) 1=18to24years 2=25to34years 3=35to44years 4=45to54years 5=55to64years 6=Age65orolder 7=Don’tknow/Notsure 8=Refusetotell [RADIOBUTTONS] C3.Whatisyourethnicity?(Pleaseselectallthatapply) 1=Asian 2=Black/AfricanAmerican 3=Hispanic 177 4=PacicIslander 5=White 6=Other(Pleasespecifybelow) 7=Don’tknow/Notsure 8=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] A1. Whichofthefollowingisatrustedandreliablesourceofinformationwhenitcomesto decidingwhetherornottovaccinate?(Pleaseselectallthatapply) 1=TVorradioannouncements 2=Celebrities 3=TheCentersforDiseaseControlandPrevention(CDC) 4=TheWorldHealthOrganization(WHO) 5=Newsontheinternet 6=Postingsonsocialmedia(Twitter,Facebook,etc.) 7=Mydoctor 8=Publichealthocials 9=Electedleaders 10=Wordofmouthfromfriends,family,co-workersorneighbors 11=Other(Pleasespecifybelow) [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] A2.Didyougetyourushotwithinthelast12months?(Pleaseselectone) 1=Yes 2=No 3=Don’tknow/Notsure 178 4=Refusetotell [RADIOBUTTONS] A3.Haveyoueverexperiencedseveresideefectsfromvaccination?(Pleaseselectone) 1=Yes(Pleasedescribebelowthevaccineyoureceivedandthesideefectsyouexperienced) 2=No 3=Don’tknow/Notsure 4=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHYES] A4. How much do you agree or disagree with the statement that “vaccines cause autism”? (Pleaseselectone) 1=Stronglydisagree 2=Disagree 3=Undecided/neutral 4=Agree 5=Stronglyagree [RADIOBUTTONS] A5.Everyonehashobbies.Nevertheless,wewouldlikeyoutoskipthisquestiontoshowthat youarereadingcarefully.Donotclickanyofthebuttonscorrespondingtohiking,watching TV,playingsports,bikeriding,readingorswimming. 1=Hiking 2=WatchingTV 3=Playingsports 4=Bikeriding 5=Reading 179 6=Swimming [RADIOBUTTONS] A6.Howmuchdoyouagreeordisagreewiththestatementthat“Naturalproductsandsub- stancesorothernon-medicalbehaviors(e.g.eatingproperlyandexercising)arebetteralterna- tivestovaccination”?(Pleaseselectone) 1=Stronglydisagree 2=Disagree 3=Undecided/neutral 4=Agree 5=Stronglyagree [RADIOBUTTONS] A7. Howmuch,ifanything,haveyoureadorheardabouttheUSNationalVaccineInjury CompensationProgram(VICP)?Haveyouheard(Pleaseselectone) 1=Alot 2=Alittle 3=Nothingatall 4=Don’tknow/Notsure 5=Refusetotell [RADIOBUTTONS] MODULEB:DiscreteChoiceExperiment Wewillpresentyouasequenceof8vaccinationscenariosandaskyoutomakeadecisionabout whetherornotyouwouldvaccinateineachsituation. 180 [RANDOMIZECHOICESWITHINEACHBLOCK] [INALLCHOICEPROFILES,INCLUDEHYPERLINKEDTEXTFOR“EFFECTIVE- NESS”:“Denedasthefractionofvaccinatedpeoplewhowillnotgetthedisease.Forexam- ple,ifadiseasehas1outof1millionriskofinfection,andthereisavaccinewith70%efective- nessinprotectingagainstthisdisease,thismeansthat70outof100peoplewhoreceivethis vaccinewillnotgetthedisease,and30outof100peoplewhoreceivethevaccinewillstillhave 1out1millionriskofinfection.”] [DCEQUESTIONSFOLLOW] [POST-DCEFOLLOW-UPQUESTIONSFOREACHBLOCKINTHEINSURANCE ARM:ONLYASKIF“NOVACCINATION”TOALL7CHOICESETS] Whatisminimumcompensationamountforwhichyouwouldbewillingtovaccinate? [INCLUDEBLANKTOBEFILLED] Whatismaximumriskofseverevaccinesideefectsforwhichyouwouldbewillingtovacci- nate? [INCLUDEBLANKTOBEFILLED] [POST-DCEFOLLOW-UPQUESTIONSFOREACHBLOCKINTHESUBSIDYARM: ONLYASKIF“NOVACCINATION”TOALL8CHOICESETS] Whatisminimumsubsidyamountforwhichyouwouldbewillingtovaccinate? [INCLUDEBLANKTOBEFILLED] Whatismaximumriskofseverevaccinesideefectsforwhichyouwouldbewillingtovacci- nate? [INCLUDEBLANKTOBEFILLED] MODULEC:Thoughtsaboutvaccination-relatedissuesandsocio-demographicinformation 181 Inthispartofthesurvey,wewillaskforyourthoughtsonsomeissuesrelatedtovaccination. Wewillalsoaskyoutoprovideusmoreinformationaboutyoursocialanddemographicsitu- ation. C4.Ingeneral,howwouldyousayyourhealthis?(Pleaseselectone)1=Excellent 2=Verygood 3=Good 4=Fair 5=Poor 6=Don’tknow/Notsure 7=Refusetotell[RADIOBUTTONS] C5.Whattypeofinsurancedoyoucurrentlyown?(Pleaseselectallthatapply) 1=Commercial/privateinsurance 2=Medicare 3=Medicaid 4=Militaryinsurance(VAoractivemilitary) 5=Other(Pleasespecifybelow) 6=Noinsurance 7=Don’tknow/Notsure 8=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] C6.Whatisyourmaritalstatus?(Pleaseselectone) 1=Single(nevermarried) 2=Married 3=Separated/Divorced 4=Widowed 182 5=Other(Pleasespecifybelow) 6=Don’tknow/Notsure 7=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] C7.Whatisthehighestdegreeorlevelofeducationyouhavecompleted?(Pleaseselectone) 1=Lessthanhighschool 2=Highschoolgraduate(includesequivalency) 3=Somecollege,nodegree 4=Associate’sdegree 5=Bachelor’sdegree 6=Doctoraldegree 7=Graduateorprofessionaldegree 8=Other(Pleasespecifybelow) 9=Don’tknow/Notsure 10=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] C8. What was your total household income before tax during the past 12 months? (Please selectone) 1=Lessthan$25,000 2=$25,000to$34,999 3=$35,000to$49,999 4=$50,000to$74,999 5=$75,000to$99,999 6=$100,000to$149,999 7=$150,000ormore 183 8=Other(Pleasespecifybelow) 9=Don’tknow/Notsure 10=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] C9.Whichofthefollowingsectorsbestdescribestheorganizationorcompanyyouworkfor? (Pleaseselectone) 1=Privatesector(e.g.mostbusinessesandindividuals) 2=Publicsector(e.g.government) 3=Not-for-protsector 4=Other(Pleasespecifybelow) 5=Don’tknow/Notsure 6=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] C10. When you think about your religious or spiritual life, which of the following do you consideryourselftobe?(Pleaseselectone) 1=Veryreligiousorspiritual 2=Somewhatreligiousorspiritual 3=Notreligiousorspiritualatall 4=Other(Pleasespecifybelow) 5=Don’tknow/Notsure 6=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] C11.Whichregionofthecountrydoyoulivein?(Pleaseselectone) 1=Midwest-IA,IL,IN,KS,MI,MN,MO,ND,NE,OH,SD,WI 2=Northeast-CT,DC,DE,MA,MD,ME,NH,NJ,NY,PA,RI,VT 184 3=Southeast-AL,AR,FL,GA,KY,LA,MS,NC,SC,TN,VA,WV 4=Southwest-AZ,NM,OK,TX 5=West-AK,CA,CO,HI,ID,MT,NV,OR,UT,WA,WY 6=Other(Pleasespecifybelow) 7=Don’tknow/Notsure 8=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] C12.Whichofthefollowingbestdescribestheareayoulivein?(Pleaseselectone) 1=Urban 2=Rural 3=Suburban 4=Other(Pleasespecifybelow) 5=Don’tknow/Notsure 6=Refusetotell [RADIOBUTTONS;INCLUDEBLANKTOBEFILLEDWITHOTHER] MODULED:Commentsaboutthesurvey D1.Inthespacebelow,pleasetellusyourgeneralimpressionsaboutthesurvey,andwhether anyquestionwasuncleartoyou. [INCLUDEBLANKTOBEFILLED] D2. Whatcharacteristicsofavaccinearemostimportanttoyou? Pleaseanswerinthespace below. [INCLUDEBLANKTOBEFILLED] 185 D3.Whichquestionswereconfusingtoyouandwouldyoureformulatethem?Pleaseanswer inthespacebelow. [INCLUDEBLANKTOBEFILLED] YOUAREABOUTTOSUBMITYOURRESPONSES Thankyoufortakingthetimetoprovideusthisimportantinput;youareabouttosubmit yourresponsestothissurvey. ConsistentwithAmazonMechanicalTurk’s(MTurk)policy,we“mayrejectyourworkifthe HITwasnotcompletedcorrectlyortheinstructionswerenotfollowed.” Pleaseverifythat youhavecompletedcorrectlythequestionsandfollowedalltheinstructions. Aftersubmittingyouranswers,yoursurveycompletioncodewillbedisplayedonthescreen. Pleaseretainthatcompletioncode,becauseyouwillneeditforpayment. Toproceedwiththesubmission,pleaseclickontheSubmitbutton. [ADD“Submit”AND“ReviewAnswers”BUTTONS] Congratulations!Youhavesuccessfullysubmittedyourresponsesthissurvey. Yoursurveycompletioncodeis:[DISPLAYCOMPLETIONCODEHERE]. To receive payment for participating, click “Accept HIT” in the Mechanical Turk window, enterthis‘CompletionCode,thenclick“Submit”. Wewillreviewyoursubmissiontodeterminewhetheryoufollowedtheinstructions,andcan receivepayment.Ifyouranswersaresatisfactory,youwillreceiveyourpaymentwithin2busi- nessdays. Thankyouforparticipatinginthissurvey! [ENDOFSURVEY] 186 A.6.1 DCEdesign UsingtheinformationinTable2.1,aswellastheSASprogramprogrambelow,wegenerated theDCEdesigninTableA.10. SASCode:DCEdesign /****************************************************************/ /* Authors: Emmanuel Drabo, Joel Hay, Neeraj Sood, Jason Doctor */ /****************************************************************/ libname dced ’/nfs/sch-home/users/edrabo/dce_study/programs’; /* Constructing the design*/ %mktruns(2**1 3**11); /* Generating the orthogonal array design */ %mktex(2**1 3**11, n=1296, seed=123182, options=nodups nohistory); proc print data=randomized; run; %mktlab(data=randomized, vars=x1-x12, out=myfinal); proc print data=myfinal; id; run; /* Avoiding dominated and unrealistic alternatives */ %macro res; bad=0; do i=1 to nalts; do j=i+1 to nalts; do k=1 to 5; if any(x[i,k]>x[j,k]) then bad=bad+1; if any(x[j,k]>x[i,k]) then bad=bad+1; end; if all(x[i,]>=x[j,]) then bad=bad+1; if all(x[j,]>=x[i,]) then bad=bad+1; end; end; %mend; /* Run the %choiceff macro; remove duplicates */ %choiceff(data=myfinal, model=class(x1-x12 / standorth), nsets=36, flags=2, seed=123182, options=relative, restrictions=res, resvars=x1-x12, maxiter=1, beta=zero,bestout=mydesres); proc sort data=mydesres; by set; run; proc print data=mydesres; by set; id set; run; %mkteval(data=mydesres, factors=x1-x12); 187 SASCode:DCEdesign(continued) /****************************************************************/ /* Examining the design: remove duplicates */ /****************************************************************/ %mktdups(generic,data=mydesres,factors=_f1-_f2 x1-x12,nalts=2,out=mynodups); proc print data=mynodups; run; %mkteval(data=mynodups, factors=x1-x12); /* Creating blocks for the design */ %let mktopts = notes; %let mktopts = version; %mktblock(data=mydesres, nalts=2, nblocks=6, actors=x1-x12, seed=123182, maxiter=100, out=dced.myoutdat); proc print data=dced.myoutdat; by block set; run; x st myoutdat.sas7bdat myoutdat.xlsx -y; /****************************************************************/ /* END OF THE PROGRAM */ /****************************************************************/ We randomly selected one choice prole in each block be duplicated with the objective of testing choice consistency. Another choice prole was duplicated for attention check. The programusedforthesamplingisprovidedbelow.FigureA.2representsanexampleofchoice prolepresentedtotherespondentsinthesubsidygroup. RCode:Samplingoftheduplicatedchoiceproles ## ******************************************************** ## Choice of the duplicated choice set for each block ## ******************************************************** set.seed(123182) dup.choice <- sample(1:6,6,replace=T) print(dup.choice) ## ******************************************************** ## Choice of the choice set to be ignored for each block ## ******************************************************** set.seed(123183) attention.choice <- sample(1:6,6,replace=T) print(attention.choice) 188 TableA.10:DCEdesign. Block Set Alt x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 1 1 1 2 3 3 2 3 1 1 3 1 2 2 3 1 1 2 2 3 3 2 3 2 3 2 3 3 3 1 1 2 1 2 3 3 3 1 1 3 1 2 1 1 2 1 2 2 2 3 3 3 1 3 2 3 3 2 2 3 1 3 1 2 2 2 2 2 2 2 3 1 1 2 3 1 3 2 2 2 2 2 2 3 2 1 3 2 1 2 1 4 1 1 3 1 1 2 1 3 3 3 1 3 1 1 4 2 1 3 1 1 2 2 1 1 2 3 1 3 1 5 1 2 1 3 3 3 1 2 1 2 3 2 1 1 5 2 2 1 3 3 3 3 1 1 1 1 2 2 1 6 1 2 2 1 1 1 3 1 3 3 3 1 1 1 6 2 2 2 1 1 1 1 2 1 2 1 2 3 2 1 1 1 1 1 3 1 2 1 1 2 1 3 2 2 1 2 1 1 1 3 1 1 2 2 1 1 1 1 2 2 1 2 2 1 1 2 2 2 2 2 1 2 1 2 2 2 2 2 1 1 2 3 1 1 1 1 3 1 2 3 1 2 2 3 2 3 1 1 3 3 3 3 3 2 3 2 2 2 3 2 3 3 3 1 2 2 1 2 2 4 1 2 1 2 1 2 3 2 3 1 3 1 1 2 4 2 2 1 2 1 2 2 3 2 2 1 3 3 2 5 1 1 2 2 3 3 2 3 2 3 2 2 3 2 5 2 1 2 2 3 3 1 2 1 1 3 2 1 2 6 1 1 3 3 2 1 3 3 3 2 2 2 3 2 6 2 1 3 3 2 1 2 2 3 3 1 1 3 3 1 1 2 2 3 3 2 3 2 1 3 1 2 1 3 1 2 2 2 3 3 2 2 1 2 2 3 3 2 3 2 1 2 3 1 3 1 1 2 2 3 2 1 2 3 2 2 2 3 1 3 1 3 3 1 1 1 3 1 3 3 1 2 1 1 2 2 3 1 1 1 3 2 3 Continued... 189 TableA.10:DCEdesign. Block Set Alt x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 3 3 2 2 1 1 2 2 2 3 3 1 1 3 2 3 4 1 1 1 2 3 1 2 3 3 1 2 3 3 3 4 2 1 1 2 3 1 1 1 2 3 1 3 1 3 5 1 2 2 3 3 1 1 1 2 2 2 2 1 3 5 2 2 2 3 3 1 3 2 3 3 2 1 3 3 6 1 1 1 3 2 3 1 3 3 2 3 2 3 3 6 2 1 1 3 2 3 2 2 1 3 1 2 1 4 1 1 2 1 2 2 1 3 2 3 1 1 1 2 4 1 2 2 1 2 2 1 1 1 1 3 2 2 1 4 2 1 1 2 3 3 3 1 1 3 1 2 1 1 4 2 2 1 2 3 3 3 3 3 2 2 1 3 3 4 3 1 1 2 3 1 3 2 1 3 3 2 2 3 4 3 2 1 2 3 1 3 3 2 1 2 3 1 2 4 4 1 2 1 2 2 1 2 1 2 3 2 3 2 4 4 2 2 1 2 2 1 2 2 3 2 1 2 1 4 5 1 1 1 1 3 2 1 3 1 1 3 1 3 4 5 2 1 1 1 3 2 1 1 2 1 1 3 1 4 6 1 2 3 2 3 1 1 1 3 1 1 3 3 4 6 2 2 3 2 3 1 2 2 2 3 3 2 2 5 1 1 1 3 1 3 2 1 2 1 3 2 2 2 5 1 2 1 3 1 3 2 3 2 2 1 3 3 3 5 2 1 1 2 2 2 3 2 3 1 1 1 3 3 5 2 2 1 2 2 2 3 3 1 3 2 3 1 2 5 3 1 2 3 1 1 1 2 1 2 1 2 1 1 5 3 2 2 3 1 1 1 1 3 3 2 1 2 1 5 4 1 1 2 1 3 2 3 1 3 2 3 3 1 5 4 2 1 2 1 3 2 1 3 2 1 3 1 3 5 5 1 1 1 3 3 1 3 3 2 3 2 2 3 5 5 2 1 1 3 3 1 1 2 1 1 1 3 2 5 6 1 2 3 1 3 3 2 1 3 2 1 2 2 Continued... 190 TableA.10:DCEdesign. Block Set Alt x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 5 6 2 2 3 1 3 3 2 2 1 3 2 3 1 6 1 1 1 3 1 3 2 3 3 2 2 2 2 1 6 1 2 1 3 1 3 2 1 2 1 2 3 3 2 6 2 1 2 3 1 1 3 2 3 3 1 1 2 2 6 2 2 2 3 1 1 3 2 2 2 2 2 1 3 6 3 1 2 1 1 3 2 2 2 3 2 2 3 1 6 3 2 2 1 1 3 2 3 1 1 3 3 2 3 6 4 1 1 2 2 1 1 1 2 3 2 2 3 3 6 4 2 1 2 2 1 1 2 3 1 3 1 1 1 6 5 1 1 3 3 2 3 1 1 3 3 1 1 3 6 5 2 1 3 3 2 3 1 2 2 1 3 3 2 6 6 1 1 3 1 2 1 3 2 2 1 3 2 3 6 6 2 1 3 1 2 1 2 3 3 2 2 1 1 Block=blockassignmentoftherespondentintheDCE. Set=uniqueidentierforthechoicesituationpresentedtotherespondent. Alt=alternativeinthechoiceset(1=vaccineAand2=vaccineB). x1=diseasetype(1=chronic,2=acute);x2=Diseasetransmissionmode(1=Directcontact,2=Vector-borne,3=Airborne);x3=Riskofinfection(1=20,000 outof1millionpeople,2=30outof1millionpeople,3=1outof1millionpeople); x4=Severesymptom(1=Life-threateningcomplication,2=Permanent handicap;3=Death);x5=Riskofseveresymptom(1=30outof100,2=2outof1000,3=1outof1million);x6=Modeofadministration(1=Needleinjec- tion,2=Intranasal,3=Oralsolution);x7=Efectiveness(1=70%,2=95%,3=99%);x8=Protectionduration–Immunity(1=1year,2=6years,3=Lifetime); x9=Severevaccinesideefect(1=Death,2=Severeneurologicalinjurythatcanbecomepermanent,3=Life-threateningallergicreactions);x10=Riskofsevere diseasesideefect(1=10outof1millionvaccinatedpeople,2=4outof1millionvaccinatedpeople,3=1outof1millionvaccinatedpeople);x11=Out-of-pocket cost(1=$0,2=$100,3=$700);x12=Compensationamount(1=$30,000or$1,000,000or$1,250,000;2=$40,000or$2,500,000or$5,000,000;3=$60,000or $3,500,000or$10,000,000). 191 FigureA.2:ExampleofaDCEchoiceproleusedinthesurvey. 192 A.7 AdditionalDCEresults Figure A.3: Conditional marginal efects of insurance on vaccine uptake by disease severity andvaccinecharacteristics. Note: Conditional marginal efect of insurance on the probability of vaccination at diferent levels of out-of- pocketcostofvaccine,vaccineecacy,riskofvaccinesideefectsandbydiseaseseverity. Abbreviation: VSE= vaccinesideefect 193 Figure A.4: Conditional marginal efects of insurance on vaccine uptake by vaccine out-of- pocketcostandindividualcharacteristics. Note: Conditional marginal efect of insurance on the probability of vaccination at diferent vaccine out-of-pocketpricelevelsandbyselectedindividualcharacteristics. 194 Figure A.5: Conditional marginal efects of insurance on vaccine uptake by vaccine ecacy andindividualcharacteristics. Note: Conditional marginal efect of insurance on the probability of vaccination at diferent levels of vaccine ecacyandbyselectedindividualcharacteristics. 195 FigureA.6:Conditionalmarginalefectsofinsuranceonvaccineuptakebyriskofvaccineside efectsandindividualcharacteristics. Note: Conditional marginal efect of insurance on the probability of vaccination at diferent levels of risk of vaccinesideefectsandbyselectedindividualcharacteristics. 196 TableA.11:Contributionsofthevaccineanddiseaseattributesbytreatmentstatus. Variables a Insurance Subsidy P-value b Estimate Marginalefect Estimate Marginalefect Constant -6.1891*** – -5.3642*** – 0.000 (0.3059) – (0.2975) – Vaccineattributes Intranasaladministration c -0.1917** -0.0173** -0.2410*** -0.0230*** 0.654 (0.0848) (0.0075) (0.0848) (0.0079) Oraladministration c -0.1840** -0.0167** -0.0269 -0.0028 0.186 (0.0803) (0.0072) (0.0811) (0.0084) Ecacy(%) 0.0504*** 0.0046*** 0.0463*** 0.0046*** 0.134 (0.0019) (0.0001) (0.0017) (0.0001) 6yearsimmunity d 0.6528*** 0.0572*** 0.6688*** 0.0667*** 0.889 (0.0839) (0.0085) (0.0844) (0.0097) Lifetimeimmunity d 1.0567*** 0.1095*** 0.8679*** 0.0938*** 0.114 (0.0798) (0.0100) (0.0804) (0.0099) VSE:Neurologicaldisorder e 0.0478 0.0048 -0.0819 -0.0086 0.299 (0.0902) (0.0090) (0.0917) (0.0096) VSE:Death e -0.4278*** -0.0355*** -0.3609*** -0.0341*** 0.725 (0.1220) (0.0098) (0.1209) (0.0111) VSErisk(permillion) -0.0550*** -0.0050*** -0.0147 -0.0015 0.004 (0.0095) (0.0009) (0.0095) (0.0009) Out-of-pocketcost($) -0.0019*** -0.0002*** -0.0022*** -0.0002*** 0.024 (0.0001) (0.0000) (0.0001) (0.0000) Compensationlevels Compensationamount($) 0.0000** 0.0000** -0.0000 -0.0000 0.014 (0.0000) (0.0000) (0.0000) (0.0000) Diseaseattributes Acutedisease f 0.2193*** 0.0199*** 0.1901*** 0.0189*** 0.824 (0.0669) (0.0060) (0.0665) (0.0066) Continued... 197 TableA.11:Contributionsofthevaccineanddiseaseattributesbytreatmentstatus. Variables a Insurance Subsidy P-value b Estimate Marginalefect Estimate Marginalefect Vector-bornedisease g 0.1023 0.0113 -0.1046 -0.0122 0.079 (0.0826) (0.0091) (0.0831) (0.0097) Airbornedisease g -0.6196*** -0.0518*** -0.6170*** -0.0595*** 0.996 (0.0816) (0.0074) (0.0801) (0.0082) Infectionrisk(per100,000) 0.0004*** 0.0000*** 0.0005*** 0.0000*** 0.136 (0.0000) (0.0000) (0.0000) (0.0000) SEV:Handicap h 0.2246*** 0.0207*** 0.1666** 0.0166** 0.597 (0.0783) (0.0071) (0.0779) (0.0077) SEV:Death h 0.2528*** 0.0231*** 0.1395 0.0141 0.339 (0.0875) (0.0077) (0.0874) (0.0087) SEVrisk(permillion) 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.860 (0.0000) (0.0000) (0.0000) (0.0000) Observations 8,874 8,748 AIC 6,768 6,805 BIC 7,066 7,103 Log-likelihood -3,342 -3,361 Pseudo-R 2 0.305 0.282 χ 2 2,930 2,641 P-value 0.000 0.000 Notes:Standarderrorsinparentheses;statisticalsignicance:***p<0.01,**p<0.05,*p<0.1.Allmodelsusethesampleofconsistentrespondents,asdenedinSection2.5.1, andcontrolforindividualcharacteristics.Thedependentvariableisthechoiceofvaccination.Abbreviations:VSE=vaccinesideefect;SEV=severityofdisease. a Thereferencealternativeinallmodelsis“Novaccination”. b P-valueofofthediferencebetweentheestimatesfortheinsuranceandsubsidygroups;estimatedbyinteractingtheinsurancedummieswiththepredictorsoftheoutcome variable. c Referencelevel:administrationbyinjection. d Referencelevel:1yearimmunity. e Referencelevel:anaphylaxis(life-threateningallergicreaction). f Referencelevel:chronicdisease. g Referencelevel:transmissionbydirectcontact. h Referencelevel:severecomplicationrequiringhospitalization. 198 TableA.12:Contributionsofthevaccineanddiseaseattributesbygender. Variables a Female Non-female P-value b Estimate Marginalefect Estimate Marginalefect No-vaccinationconstant -5.3192*** – -6.4387*** – 0.000 (0.2787) – (0.3345) – Insurance -0.0361 -0.0035 0.1664*** 0.0157*** 0.036 (0.0594) (0.0058) (0.0640) (0.0061) Vaccineattributes Intranasaladministration c -0.3253*** -0.0305*** -0.0964 -0.0089 0.057 (0.0819) (0.0075) (0.0880) (0.0080) Oraladministration c -0.2072*** -0.0203*** -0.0017 -0.0002 0.072 (0.0788) (0.0076) (0.0828) (0.0079) Ecacy(%) 0.0471*** 0.0046*** 0.0498*** 0.0047*** 0.310 (0.0017) (0.0001) (0.0019) (0.0002) 6yearsimmunity d 0.7549*** 0.0724*** 0.5738*** 0.0523*** 0.116 (0.0819) (0.0092) (0.0871) (0.0090) Lifetimeimmunity d 1.0274*** 0.1101*** 0.9084*** 0.0949*** 0.267 (0.0778) (0.0098) (0.0829) (0.0101) VSE:Neurologicaldisorder d 0.0446 0.0047 -0.0716 -0.0072 0.378 (0.0877) (0.0092) (0.0947) (0.0094) VSE:Death e -0.3892*** -0.0346*** -0.3918*** -0.0346*** 0.971 (0.1177) (0.0101) (0.1258) (0.0108) VSErisk(permillion) -0.0249*** -0.0024*** -0.0480*** -0.0045*** 0.085 (0.0092) (0.0009) (0.0098) (0.0010) Out-of-pocketcost($) -0.0023*** -0.0002*** -0.0018*** -0.0002*** 0.001 (0.0001) (0.0000) (0.0001) (0.0000) Compensationlevels Compensationamount($) -0.0000 -0.0000 0.0000 0.0000 0.129 (0.0000) (0.0000) (0.0000) (0.0000) Continued... 199 TableA.12:Contributionsofthevaccineanddiseaseattributesbygender. Variables a Female Non-female P-value b Estimate Marginalefect Estimate Marginalefect Diseaseattributes Acutedisease f 0.1681*** 0.0162*** 0.2505*** 0.0234*** 0.374 (0.0640) (0.0061) (0.0699) (0.0065) Vector-bornedisease g 0.0225 0.0026 -0.0337 -0.0037 0.655 (0.0807) (0.0095) (0.0852) (0.0093) Airbornedisease g -0.6793*** -0.0614*** -0.5710*** -0.0510*** 0.304 (0.0779) (0.0076) (0.0845) (0.0080) Infectionrisk(per100,000) 0.0005*** 0.0000*** 0.0004*** 0.0000*** 0.275 (0.0000) (0.0000) (0.0000) (0.0000) SEV:Handicap h 0.1406* 0.0138* 0.2639*** 0.0247*** 0.284 (0.0750) (0.0073) (0.0817) (0.0075) SEV:Death h 0.2102** 0.0201** 0.2004** 0.0192** 0.891 (0.0844) (0.0079) (0.0913) (0.0085) SEVrisk(permillion) 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.051 (0.0000) (0.0000) (0.0000) (0.0000) Observations 9,612 8,010 AIC 7,344 6,230 BIC 7,646 6,523 Log-likelihood -3,630 -3,073 Pseudo-R 2 0.295 0.292 χ 2 3,034 2,536 P-value 0.000 0.000 Notes: Standarderrorsinparentheses; statisticalsignicance: ***p<0.01, **p<0.05, *p<0.1. Allmodelsusethesampleofconsistentrespondents, asdenedinSec- tion2.5.1,andcontrolforindividualcharacteristics.Thedependentvariableisthechoiceofvaccination.Abbreviations:VSE=vaccinesideefect;SEV=severityofdisease. a Thereferencealternativeinallmodelsis“Novaccination”. b P-valueofofthediferencebetweentheestimatesforthetwogroups;estimatedbyinteractingthegen- derdummieswiththepredictorsoftheoutcomevariable. c Referencelevel:administrationbyinjection. d Referencelevel:1yearimmunity. e Referencelevel:anaphylaxis(life-threateningallergicreaction). f Referencelevel:chronicdisease. g Referencelevel:transmissionbydirectcontact. h Referencelevel:severecomplicationrequiringhospitalization. 200 References [1] Berkowitz,B.andGami,L. Ebolaspreadsslower,killsmorethanotherdiseases. The Washington Post,Oct142014. AccessedFeb15,2016. [2] Center for Disease Control and Prevention. HIV in the United States: At a glance. Available at http://www.cdc.gov/hiv/resources/factsheets/PDF/HIV_at_a_ glance.pdf,2012. AccessedJun05,2013. [3] Centers for Disease Control and Prevention. Severe Acute Respiratory Syndrome (SARS). Available athttp://www.cdc.gov/sars/about/faq.html, Jul 2 2012. Accessed Feb15,2016. [4] CentersforDiseaseControlandPrevention. Howuspread. Availableathttp://www. cdc.gov/u/about/disease/spread.htm,Aug2014. AccessedonDec15,2015. [5] Centers for Disease Control and Prevention. Middle East Respiratory Syndrome (MERS). Available at http://www.cdc.gov/coronavirus/mers/us.html, Dec 8 2015. AccessedFeb15,2016. [6] CentersforDiseaseControlandPrevention. PossibleSide-efectsfromVaccines. Avail- able at http://www.cdc.gov/vaccines/vac-gen/side-efects.htm, Nov 5 2015. Accessed Feb15,2016. [7] CentersforDiseaseControlandPrevention. ReviewofHuman-to-HumanTransmis- sionofEbolaVirus.Availableathttp://www.cdc.gov/vhf/ebola/transmission/human- transmission.html,Oct12015. AccessedFeb15,2016. [8] CentersforDiseaseControlandPrevention.ReportedTuberculosisintheUnitedStates, 2014.Atlanta,GA:USDepartmentofHealthandHumanServices,CDC,October2015. AccessedonDecember15,2015. 201 [9] Centers for Disease Control and Prevention. Mumps. Available at http://www.cdc. gov/mumps/index.html,Feb92016. AccessedFeb15,2016. [10] CentersforDiseaseControlandPrevention.MeaslesCasesandOutbreaks.Availableat http://www.cdc.gov/measles/cases-outbreaks.html,Feb102016. AccessedFeb15,2016. [11] Clark,M.D.etal. DiscreteChoiceExperimentsinHealthEconomics:AReviewofthe Literature. Pharmacoeconomics,32(9):883–902,Sep2014. [12] Cook, D.R. and Nachtsheim, C.J. Comparison of algorithms for constructing exact D-optimaldesigns. Technometrics,22(3),1980. [13] Cook,K.M.andEvans,G. TheNationalVaccineInjuryCompensationProgram. Pedi- atrics,127(Suppl1):S74–7,2011. [14] Davis,W.N.Nolongerimmune.AmericanBarAssociationJournal,92(7):19–43,2006. [15] deBekker-Grob,E.W.,Ryan,M.andGerard,K. Discretechoiceexperimentsinhealth economics:Areviewoftheliterature. Health Economics,21(2):145–72,Feb2012. [16] Douclef, M. The Ebola Outbreak: What’s My Risk Of Catching Ebola? National Public Radio,Oct252014. [17] Dye,C.etal. Globalburdenoftuberculosis:estimatedincidence,prevalence,andmor- talitybycountry. Journal of the American Medical Association,282(7):677–86,1999. [18] Fedorov,V.V. Theory of Optimal Experiments Designs. Elsevier,1972. [19] Fielder,J.etal. Ahightuberculosiscase-fatalityrateinasettingofefectivetuberculosis control:implicationsforacceptabletreatmentsuccessrates. The International Journal of Tuberculosis and Lung Disease,6(12):1114–17,2002. 202 [20] Gennarelli,T.andWodzin,E. The Abbreviated Injury Scale 2005 – Update 2008. Bar- rington,IL:AssociationfortheAdvancementofAutomotiveMedicine,2008. [21] Hanna,D.B.etal.DisparitiesamongUSstatesinHIV-relatedmortalityinpersonswith HIVinfection,2001-2007. AIDS,26(1):95–103,Jan22012. [22] Heatwole,N.TheValueofNon-FatalInjuriesofVariousSeveritiesforUseinTerrorism- Related Policy Analyses. Available at https://benetcostanalysis.org/publicFiles/ Heatwole-VSI.pdf,Mar2014. AccessedFeb15,2016. [23] Huber,J.andZwerina,K. Theimportanceofutilitybalanceinecientchoicedesigns. Journal of Marketing Research,33(3):307–17,1996. [24] Johnson, R.F. et al. Constructing Experimental Designs for Discrete-Choice Experi- ments: ReportoftheISPORConjointAnalysisExperimentalDesignGoodResearch PracticesTaskForce. Value in Health,16(1):3âĂŞ–13,2013. [25] Kahneman, D. and Tversky, A. Prospect Theory - Analysis of Decision under Risk. Econometrica,47(2):263–91,1979. [26] Kelly,H.andValenciano,M. Estimatingtheefectofinuenzavaccines. Lancet Infec- tious Diseases,12(1):5–6,Jan2012. [27] Klaus, Z., Huber, J.andKuhfeld, W.F. A General Method for Constructing Efficient Choice Designs. SASInstituteInc.,2010. AccessedFeb16,2016. [28] Lancaster,K.J. Anewapproachtoconsumertheory. Journal of Political Economy,74 (2):132–57,1966. [29] Larson, H.J. et al. Understanding vaccine hesitancy around vaccines and vaccination fromaglobalperspective: Asystematicreviewofpublishedliterature,2007-2012. Vac- cine,32(19):2150–9,Apr172014. 203 [30] Louviere,J.J.,Flynn,T.N.andCarson,R.T.DiscreteChoiceExperimentsAreNotCon- jointAnalysis. Journal of Choice Modelling,3(3):57–72,2010. [31] Osterholm, M.T. et al. Ecacy and efectiveness of inuenza vaccines: A systematic reviewandmeta-analysis. Lancet Infectious Disease,12(1):36–44,Jan2012. [32] Pear, R. Threats and responses: Legal risks; for victims of vaccine, win- ning case will be hard. The New York Times, Dec 14 2002. Available at http://www.nytimes.com/2002/12/14/us/threats-responses-legal-risks-for-victims- vaccine-winning-case-will-be-hard.html. [33] Petrucelli, E., States, J.D. and Hames, L.N. The abbreviated injury scale: Evolution, usageandfutureadaptability. Accident Analysis & Prevention,13(1):29–35,Mar1981. [34] Russell, R. et al. Diferences in mortality predictions between Injury Severity Score triplets:Asignicantaw. Journal of Trauma,56(6):1321–4,Jun2004. [35] Sadique,M.Z.etal.Theefectofperceivedrisksonthedemandforvaccination:Results fromadiscretechoiceexperiment. PLoS One,8(2),Feb82013. [36] SARS Investigative Team. Revised US Surveillance Case Denition for Severe Acute RespiratorySyndrome(SARS)andUpdateonSARSCases—UnitedStatesandWorld- wide,December2003. Morbidity and Mortality Weekly Report,52(49):1202–6,Dec12 2003. [37] Segui-Gomez,M.andLopez-Valdes,F.J. InjurySeverityScaling. InLi,G.andBaker, S.P., editors, Injury Research: Theories, Methods, and Approaches, chapter 14, pages 281–95.NewYork,NY:Springer,2012. [38] Straetemans,M.etal. Assessingtuberculosiscasefatalityratio: Ameta-analysis. PLoS One,6(6):e20755,2011. 204 [39] Train,K. Discretechoicemethodswithsimulation.Cambridge,UK:CambridgeUniver- sityPress,2edition,2009. [40] UnitedStatesCourtsofFederalClaims.USCourtofFederalClaimsOpinions.Available athttp://www.uscfc.uscourts.gov,Feb102016. AccessedFeb15,2016. [41] UnitedStatesGovernmentAccountabilityOce.VaccineInjuryCompensation:Most Claims Took Multiple Years and Many Were Settled through Negotiation. Technical report,Nov2014. Availableathttp://www.gao.gov/assets/670/667136.pdf. [42] US Department of Health & Human Services. AIDS.gov. Available at https://www.aids.gov/hiv-aids-basics/just-diagnosed-with-hiv-aids/understand- your-test-results/cd4-count/index.html,2010. AccessedJune05,2013. [43] USDepartmentofHealthandHumanServices. VaccinesLicensedforImmunization and Distribution in the US with Supporting Documents. Available at http://www. vaccines.gov/more_info/approvals/index.html,Apr162015. AccessedFeb15,2016. [44] USDepartmentofTransportation.GuidanceonTreatmentoftheEconomicValueofa StatisticalLife. Availableathttps://www.transportation.gov/sites/dot.gov/les/docs/ VSL2015_0.pdf,Jun172015. AccessedFeb4,2016. [45] Willis,H.H.andLaTourrette,T.Usingprobabilisticterrorismriskmodelingforregula- torybenet-costanalysis:applicationtotheWesternhemispheretravelinitiativeinthe landenvironment. Risk Analysis,28(2):325–39,Apr2008. [46] Wilson, K. and Keelan, J. The case for a vaccine injury compensation program for Canada. Canadian Journal of Public Health,103(2):122–4,Mar-Apr2012. [47] WorldHealthOrganization. Consensusdocumentontheepidemiologyofsevereacute respiratorysyndrome(SARS).Technicalreport,WHO,DepartmentofCommunicable 205 Disease Surveillance and Response, 2003. Available at http://www.who.int/csr/sars/ en/WHOconsensus.pdf. [48] WorldHealthOrganization.Internationaltravelandhealth:Rubella.Availableathttp: //www.who.int/ith/diseases/rubella/en/,Feb2016. AccessedFeb15,2016. 206 AppendixB HIVModelingApproach:Mathematical Details B.1 Introduction WeconstructedanHIVepidemicmodelfortheLosAngeles(LA)county.Chapter3presented theresultsfromthesimulationofthemodelforthemenwhohavesexwithmen(MSM)pop- ulation.However,withappropriatedata,themethodologiesandlogicunderlyingthemodel canbeeasilyadaptedtotheheterosexualpopulationandothergroupsandpopulations.Inthis appendix, wepresenttechnicaldetailsfortheMSMmodelanddescribethecomputational andmathematicalmethodologiesusedtoobtainparameterestimatesaswellastheresultspre- sentedinChapter3. Therestofthisappendixproceedsasfollows: InSectionB.2, wepresentourmathematical modelanddescribehowthemodelsimulatesdiseaseprogression, diseasetransmission, and the dynamics of the emergence of acquired and transmitted multi drug-resistant HIV in a population.Thissectionalsoformulatestheordinarydiferentialequations(ODEs)describ- ingthemean-eldpopulationdynamics. SectionB.3discussesthederivationsoftheparam- eters used in the model. In Section B.4 we discuss the sampling methodology used to cali- brateourmodel.InSectionB.5wediscussthedatausedforestimatingthepopulationineach compartmentatthestartofthesimulation(initialconditions).InSectionB.6wediscussthe modelcalibrationwhichusesLosAngelesCountyprevalencedatainthediferentstagesof HIV/AIDSfrom2000to2009.Inadditiontothemodelcalibrationweshowasimplesensi- tivityanalysisoftheinputparameters. InSectionB.7wediscusshowwemodifyourmodel 207 toinvestigateefectsofthetest-and-treatpolicywherebyHIVtestingratesareincreasedand newlydiagnosedinfectedindividualsareputonHAARTtreatmentsoonafter.InSectionB.8 wediscussthemethodologyweusetodoasensitivityanalysisofourtest-and-treatsimulation. Finally,inSectionB.9wepresentstheresultsofourmodelwithoutallowingforearlytreat- menttoprovideinsightintowhatshareofthebenetfromtest-and-treatiscausedbyearly treatmentandincreasedtesting,respectively. B.2 HIVepidemicsimulationmodelstructure OursimulationmodelfollowsthegeneralapproachusedinpreviousHIVmodeling[6,10, 18,35,37,43].Inparticular,wecloselyfollowthemodelingapproachtakenbyDr.Blowerand herresearchteam[37].WeconsideracompartmentalHIVmodelwherebyindividualsinour populationaredividedintoeight(8)keyHIVinfectionstatuses:susceptibleorHIVnegative (S),infectedintheprimarystageofinfection(P),infectedandunawareofbeinginfected(I), diagnosedinfectedbutthatarenottreatment-eligible(J),infectedandtreatment-eligible(E), thosethatarecurrentlytreated(T),thosethatprogresstoAIDSwithouttreatment(A)and thosethathaveAIDSbutarecurrentlytreated(T A ). Inourmodel,treatmenteligibleindi- viduals(i.e.,thoseinstatesEandT)arethoseadultsthathavemetstandardCDCguidelines fortreatmentwithHAART.Theseindividualsincludethosethathaveadvanceddiseaseand havetheirCD4cellcountlessthan350cellspermicro-liter.Theyalsoincludethosethat,due tobeingontreatmentinthepast,havemanagedtoincreasetheirCD4countcellstemporarily above350cellspermicro-liter. WeuseaCD4cellcountof350asourcutofsincethisisthe guidelineforthemajorityoftheyearsonwhichwecalibrateourmodel.Outoftheseeight(8) states,seven(7)representHIVpositivestatuses(allbutS).Weconsiderpopulationsinfected withadrugsensitiveHIVstrain(subscripts)andthoseinfectedwiththemulti-drug-resistant (MDR)HIVstrain(subscriptr).Thus,ourmodelismadeoffteen(15)compartments,the 208 susceptibleorHIVnegative,sevendrugsensitiveHIVinfectedpopulationsandsevendrug resistant HIV infected populations. The compartmental model follows the dynamics illus- tratedinChapter3,Figure3.1. TheillustrateddynamicsshownaboveareintegratedviathefollowingsetofOrdinaryDifer- entialEquations(ODEs): ˙ S = π− (μ +λ s +λ r )S (B.1) ˙ P s = λ s S− (μ +ρ)P s (B.2) dI s dt = ρP s − (μ +χ s +ω +ω A )I s (B.3) ˙ J s = ωI s − (μ +ν s )J s (B.4) ˙ E s = χ s I s +ν s J s +gT s +qE r − (μ +γ Es +σ)E s (B.5) ˙ T s = σE s − (μ +γ Ts +g +r)T s (B.6) ˙ A s = ω A I +γ Es E s +g A T As +qA r − (μ +γ As +σ A )A s (B.7) ˙ T As = γ Ts T s +σ A A s − (μ +γ TAs +g A +r)T As (B.8) ˙ P r = λ r S− (μ +ρ)P r (B.9) ˙ I r = ρP r − (μ +χ r +ω +ω A )I r (B.10) ˙ J r = ωI r − (μ +ν r )J r (B.11) ˙ E r = χ r I r +ν r J r +gT r − (μ +γ Er +σ +q)E r (B.12) ˙ T r = σE r +rT s − (μ +γ Tr +g)T r (B.13) ˙ A r = ω A I r +γ Er E r +g A T Ar − (μ +γ Ar +σ A +q)A r (B.14) ˙ T Ar = γ Tr T r +σ A A r +rT As − (μ +γ TAr +g A )T Ar (B.15) 209 B.2.1 Diseaseprogression Thetransitionratesinourmodelrepresenttheinverseofthedurationinyearsspentineach compartment. Peopleenterthemodelintothesusceptiblecompartmentatrateπ,givenby the yearly number of individuals that become sexually active (apart fromπ all rates in our modelaregivenpercapita). Thepopulationcompartmentcontractionduetobackground mortality is characterized by the per capita rateμ. The population in the susceptible com- partmentbecomesinfectedwithHIVatrateλ s andλ r ,dependingonwhethertheybecome infectedwithadrugsensitiveordrugresistantstrain,atwhichpointtheyentertheprimary phaseofinfection(P). Newlyinfectedindividualsintheprimaryphasearecharacterizedby largetransmissibilities[22].However,thisisashortphaseandtheyquicklytransitiontostage I atrateρ, atwhichpointtheyarestillunawareoftheirHIV+status. Infectedindividuals thatarediagnosedwithHIV(becomeaware)enterstagesJ andE atratesω andχrespec- tivelydependingonwhethertheirCD4cellcountisabove350(stageJ)orbelow350(stage E). Those that are aware of being infected but not yet treatment eligible (J) transition to stageE atrateν. Theratesχ,ω,andν areintrinsicallyrelated. Theparameterχessentially representsthebiologicalrateoftheCD4cellcountdroppingbelow350(transitioningtostage E)afterreachingstageI,whereasωandνrepresentratesthatarebasedonpolicy.Theparam- eterωistherateatwhichpeopleinstageI arediagnosedwithHIV,whichchangesbasedon theintensityoftestingprograms,andνistherateatwhichindividuals’CD4cellcountsdrop below350aftertheyarediagnosed.Ifpeoplearediagnosedwithgreaterfrequency,thenωwill belarger(thedurationinIwillbeshorter)andpeoplewillenterstageJwithahigherCD4cell count.ThismeansitwilltakelongertoprogresstostageEfromstageJ,thusνwillbesmaller (thedurationinJ willbelonger).However,thesamebiologicalprocessoccurswhenanindi- vidualpassesthroughω andν aswhenanindividualpassesthroughχ. Sincetheinverseof 210 eachrateisthetimedurationinyearsineachcompartment,thefollowingrelationshipmust hold. 1/χ = 1/ω + 1/ν. (B.16) It is also possible to progress from stage I directly to stage A if an individual that is unaware of her/his status progresses to AIDS prior to being diagnosed. This process occurs at rate ω A . This rate can be quite small as it is unlikely that individuals rst become aware of their status after they have developed AIDS. Once an individual is in stageE, he/she can initiate antiretroviral treatment and progress to stageT at rateσ. However, due to a variety of factors, including poor access to medical care, many treatment-eligible individuals do not go on treatment, in which case they progress to AIDS (stage A) at rate γ E . We assume that progression to AIDS is an irre- versible process and that once adults have AIDS they have added mortality rates γ TA and γ A depending on whether or not they are on treatment (T A or A). If an individ- ual initiates treatment and enters stage T, he/she can either progress to AIDS while on treatment (stageT A ) at rateγ T , or quit taking treatment and retreat to stageE at rateg. OurmodeltracksbothacquiredandtransmittedMDR[10,36,43].Individualsontreatment may acquire MDR at a rate r. This rate depends upon the average adherence (g), viral replication rate, and the specic mutations that arise [9]. Adults that acquire MDR and gooftreatmentmay, givenenoughtime, revertbacktohavingjustthedrugsensitiveHIV strain. This reversion occurs at a rate q and applies to those adults with HIV and those previously on treatment who have developed AIDS, and are currently untreated. Finally, individuals that acquire MDR can then potentially transmit the drug resistant HIV strain tothesusceptibleindividuals. Thesenewlyinfectedadultswouldthenprogressthroughthe correspondingMDRinfectionsstagescharacterizedbydiferentprogressionrates. 211 B.2.2 DiseaseTransmission DrugsensitiveanddrugresistantHIVincidenceismodeledbythetransitionofindividuals fromthesusceptiblepopulationtotheprimaryinfectionstates(P s andP r )andoccursata rateλ s andλ r respectively.Theseareknownastheforceofinfectionandaregivenby λ s = X Xs c X ˜ β Xs X s /N =c X Xs β Xs X s /N, (B.17) λ r = X Xr c X ˜ β Xr X r /N =c X Xr β Xr X r /N, (B.18) whereN representsthetotalpopulation(i.e.,thesumofthesusceptibleandallinfected,both drugsensitiveandresistantstates).Thecontactratec X representstheaveragenumberofnew sexpartnersperyearwithwhichatypicalindividualinthepopulationbelongingtodisease stageX engagesinunprotectedsex.Thisisgivenby c X = cη X , (B.19) whereη X isascalingfactorthatchangesc X dependingontheinfectionstatusX. Notethat inEquationsB.17andB.18wesplitthefactorc X intocη X andredenetheparameterβ X as beingidenticaltoη X ˜ β X .Theinfectiontransmissibilities,β X areprobabilitiesperyearandper HIVserodiscordantpartnershipofinfection. Theseper-partnershiptransmissibilities(β X ) arerelatedtotheper-acttransmissibilities(α X )whicharedeterminedbyanalyzingviralload datadescribedinSectionB.3.1.1.FollowingtheapproachtakenbySmithetal.[36],wedeter- minetheperactprobabilityoftransmissionfromtherelationshipbetweentheviralloadv X andbaselinevaluesofper-acttransmissibilitiesα Base X andviralloadv Base X (0.0018and12,500 respectively)usingtherelationshipfoundbyQuinnetal.[32]: α X = 2.45 log 10 v X /v Base X α Base X , (B.20) 212 The per-partnership probabilities of transmission are then estimated based on the per act probabilityoftransmissionbyusingabinomialmodel β X =η X 1− (1−α X ) ζ f X ≈η X ζ f X α X , (B.21) wheref representstheaveragenumberofunprotectedsexactsanadulthasper-partnership, andζ X is a scaling factor that modies this number depending on the infection state. The expression forβ X can also be approximated as illustrated in Equation B.21 by a rst order Taylorexpansion,sincethevaluesofα X aresmall. RelativelylittleisknownaboutthetnessofresistantstrainsofHIVinvivo.However,invitro experimentshaveshownthatthereplicationratesofresistantstrainsaregenerallylessthanthat ofwild-typestrains[11,27,46]. Weusetheseresultstocomputemodiedtransmissibilities fordrugresistantstrainsgivenbyEquationB.22,whereh r isaproportionalityconstant[49]: β Xr = h r β Xs . (B.22) B.2.3 Modeloutput ThemodeltracksthedynamicsofallfteenpopulationcompartmentsandthustracksHIV andAIDSprevalences. AsshownbyEquationsB.1-B.15,theprevalencesatthestartofthe integrationspecifythefuturedynamicsoftheepidemic.Acomputationalintegrationofthese ODEsbystandardnumericaltechniques[44]providesthedynamics.Inadditiontothepreva- lences, we track a few cumulative incidences: Total number of infected with drug sensitive strain (I s ); Total number of infected with MDR strain (I r ); Total number of HIV cases 213 thatprogresstoAIDS(A);Totalnumberofinfectedindividualsthatgetdiagnosedwithhav- ingHIV(H );Totalnumberoftimesindividualsentertreatment(T );andTotalnumberof deaths(D).ThesetofODEsdescribingthedynamicsaregivenby: dI s dt = λ s S, (B.23) dI r dt = λ r S, (B.24) dA dt = ω A (I s +I r ) +γ Es E s +γ Er E r +γ Ts T s +γ Tr T r , (B.25) dH dt = (ω +ω A +χ)(I s +I r ), (B.26) dT dt = σ(E s +E r ) +σ A (A s +A r )−g(T s +T r ) +g A (T As +T Ar ), (B.27) dD dt = γ As A s +γ TAs T As +γ Ar A r +γ TAr T Ar . (B.28) Note that cumulative incidences do not afect the dynamics of the model. These are all set tozeroatthebeginningoftheintegration. Takingthediferencebetweensuccessiveyearsof thesecumulativeincidencesprovidesthechangewithinayear:forexamplethetotalnumber ofdeathsduetoAIDSbetweenyeartandt + 1isgivenby ΔD(t) =D(t + 1)−D(t). B.3 Parameterranges Weestimateuncertaintyrangesforourmodelparametersbasedonliteraturereviews,expert opinions,andoursubjectiveassessmentoftherange.Parameterrangesareusedforrandomly samplingoveruncertaintyrangesusingaLatinHypercubeSampling(LHS)design[20,21] described in Section B.4. Using the LHS we carry out 2 sets of 500,000 simulations with independently sampled combinations of parameter values. The estimation methodologies forkeyparameterrangesaredescribedbelow. 214 B.3.1 Perpartnershiptransmissibilities,β Asnotedabove,theperpartnershipprobabilityoftransmissionparameter(β)isacombina- tionofeachcompartment’srespectivetransmissionprobability(α),scalingfactor(η X ),and numberofsexactsperpartnerperyear(ζ f X )asshownbyEquationB.21.Belowwedetailhow eachofthesethreeparametersarecalculated. B.3.1.1 Infectiousnesspersexualact,α TheinfectiousnesspersexualactisrepresentedbyαinEquationB.21.Aspreviouslydiscussed, tocalculateα,wefollowSmithetal.[36]andassumeabaselineviralloadof12,500(η X (Base) ) with a corresponding baseline per-act infectivity of 0.0018 (α(w)) [32]. Viral load infor- mationforallHIVstagesofourmodelexceptforstageP wasobtainedfromtheHIVCost andServicesUtilizationStudy(HCSUS)data[33].TheHCSUSdatasetincludesanarrayof informationonHIVpositiveindividualsincare,includingviralload,CD4cellcount,treat- mentstatus, sexualactsperyear, condomuse, andnumberofsexualpartners. Tocalculate viralloads,wediferentiatethestagesofourmodelbasedonrespectiveCD4cellcount,treat- mentstatus,orAIDSstatus,andcalculatemeanviralloadsseparatelyforeachstage. People in the primary stage, stageP, are unaware of their status and therefore do not exist in the HCSUS data. Estimates of theα parameters in this stage were retrieved from Smith et al. [36].AlthoughindividualsinstageIarealsounawareoftheirinfectiousstatus,thebiological phaseofthediseaseandthereforetheassumedCD4cellcountforthisisidenticaltostageJ. WethereforeusethesameviralloadandsubsequentlythesamealphavalueforstagesI and J. Lowerandupperboundsforthesestageswerecalculatedbytakingthelowerandupper endsofthe95%condenceintervaloftheviralloadestimatefoundintheHCSUSdata.Viral loadsandalphasforpeoplewithAIDS,stageA,areassumedtobethesameasintheprimary stage[17]. 215 Wecalculateviralloadsforpeopleontreatment(StageT)andwithAIDSandontreatment (StageT A )usingtheclinicaltrialsestimatesfromGulicketal.[19],whoreportedadecreasein viralloadof 1.77 log 10 copiespermilliliterasaresultofHAART.Thisindicatesadecrease inviralloadfromstageEtoT andfromstageAtoT A of 1.77 log 10 copiespermilliliter.To calculatethis,weconvertthestageEandstageAviralloadsinto log 10 copiesandcalculatethe necessaryviralloadsforstagesT andT A toreachadecreaseof 1.77 log 10 copiespermilliliter, asshownbelowinEquationB.29andEquationB.30: log 10 (viralloadStageE)− log 10 (viralloadstageT ) = 1.77 (B.29) log 10 (viralloadStageA)− log 10 (viralloadstageTA) = 1.77 (B.30) To generate upper and lower bounds of the ranges, we repeat the above process, using the upperandlowerendsoftheGulicketal.[19]condenceinterval,whichare 1.88 log 10 copies (lowerbound)and 1.66 log 10 copies(upperbound). Rangesfortheαineachstagearedis- playedinTableB.1. TableB.1:Rangesoftheαparametervaluesbydiseasestage. Stage Low Mode High P 0.0031 0.0108 0.0185 I 0.0020 0.0038 0.0060 J 0.0020 0.0038 0.0060 E 0.0029 0.0044 0.0054 A 0.0031 0.0108 0.0185 T 0.0005 0.0009 0.0012 TA 0.0009 0.0011 0.0014 216 B.3.1.2 Scalingfactors,η X Wecalculateascalingfactor,η X ,whichrepresentstheaveragenumberofsexualpartnersper year in the population in each disease stage, relative to the susceptible population. To esti- mateη X werstestimatethenumberofpartnersfortheHIVdiagnosedpopulationusing theHCSUSdata.ForstagesJ,EandA,weusetheresponsestoaquestionfromtheHCSUS survey,whichasksparticipantsabouttheirnumberofsexualpartnersinthepreviousyear.We taketheaveragewithineachstage,leavingoutoutliersthataregreaterthanorequalto100, therebyexcluding7observations. Forthetreatmentstages(T andT A )weusetheinference from Lakdawallaetal.[26],andassumethatnumberofpartnersincreasestothesamelevel asstageI aftertreatmentinitiation. Next,wecalculatethenumberofpartnersfortheMSM susceptiblepopulationusingresponsestoaGeneralSocialSurvey(GSS)[38]questionwhich askseachparticipantaboutthenumberofpartnerstheyhavehadinthelastyear.Inorderto estimatethescalingfactor,wedividethenumberofpartnersdeterminedforeachstagebythe numberofpartnersforthesusceptiblepopulations. Thescalingfactorη X fortheinfection stagesP andI areη P =η I = 1sincetheseindividualsareunawareoftheirinfectiousstatus, andarethusassumedtobehaveasthesusceptiblepopulation. Finally,lowerboundsforthe scalingfactorwerecalculatedusingthelowerendsofthe95%condenceintervalaroundthe numberofpartnersestimates;theupperboundswereallsettoone. 217 TableB.2:Rangesofthescalingfactorparameter(η X ),bydiseasestage. Stage Lower Mode Upper P 1 1 1 I 1 1 1 J 0.76 0.83 1 E 0.76 0.83 1 A 0.70 0.76 1 TA 1 1 1 T 1 1 1 B.3.1.3 Numberofunprotectedsexactsperpartner,ζ f X WecalculatenumberofunprotectedsexactsperpartnershipforstagesP andIusingestimates fromMosherandA.Chandro[30]andtheGSSdata[38].WeusetheMosherandA.Chandro [30]estimatestocreatethenumberofsexactsperyear.Weuseresponsestoaquestioninthe GSSsurveypertainingtothenumberofpartnersinthelastyear,andcondomuseinthemost recentsexualencounter. ThenumberofsexualactsreportedbyMosherandA.Chandro[30]wasmultipliedbythe percentoftimesacondomwasnotusedtodeterminetheannualnumberofriskysexacts. This number was then divided by the number of partners to estimate the total number of riskyactsperpartnershipperyear.Thelowerandupperrangeswerecalculatedusingthelower andupperendsofa95%condenceintervalforallthreeparameters(condomuse,numberof partners,andnumberofsexacts)andarepresentedinTableB.3. ForstageJ,weusetheresultsofMarksetal.[28],whoconductedameta-analysisoftheefect ofHIVtestingonsexualbehavior,andfoundriskysexactstodecreaseby68%afteranHIV 218 positivepersonbecameawareofher/hisHIVpositivestatus[28]. Toimplementthisefect, weusethevalueforf thatensuresthattheproductoff andthescalingfactorforthenumber ofpartnersinstageJ is68%lowerthantheequivalentestimateinstageI.Fortheupperand lowerboundsweusethe95%condenceintervalof56%and76%reductions,respectivelyin Marksetal.[28].WeassumestagesEandJ tohavesimilarrisksinsexualacts. ForstageA, weassumeanadditional10%reductionintheriskpersexualact, overtheesti- matedreductionsinstagesJandE.Thisextrareductionstemsfromthefactthatthereported numberofpartnersforpeoplewithAIDSintheHCSUSdatais10%lowerthanthatreported byindividualsinstageE.Weestimatedtheupperandlowerboundsbasedon10%reductions oftheupperandlowerboundsforstageJ. For stagesT andT A , we used the result from Lakdawalla et al. [26] who estimated a 133% increaseinnumberofpartnersasaresultoftreatment.Tocapturethisefectwemultiplythe productoff andthescalingfactorforstagesE andAby 2.33(a133%increase)andusethe f valuethatproducesthissameproductintheT andT A stages,respectively. Fortheupper andlowerboundswesubstitutethe133%increasefromE toT fortheupperandlowerends ofthecondenceintervalof 1.18− 1.48estimatedbyLakdawallaetal.[26]. 219 TableB.3:Unprotectedsexactsperpartnerperyear,ζ f X ,bystage Stage Lower Mode Upper P 14.84 16.93 19.44 I 14.84 16.93 19.44 J 4.67 6.53 8.55 E 04.67 6.53 8.55 A 4.60 6.43 7.70 TA 6.99 11.36 19.09 T 7.76 12.62 21.22 Therespectivelowerandupperboundsofalphas(α), scalingfactors(η X ), andsexactsper partnerperyear(ζ f X )arecollectivelyinsertedintoEquationB.21foreachstagetoproducethe betaranges.ThenalrangeforeachstageisshowninTableB.4. TableB.4:Derivedperpartnershiptransmissibilities. Parameter Mode Lower Upper PDF Reference β P 0.1679 0.0225 0.6089 PERT Derivation β I 0.0632 0.0146 0.2208 PERT Derivation β J 0.0207 0.0036 0.1004 PERT Derivation β E 0.0237 0.0051 0.0906 PERT Derivation β A 0.0516 0.0050 0.2678 PERT Derivation β TA 0.0129 0.0002 0.0808 PERT Derivation β T 0.0113 0.0001 0.0511 PERT Derivation Toobtaintheequivalentdrugresistanttransmissibilities,wemultiplythevaluesandranges inTableB.4bytheparameterh r whosevalueandrangeisgiveninTableB.5. 220 TableB.5:MDRtransmissibilitymultiplicativefactor. Parameter Mode Lower Upper PDF Reference Table h r 0.1000 0.1000 0.2000 PERT [36] S3 B.3.1.4 Contactrate,c Alongwiththeper-partnershiptransmissibilityβ,diseasetransmissionisalsodrivenbythe contactratec,thenumberofsexualpartnersperyearinthesusceptiblepopulation.Therange forthecontactrateisshowninTableB.6. Theparameterctakesamedianvalueof 2.5with arangeof 0.5to 12.0(seeTableB.6). ThismeansthatonaverageanMSMhas 2.5newsex partners each year. These values were chosen based on multiple data sources for the MSM population. TableB.6:Parametersspecifyingcontactrange. Parameter Mode Lower Upper Distribution Source c msm 2.5 0.5 12.0 PERT [23–25,31,36,37,48] B.3.2 Populationparameters Thetwoparametersthatcontrolthepopulationofthemodelaretheinowofnewsexually active adults, π, and the mortality rate, μ. The inow population is comprised of 14 year oldsthatwillenterthepopulationassexuallyactive. WeusetheRANDCaliforniaStatistics PopulationandDemographic(RCSPD)datatoestimatethisparameter.Duetosmallsample variability in the RCSPD, the data are only available in ve-year age increments. Thus, to obtainjustthe14yearoldpopulationweusethe10to14yearoldagegroupanddividethe populationbyve. Thisinowismultipliedbytheproportionofthepopulationestimated 221 tobeMSM(seeTableB.9)toobtaintheinowofnewsexuallyactiveMSMs. Theaverage backgroundmortalityrateμiscalculatedasaweightedsumofthemortalityratescollected fromtheUScensusdata.ValuesfortheseparametersareshowninTableB.7. TableB.7:Parametersspecifyingthepopulationofthemodel. Parameter Mode Lower Upper Distribution Source π 2,200,000 2,178,000 242,000 Uniform [41] μ 0.001 0.00099 0.0011 Uniform [41] B.3.3 Diseaseprogressionparameters Progression rates through the diferent HIV stages are estimated based on inverting the expected duration that typical adults spend in the diferent stages before transitioning to a subsequentstage. Thesevalueswereestimatedfromtherelevantliteratureandexpertopin- ion. Uncertainty range for these parameters along with sources for estimation are given in TableB.8. 222 TableB.8:Transitionrates Parameter Mode Lower Upper PDF Reference Table ρ 10.4286 6.6364 24.3333 PERT [36] S10 ω hetero 0.2326 0.1709 1.3333 PERT Derivation ω msm 0.3030 0.1709 1.3333 PERT [36] S10 ω A 0.0550 0.0289 0.1503 PERT [39]andDerivation Table2 χ 0.1667 0.1429 0.2000 PERT [36] S10 ν hetero 0.5882 0.2353 0.8696 PERT Derivation ν msm 0.3704 0.2353 0.8696 PERT [36] S10 γ Es 0.3509 0.2632 0.5263 Uniform [29,45]andDerivation γ Er 0.7018 0.5263 1.0526 Uniform ExpertOpinion γ Ts 0.0606 0.0455 0.0909 Uniform [36] S10 γ Tr 0.1714 0.1000 0.6000 Uniform [7,10] Table1 σ 0.6486 0.3333 12.0000 Uniform ExpertOpinion σ A 1.3001 0.6667 26.0714 Uniform ExpertOpinion g 0.0813 0.0000 0.1625 Uniform [36] S4 g A 0.0406 0.0000 0.0813 Uniform ExpertOpinion γ As 0.9600 0.5000 12.0000 Uniform [29]andExpertOpinion γ Ar 1.9200 1.0000 24.0000 Uniform ExpertOpinion γ TAs 0.1333 0.0769 0.5000 Uniform [7,10,34,36,50] S4;Table1 γ TAr 0.2667 0.1538 1.0000 Uniform ExpertOpinion r 0.0310 0.0000 0.0620 Uniform [36,42] S10 q 0.0155 0.0000 0.0310 Uniform [12]andExpertOpinion 223 B.4 Samplingstyle WeuseaLatinHypercubeSampling(LHS)designtocalibrateourmodel.Foreachparameter, wespecifyaprobabilitydistributionwhensamplingwithinitsuncertaintyrangeofvalues.We assumetwodiferentprobabilitydistributions:uniformandbetaPERT(ProgramEvaluation andReviewTechnique). Forparametersforwhichthereislargeuncertaintyinthevalueof theparameterweuseauniformdistribution. Typically,samplingfromthebetadistribution requiresminimumandmaximumvalues(x min andx max )andtwoshapeparameters,v and w. ThebetaPERTdistributionusesthemodeormostlikelyparameter(x mode )togenerate theshapeparametersv andw ofabetadistribution. Anadditionalscaleparameterλscales theheightofthedistribution; thedefaultvalueforthisparameterisfour(4). InthePERT distribution,themeanμiscalculatedas μ = x min +x max +λx mode λ + 2 , (B.31) andisusedtocalculatethevandwshapeparameters v = (μ−x min )(2x mode −x min −x max ) (x mode −μ)(x max −x min , (B.32) w = (x max −μ)(2x mode −x min −x max ) (x mode −μ)(x max −x min . (B.33) B.5 Initialconditionsandbaselinedata To calibrate our model we identify simulations that match the HIV/AIDS epidemic from 2000-2009forMSMsinLAC.Inordertohaveabaselinereference, weuseLACdatafrom the years 2000 to 2009 to populate each compartment of our model. The populations for 224 the rst reference year are the initial conditions for the model simulation. The process for estimatingthereferencedataandtheinitialconditionsaredescribedbelow. B.5.1 MSMpopulation WeconstructthedemographicvariablesusingdatafromtheRCSPDdatabase[40]. These dataarebasedonannualestimatesoftheresidentpopulationineachcountyintheUnited States based on the 2000 US Census, and are available at the Metropolitan Statistical Area (MSA) level. They have been constructed by the Population Estimates Program of the US CensusBureauandtheNationalCenterforHealthStatistics(NCHS).Theseriesaggregates data by geographic area (US, statewide, regions and MSAs, and counties), gender (female, male),race(non-HispanicWhite;White,includingHispanics;HispanicWhite;Hispanicsof anyrace;Blacks;AmericanIndian,Eskimo,orAleut;andAsianorPacicIslander),andage group(inve-yearincrements)andcoversthe1970-2009period. Forourpurposes,weusetheannualestimatesofthemalepopulation,ages15through64years, intheLACMSA.TheMSMpopulationisasubsetofthetotalmalepopulation. Toobtain thissubset,wemultiplythetotalmalepopulationinaparticularagegroupbythepercentage of the population that identies as MSM, as specied in Table B.9. We note that, bisexual identiersarediculttodistinguishinthedata,andthereissomeoverlapinthispopulation. WethensumalltheagegroupstogetthetotalMSMpopulationbyyear. 225 TableB.9:Initialpopulations SexualOrientation Ages PercentIdentify Source MSM 15-19 3.1 [47] MSM 20-24 4.3 [47] MSM 25-34 3.7 [47] MSM 35-64 4 [14] WepartitionthistotalMSMpopulationintotheappropriatecompartmentsusingdataper- centagesfromtheLACSurveillanceSummaries[2–5]. ThetotalpopulationsintheRCSPD LACMSAandtheSurveillanceSummariesdonotperfectlyalign.Thetotalnumberofpeo- plelivingwithHIV/AIDS(PLWHA)populationintheSurveillanceSummariesislargerthan thatfromtheRCSPD.Toadjustforthiswecreateayearlyadjustmentfactortoweightthe LAC MSA data. Such factors allow us to match the total HIV and AIDS infection rates reportedintheSurveillanceSummary. CalculationsofthetotalAIDSandnon-AIDSHIV casesareexplainedinthesubsequentsections. B.5.2 Inow-14yearolds AsdescribedinSectionB.3.2,theinowpopulationiscomprisedof14yearoldsthatwillenter thepopulationassexuallyactiveatisestimatedusingtheRCSPD. B.5.3 AIDSprevalence ToobtainthetotalAIDScaseswhichwesubsequentlydivideintothecompartmentsT and T A ,weuseasmuchdataaspossiblefromthemostrecentSurveillanceSummary,namelythe 2011SurveillanceSummary[5].However,foryears2000to2002,weoftenneedtorelyondata 226 inearlierSurveillanceSummaries[2,4]sincethe2011SurveillanceSummaryusuallycontains dataonlyfrom2003to2010. Toobtain thetotalannualMSM AIDScases, weusedatafrom theMSMand MSM/IDU categories in the Surveillance Summaries identied in Table B.10. We leave out AIDS cases frombloodtransfusions/hemophiliaandperinatalexposuresinceourmodeldoesnotcapture thesedynamics(thesecasesaccountforlessthanonepercentofallcases).Thereisalineartrend forthetotalnumberofcases(MSMplusheterosexualcases);however,wenoticeakinkprior to2003intheindividualMSMandheterosexualtimeseries.BoththeMSMandheterosexual seriesarelinearfrom2003to2010leadingustobelievethedistributionpriorto2003tobe slightlyof.Tocorrectforthisdistributionalodditypriorto2003,weusethepercentofcases thatareMSMfrom2004to2010tolinearlyimputethepercentagesforyears2000to2003 (alsoshowninTableB.10).WeusetheimputedMSMseriesasourannualtotal.Anoverview ofthedivisionofAIDScasesisshowninFigureB.1. 227 TableB.10:TotalMSMAIDScasesinLAbyyear. Year ReportedCases ImputedCases Source 2000 12,182 13,283 [2],Table18 2001 12,768 14,058 [2],Table18 2002 13,704 15,175 [2],Table18 2003 16,194 16,188 [4],Table20 2004 16,917 16,917 [4],Table20 2005 17,596 17,596 [4],Table20 2006 18,256 18,256 [4],Table20 2007 18,848 18,848 [4],Table20 2008 19,719 19,719 [4],Table20 2009 20,577 20,577 [4],Table20 2010 21,086 21,086 [4],Table20 228 FigureB.1:DivisionofMSMAIDScasesintocompartments. TOTAL AIDS CASES (MSM and Heterosexual) 2000-‐2001: January 2008 Surveillance Summary Table 18 2002: January 2010 Surveillance Summary Table 17 2003-‐2010: January 2011 Surveillance Summary Table 20 Annual MSM AIDS CASES, Stages A S ,TA S , A R ,TA R MSM and MSM/IDU Table Numbers MSM AIDS ON TREATMENT, Stages TA S and TA R ~80% of total AIDS cases MSM NO TREATMENT, Stages A S and A R ~20% of total, (1-‐0.8) MSM NO TREATMENT DRUG SENSITIVE Stage A S 97% no Treatment ~19.4% of Total MSM NO TREATMENT DRUG RESISTANT Stage A R 3% no Treatment ~0.6% of Total MSM TREATMENT DRUG RESISTANT Stage TA R 3% on Treatment ~2.5% of Total MSM TREATMENT DRUG SENSITIVE Stage TA S 97% on Treatment, ~77.5% of total B.5.3.1 DivisionofAIDScasesintotreatmentandnotreatment–compartmentsT andT A TherstpartitionoftheMSMAIDScasesconsistsofsplittingitintothosethatarereceiving treatment(T A )andthosenotontreatment(A).Weassumethatmostpeoplediagnosedwith AIDS will be on treatment yet some will opt not to go onto treatment for reasons such as drug-sideefects. WealsoexpectthattheindividualsnewlydiagnosedwithAIDSwillhavea slightlyhigherrateofreceivingtreatment.InadditiontothetotalAIDScasesfromTableB.10, theSurveillanceSummariesprovidethenumberof new AIDScases. Foryear2000, weuse Table B.5 from the 2008 Surveillance Summary [2]; for year 2001, Table B.5 from the 2010 Surveillance Summary; and for 2002 to 2009, Table B.16 from the 2011 Surveillance Sum- mary [5]. The total number of AIDS cases contains the number of new AIDS cases. We 229 assumethatthenumberofnewAIDScasesarethosethatarerecentlydiagnosed(withinthe year).Wefurtherassumethatroughly80%ofpeoplediagnosedwithAIDSareontreatment treatment[8,16]. B.5.3.2 DrugsensitiveanddrugresistantAIDSpopulations ThenalstageofdivisionfortheMSMAIDSpopulationisbetweenthedrugresistantand drugsensitivepopulations. BydividingtheAIDSontreatmentandAIDSnotontreatment into drug sensitive and drug resistant populations, four sub-populations are produced: (1) AIDSontreatment,drugsensitive,(2)AIDSontreatment,drugresistant,(3)AIDSnoton treatment, drug sensitive, and (4) AIDS not on treatment, drug resistant. We assume that MDR will be rather low and so for both the treatment and no treatment populations we estimatethat3.1%ofthepopulationswillhaveMDR.Thisvalueisthemeanoftherangefor multi-drugresistance(0-6.2%)estimatedbyvandeVijveretal.[42]. B.5.4 Non-AIDSHIVprevalence UnliketheAIDSdata,non-AIDSHIVdatawereneithercollectedbyLACnorreportedin theSurveillanceSummariesuntilabout2002. Code-andname-basedHIVreportingbegan in2002and2006,respectively[5]. Non-AIDSHIVdatafrom2008,2009,and2010arealso preliminary. Thedatacollectedbetween2007and2010onHIVcontainonlynewdiagnoses butnotthetotalpopulationlivingwithnon-AIDSHIV.FromtheseriesofSurveillanceSum- maries, we notice that preliminary data can change quite signicantly as it is validated and nalized(e.g.,the“Other”AIDScategoryinTable17ofthe2010SurveillanceSummary[4] is much larger than the “Other” AIDS category in Table 20 of the 2011 Surveillance Sum- mary[5]).Usingtheavailabledata,weimputedthenumbersofHIVcaseswhenunavailable. AswiththeAIDScalculations, weignoredHIVinfectionsoriginatingfrombloodtransfu- sionsandperinatalexposure. 230 B.5.4.1 ConstructingknownHIVcases–compartmentsJ,E,andT Therstnon-AIDSHIVdataappearinTable21inthe2008SurveillanceSummaryandthis tablecontainsdataoncumulativeHIVcasesfromJuly1,2002toApril17,2006.Weusethisas aproxyforthenumberofpersonslivingwithHIVinyear2003.TogettheHIVcasespriorto 2006,wemultiplythepercentageofthepopulationin2003thatareawareoftheirHIVstatus bytheannualpopulationsofthetotalreportednumberofpeoplelivingwithHIV/AIDSfrom 2000to2006. Similarlyforyear2007,wemultiplythepercentageofthepopulationthatare awareoftheirHIVstatusin2008bythe2007population.Thisproducestheawarenon-AIDS HIVcaseswhicharethesumofpopulationsincompartmentsJ,E,andT. 231 TableB.11:TotalawareMSMHIVcasesinLAbyyear. Year Population ReportedHIV %withHIV ImputedHIV Source 2000 176,969 – 0.0527 9,319 [2],Table21using2003population 2001 179,327 – 0.0527 9,443 [2],Table21using2003population 2002 181,514 – 0.0527 9,558 [2],Table21using2003population 2003 183,832 9,680 0.0527 9,680 [2],Table21 2004 185,893 – 0.0527 9,789 [2],Table21using2003population 2005 187,206 – 0.0527 9,858 [2],Table21using2003population 2006 188,335 – 0.0527 9,858 [2],Table21using2003population 2007 189,069 – 0.0640 12,100 [3],Table23using2008population 2008 190,186 12,172 0.0640 12,100 [3],Table23 2009 190,967 13,316 0.0697 13,316 [4],Table22 2010 190,699 13,343 0.0700 13,343 [4],Table13 232 B.5.4.2 Divisionofawaretreatmenteligible(EandT)andineligible(J) NextwedividetheHIVawarecasesbythosethatareeligiblefortreatmentandthosewhoare not. Before reaching the treatment stage (T) all individuals must rst be treatment eligible (i.e., each person must transit through stageE before entering stageT). We use CD4 cell countstodividetheawareHIVcasesintoJ andE +T. WeassumeindividualswithCD4 cellcountsabove350tobeineligiblefortreatment;wefurtherassumeindividualswithCD4 countsbelow350tobeeligiblefortreatment(thiswasthetreatmentguidelineformostyears usedforthecalibration).Weusethe2005CDCstatistics(themiddleofourmodelingperiod) to estimate the percentage of the non-AIDS HIV aware population with CD4 cell counts greaterthan350andbetween200and350[1]. Aapproximately72.55%ofthepopulationin the combined non-AIDS HIV aware compartments had CD4 cell counts greater than 350, henceweensuredthat72.55%oftheannualawarepopulationswereincompartmentJ with theremaining37.46%wereintheEorT compartments(referredtoasE +T). B.5.4.3 DivisionofawaretreatmenteligibleEandawareontreatmentT To divide theE + T population into separate compartments, we we assume that a larger proportionofpatientswithfullblownAIDSwouldseektreatment,comparedtosero-positive individualswithoutAIDS.Wefurtherassumethatapproximately80%and60%ofAIDSand sero-positivepatients,respectively,willbeontreatment.Theestimateof60%isbasedonthe sameestimatesusedintheAIDSdivision[8,16]. B.5.4.4 CalculationoftheHIVunawarecases,P andI The Surveillance Summaries provide data for the number of known HIV cases; however, thoseunawareoftheirinfectionstatusalsoneedtobeconsideredandestimated. TheLAC 233 DepartmentofHIVandSTDPreventionandtheCDCestimatethattheunawarepopula- tionaccountsforapproximately25%ofHIVcases[13].Thus,toobtaintheunknownnumber P +I,weusethefollowingformula: P +I = (HIVAware+AIDS)× 0.25 0.75 (B.34) InordertosplitthecompartmentsP andI weusethemodeoftheparametersrepresenting thetransitionratebetweenthesetwocompartments,asshowninTableB.8. P = 1 ρ × 365 (B.35) I = 1 ω × 365 (B.36) Unaware = I +P = 1 ω + 1 ρ ! × 365 (B.37) WedividethecombinedunawarepopulationbetweentheP andI compartmentsbasedon thepercentageoftimeindividualsspendineachcompartment(thedurationistheinverseof therate).Thisisabout35daysinstageP and1,205daysinstageI. B.5.4.5 DrugsensitiveanddrugresistantHIVpopulations Aswiththedrug-resistantanddrug-sensitiveAIDSpopulations,wedivideeachoftheHIV compartmentsintodrugsensitiveanddrugresistantcompartments.Weusethesamevalueof 3.1%aswiththeAIDScompartments[42]. 234 B.6 Modelcalibration Asmentionedabove,wecalibrateourmodeltomatchtheavailabledataforLACfrom2000- 2009. Inordertodothiswerstrun500,000LHSsimulations. Eachsimulationindepen- dentlysamplesasetofparametersbasedontheirspeciedrangeanddistribution(TableB.7- TableB.8)toproduceoutputsfortheyearsspecied. Onceallsimulationsarecompiled,we then drop simulations that do not t the following constraints: (a) an HIV+ but unaware populationof21-25%ofthetotalHIV/AIDSpopulation[13],and(b)anMDRprevalenceof 2-5%ofthetotalpopulation[36,42].Throughthisprocess,wedroppedabout50%ofthesim- ulations.Wethencomparedeachoftheremainingsimulationstotheproducedbaselinedata (Section B.5), and calculated the annual diferences between the total number of PLWHA fromthebaselineandthetotalnumberofPLWHAproducedbythemodel(referredtoas theerror). Foreachsimulation,wethencomputedthesumofthesquarederrors(SSE)acrossallyears. ThesimulationsthatproducethelowestSSEaretherunsthatmostaccuratelymatchthebase- linedata. Wethenidentiedthebest1,000simulations(thosewiththelowestSSE)tocreate new,narrowerrangesfromwhichtore-sample. Thisapproachensuresthatoursubsequent setofsimulationsonlysamplesparametervaluesthatwerepreviouslyincludedinrunsthat accuratelymatchedthebaselinedataonthetotalHIV/AIDSprevalence. Figure??depictsin factthesumofsquarederrorsandlogsumofsquarederrorsbyrank.Theseplotssuggestthat agrowthinSSEoccursatlowerranks. TheSSEsandlogSSEsalsoappearbemonotonically increasingintheranks,suggestingthatwecouldhavechosenmorethanthetop1000runsand stillobtaingoodtstothedata. 235 FigureB.2:SSEsandLogSSEsagainsttheSSEranks 0 100 200 300 400 500 0 20000 40000 60000 80000 SSE Rank (000) SSE (000,000) (a) Sumofsquarederrors(SSEs)byrank 0 100 200 300 400 500 5 6 7 8 9 10 11 SSE Rank (000) Log 10 (SSE) (b) LogSSEsbyrank Inordertoaddmoregranularitytoourcalibration,werunanadditional500,000LHSsim- ulations,samplingfromthenew,narrowerparameterrangesproducedbythetop1,000runs fromtheprocessabove. Wedroppedallsimulationsthatdidnotttheconstraints(a)and (b) outlined above. This process allowed us to drop about 20% of the simulations. Since theLACSurveillanceReportstrackthetotalPLWHApopulationbyspecifyingawaresero- positive patients without full-blown AIDS and those with full-blown AIDS, we calibrated onthesecategoriestoaddmoredetailtoourcalibration.Todothis,weonlykeptsimulations that have an output with total AIDS cases that are less than 5% diferent from total AIDS casesreportedinthe2009surveillancedata. Wechoseonlytheyear2009sincereportinghas improvedwithtime,and2009isthemostrecentyearwithhighlyaccuratereporting. Then, withintheremainingsimulationsweonlykeptsimulationsthatproducedatotalofnon-AIDS HIVawarecasesthatwerelessthan5%diferentfromthetotalnon-AIDSHIVcasesreported inthe2009surveillancedata(allcasesreportedinthesurveillancedataareawareoftheirinfec- tionstatus). Finally,oftheremainingsimulations, wechosethesimulationwiththelowest SSEacrossallyearsforthetotalPLWHApopulation,andidentiedthisasthe“BestRun”.We 236 usedtheparametersproducedbythisrunasthemainparametersofthemodel.Theparame- tersthatmostinuencedthedynamicsofthemodelwereβ I ,C msm ,β TA ,β J ,γ TAs ,andβ T . Thecalibratedparametervaluesalongwiththenarrowerrangesforthesecondsetof500,000 runsaredisplayedinTableB.12.Aplottedcomparisonbetweenthebaselinedataandourbest runcanbeseeninFigure3.2(Chapter3). TableB.12:Calibratedparametervaluesandnarrowedparameterranges. Parameter CalibratedValue NarrowerRangesFromTheTop1,000Runs Mean Min Max π 1.0324 1.0001 0.9022 1.0979 ρ 11.0136 11.8296 6.7713 22.1852 ω msm 0.2269 0.2611 0.1799 0.4524 ω A 0.0523 0.0640 0.0294 0.1387 χ 0.1700 0.1665 0.1450 0.1956 ν 0.6779 0.4648 0.2501 0.8302 γ Es 0.6658 0.3694 0.2674 0.6693 γ Er 1.3080 0.7325 0.5314 1.3386 γ Ts 0.0777 0.0639 0.0468 0.0875 γ Tr 0.1947 0.2388 0.1045 0.5370 σ 0.4040 1.1142 0.3353 6.8907 σ A 10.8637 4.9983 0.6766 20.6688 g 0.1160 0.0908 0.0234 0.1576 g A 0.0314 0.0408 0.0024 0.0774 μ 0.0004 0.0003 0.0003 0.0004 γ As 0.5427 2.6022 0.5093 7.8389 γ Ar 1.7016 6.7523 1.1098 19.8800 γ TAs 0.1187 0.2063 0.0795 0.4120 γ TAr 0.4891 0.4359 0.1643 0.8296 r 0.0278 0.0214 0.0061 0.0535 q 0.0043 0.0154 0.0006 0.0307 h r 0.1232 0.1176 0.1000 0.1756 C msm 4.5046 4.3037 2.2798 8.4753 β P 0.1089 0.2297 0.0364 0.5296 β I 0.0505 0.0748 0.0180 0.1900 β J 0.0185 0.0340 0.0048 0.0837 β E 0.0170 0.0338 0.0068 0.0839 β T 0.0159 0.0116 0.0002 0.0375 β A 0.0677 0.0944 0.0207 0.2398 β TA 0.0275 0.0266 0.0024 0.0653 237 B.6.1 Calibrationsensitivityanalysis Weconductedasensitivityanalysisonall31parametervalues,varyingsequentiallyeachparam- eteratatimebetweenalowerandupperbound,whilekeepingallotherparametersxedto their“bestrun”valuesfromtherststageofthecalibrationstep.Thisanalysiswasconducted todeterminetheparametersthatmostinuencedthemodeldynamic.Weformallyrepresent thisasfollows: S(x) = f(x j i |x {−i} ),∀i∈{1,··· , 31},andj∈{L,M,U}, (B.38) whereS(·)representstheestimateofthediferentialequationatx = (x 1 ,··· ,x 31 ),iisthe ithparameter,andjthelowerbound,medianandupperboundofparameterx i . Wecomputedthesizesoftherangesthateachparameterproducedaroundthebestrun.The sizeswerecomputedusingthesumofthesquarederrors,wherethe“error”foreachyearisthe deviationofthesimulateddatafromthebaselinedata.Theresultsfromthisanalysisaresum- marizedinFigureB.3.Thedarker(lighter)colorsindicatethemost(least)sensitiveparameters inthecalibrationforagivencompartment.Whiletheactualvaluesoftheserangesdonothave anyparticularinterpretation,theyserveasarankingmeasureoftheinuenceofeachparame- teroneachcompartment,withlargervaluesindicatinggreaterinuence.Overall,theseresults suggestthattheparametersthatinuencedthemostthedynamicofthemodelwereβ I ,C msm , β TA ,β J ,γ TAs ,andβ T . 238 FigureB.3:Sensitivitymapfortheparametercalibration. Total I J E A T TA T + TA AIDS JET Compartment β I C msm β TA σ β J ω ν γ TAs σ A β T ω A γ Ts β P ρ β E r χ g γ Es γ As γ Tr β A g A π γ TAr μ h γ Er γ Ar q Parameter 239 B.7 Test-and-treatmodel InordertosimulateapolicychangeofincreasingHIVtestingandallowingforearlytreatment (testandtreat)weextendedtheHIVEpidemicSimulationModeldescribedabovetoincor- poratetwoadditionalcompartments,TJ s andTJ r ,whichrepresentthenon-drugresistant andMDRinfectedindividuals,respectively,whoinitiateARTtreatmentearly(withCD4cell countsabove350).Thesecompartmentsarecharacterizedashavinglowerratesoftransmissi- bility(denedasβ TJ )relativetostageJ [15].Theaugmentedcompartmentalmodelfollows thedynamicsillustratedinFigureB.4.Thisgureonlydepictsthecompartmentsforindivid- ualscarryingthedrugsensitivestrain;however,thecompartmentsandowsforindividuals carryingthedrugresistantstrainmirrorthoseofFigureB.4.Individualwhoinitiatetreatment earlytransitiontostageTJ atrateσ TJ .OnceintheTJ compartment,peoplecaneitherter- minatetreatmentwithaCD4cellcountstillabove350andtransitionbacktostageJ atrate g TJ oriftreatmentisnotefective,theycanterminatetreatmentwithaCD4cellcountbelow 350andtransitiontostageE atrateθ. PeopleinstageTJ s canalsodevelopMDRthesame waytheydoinstageT s andtransitiontostageTJ r atrater. 240 FigureB.4:Schematicsofthetest-and-treatscenariomodel. Note: EstimatesofparametersanddatasourcesenteringthemodelareprovidedinTableB.12andTableB.13, anddiscussedinSectionB.7.1. ThefollowingODEsintegratethesedynamicsforthetest-and-treatscenario: dS dt = π− (μ +λ s +λ r )S (B.39) dP s dt = λ s S− (μ +ρ)P s (B.40) dI s dt = ρP s − (μ +χ s +ω +ω A )I s (B.41) dJ s dt = ωI s +g TJ TJ s − (μ +ν s +σ TJ )J s (B.42) dTJ s dt = σ TJ J s − (μ +g TJ +θ +r)TJ s (B.43) dE s dt = χ s I s +ν s J s +θTJ s +gT s +qE r − (μ +γ Es +σ)E s (B.44) dT s dt = σE s − (μ +γ Ts +g +r)T s (B.45) dA s dt = ω A I +γ Es E s +g A T As +qA r − (μ +γ As +σ A )A s (B.46) dT As dt = γ Ts T s +σ A A s − (μ +γ TAs +g A +r)T As (B.47) dP r dt = λ r S− (μ +ρ)P r (B.48) 241 dI r dt = ρP r − (μ +χ r +ω +ω A )I r (B.49) dJ r dt = ωI r +g TJ TJ r − (μ +ν r +σ TJ )J r (B.50) dTJ r dt = σ TJ J r +rTJ s − (μ +g TJ +θ)TJ r (B.51) dE r dt = χ r I r +ν r J r +θTJ r +gT r − (μ +γ Er +σ +q)E r (B.52) dT r dt = σE r +rT s − (μ +γ Tr +g)T r (B.53) dA r dt = ω A I r +γ Er E r +g A T Ar − (μ +γ Ar +σ A +q)A r (B.54) dT Ar dt = γ Tr T r +σ A A r +rT As − (μ +γ TAr +g A )T Ar (B.55) B.7.1 Test-and-treatmodelparametervalues Thenewparametersofthetestandtreatmodel,σ TJ ,g TJ ,θ,andβ TJ (alldescribedabove)are basedonpreviouslycalibratedparametersandliteraturereview.Theparameterσ TJ isxedat thesamecalibratedvalueforσ(therateoftreatmentinitiationaftertheCD4cellcountdrops below 350). This assumes that once eligible, people in stagesE andJ will begin treatment atthesamerate. Wedenetheθandg TJ parameterstoberelatedtog. Weassumethatthe adherencerateforpeoplethatinitiatetreatmentearly(TJ)isidenticaltothatofpeoplewho initiatetreatmentlate(T).However,acertainportionofthosewhostarttreatmentearlyand do not adhere will terminate treatment while their CD4 cell count is above 350 (transition throughg TJ ). Further,acertainportionwillterminatetreatmentwiththeirCD4cellcount below350(transitionthroughθ).Thereforeweassumethefollowingrelationships: θ = θ f ·g (B.56) g TJ = (1−θ f )·g, (B.57) 242 whereθ f takesavalueintherange [0, 1],anddeterminestherespectivesharesthattransition throughθandg TJ .Weassumethat10%ofthoseinTJ willterminatetreatmentwithaCD4 cellcountbelow350andthereforewesetθ f to 0.1. Weassumeβ TJ toberelatedtotheβ J throughaproportionalityconstantβ f TJ ,suchthat β TJ = β f TJ ·β J . (B.58) BasedonCohenetal.[15],weensurethatβ TJ is96%lowerthanβ J andthereforesetβ f TJ to 0.04. B.7.2 Baselinetest-and-treatscenarioparameters Asdescribedinthemaintext,wechooseabaselinetestandtreatscenariowherewemanipulate theσ TJ ,σ,andωparameterssothattheyreectatestandtreatpolicyinwhichHIVpositive people are tested annually on average, and initiate treatment 2.5 years after they have been diagnosedwithHIV.Exceptforthenewparametersσandω,andthetest-and-treatparameters outlined above, all parameters for this scenario take the calibrated values described earlier. Further,sinceωchanges,ν isalsoadjustedtomaintaintherelationshipbetweenω,ν andχ, asdescribedinSectionB.2.1. Thevaluesofthenewparametersforthebaselinetestandtreat scenarioaredisplayedinTableB.13. 243 TableB.13:Parametersforthebaselinetest-and-treatscenario. Parameters Values Denition σ TJ 0.4040 Rate at which HIV positive people begin treatment after diagnosis (rateofearlytreatment). ω 1.0000 RateatwhichHIVpositivepeoplearetested. θ f 0.1000 Portion of the population in TJ that do not adhere to treatment whentheirCD4cellcountisbelow350. θ 0.0116 RateatwhichpeopleinTJ donotadheretotreatmentwhentheir CD4cellcountisbelow350. g TJ 0.1044 RateatwhichpeopleinTJ donotadheretotreatmentwhentheir CD4cellcountisabove350. β f TJ 0.0400 Percent relationship between the per-partnership transmissibility in stageTJ relativetostageJ. β TJ 0.0007 Per-partnershiptransmissibilityinstageTJ. B.8 Sensitivityanalysisoftheimpactoftest-and-treat Weconductaseriesofsensitivityanalysestodeterminetheimpactofvariousassumptionsand modelparametersontheoutcomesmeasured. B.8.1 One-waysensitivityanalysis We rst conduct a one-way sensitivity analysis to determine how the model results are impactedbyeachparameterindividually.Inparticular,wewanttounderstandwhichparam- etersmostinuencetheimpactoftest-and-treatstrategyonthenumberofnewinfectionsand theprevalenceofMDR.Todothis,weimposedoneachparameteradisturbanceδ = 10% aboveand belowthe baselinetest andtreat valueswhile holdingall otherparameters xed. 244 Weusedthesenewparametervaluestoestimatetheefectsoftest-and-treatonthenewinfec- tionsandMDR.TableB.14reportstheresultsforthenewinfections. Column1reportsthe parametername,columnstwoandthreereportthereductioninnewinfectionsduetotest- and-treatatupperandlowerdisturbancelevels,respectively,andcolumnfourcomparesthe reductionsinnewinfectionsreportedincolumnstwoandthreetoourbaselinetestandtreat results(33.79%reduction). Overall, we nd that the results for the new infections are most sensitive to the parameters thatrepresentsexualbehavior(theβsandcparameters).ThedeathrateforpeopleintheT A compartment,γ TA , alsohasarelativelylargeefectsincethiscompartmenthasalargepop- ulation. However,ourresultsareratherstableevenwiththese10%changes,withthelargest impactofany10%changeinourbaselineparametersresultinginonlya 1.82percentagepoint changeinnewinfectionsestimates. TableB.16reportssimilarresultsforMDR,forwhichweestimatedabaselineresultof 9.06% prevalencein2023.Thisresultwasmostsensitivetotheassumedresistancerateaswellastothe parameterγ TA .Similartonewinfections,theresultsforMDRwiththesedisturbanceswere ratherstable,withthelargestimpactofany10%changeinourbaselineparametersresulting inonlya 1.82percentagepointchangeinourresultsfornewinfections. Wealsopresentresultsoftheone-waysensitivityanalysisgraphicallyinFigureB.5. Weplot therunsthatproducedthelargestdiferencefromourbaselinerun(maximumandminimum) thatresultedfromthe10%disturbanceanalysisonthesameplotasourbaselinerunforeach outcome.Thisprovidesasenseofanupperandlowerboundofthetest-and-treatimpact. 245 TableB.14:Percentreductioninnewinfectionsby2023. Parameter Upper(%) Lower(%) MaxDif a β I 35.38 31.97 1.828 γ TAs 34.77 32.75 1.043 β TA 32.79 34.80 1.005 β J 33.17 34.41 0.622 c 33.26 33.50 0.530 π 34.09 33.48 0.314 r 34.09 33.50 0.296 γ Es 34.04 33.52 0.279 ω A 33.61 33.98 0.182 β T 33.65 33.94 0.144 σ 33.69 33.91 0.113 γ Ts 33.85 33.74 0.058 γ As 33.85 33.74 0.053 θ 33.74 33.85 0.050 σ A 33.76 33.84 0.043 g 33.83 33.76 0.035 ρ 33.82 33.76 0.033 γ TAr 33.81 33.77 0.023 β P 33.77 33.81 0.021 h r 33.78 33.81 0.016 g A 33.80 33.79 0.005 γ Tr 33.80 33.79 0.005 β E 33.79 33.79 0.002 γ Ar 33.80 33.79 0.001 γ Er 33.80 33.79 0.001 β A 33.79 33.79 0.000 q 33.79 33.80 0.000 μ 33.79 33.80 0.000 a Maximumdiferencefromthebaselineresults(percentagepoints) 246 TableB.15:PercentoftheHIV/AIDSpopulationwithMDRin2023. Parameter Upper(%) Lower(%) MaxDif a r 9.48 8.63 0.430 γ TAr 8.80 9.37 0.315 γ TAs 9.24 8.87 0.185 γ Tr 8.98 9.15 0.091 σ 9.13 8.98 0.079 g 8.99 9.14 0.078 C msm 9.04 9.12 0.066 β TA 9.02 9.11 0.053 h r 9.10 9.02 0.041 β I 9.10 9.02 0.040 γ Er 9.03 9.09 0.030 γ Ts 9.03 9.08 0.025 pi 9.04 9.08 0.023 β J 9.04 9.08 0.021 ω A 9.04 9.08 0.020 θ 9.04 9.08 0.020 γ Ar 9.04 9.07 0.014 γ As 9.07 9.05 0.009 ρ 9.07 9.05 0.009 β T 9.05 9.06 0.006 σ A 9.06 9.05 0.005 γ Es 9.05 9.06 0.004 β P 9.06 9.06 0.002 β E 9.06 9.06 0.002 g A 9.06 9.06 0.001 q 9.06 9.06 0.000 β A 9.06 9.06 0.000 μ 9.06 9.06 0.000 a Maximumdiferencefromthebaselineresults(percentagepoints). 247 FigureB.5:Baselinetest-and-treatscenariocomparedtothestatusquo. 2012 2014 2016 2018 2020 2022 2024 0 10 20 30 40 50 60 Y ear Thousands [000] Status Quo Test and Treat Status Quo lower/upper bound Test and Treat lower/upper bound (a) Cumulativenewinfections:Test-and-treatsimula- tionbasedontheparameterC msm ;upperandlower boundsforbothstatusquoandtestandtreatscenario determinedbytheparameterβ I . 2012 2014 2016 2018 2020 2022 2024 0 10 20 30 40 50 60 Y ear Thousands [000] Status Quo Test and Treat Status Quo lower/upper bound Test and Treat lower/upper bound (b) CumulativenewAIDS:Bestt,upperandlower boundsforbothstatusquoandtestandtreatscenario determinedbytheparameterC msm . 2012 2014 2016 2018 2020 2022 2024 0 10 20 30 40 50 60 Y ear Thousands [000] Status Quo Test and Treat Status Quo lower/upper bound Test and Treat lower/upper bound (c) Cumulative deaths: Test-and-treat simulation based on the parameter C msm ; upper and lower boundsforbothstatusquoandtestandtreatscenario determinedbytheparameterβ I . 248 B.8.2 Multi-waysensitivityanalysis Next,weallowallparameterstovarysimultaneously.Todothis,weusethe10%disturbance aboveandbelowthecalibratedvalueasrangefromwhichwesample100,000newparameter sets. Next, for each parameter set, we conduct a test-and-treat simulation and a status quo simulationandcalculatetheepidemiologicalimpactoftest-and-treatforeachset.Thedistri- butionofpercentreductionsforall100,000runswithcondenceintervalsarepresentedfor eachepidemiologicaloutcomeinFigureB.6.Wecanseethatthedistributionsarerathertight with 95% condence intervals ranging from a 30-37% reduction in new infections, a 16-21% reductionindeaths,anda36-41%reductioninnewAIDScases. 249 FigureB.6:Multi-waysensitivityanalysis–distributionsofthe%reductioninnewinfections, deathsandnewAidscases. % Reduction in New Infections Frequency 28 30 32 34 36 38 40 0 500 1000 1500 2000 2500 Best Run Estimate 95% Confidence Interval Bound (a) Distribution of % reduction in new infections frommulti-waysensitivtyanalysis. % Reduction in Deaths Frequency 16 18 20 22 0 1000 2000 3000 4000 Best Run Estimate 95% Confidence Interval Bound (b) Distribution of % reduction in deaths from multi-waysensitivtyanalysis. % Reduction in New AIDS Cases Frequency 36 38 40 42 0 500 1000 1500 2000 2500 3000 3500 Best Run Estimate 95% Confidence Interval Bound (c) Distribution of % reduction in new AIDS cases frommulti-waysensitivtyanalysis. 250 B.8.3 Sensitivitytovariationsintheinfectiousnessassumption Ourtest-and-treatscenarioismodeledontheassumptionofa96%reductionininfectious- nessduetoART[15]. Thisestimate,however,isbasedonstudiesofserodiscordantcouples andmaynotapplytotheMSMpopulation. SincetheecacyofARTmaydiferbetween MSMandheterosexualpopulations, weconductasensitivityanalysisonourparameterfor thereductionininfectiousness. Wetestthesensitivitytotheresultsofourassumptionofa 96% reduction in infectiousness from ART using the upper and lower ends of the Cohen et al. [15] 95% condence interval, 99% and 73% reductions, respectively. We also conduct athirdsimulationassuminga50%reductionforadditionalcontrast. Wepresentresultsfor newinfectionsunderdiferentinfectiousnessassumptionsgraphicallyinFigureB.7. Wecan see that results do not change signicantly, even under the assumption of a 50% reduction. This is largely attributable to the fact that much of the efect of the test and treat policy is derivedfromtheincreaseintestingaswillbeshowinSectionB.9. 251 FigureB.7:Robustnessanalysisaroundtheinfectiousnessparameter. 2012 2014 2016 2018 2020 2022 2024 0 10 20 30 40 50 Y ear Thousands [000] Status Quo 99% reduction 96% reduction 73% reduction 50% reduction B.8.4 Calibrationvs.literaturebasedparameterestimates One advantage of simulating the HIV/AIDS epidemic in Los Angeles is that there is high qualitydataavailablethatcanusedforcalibration. Manysettingsdonothavethisluxury,in whichcaseparametersaregenerallychosenbasedonclinicalliterature.Initially,weconstruct parameterrangesforsamplingbasedonclinicalliteratureandrenetheseestimatesusingthe calibrationmethoddescribedinSectionB.7. Toassesstheimportanceofcalibrationrelative tochoosingparametersbasedonliterature,werunasimulationusingparametersthatrepre- sentourbestestimatebasedontheavailableliterature. Wepresentoutcomesforthissimu- lationalongwithourcalibratedbaselineoutcomesfromthemainpaperinTableB.16. This 252 tablehighlightstheimportanceofcalibration. Wecanseethatifwehadnotcalibratedthe model,wewouldhavedramaticallyunderestimatedabsoluteepidemiologicalbenetsonall threeoutcomes.Althoughwewouldhaveachievedrelativelysimilarresultsinpercentreduc- tions in new infections (a slight overestimation of 1.19 percentage points), we would have underestimatedpercentreductionsindeathsandnewAIDScasesby 6.45percentagepoints and 14.32percentagepoints,respectively. 253 TableB.16:Comparisonofmodelresultsusingliterature-orcalibration-basedparametervalues. Cumulativefrom2013-2023 NewInfections Deaths NewAidsCases %Reduction Absolute Reduction %Reduction Absolute Reduction %Reduction Absolute Reduction Initialliterature-basedvalues 34.98% 2,674 12.46% 1,763 24.70% 2,921 Calibratedvalues 33.79% 18,272 18.91% 7,956 39.02% 19,083 Diference a 1.19% 15,598 6.45% 6,193 14.32% 16,162 a Diferenceinbaselinescenariotestandtreatresultsforcalibratedvs.non-calibratedparametervalues. 254 B.9 Resultswithoutearlytreatment Inadditiontotheresultspresentedinthemaintextofthepaper,wealsorunaseriesofsim- ulationsthatrepresentapolicyofscalinguptestingprogramsonly,withoutallowingpeople tostarttreatmentearly(nochangesintreatmentguidelines).Thissetofsimulationsisessen- tiallythecalibratedmodelshowninFigure3.1(Chapter3;noTJ compartmentsareincluded intheseruns)withtheexceptionthatweincreaseωtosimulateanincreaseintestingservices. Wealsomanipulateσ tosimulateamoreaggressiveattempttoputpeoplewithaCD4cell countbelow350ontreatment.Werunallscenarios,asshowninChapter3,howeverinthese scenariosthetimetoinitiatetreatmentbeginsafterbecomingtreatmenteligibleundertradi- tionalguidelines(theCD4cellcountdropsbelowacertainlevel),whereasinthescenariosin themaintextthetimetoinitiatetreatmentbeginsafterdiagnosis. Theresultsforthesesim- ulationsarepresentedinTableB.17. Wecanseethatbyrunningthesamescenarioswithout allowingforearlytreatment,wegetabouthalfofthebenetwithvirtuallynochangeinMDR comparedtotheresultsproducedbythetestandtreatscenariospresentedinthepaper. For example, underourbaselinetestandtreatscenariointhemaintextwendthatnewinfec- tionsarereducedby34%;deathsarereducedby19%;andnewAIDScasesarereducedby39%. However,inthesamescenariowithoutallowingforearlytreatment,thesereductionsare19%, 10%,and19%,respectively. 255 TableB.17:Resultswithoutearlytreatment. Scenario Cumulativefrom2013-2023 %oftheTotalHIV/AIDSPopulationin2023 NewInfections (%Reduction) Deaths (%Reduction) NewAIDSCases (%Reduction) %MDR a %Unaware %Treated Status Quo 54,067 42,083 48,907 4.79% 20% 64% Treatment 2.5 Years TestEvery1Year 43,735(19%) 38,070(10%) 39,647(19%) 4.72% 6% 60% TestEvery2Years 46,314(14%) 39,776(5%) 43,266(12%) 4.92% 9% 63% TestEvery3Years 47,893(11%) 40,924(5%) 45,713(7%) 5.07% 11% 66% TestEvery4.4Years 54,067(0%) 42,083(0%) 48,907(0%) 4.79% 20% 64% Treatment 1 Year TestEvery1Year 42,901(21%) 35,473(16%) 34,714(29%) 5.65% 6% 64% TestEvery2Years 45,376(16%) 36,972(12%) 37,891(23%) 5.89% 9% 68% TestEvery3Years 46,883(13%) 37,974(10%) 40,029(18%) 6.06% 11% 70% TestEvery4.4Years 52,972(2%) 39,113(7%) 43,048(12%) 5.72% 19% 69% Treatment 6 Months TestEvery1Year 42,234(22%) 33,345(21%) 34,714(37%) 6.41% 5% 67% TestEvery2Years 44,627(17%) 34,677(18%) 37,891(31%) 6.68% 8% 71% TestEvery3Years 46,078(15%) 35,562(15%) 40,029(28%) 6.88% 10% 74% TestEvery4.4Years 52,094(4%) 36,681(13%) 43,048(22%) 6.50% 18% 72% a MDR=multipledrugresistance. 256 B.10 Robustnessanalysisunderresistance Forillustrativepurposes,wealsosimulatedtheepidemicunderthebaselinetestandtreatsce- narioandthestatusquotodeterminewhetherornottheincreaseinMDRcouldofsetthe benetsoftestingandtreating.TheresultsinFigureB.8suggestthatthetestandtreatpolicy stillofersoferssubstantialepidemiologicalbenets,evenata23%MDRlevel,asshownby thestatusquoandtest-and-treattrendscontinuingtodivergeforalloutcomes. Itisimpor- tanttonotethatthisresultonlyillustratesthattherearestillbenetsoftest-and-treateven inthepresenceofhighlevelsofMDR.Thepredictivepowerofthemodelwithsuchalong timehorizonishoweverlimited,asitreliesontherigidassumptionofcurrenttrendsholding. For example, the dynamic of the epidemic is subject to the introduction of new treatment technologiesandtherapiesinthemarket,aswellastochangesinperceptionsandbehaviors towardsHIV/AIDS. 257 FigureB.8:Robustnessanalysis–baselinetest-and-treatscenariocomparedtothestatusquo. 2010 2020 2030 2040 2050 2060 2070 0 50 100 150 200 250 300 Cumulative New Infections Y ear Thousands [000] 2010 2020 2030 2040 2050 2060 2070 0 50 100 150 200 250 300 Cumulative Deaths Y ear Thousands [000] 2010 2020 2030 2040 2050 2060 2070 0 50 100 150 200 250 300 Cumulative New AIDS Cases Y ear Thousands [000] 2010 2020 2030 2040 2050 2060 2070 5 10 15 20 25 Multi−Drug Resistance Y ear Percent [%] 2010 2020 2030 2040 2050 2060 2070 0 5 10 15 20 25 Unaware Y ear Percent [%] 2010 2020 2030 2040 2050 2060 2070 60 65 70 75 80 Treatment Y ear Percent [%] Test and Treat Status Quo 258 References [1] CDCSurveillanceData. CDC Advance Data Table 7,11(2),2005. [2] HIV Epidemiology Program, Los Angeles County Department of Public Health. HIV/AIDS Surveillance Summary,pages1–29,Jan2008. [3] HIV Epidemiology Program, Los Angeles County Department of Public Health. HIV/AIDS Surveillance Summary,pages1–33,Jan2009. [4] HIV Epidemiology Program, Los Angeles County Department of Public Health. HIV/AIDS Surveillance Summary,pages1–33,Jan2010. [5] HIV Epidemiology Program, Los Angeles County Department of Public Health. HIV/AIDS Surveillance Summary,pages1–32,Jan2011. [6] Anderson,R.M.andMay,R.M. Infectious diseases of humans : Dynamics and control. Oxford,UK:OxfordUniversityPress,1991. [7] Baggaley, R.F., Ferguson, N.M. and Garnett, G.P. The epidemiological impact of antiretroviral use predicted by mathematical models: A review. Emerging Themes in Epidemiology,2(9):1–18,Sep2005. [8] Bingham,T.A.StateoftheHIVEpidemicAmongHIVMeninLosAngeles.TheBody, 2003. [9] Blower, S.etal. Theantiretroviralrolloutanddrug-resistantHIVinAfrica: Insights fromempiricaldataandtheoreticalmodels. AIDS,19(1):1–14,Jan2005. [10] Blower,S.M.,Gershengorn,H.B.andGrant,R.M. ATaleofTwoFutures: HIVand AntiretroviralTherapyinSanFrancisco. Science,287(5453):650–54,Jan2000. 259 [11] Bouchonnet, F. et al. Quantication of the efects on viral DNA synthesis of reverse transcriptasemutationsconferringhumanimmunodeciencyvirustype1resistanceto nucleosideanalogues. Journal of Virology,79(2):812–22,Jan2005. [12] Brenner,B.etal. Persistenceofmultidrug-resistantHIV-1inprimaryinfectionleading tosuperinfection. AIDS,18(12):1653–60,Aug2004. [13] CentersforDiseaseControlandPrevention. CDCHIV/AIDSFACTS.NewEstimates ofUSHIVPrevalence,2006. 2008. [14] CentersforDiseaseControlandPrevention. CDCFactSheet:HIVandAIDSAmong Gay and Bisexual Men. Available at http://www.cdc.gov/nchhstp/newsroom/docs/ FastFacts-MSM-FINAL508COMP.pdf,Sep2010. [15] Cohen,M.S.etal. PreventionofHIV-1infectionwithearlyantiretroviraltherapy. New England Journal of Medicine,365(6):493–505,Aug2011. [16] Comulada,W.S.etal.UseofHAARTamongyoungpeoplelivingwithHIV.American Journal of Health Behavior,27(4):389–400,2003. [17] G. Pantaleo, G. Graziosi, A.S.F. The Immunopathogenesis of Human Immunode- ciencyVirusInfection. New England Journa of Medicine,328(5):327–35,1993. [18] Garnett,G.P.andAnderson,R.M. FactorscontrollingthespreadofHIVinheterosex- ualcommunitiesindevelopingcountries:patternsofmixingbetweendiferentageand sexualactivityclasses. Philosophical transactions of the Royal Society of London. Series B, Biological Sciences,342(1300):137–59,Oct1993. [19] Gulick, R. et al. Treatment with Indinavir, Zidovudine, and Lamivudine in Adults withHumanImmunodeciencyVirusInfectionandPriorAntiretroviralTherapy.New England Journa of Medicine,337(11):734–9,1997. 260 [20] Helton,J.C.andIman,R.L.Sensitivityanalysisofamodelfortheenvironmentalmove- mentofradionuclides. Health Physics,42(5):565–84,May1982. [21] Hoare, A., Regan, D.G. and Wilson, D.P. Sampling and sensitivity analyses tools (SaSAT)forcomputationalmodelling. Theoretical Biology and Medical Modelling,5: 4,2008. [22] Hollingsworth,D.T.,Anderson,R.M.andFraser,C. HIV-1transmission,bystageof infection. Journal of Infectious Diseases,198(5):687–93,Sep2008. [23] Hong-Van Tieu, Christopher Murril, G.X.B.A.K. Sexual Partnering and HIV Risk amongBlackMenWhoHaveSexwithMen:NewYorkCity. JournalofUrbanHealth, 87(1):113–21,2009. [24] Koblin,B.A.etal.High-RiskBehaviorsAmongMenWhoHaveSexWithMenin6US Cities: BaselineDatafromtheEXPLOREStudy. American Journal of Public Health, 93(6):926–32,Jun2003. [25] Koblin,B.A.etal. RiskfactorsforHIVinfectionamongmenwhohavesexwithmen. AIDS,20(5):731–9,2006. [26] Lakdawalla, D., Sood, N. and Goldman, D. HIV Breakthroughs and Risky Sexual Behavior. Quarterly Journal of Economics,121(3):1063–102,2006. [27] Mammano,F.,Petit,C.andClavel,F.Resistance-associatedlossofviraltnessinhuman immunodeciencyvirustype1: Phenotypicanalysisofproteaseandgagcoevolutionin proteaseinhibitor-treatedpatients. Journal of Virology,72(9):7632–7,Sep1998. [28] Marks,G.,Crepaz,N.andJanssen,R.S.Meta-AnalysisofHigh-RiskSexualBehaviorin PersonsAwareandUnawareTheyareInfectedWithHIVintheUnitedStates: Impli- cationsforHIVPreventionPrograms. AIDS,39(4):446–53,2005. 261 [29] Morgan,D.etal. HIV-1infectioninruralAfrica: Isthereadiferenceinmediantime to AIDS and survival compared with that in industrialized countries? AIDS, 16(4): 597–603,2002. [30] Mosher,W.D.andA.Chandro,J.J. Sexualbehaviorandselectedhealthmeasures:Men andwomen15–44yearsofage,UnitedStates2002. Advance Data,(362),2005. [31] Osmond,D.H.etal.ChangesinprevalenceofHIVinfectionandsexualriskbehaviorin menwhohavesexwithmeninSanFrancisco:1997–2002. American Journal of Public Health,97(9):1677–83,Sep2007. [32] Quinn,T.etal. Viralloadandheterosexualtransmissionofhumanimmunodeciency virustype1. New England Journa of Medicine,342(13):921–9,2000. [33] RandCorporationandAgencyforHealthcareResearchandQuality(AHRQ). HIV Cost and Services Utilization Study (HSCUS), 1994-2000. Available at http://www. rand.org/health/projects/hcsus.html. [34] Schackman,B.R.etal.Thelifetimecostofcurrenthumanimmunodeciencyviruscare intheUnitedStates. Medical Care,44(11):990–7,Nov2006. [35] Smith,R.J.andBlower,S.M. Coulddisease-modifyingHIVvaccinescausepopulation- levelperversity? Lancet Infectious Diseases,4(10):636–9,Oct2004. [36] Smith,R.J.etal.EvolutionaryDynamicsofComplexNetworksofHIVDrug-Resistant Strains:TheCaseofSanFrancisco. Science,327:697–701,Feb52010. [37] Smith, R.J. et al. Supplemental Information for Evolutionary Dynamics of Complex NetworksofHIVDrug-ResistantStrains:TheCaseofSanFrancisco.Science,327:1–40, Feb52010. 262 [38] Smith,T.W.etal.GeneralSocialSurveys1972-2010.Availableathttp://www3.norc.org/ gss+website,2011. [39] Sykes,B.L.andPiquero,A.R. StructuringandRe-CreatingInequality:HealthTesting Policies,Race,andtheCriminalJusticeSystem. Annals of the American Academy of Political and Social Science,623(214):214–27,May2009. [40] The RAND Corporation. Rand California Population and Demographic Statistics. Availableathttp://ca.rand.org/stats/popdemo/popraceage.html,Nov2010. [41] US Census Bureau, Survey: American Community Survey. United States S0101. Age and Sex Data Set: 2006-2008 American Community Survey 3-Year Estimates. Avail- able at http://factnder.census.gov/servlet/STTable?_bm=y&-geo_id=01000US&- qr_name=ACS_2008_3YR_G00_S0101&-ds_name=ACS_2008_3YR_G00_,2010. [42] van de Vijver, D.A., Wensing, A.M. and Boucher, C.A. The Epidemiology of Transmission of Drug Resistant HIV-1. In Thomas Leitner T, Foley B, H.B.M.P.M.F.M.J.W.S.K.B.,editor,HIV Sequence Compendium 2006/2007,pages17– 36.LosAlamos,NM:TheoreticalBiologyandBiophysicsGroup,LosAlamosNational Laboratory,2007. [43] Vardavas,R.andBlower,S. TheemergenceofHIVtransmittedresistanceinBotswana: “WhenwilltheWHOdetectionthresholdbeexceeded?”. PLoS One,2(1):e152,2007. [44] Vetterling, W.T. Numerical recipes in C++ Example book. (C++). Cambridge, UK: CambridgeUniversityPress,2002. [45] Veugelers,P.J.etal.DeterminantsofHIVDiseaseProgressionamongHomosexualMen RegisteredintheTricontinentalSeroconverterStudy. American Journal of Epidemiol- ogy,140(8):747–58,1994. 263 [46] Wensing, A.M. and Boucher, C.A. Worldwide transmission of drug-resistant HIV. AIDS Reviews,5(3):140–55,2003. [47] Willaim D. Mosher, Anjani Chandra, J.J. Sexual Behavior and Selected Health Mea- sures: MenandWomen15-44YearsofAge,UnitedStates,2002. CDC Advance Data From VItal and Health Statitistics,(362):1–57,Sep2005. [48] Wilson,D.P. ModellingBasedonAustralianHIVNoticationsDataSuggestsHomo- sexual Age Mixing is Primarily Assortative. Journal of Acquired Immune Deficiency Syndrome,51(3):356–60,July12009. [49] Wilson, D.P., Kahn, J. and Blower, S.M. Predicting the epidemiological impact of antiretroviral allocation strategies in KwaZulu-Natal: The efect of the urban-rural divide. Proceedings of the National Academy of Sciences,103(38):14228–33,Sep2006. [50] Zaric,G.S.etal.TheEfectsofProteaseInhibitorsontheSpreadofHIVandtheDevel- opmentofDrug-ResistantHIVStrains:ASimulationStudy.Simulation,71(4):262–75, Oct1998. 264 AppendixC HIVEpidemicModelingand Costs-efectivenessEstimationApproaches C.1 Introduction Theaimofthissupplementaryappendixistoprovidedetailsonourmodelingapproachand our cost-efectiveness analysis methodology. A compartmental mathematical model of the humanimmunodeciencyvirus(HIV)/AIDSepidemicisdevelopedtoestimatethecumula- tiveincidenceof HIVinfectionsamongtheLosAngeles County(LAC)menwhohave sex withmen(MSM).Aneconomicmodelisdevelopedtoestimatethecostandefectivenessof variousHIVpreventionandtreatmentstrategiesinalifetimehorizonandfromaUSsocietal perspective. This appendix is organized as follows: Section C.2 describes the structure of our HIV epi- demic simulation model. Section C.3 describes the structure of the economic model. Sec- tionC.4brieypresentsourbasecaseresults. SectionC.5discussesourmethodfortheseries ofsensitivityanalysesconductedtoassesstherobustnessofourestimates.Finally,SectionC.6 describesourapproachforgatheringtheclinicalandpolicyevidencesupportingthemodels. C.2 HIVepidemicsimulationmodelstructure WeextendedtheHIVepidemicsimulationmodelof Soodetal.[80],whichsimulatedthe efectofTestingandTest-and-TreatintheLACMSMpopulation.Ourextensionofthemodel helpstosimulatetheefectofpre-exposureprophylaxis(PrEP)onthecumulativeincidence 265 ofHIVintheLACMSMpopulation.BesidestheadditionofPrEP,thecurrentmodelstrati- estheinfectedpopulationbyunawarenessofserostatusthroughtesting.Specically,weadd theSJ,PJ k ,EJ k , andAJ k compartments to denote respectively the susceptible individ- ualsawareoftheirseronegativestatus,theinfectedindividualsawareoftheirserostatusand whoareintheprimary,symptomatic,andAIDSstagesofHIVinfection.Wefurtheraddthe TPJ k compartmenttodenoteinfectedindividualsintheprimarystageofinfectionwhoare treatedwithART.Finally,weincludetheSPrEP,PPrEP k ,andIPrEP k compartments todenotesusceptibleandinfectedPrEPusersintheprimaryandsymptomaticHIVstages.In thesenotations,ktakesthevaluessorr,whichrepresentthedrug-sensitiveanddrug-resistant strata, respectively. Figure 4.1 illustrates the disease transmission and progression dynamics forthedrug-sensitivestratum. Thedrug-resistantstratumisonlyamirrorimageofthisdia- gram.MoredetailsonthetheoreticalfoundationsforthemodelareprovidedinAppendixB. TableC.1providesaglossaryofthenotationsandsymbolsusedinthisstudy,withtheirde- nitions. 266 TableC.1:Glossary–Denitionsofthemodelinputparameters. Parameters Denition Model Compartments S UninfectedMSMs,unawareofserostatus SJ UninfectedMSMs,awareofserostatus SPrEP UninfectedMSMs,awareofserostatusandtreatedwithPrEP P s , P r InfectedMSMs,unawareofserostatus(primarystage) PJ s , PJ r InfectedMSMs,awareofserostatus(primarystage) PPrEP s , PPrEP r InfectedMSMstreatedwithPrEP,unawareofserostatus(primarystage) TPJ s , TPJ r InfectedMSMstreatedwithART,awareofserostatus(primarystage) I s , I r InfectedMSMs,unawareofserostatus(asymptomaticstage) J s , J r InfectedMSMs,awareofserostatus(asymptomaticstage) IPrEP s , IPrEP r InfectedMSMstreatedwithPrEP,unawareofserostatus(asymptomaticstage) TJ s , TJ r InfectedMSMstreatedwithART,awareofserostatus(asymptomaticstage) E s , E r InfectedMSMsunawareofserostatus(symptomaticstage) EJ s , EJ r InfectedMSMsawareofserostatus(symptomaticstage) T s , T r InfectedMSMstreatedwithART,awareofserostatus(symptomaticstage) A s , A r InfectedMSMsunawareofserostatus(AIDSstage) AJ s , AJ r InfectedMSMsawareofserostatus(AIDSstage) TA s , TA r InfectedMSMstreatedwithART,awareofserostatus(AIDSstage) Demographic Parameters π 0 AnnualrateofinowofLACmalesintothesusceptiblecompartment π f MultiplicativefactorofthemaleentryratefortheMSMpopulation μ AnnualrateofnaturaldeathintheMSMpopulation Sexual Behavior Parameters τ C Reductioninsexualbehaviorowingtotestingandcounseling C mix Sexualmixingrate Per-Partnership Probabilities of HIV Transmission β Ps , β Pr InfectedMSMs,unawareofserostatus(primarystage) β PJs , β PJr InfectedMSMs,awareofserostatus(primarystage) β PPrEPs , β PPrEPr InfectedMSMstreatedwithPrEP,unawareofserostatus(primarystage) β TPJs , β TPJr InfectedMSMstreatedwithART,awareofserostatus(primarystage) Continued... 267 TableC.1:Glossary–Denitionsofthemodelinputparameters. Parameters Denition β Is , β Ir InfectedMSMs,unawareofserostatus(asymptomaticstage) β Js , β Jr InfectedMSMs,awareofserostatus(asymptomaticstage) β IPrEPs , β IPrEPr InfectedMSMstreatedwithPrEP,unawareofserostatus(asymptomaticstage) β TJs , β TJr InfectedMSMstreatedwithART,awareofserostatus(asymptomaticstage) β Es , β Er InfectedMSMsunawareofserostatus(symptomaticstage) β EJs , β EJr InfectedMSMsawareofserostatus(symptomaticstage) β TEJs , β TEJr InfectedMSMstreatedwithART,awareofserostatus(symptomaticstage) β As , β Ar InfectedMSMsunawareofserostatus(AIDSstage) β AJs , β AJr InfectedMSMsawareofserostatus(AIDSstage) β TAs , β TAr InfectedMSMstreatedwithART,awareofserostatus(AIDSstage) HIV Testing Parameters φ AnnualrateofHIVtesting(individualsnotusingPrEP) φ PrEP AnnualfrequencyofHIVtestingandotherrecommendedscreenings(PrEPinitiators) 1/ψ AveragedurationofserostatusidenticationforuninfectedindividualssincetheirlastHIVtest Annual Rates of Serostatus Identification ω S Uninfected ω SPrEP Uninfected,treatedwithPrEP ω P Infectedinprimarydiseasestage ω PPrEP Infectedinprimarydiseasestage,treatedwithPrEP ω I Infectedinasymptomaticdiseasestage ω IPrEP Infectedinasymptomaticdiseasestage,treatedwithPrEP ω E Infectedinsymptomaticdiseasestage ω A InfectedinAIDSstage Annual Rates of HIV Disease Progression ρ ProgressionratefromprimarytoasymptomaticHIV χ s , χ r ProgressionratefromasymptomatictosymptomaticHIV(untreated,unawareHIV+MSMs) ν s , ν r ProgressionratefromasymptomatictosymptomaticHIV(untreated,awareHIV+MSMs) θ ProgressionratefromasymptomatictosymptomaticHIV(ART-treatedMSMs) γ Es , γ Er ProgressionratetoAIDS(treatment-eligibleMSMsunawareoftheirHIV+status:CD4) γ EJs , γ EJr ProgressionratetoAIDS(treatment-eligibleMSMsawareoftheirHIV+status:CD4) Continued... 268 TableC.1:Glossary–Denitionsofthemodelinputparameters. Parameters Denition γ Ts , γ Tr ProgressionratetoAIDS(ART-treatedMSMs) HIV/AIDS-Related Mortality γ AJs , γ AJr AnnualHIV-relateddeathrate,untreatedAIDS γ TAs , γ TAr AnnualHIV-relateddeathrate,ART-treatedAIDS Annual Rates of ART and PrEP Initiation σ ARTinitiationrate,symptomaticstage σ SPrEP PrEPinitiationrate,uninfected σ PPrEP PrEPinitiationrate,primarystage σ IPrEP PrEPinitiationrate,asymptomaticstage σ TPJ ARTinitiationrate,primarystage σ TJ ARTinitiationrate,asymptomaticstage σ A ARTinitiationrate,AIDSstage Treatment Adherence Parameters g AnnualrateofARTdiscontinuation,symptomaticstage g SPrEP AnnualrateofPrEPdiscontinuation,uninfected g PPrEP AnnualrateofPrEPdiscontinuation,primarystage g IPrEP AnnualrateofPrEPdiscontinuation,asymptomaticstage g TPJ AnnualrateofARTdiscontinuation,primarystage g TJ AnnualrateofARTdiscontinuation,asymptomaticstage g A AnnualrateofARTdiscontinuation,AIDSstage Reduction in Sexual Infectivity Due to ART and PrEP τ ART ReductioninsexualinfectivityowingtoART τ PrEP ReductioninsexualinfectivityowingtoPrEP Drug Resistance Parameters r AnnualrateofacquiredMDRinARTusers r PrEP AnnualrateofacquiredMDRinPrEPusers h r MDRtransmissibilitymultiplicativefactor q AnnualrateofmutationHIVfromtheacquired-resistantstraintothedrug-sensitivestrain Screening Parameters Pre A Antibodytestsensitivity(pre-seroconversion) Continued... 269 TableC.1:Glossary–Denitionsofthemodelinputparameters. Parameters Denition Post A Antibodytestsensitivity(post-seroconversion) ξ A Antibodytestspecicity NAAT NucleicAcidAmplicationtestsensitivity ξ NAAT NucleicAcidAmplicationtestspecicity 270 C.2.1 Systemofordinarydiferentialequations. Thediseasedynamicsarecapturedbythefollowingsystemofordinarydiferentialequations (ODEs).TheyaresimilartoEquationB.1-EquationB.15inAppendixB,withtheadditionof theequationsthatcapturetheefectofthePrEPpolicyandthoseoftheserostatusawareness stratication. 271 ˙ S = (π +g PrEP SPrEP +ψSJ)− (μ +λ s +λ r +ω S )S (C.1) ˙ SJ = (ω S S +ω SPrEP SPrEP )− (μ +ψ +λ Js +λ Jr +σ SPrEP )SJ (C.2) ˙ SPrEP = σ SPrEP SJ− (μ +λ sPrEP +λ rPrEP +gSPrEP +ω SPrEP )SPrEP (C.3) ˙ P s = (λ s S +λ Js SJ +g PPrEP PPrEP s )− (μ +ρ +σ PPrEP +ω P )P s (C.4) ˙ P r = (λ r S +λ Jr SJ +g PPrEP PPrEP r )− (μ +ρ +σ PPrEP +ω P )P r (C.5) ˙ PJ s = (ω P P s +ω PPrEP PPrEP s )− (μ +ρ)PJ s (C.6) ˙ PJ r = (ω P P r +ω PPrEP PPrEP r )− (μ +ρ)PJ r (C.7) ˙ PPrEP s = (λ sPrEP SPrEP +σ PPrEP P s )− (μ +g PPrEP +ρ PrEP +ω PPrEP +r PrEP )PPrEP s (C.8) ˙ PPrEP r = (λ rPrEP SPrEP +σ PPrEP P r +r PrEP PPrEP s )− (μ +g PPrEP +ρ PrEP +ω PPrEP )PPrEP r (C.9) ˙ I s = (ρP s +g IPrEP IPrEP s )− (μ +χ s +ω I +σ IPrEP )I s (C.10) ˙ I r = (ρP r +g IPrEP IPrEP r )− (μ +χ r +ω I +σ IPrEP )I r (C.11) ˙ J s = (ω I I s +ρPJ s +g TJ TJ s +ω IPrEP IPrEP s )− (μ +ν s +σ TJ )J s (C.12) ˙ J r = (ω I I r +ρPJ r +g TJ TJ r +ω IPrEP IPrEP r )− (μ +ν r +σ TJ )J r (C.13) ˙ IPrEP s = (ρ PrEP PPrEP s +σ IPrEP I s )− (μ +g IPrEP +ω IPrEP +r PrEP )IPrEP s (C.14) 272 ˙ IPrEP r = (ρ PrEP PPrEP r +σ IPrEP I r +r PrEP IPrEP s )− (μ +g IPrEP +ω IPrEP )IPrEP r (C.15) ˙ E s = (χ s I s +qE r )− (μ +γ Es +ω E )E s (C.16) ˙ E r = χ r I r − (μ +γ Er +ω E +q)E r (C.17) ˙ EJ s = (ν s J s +gT s +ω E E s +qEJ r )− (μ +γ EJs +σ)EJ s (C.18) ˙ EJ r = (ν r J r +gT r +ω E E s )− (μ +γ EJr +σ +q)EJ r (C.19) ˙ TJ s = σ TJ J s − (μ +g TJ +θ +r TJ )TJ s (C.20) ˙ TJ r = (σ TJ J r +r TJ TJ s )− (μ +g TJ +θ)TJ r (C.21) ˙ T s = (σEJ s +θTJ s )− (μ +γ Ts +g +r)T s (C.22) ˙ T r = (σEJ r +θTJ r +rT s )− (μ +γ Tr +g)T r (C.23) ˙ A s = (γ Es E s +qA r )− (μ +γ As +ω A )A s (C.24) ˙ A r = γ Er E r − (μ +γ Ar +ω A +q)A r (C.25) ˙ AJ s = (ω A A s +γ EJs EJ s +g A TA s +qAJ r )− (μ +γ AJs +σ A )AJ s (C.26) ˙ AJ r = (ω A A r +γ EJr EJ r +g A TA r )− (μ +γ AJr +σ A +q)AJ r (C.27) ˙ TA s = (γ Ts T s +σ A AJ s )− (μ +γ TAs +g A +r)TA s (C.28) ˙ TA r = (γ Tr T r +σ A AJ r +rTA s )− (μ +γ TAr +g A )TA r (C.29) 273 Webeginthesimulationinyear2000with16,6307MSMinLACandweestimatethat18.1% of the men are HIV-positive, and that nearly 25.0% of the infected ignore their serostatus (TableC.2). TableC.2:InitialpopulationofMSMinLosAngelescountyinyear2000 Variable Count(%) LACMSMpopulation 176,971 Totalinfected,n(%ofLACMSMpopulation) 30,137(17.0) Infectedunaware,n(%ofTotalinfected) 7,534(25.0) InfectedtreatedwithART,n(%ofTotalinfected) 12,199(40.5) TotalAIDS,n(%ofTotalinfected) 13,283(44.1) AIDStreatedwithART,n(%ofTotalAIDS) 10,664(80.3) In our base case analysis, we consider four standalone intervention strategies: (i) the status quostrategyconsistingoftestingtheMSMpopulationatcurrentrates,andinitiatingART onlyinthetreatment-eligibleinfectedMSMawareoftheirserostatus(i.e.thosewithCD4≤ 500whoareawareoftheirdiseasestatus); (ii)thetestingstrategy, whichconsistsoftesting moreMSMeachyear,andinitiatingARTonlyinthetreatmenteligibleMSM;(iii)thetest- and-treatstrategywhichconsistsoftestingtheMSMpopulation,andimmediatelyinitiating treatmentinnewlyidentiedHIV+MSM,regardlessoftheirCD4count;andnally(iv)the PrEPstrategy, whichconsistsofHIVtestingevery6monthsandPrEPstartevery4yearin uninfectedMSM. C.2.2 Populationdata The demographic data used to populate the model were obtained from the LAC Depart- mentofPublicHealthSurveillancereportsandtheRANDCaliforniadatabase(TableC.3). Themethodologiesusedtoestimatethecompartment-specicpopulationsizes,aswellasthe 274 uninfected and infected populations, are described in Appendix B. US life tables for both HIV-positive and uninfected individuals (Table C.4) were used to estimate the average life expectanciesforthehealthyanduninfectedpopulations. TableC.3:CountofLACmalepopulationbyyearandagegroup AgeGroup,y Year 2009 2010 2011 15-19 367,885 383,496 374,014 20-24 356,308 384,052 393,682 25-29 411,296 386,476 392,753 30-34 389,017 361,299 367,142 35-39 373,417 355,324 347,877 40-44 360,994 357,731 360,816 45-49 356,694 349,242 347,199 50-54 315,382 324,416 328,622 55-59 257,673 271,045 279,500 60-64 198,635 214,313 224,038 65-69 139,659 149,235 155,007 70-74 106,610 110,095 112,831 75-79 81,904 82,513 83,700 80-84 59,908 61,286 62,139 >85 50,108 53,387 55,934 275 Table C.4: Average life expectancies for healthy and HIV-positive male populations in the UnitedStates. Uninfected HIV-positive Age(y) LifeExpectancy(y) AgeGroup(y) LifeExpectancy(y) 0 76.30 0-1 74.30 1 75.80 1-5 73.87 5 71.90 5-10 69.98 10 66.90 10-15 65.04 15 62.00 15-20 60.11 20 57.20 20-25 55.38 25 52.50 25-30 50.74 30 47.90 30-35 46.05 35 43.20 35-40 41.35 40 38.60 40-45 36.72 45 34.00 45-50 32.21 50 29.60 50-55 27.86 55 25.50 55-60 23.67 60 21.50 60-65 19.74 65 17.80 65-70 16.11 70 14.30 70-75 12.80 75 11.00 75-80 9.89 80 8.20 80-85 7.44 85 5.90 85-90 5.47 90 4.10 90-95 3.95 95 2.90 95-100 2.82 100 2.10 >100 2.03 C.2.3 Age-weightedaveragelifeexpectancyfor15-to65-year-oldMSM. Weestimatedtheage-weightedaveragelifeexpectanciesforhealthyandinfectedMSMinLAC accordingtothefollowingformula: e 14−65y = X g X g e g X g X g , (C.30) wheree 15−65y denotestheestimatedage-weighedaveragelifeexpectancyforthe15-to65-year- oldMSM,andX g ande g indicatethenumberofindividualsandtheaveragelifeexpectancy inagegroupg,respectively.Inourestimationapproach,weassumedthatthelifeexpectancy 276 fortheMSMequalsthatofanymaleinLAC.Inotherwords,weassumedsexualorientation hasnoefectonlifeexpectancy. WealsousedYear2011data;thechoiceofyearhadnegligible efectontheaveragelifeexpectancyestimates. C.2.4 Modelinputparametervalues Themodelinputparametervaluesanduncertaintyrangeswerederivedfromthepublished literature,followingasystematicreviewapproach(seeSection5forthedescriptionofliterature reviewmethodology). Wecollectedestimatesontheepidemic,cost,efectiveness,andpolicy variables. Inwhatfollows,wedescribeourmethodologyforestimatingtheepidemicmodel inputparametervaluesandranges. Consistentwiththeepidemiologyliterature,therateof occurrenceofaneventwasdenedastheinverseoftheaveragedurationinthediseasestate beforetheeventoccurs[8].Equivalently,wedenedtheaveragedurationT s indiseasestates tobetheinverseofthesumoftheoutowratesfromthatstate.3Forexample,whentherate parameterω SPrEP takesthevalueof4.8peryear,thisindicatesthatittakes,onaverage,1/4.8 years,or2.5months,foranindividualintheSPrEPcompartmenttobetestedandtransition totheSJcompartment.AcompletelistofparametersnotassociatedwiththePrEPextension isalsoavailableinChapter3andin Soodetal.[80]. C.2.4.1 Methodforestimatingratesandprobabilities Weassumedthattheoutowsfromanygivencompartmentinourmodelhadtheexponen- tial distribution in times of outow, consistent with prior literature [79], and modeled the transitionprobabilitiespasfollows: p = h(r,t) =P (t ∗ ≤t,r) = Z t 0 f(s,r)ds = 1−e −rt , (C.31) 277 wheref(t,r) =re −rt istheprobabilitydensityfunctionoftheexponentialfunction,t≥ 0 isthetime, andr istherateparameter. Torecovertherateparameterr, weappliedthelog transformation[79]: r = g(p,t) =−t −1 ln(1−p) (C.32) C.2.4.2 Rateofentryofthesusceptiblepopulation FollowingtheapproachinAppendixBandin Soodetal.[81],wecalculatedtheannualrateof inowofnewsusceptibleindividualsintothesusceptiblecompartmentSasaproductofthe totalMSMinowandtheentryratemultiplicativefactorforthe15-to65-year-oldpopulation, usingtheformulainEquationC.33andtheestimatesinTableC.5. π = π 0 π f (C.33) TableC.5:ParametersspecifyinginowofnewMSMinthemodel. Parameters Value Range Reference π 0 :TotalMSMannualinow 3,484 - [80,91] π f :Entryratemultiplicativefactor 1.0324 0.9022-1.0979 [80,91] C.2.4.3 PrEPecacy TheecacyandlevelsofprotectionofPrEPrelatestothelevelofadherencetothetreatment regimen.Althoughtheheterogeneityinproductuseisstillunderinvestigation,currentresults suggestthatPrEPecacydependsonthelevelofadherencetothetreatmentregimen[3,12, 20,22,29,34,45,58,86,93]. Inthesestudies,adherencewasmeasuredbytheconcentration oftenofovirintheblood,whichisoneobjectivemarkerofadherence[20]. 278 TableC.6: Relationshipbetweenadherence toPrEP,PrEPecacy,andlevelsofprotection ofPrEP Study Size Duration b Adherence Ecacy Protection Value Range Value Range PinP-TDF/FTC[3] 4,758 36 0.81 0.75-0.83 0.75 0.55-0.87 0.90 PinP-TDF[3] 4,758 36 0.81 0.67-0.83 0.67 0.44-0.81 0.80 TDF2-TDF/FTC[86] 1,219 180 0.80 - 0.63 0.22-0.83 0.70 BKK-TDF[12] 2,413 84 0.67 - 0.49 0.10-0.72 0.70 iPrEX-TDF/FTC[29] 2,499 132 0.51 - 0.44 0.15-0.63 0.90 a Measuredbydrugconcentrationintheblood. b Numberofweeks. TableC.6suggeststhatthereexistsgreatvariabilityinadherencelevelsacrossstudies. There also appears to be a strong correlation between adherence to PrEP and the ecacy of PrEP[20].Followingthisbodyofliterature,wemodeltherelationshipbetweenadherenceto PrEP,asmeasuredbytenofovirbloodlevelsinnon-seroconverters,andPrEPecacy,inorder todynamicallycapturetheefectofadherenceonHIVprevalenceandcosts. Basedondata pointsfromrecentRCTstudies,wendtherelationshipbetweenPrEPecacyandadherence toPrEPtobeapproximatedby: τ PrEP (κ PrEP ) = 4.2709κ 2 PrEP − 4.7793κ PrEP + 1.768. (C.34) Forthepurposesofourstudy,weadoptedaconservativeapproachandusedtheadherence estimatefromtheiPrEXstudyandtherangeofadherencelevelsofallthereviewedstudies (TableC.7). Real-worldadherencelevelsremainopenforfuturestudiesbutarelikelytobe lowerthantheadherencelevelsobservedinexperimentalenvironments. 279 TableC.7:AdherencetoPrEP:Estimatesbasedontenofovirconcentrationinbloodofnon- seroconverters. Parameters Value Range Reference κ PrEP :AdherencetoPrEP 0.5100 0.3000-0.8300 [29] C.2.4.4 Calculationoftheperpartnershiptransmissibilityparameters:βparameters FollowingtheapproachinChapter3(Soodetal.[81]),wecalculatedtheperpartnershiptrans- missibilityparametervaluesforthedrug-resistantstratumasfollows: β kr = β ks h r , ∀k∈{P,I,EJ,T,AJ,TA}, (C.35) wheretheβ ks parametersrepresentthetransmissibilitypersexualpartnershipwithaninfected individualinthedrug-sensitivestratumincompartmentk,andh r isamultiplicativefactor accountingfortheefectofdrugresistance(TableC.8andTableC.9). TableC.8:Perpartnershiptransmissibilities Parameters Value Range Reference β Ps :Primarystage 0.1089 0.0364-0.5296 [80],Chapter3 β Is :Asymptomaticstage(unaware) 0.0505 0.0180-0.1900 [80],Chapter3 β Js :Asymptomaticstage(aware) 0.0185 0.0048-0.0837 [80],Chapter3 β EJs :Untreatedsymptomaticstage(aware) 0.0170 0.0068-0.0839 [80],Chapter3 β Ts :ART-treatedsymptomaticstage 0.0159 0.0002-0.0375 [80],Chapter3 β AJs :UntreatedAIDSstage(aware) 0.0677 0.0207-0.2398 [80],Chapter3 β TAs :ART-treatedAIDSstage 0.0275 0.0024-0.0653 [80],Chapter3 TableC.9:MDRtransmissibilitymultiplicativefactor Parameters Value Range Reference h r :MDRtransmissibilitymultiplicativefactor 0.1232 0.1000-0.1756 [77,80],Chapter3 280 Inaddition, wedenedβ PPrEPs andβ IPrEPs todenotetheperpartnershiptransmissibili- tiesintheprimaryandasymptomaticunawarestagesforindividualsreceivingPrEP;wealso denedβ PJs ,β TJs ,β Es ,andβ As todenotetheperpartnershiptransmissibilitiesinthepri- mary aware stage, the treated asymptomatic HIV stage under Test-and-Treat, the unaware anduntreatedsymptomaticHIVstage,andtheunawareuntreatedAIDSstage,respectively. TheseparameterswereestimatedusingthevaluesinTableC.8andTableC.9,followingthe equationsbelow: β PPrEPs = (1−τ PrEP )· (1−τ C )·β Ps (C.36) β IPrEPs = (1−τ PrEP )· (1−τ C )·β Is (C.37) β PJs = (1−τ C )·β Ps (C.38) β TJs = (1−τ ART )·β Js (C.39) β Es = (1−τ C )·β EJs (C.40) β As = (1−τ C )·β AJs . (C.41) UsingtheestimatesinTableC.8andthoseestimatedviaEquationC.36-EquationC.41,we estimatedthetransmissibilityratesfollowingtheformulaeinEquationB.17-EquationB.18 inAppendixB[81]: λ s (t) = C mix · P Xs β Xs X s (t) N(t) (C.42) λ r (t) = C mix · P Xr β Xr X r (t) N(t) , (C.43) 281 whereN(t) = X Xs X s (t) + X Xr X r (t),t denotes time, andC mix is the sexual mixing rate (Table C.10). This approach assumes proportional sexual mixing. We further calculate the transmissionratesforthesusceptiblepopulationreceivingPrEPasfollows: λ sPrEP (t) = λ s (t)· [1−τ(κ PrEP )] (C.44) λ rPrEP (t) = λ r (t)· [1−τ(κ PrEP )], (C.45) whereτ(·)mapsadherencetoecacyasinEquationC.34. TableC.10:ParameterSpecifyingContactRange Parameters Value Range Reference C mix :Sexualmixingrate 4.5046 2.2798-8.4753 [48,49,65,77,78,80,89,97] TableC.11:ReductionsinsexualinfectivityandriskybehaviorsowingtoART,PrEPandcoun- seling Parameters Value Range Reference τ C :Reductioninsexualbehaviorowingtotestingandcounseling 0.2000 0.0000-0.5000 [43,56,57,72,74,83] τ ART :ReductioninsexualinfectivityowingtoART 0.9000 0.5000-0.9900 [1,14,15,43,44,55,56,59,70,74,85, 95,98] τ PrEP :ReductioninsexualinfectivityowingtoPrEP 0.9200 0.05000-0.9900 [4,43] C.2.4.5 Testsensitivityandspecicity TableC.12reportsthesensitivityandspecicityestimatesandrangesofvarioustestsusedin HIV screening, the monitoring of viral load, CD4 count, and resistance to treatment regi- mens. 282 TableC.12:Testsensitivityandspecicity. Parameters Value Range Reference Antibodytest Pre A :Preseroconversionsensitivity 0.5819 0.2650-0.8990 [4,66,97] Post A :Postseroconversionsensitivity 0.9895 0.9740-0.9990 [4,66,97] ξ A :Specicity 0.9936 0.9770-1.0000 [4,66,97] Nucleicacidamplicationtest(NAAT) NAAT :Sensitivity 0.9527 0.9260-0.9740 [4,66,97] ξ NAAT :Specicity 0.9863 0.9690-0.9960 [4,66,97] ARTgenotypicresistancetest ART R :Sensitivity 0.8670 0.5630-0.9820 [33] ξ ART R :Specicity 0.8845 0.6210-0.9800 [33] PrEPgenotypicresistancetest PrEP R :Sensitivity 0.8620 0.5630-0.9820 [33] ξ PrEP R :Specicity 0.8650 0.6210-0.8920 [33] FollowingthetestingalgorithmrecommendedbytheCentersforDiseaseControlandPre- vention(CDC)[10],andusingtheestimatesinTableC.12. Weestimatedtheprobabilitiesof truepositive,truenegative,falsepositiveandfalsenegativetestresultsasfollows: TP k = P (Test+|HIV+) = ( k A ) 2 + k A NAAT (C.46) FP = P (Test+|HIV-) = (1−ξ A )ξ A (1−ξ NAAT ) + (1−ξ A ) 2 (C.47) TN = P (Test-|HIV-) =ξ A + (1−ξ A )ξ A ξ NAAT (C.48) FN k = P (Test-|HIV+) = (1− k A ) + k A (1− k A )(1− NAAT ), (C.49) wherethesuperscriptkdenotesthepre-andpost-seroconversionperiods.Weassumedazero probabilityoftruenegativeinthepost-seroconversionperiod. 283 C.2.4.6 Ratesofidenticationthroughtesting CDC recommends that all sexually active MSM be tested at least once a year for HIV and otherSTIs[28]. Forthepurposesofthisanalysis,wedenetheprobabilityofbeingidenti- edthroughtestingastheprobabilityoftruepositivetimestheprobabilityoftesting. The probabilityoftestingiscalculatedbyapplyingEquationC.31tothetestingrateφ(TableC.13). TherateoftestingisthendeterminedbyapplyingEquationC.32totheprobabilityofbeing identied.TheseexpressionsarecapturedbelowbyEquationC.50-EquationC.57: ω S = g(h(φ S−SJ , 1)·TN,t) (C.50) ω SPrEP = g(h(φ SPrEP−SJ , 1)·TN·h(g SPrEP , 1),t) (C.51) ω P = g(h(φ P−PJ , 1)·TP Pre ,t) (C.52) ω PPrEP = g(h(φ PPrEP−PJ , 1)·TP Pre ,t) (C.53) ω I = g(h(φ I−J , 1)·TP Post ,t) (C.54) ω IPrEP = g(h(φ IPrEP−J , 1)·TP Post ,t)·h(g IPrEP , 1),t) (C.55) ω E = g(h(φ E−EJ , 1)·TP Post ,t) (C.56) ω A = g(h(φ A−AJ , 1)·TP Post ,t), (C.57) whereφ S−SJ = φ P−PJ = φ P−PPrEP = φ I−J = φ I−IPrEP = φ E−EJ = φ A−AJ = φ, andφ SPrEP−SJ =φ PPrEP−PJ =φ IPrEP−J =φ PrEP . TableC.13:RateofHIVscreeninginthepopulation Parameters Value Range Reference φ:HIVscreeningrateinunawareHIV+(non-AIDS)individuals 0.2269 0.1799-0.4524 [80,81] φ PrEP :HIVscreeningrateinPrEPusers 4.0000 - [80] 284 C.2.4.7 ARTcoverageandinitiationrates Weestimatethateachyear,nearlyallMSM(99.9%;range: 49-100%)intheAIDSstagewho are aware of their disease status will initiate ART. We also estimate nearly 33% (range: 28- 93%)ofeligibleMSMwithknownHIVstatus(i.e. CD4≤ 500cellsperLandintheEJ compartment)willinitiateARTannually. Wefurtherestimatethat50%(range: 25-86%)of MSMwithknownHIVstatusinitiateARTearly(i.e.atCD4> 500cellsperL). C.2.4.8 RatesofPrEPinitiation:σ PrEP parameters WeassumeinourmodelthatPrEPisoferedtohigh-riskMSMonly,becausecurrentguide- linesrecommendPrEPonlyforthatpopulation.Werecognizethatinpractice,somelow-risk MSM might be using PrEP, although that population is likely to be very small in compar- ison the targeted population. Given that PrEP has recently been approved and has not yet beenwidelyadopted,itisdiculttoestimateitsadoptionrate.However,recentstudieshave examinedPrEPacceptabilitybyMSMinvariouslocations,includingCalifornia[16,51,53,62]. In general, these studies report high willingness to use PrEP in the MSM population, with estimatesrangingfrom50%to90%. OtherstudieshaveestimatedPrEPuseinexperimental settings, andreportedhighPrEPuptake[17,52? ]. However, statedwillingness-to-useesti- matesanduptakeratesobservedinexperimentalsettingslikelyoverestimateactualuse,and many physicians and MSM lack knowledge about PrEP [30, 60]. For example, recent data indicatethat10%ofMSMinSanFranciscohadadoptedPrEPby2014[30,35].Therefore,we usethisconservativePrEPadoptionestimateinourbasecaseanalysis(TableC.14).Inscenario analyses,thisproportionischanged.UsingthevaluesinTableC.14,weestimatedtheratesof PrEPinitiationaccordingtothefollowingequations: σ SPrEP = g(σ PrEP , 1) (C.58) σ PPrEP = g(h(ω P , 1)σ PrEP , 1) (C.59) 285 σ IPrEP = g(h(ω I , 1)σ PrEP , 1), (C.60) whereh(·)andg(·)aredenedasinEquationsC.31-C.32. TableC.14:ProbabilityofPrEPinitiation Parameters Value Range Reference φ: AnnualprobabilityofPrEPinitiationin susceptiblepopulation 0.094 0.085-0.104 [28,51,53,62];calculated C.2.4.9 ProgressionratefromtreatedasymptomatictotreatedsymptomaticHIV WeestimatedtherateofprogressionfromthetreatedasymptomaticHIVstagetothetreated symptomaticHIVstageasfollows: θ = γ Ts (θ f − 1) , (C.61) whereθ f andγ Ts aredenedinTableC.16. Inthisexpression,itisassumedthattheaverage durationinthetreatedasymptomaticHIVstageequalstheaveragedurationfromthetreated symptomaticHIVstagetothetreatedAIDSstage,minustheaveragedurationinthetreated symptomaticstage. TableC.15:DurationofsymptomaticHIVtreatedwithART. Parameters Value Range Reference θ f : Average duration from the treated symp- tomaticHIVstagetothetreatedAIDSstage 18 12-30 [15, 30, 35, 43,97] γ Ts :ProgressionratetoAIDSinART-treatedindi- viduals 0.0777 0.0468-0.0875 [80,81] 286 C.2.5 Modelcalibration Tocalibratethemodelparameters,weusedaLatinhypercubesampling(LHS)design,which sampled sets of parameters based on probability distributions within uncertainty ranges. We then identied the set of parameters that produced a simulation projection that most closelymatchedtheepidemiccharacteristicsinLAC,asreportedinthesurveillancedata.The methodologyforthemodelcalibrationisfullydescribedinAppendixBandinSoodetal.[81]. FigureC.1representsourreproductionofthecalibrationresultsofthemodel(alsoshownin Chapter3andinSoodetal.[80]).Themodelwaswellcalibratedtothereportedsurveillance data. C.2.6 ProjectedHIVincidence Usingthismodel,wecomputedtheprojectedHIVincidencewithvariousHIVprevention strategies.FigureC.2depictstheprojectedcumulativeincidenceofHIVfortheprogramdura- tionforthebasecaseandselectedstrategiesontheecientfrontier. Ascanbenoticed,the Test-and-TreatandPrEPstrategiesyieldthelargestreductionsinnewHIVincidenceoverthe program duration. Enhancing these strategies with more aggressive testing, and ART and PrEPcoveragefurtherincreasestheirepidemiologicalimpacts. 287 FigureC.1:Modelcalibrationresult:simulatedvs.surveillancedata. 2000 2002 2004 2006 2008 2010 10 15 20 25 30 35 Y ear Number of Living Cases ('000s) HIV/AIDS Aware AIDS Non−AIDS Aware Surveillance Simulation ThecalibratedmodelaccuratelypredictstrendsinthecumulativeincidencesofAIDSandnon-AIDSHIV-aware casesinLACduringthe2000-2010period(Chapter3;[80]) 288 FigureC.2:ProjectedcumulativeHIVincidenceforthestatusquo,andbaselineTesting,Test- and-Treat,andPrEPscenarios. 0 20 40 60 80 100 Y ear Cumulative HIV Incidence ('000s) 99,874 95,542 52,115 40,992 22,573 2013 2018 2023 2028 2033 Status Quo, SQ (Test 4.4 y + ART 2.5 y at CD4 ≤ 500; strategy 1) Test−and−treat, TT (SQ + Immediate Early ART at CD4>500; strategy 2) Enhanced TT (TT + Test 6 mo; strategy 6) PrEP (TT + Test 6 mo + PrEP 4 y; strategy 7) Enhanced PrEP (PrEP + Test 3 mo + Immediate PrEP; strategy 13) C.3 Economicmodelstructure C.3.1 Calculationofcosts C.3.1.1 Costestimatesfromtheliteratureandgovernmentfeeschedules We obtained disease-state specic medical cost data from the published literature and gov- ernment fee schedules. The cost data and relevant sources are summarized in Table C.16. All cost estimates were converted into 2013 US dollars using the appropriate ination fac- tor [61]. We included in this analysis the costs related to goods and services used to deliver medicalcare.Theseincludephysicianvisits;drugs(ART,PrEP);managementofopportunis- ticinfections;testsforHIV(eg,enzymeimmunoassay[EIA],enzyme-linkedimmunosorbent 289 assay[ELISA];rapidHIVtest,conrmatorytestingusingNAAT);STIs;bloodureanitrogen concentrationandserumcreatininelevels; CD4countandviralloadmonitoring; resistance test;andpretestandposttestcounselingandlinkagetocare. Weexcludeddirectnonmedical costsbecausetheyaresmallandsimilaracrossthealternativeinterventionsbeingassessed[61]. Directnonmedicalcostsweredenedasthosebornebythesubjectbeyondthehealthcareset- ting(eg,transportationcosts,otherout-of-pocketexpenses,andresourcesfromotheragen- cies). Wecalculatedtheindirectcostsassociatedwithinformalcaregiversupportandunpaid helpbyfamilyandfriendsusingestimatedaveragehomehealthcarecosts[23],aswellasesti- matesofAIDSpatients’homecareutilization[9,24],weightedbythenationalaveragehourly compensationratesofhomehealthandpersonalcareaides(2010US$11.63/hour)[69]. We excluded changes in caregivers’ quality of life due to their caregiving activities, because the denominatorshouldexclusivelyincludehealth-related(notcare-related)qualityoflife[39]. Wealsoexcludedfromthenumeratorotherindirectcostsrelatedtothevalueoftheindivid- ual’sforgoneproductivityowingtotheillness-relatedmorbidityandmortality,becausethey arealreadycapturedinthedenominator(quality-adjustedlifeyears[QALYs])[19]. 290 TableC.16:Costparameters. Parameters Value Range Reference Annual HIV-related health care costs ($) AcuteHIV a 30 10-500 [19,39,42,75,97] AsymptomaticHIV,untreated b 4,130 3,000-6,000 [5,43,94,97] SymptomaticHIV,untreated b 6,934 5,000-9,000 [5,43,94,97] SymptomaticHIV-treatedwithART,excludingARTcosts b 6,181 5,000-7,000 [5,43,94,97] AIDS,untreated b,c 21,863 15,000-25,000 [5,7,23,26,27,40,43,61,92, 97] AIDS-treatedwithART,excludesARTcosts b,c 9,950 6,000-17,000 [23,26,27,40,43,54,61,72, 92,94,97] Annualcostofantiretroviraltherapy(ART) b 15,000 13,520-17,109 [7,40,43,54,72] Cost of PrEP (cost per test, refill, or visit; $) Truvada(tenofovir,TDF/emtricitabine,FTC):30-dsupply b 776 672-925 [40,68,76] Sexuallytransmittedinfection(STI)testing:costpertest b 54 25-75 [87,97] BUN d concentrationandserumcreatinineleveltesting:costpertest b 23 10-40 [87,97] Physicianvisit:costpervisit b 100 10-200 [87,97] Cost of HIV testing (cost per test; $) Costofinitialtest:3rd/4thgenerationtest a,e 19 9-45 [13] CostofconrmatorytestingorHIVRNAtest a,f 48 16-158 [13] CostofCD4cellcountmonitoring a,g 52 10-87 [13] CostofHIVgenotypetest h 177 54-239 [13] Cost of counseling (cost per visit; $) Pretestcounseling b 13 0-100 [13,25,97] PosttestcounselingforHIV-negativeindividuals b 7 0-50 [13,25,97] Posttestlinkage/counselingforHIV-positiveindividuals b 14 0-100 [13,25,97] Othercostsandcost-relatedparameters CostofHIVdiagnosis($) b 500 125-1,200 [87,97] Annualcostdiscountrate 0.03 0.0000-0.0500 [15,19,67,97] 2013to2010inationfactor 1.0900 - [18] a USdollarsexpressedin2010value.; b USdollarsexpressedin2013value.; c Inclusiveofinformalsupportcosts.; d Bloodureanitrogen; e EIA/ELISA(CPT:86703,G0432,G0433,87389)orrapidHIVtest(CPT:G0345); f NAATtestforHIVRNA(CPT:87535); g CPT:86359,86360,86361; h CPT:87901and87906 291 C.3.1.2 Totallifetimecosts Weestimatedthediscountedcostsfortheentirepopulationasthesumoftheannualhealth carecostsforallindividualsplusthecostoftheprogram,whichincludesthecostsofantiretro- viralorPrEPtreatment,andthetotalscreeningandcounselingcostsfortheinterventiondura- tion.Specically,weestimatedthecostsas: Costs = C HC +C ART +C PrEP +C CD4/RNA +C HIVT +C HIVD + C PrEPM +C V +C RT , (C.62) where C HC = Z +∞ 0    e −rt X ∀s∈S 1 C HC s X s (t)    dt (C.63) C ART = Z +∞ 0    e −rt X ∀s∈S 2 C ART s X s (t)    dt + Z T 0    e −rt X ∀s∈S 3 C ART s X s (t)    dt (C.64) C PrEP = Z T 0    e −rt X ∀s∈S 2 C PrEP s X s (t)    dt (C.65) C CD4/RNA = Z +∞ 0    e −rt X ∀s∈S 5 C CD4/RNA s X s (t)    dt + Z T 0    e −rt X ∀s∈S 6 C CD4/RNA s X s (t)    dt (C.66) C HVT = Z +∞ 0    e −rt X ∀s∈S 4 ∪S 6 C HVT− s X s (t)    dt + Z +∞ 0    e −rt X ∀s∈S 7 C HVT− s +ω s C HVT+ s X s (t)    dt + Z T 0    e −rt X ∀s∈S 8 C HVT− s +ω s C HVT+ s X s (t)    dt (C.67) 292 C HIVD = Z +∞ 0    e −rt X ∀s∈S 7 C HIVD s X s (t)    dt + Z T 0    e −rt X ∀s∈S 8 C HIVD s X s (t)    dt (C.68) C PrEPM = Z +∞ 0    e −rt X ∀s∈S 4 C PrEPM s X s (t)    dt (C.69) C V = Z +∞ 0    e −rt X ∀s∈S 5 C V s X s (t)    dt + Z T 0    e −rt X ∀s∈S 4 ∪S 6 C V s X s (t)    dt (C.70) C RT = Z +∞ 0    e −rt X ∀s∈S 5 C RT s X s (t)    dt + Z T 0    e −rt X ∀s∈S 6 C RT s X s (t)    dt (C.71) anddenoterespectivelythe(i)thehealthcarecosts,(ii)thecostofART,(iii)thecostofPrEP, (iv)theunitcostofCD4/RNAmonitoring,(v)theunitcostofHIVtesting,(vi)thecostof diagnosingHIV,(vi)thecostofmonitoringPrEPusers,(vii)thecostsofphysicianvisits,and (viii)theunitcostofHIVgenotypictest,and S 1 = {P k ,PJ k ,PPrEP k ,I k ,J k ,IPrEP k ,E k ,EJ k ,A k ,AJ k ,T k ,TA k } k∈{s,r} (C.72) S 2 = {T k ,TA k } k∈{s,r} (C.73) S 3 = {TJ s ,TJ r } (C.74) S 4 = {SPrEP,PPrEP k ,IPrEP k } k∈{s,r} (C.75) S 5 = {PJ k ,J k ,EJ k ,T k ,AJ k ,TA k } k∈{s,r} (C.76) S 6 = {P k ,I k ,E k ,A k } k∈{s,r} (C.77) S 7 = {PPrEP k ,IPrEP k } k∈{s,r} (C.78) 293 anddenoterespectivelythesetsofdiseasestatesinwhichexpenditureisincurred.Inaddition, intheaboverelations,C s j denotestheannualcostofjindiseasestates,T representsthe20- yearprogramduration,andrdenotestheannualcostdiscountrate.[27,39] C.3.2 CalculationofQALYs C.3.2.1 Health-relatedQOLsummaryscore Health-related quality of life (HRQOL) scores for the general US male population were obtainedfromthepublishedliterature[32].TheseutilityscoresweremeasuredontheEQ-5D scaleandwerestratiedbyageandsex.Thereporteddatacontainsboththepopulationmean estimateandtheassociateduncertaintyranges(TableC.17). Table C.17: US male population age- and sex-stratied for each continuous health-related QOLsummaryscore(EQ-5DUS)[32]. AgeGroup(y) Median Mean Lower95%CI Upper95%CI 20-29 1.000 0.928 0.922 0.934 30-39 1.000 0.918 0.912 0.925 40-49 1.000 0.887 0.880 0.894 50-59 0.827 0.861 0.853 0.870 60-69 0.827 0.840 0.827 0.852 70-79 0.827 0.802 0.788 0.816 80-89 0.816 0.782 0.757 0.807 Giventhatthereporteddataonlycoversthe20-to89-year-oldUSmalesby5-yearbins, we use an extrapolation method to estimate the scores for the 15- to 19-year-old group. This is donethroughasimplelineartoftheagegroupmeanHRQOLscoreonthemidpointofthe agebin.WefurtherassumethattheHRQOLsummaryscoresforindividualsolderthan89is approximatedbythatofthoseinthe80-to89-year-oldgroup. 294 UsingthedatainTableC.17,theassumptionsmadeabouttheHRQOLestimatesforthe15-to 19-year-oldand89+-year-oldgroups,andtheagedistributionoftheLACMSMpopulation, weestimatedtheHRQOLfactor(denotedbyHRQOL f )forarepresentativeindividualin thepopulation.WecomputedtheHRQOLfactorastheweightedaverageoftheage-adjusted HRQOLsummaryscoredineachagegroupg,asformulatedinthefollowingequation: HRQOL f = X g∈G w g HRQOL g , (C.79) whereHRQOL g andw g denote respectively the HRQOL and the share of MSM in age groupg,respectively. C.3.2.2 Disease-stateQOLutilityscoreweights Alldisease-statequalityoflife(QOL)utilityscoreweightswerederivedfromthepublished literature(TableC.18). 295 TableC.18:Disease-stateQOLutilityweightsandotherefectivenessparameters. Parameters Value Range Reference DiseaseStateQOLUtilityWeights Uninfected(noPrEP) 1.0000 - [15,19,97] Uninfected(PrEP) 1.0000 0.9000-1.0000 [29,32,97] AcuteHIV,unidentied 0.9200 0.7300-0.9700 [6,19,39,42,47,53,63,64,75,97] AcuteHIV,identied 0.8600 0.6800-0.9100 [6,19,39,42,47,53,63,64,72,75,90, 97] AcuteHIV,treatedwithART 0.8800 0.6800-0.9400 [37,50] AsymptomaticHIV,unidentied 0.9100 0.8500-0.9500 [19,72,90,97] AsymptomaticHIV,identied(Year1) 0.8400 0.8400-0.9500 [15,67,72,75,97] AsymptomaticHIV,identied(Year2+) 0.8900 0.8500-0.9500 [15,19,67,72,97] AsymptomaticHIV,treatedwithART 0.9100 0.8500-0.9500 [15,19,50,67,97] SymptomaticHIV,unidentied 0.8000 0.7000-0.8000 [15,19,31,36,72,97] SymptomaticHIV,identied 0.7200 0.7000-0.8000 [15,19,90,97] SymptomaticHIV,treatedwithART 0.8300 0.7800-1.0000 [15,19,31,43,67,72,97] AIDS,unidentied 0.7200 0.2400-0.8000 [15,43,67,72,97] AIDS,identied 0.7200 0.6000-0.7500 [15,31,36,43,97] AIDS-treatedwithART 0.8200 0.8200-0.8700 [15,31,43,97] OtherEfectivenessParameters QOLdecrementfactorforfalse-positiveresult 0.1200 0.0000-0.4800 [72,84,90,97] QOLdecrementfactorowingtoresistancetoARTorPrEP 0.0000 0.0000-0.0100 Assumed AnnualQOLdiscountrate 0.0300 0.0000-0.0500 [15,19,67,97] 296 C.3.2.3 Age-adjustedhealth-relatedQOLutilityscoresforeachdiseasestate WecombinedtheseutilityscoreweightswiththeHRQOLfactortoestimatetheage-adjusted health-related QOL utility scores for each disease state. Specically, we compute the age- weightedaveragehealth-relatedQOLutilityscores(q s )foragivendiseasestatesastheproduct oftheHRQOLfactor(HRQOL f )andthedisease-stateQOLutilityweight(QOL s ): q s = HRQOL f QOL s . (C.80) C.3.2.4 TotalQALYs We estimated the total health benets for the entire population in discounted QALYs and assumedalifetimehorizontoaccountforthehealthbenetsoccurringbeyondtheinterven- tionduration.Specically,weestimatedtheQALYsasfollows: QALYs = Z +∞ 0 e −rt X ∀s∈S q s X s (t)dt, (C.81) whereq s andX s denote the health-related QOL and the number of individuals in disease states,respectively,andristheQOLannualdiscountrate. C.3.2.5 Calculationoftheincrementalcost-efectivenessratio The incremental cost-efectiveness ratio (ICER) of 2 interventions i and j (denoted by ICER ij )wascomputedastheratiooftheincrementalcosttotheincrementalefectiveness oftheseinterventions: ICER ij = Costs j −Costs i QALYs j −QALYs i . (C.82) 297 Thenumeratorinthisexpressionisthemarginalcostofmovingfromstrategyitostrategy j,whereasthedenominatorcapturesthemarginalQALYgainofmovingfromstrategyito strategyj. InordertocalculatetheICERsfortherationaldecisionsontheecientfrontier, we follow the method in Drummond et al. [21]. We begin by rst ranking the discounted costsinascendingorder.Next,weexcludeallstronglydominatedstrategies,whicharestrate- giesthataremoreexpensive,butlessefectivethanotherstrategiesinthechoiceset. Among the reduced set of strategies, we next determine those that are extendedly dominated. An extendedlydominatedstrategyisonethathasanICERgreaterthanthatofamoreefective strategy. C.4 Results Wesimulatedatotalof623strategies(TableC.19),whicharevariantsandcombinationsofthe Testing,Test-and-Treat,andPrEPstrategies,wherebytheintensityoftesting,ARTcoverage andPrEPuptakearechanged. 298 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies StatusQuo,SQ(Test4.4y+ART2.5yatCD4≤ 500) SQ+EarlyART2.5y+PrEP3y Testing(SQ+Test1y) SQ+EarlyART2.5y+PrEP4y SQ+EarlyART1mo SQ+EarlyART2.5y+PrEP4.4y SQ+EarlyART1mo+ImmediatePrEP SQ+EarlyART2.5y+PrEP6mo SQ+EarlyART1mo+PrEP1mo SQ+EarlyART3mo SQ+EarlyART1mo+PrEP1.2y SQ+EarlyART3mo+ImmediatePrEP SQ+EarlyART1mo+PrEP2y SQ+EarlyART3mo+PrEP1mo SQ+EarlyART1mo+PrEP3mo SQ+EarlyART3mo+PrEP1.2y SQ+EarlyART1mo+PrEP3y SQ+EarlyART3mo+PrEP2y SQ+EarlyART1mo+PrEP4y SQ+EarlyART3mo+PrEP3mo SQ+EarlyART1mo+PrEP4.4y SQ+EarlyART3mo+PrEP3y SQ+EarlyART1mo+PrEP6mo SQ+EarlyART3mo+PrEP4y SQ+EarlyART1y SQ+EarlyART3mo+PrEP4.4y SQ+EarlyART1y+ImmediatePrEP SQ+EarlyART3mo+PrEP6mo SQ+EarlyART1y+PrEP1mo SQ+EarlyART3y SQ+EarlyART1y+PrEP1.2y SQ+EarlyART3y+ImmediatePrEP SQ+EarlyART1y+PrEP2y SQ+EarlyART3y+PrEP1mo SQ+EarlyART1y+PrEP3mo SQ+EarlyART3y+PrEP1.2y SQ+EarlyART1y+PrEP3y SQ+EarlyART3y+PrEP2y SQ+EarlyART1y+PrEP4y SQ+EarlyART3y+PrEP3mo SQ+EarlyART1y+PrEP4.4y SQ+EarlyART3y+PrEP3y SQ+EarlyART1y+PrEP6mo SQ+EarlyART3y+PrEP4y SQ+EarlyART2y SQ+EarlyART3y+PrEP4.4y SQ+EarlyART2y+ImmediatePrEP SQ+EarlyART3y+PrEP6mo SQ+EarlyART2y+PrEP1mo SQ+EarlyART4y+PrEP4.4y SQ+EarlyART2y+PrEP1.2y SQ+EarlyART6mo SQ+EarlyART2y+PrEP2y SQ+EarlyART6mo+ImmediatePrEP SQ+EarlyART2y+PrEP3mo SQ+EarlyART6mo+PrEP1mo SQ+EarlyART2y+PrEP3y SQ+EarlyART6mo+PrEP1.2y SQ+EarlyART2y+PrEP4y SQ+EarlyART6mo+PrEP2y Continued...Note:Strategiesinboldfontarethoseonthefrontier,orassessedinthebasecaseanalysis. 299 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies SQ+EarlyART2y+PrEP4.4y SQ+EarlyART6mo+PrEP3mo SQ+EarlyART2y+PrEP6mo SQ+EarlyART6mo+PrEP3y SQ+EarlyART2.5y SQ+EarlyART6mo+PrEP4y SQ+EarlyART2.5y+ImmediatePrEP SQ+EarlyART6mo+PrEP4.4y SQ+EarlyART2.5y+PrEP1mo SQ+EarlyART6mo+PrEP6mo SQ+EarlyART2.5y+PrEP1.2y SQ+ImmediatePrEP SQ+EarlyART2.5y+PrEP2y SQ+PrEP1mo SQ+EarlyART2.5y+PrEP3mo SQ+PrEP1.2y SQ+PrEP2y SQ+Test1y+EarlyART2.5y+PrEP1.2y SQ+PrEP3mo SQ+Test1y+EarlyART2.5y+PrEP2y SQ+PrEP3y SQ+Test1y+EarlyART2.5y+PrEP3mo SQ+PrEP4.4y SQ+Test1y+EarlyART2.5y+PrEP3y SQ+PrEP6mo SQ+Test1y+EarlyART2.5y+PrEP4y SQ+Test1mo+EarlyART3mo SQ+Test1y+EarlyART2.5y+PrEP4.4y SQ+Test1y+EarlyART1mo+ImmediatePrEP SQ+Test1y+EarlyART2.5y+PrEP6mo SQ+Test1y+EarlyART1mo+PrEP1mo SQ+Test1y+EarlyART3mo SQ+Test1y+EarlyART1mo+PrEP1.2y SQ+Test1y+EarlyART3mo+ImmediatePrEP SQ+Test1y+EarlyART1mo+PrEP2y SQ+Test1y+EarlyART3mo+PrEP1mo SQ+Test1y+EarlyART1mo+PrEP3mo SQ+Test1y+EarlyART3mo+PrEP1.2y SQ+Test1y+EarlyART1mo+PrEP3y SQ+Test1y+EarlyART3mo+PrEP2y SQ+Test1y+EarlyART1mo+PrEP4y SQ+Test1y+EarlyART3mo+PrEP3mo SQ+Test1y+EarlyART1mo+PrEP4.4y SQ+Test1y+EarlyART3mo+PrEP3y SQ+Test1y+EarlyART1mo+PrEP6mo SQ+Test1y+EarlyART3mo+PrEP4y SQ+Test1y+EarlyART1y SQ+Test1y+EarlyART3mo+PrEP4.4y SQ+Test1y+EarlyART1y+ImmediatePrEP SQ+Test1y+EarlyART3mo+PrEP6mo SQ+Test1y+EarlyART1y+PrEP1mo SQ+Test1y+EarlyART3y SQ+Test1y+EarlyART1y+PrEP1.2y SQ+Test1y+EarlyART3y+ImmediatePrEP SQ+Test1y+EarlyART1y+PrEP2y SQ+Test1y+EarlyART3y+PrEP1mo SQ+Test1y+EarlyART1y+PrEP3mo SQ+Test1y+EarlyART3y+PrEP1.2y SQ+Test1y+EarlyART1y+PrEP3y SQ+Test1y+EarlyART3y+PrEP2y Continued...Note:Strategiesinboldfontarethoseonthefrontier,orassessedinthebasecaseanalysis. 300 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies SQ+Test1y+EarlyART1y+PrEP4y SQ+Test1y+EarlyART3y+PrEP3mo SQ+Test1y+EarlyART1y+PrEP4.4y SQ+Test1y+EarlyART3y+PrEP3y SQ+Test1y+EarlyART1y+PrEP6mo SQ+Test1y+EarlyART3y+PrEP4y SQ+Test1y+EarlyART2y SQ+Test1y+EarlyART3y+PrEP4.4y SQ+Test1y+EarlyART2y+ImmediatePrEP SQ+Test1y+EarlyART3y+PrEP6mo SQ+Test1y+EarlyART2y+PrEP1mo SQ+Test1y+EarlyART6mo SQ+Test1y+EarlyART2y+PrEP1.2y SQ+Test1y+EarlyART6mo+ImmediatePrEP SQ+Test1y+EarlyART2y+PrEP2y SQ+Test1y+EarlyART6mo+PrEP1mo SQ+Test1y+EarlyART2y+PrEP3mo SQ+Test1y+EarlyART6mo+PrEP1.2y SQ+Test1y+EarlyART2y+PrEP3y SQ+Test1y+EarlyART6mo+PrEP2y SQ+Test1y+EarlyART2y+PrEP4y SQ+Test1y+EarlyART6mo+PrEP3mo SQ+Test1y+EarlyART2y+PrEP4.4y SQ+Test1y+EarlyART6mo+PrEP3y SQ+Test1y+EarlyART2y+PrEP6mo SQ+Test1y+EarlyART6mo+PrEP4y SQ+Test1y+EarlyART2.5y SQ+Test1y+EarlyART6mo+PrEP4.4y SQ+Test1y+EarlyART2.5y+ImmediatePrEP SQ+Test1y+EarlyART6mo+PrEP6mo SQ+Test1y+EarlyART2.5y+PrEP1mo SQ+Test1y+ImmediatePrEP SQ+Test1y+PrEP1mo SQ+Test2y+EarlyART2.5y SQ+Test1y+PrEP1.2y SQ+Test2y+EarlyART2.5y+ImmediatePrEP SQ+Test1y+PrEP2y SQ+Test2y+EarlyART2.5y+PrEP1mo SQ+Test1y+PrEP3mo SQ+Test2y+EarlyART2.5y+PrEP1.2y SQ+Test1y+PrEP3y SQ+Test2y+EarlyART2.5y+PrEP2y SQ+Test1y+PrEP4.4y SQ+Test2y+EarlyART2.5y+PrEP3mo SQ+Test1y+PrEP6mo SQ+Test2y+EarlyART2.5y+PrEP3y SQ+Test2y SQ+Test2y+EarlyART2.5y+PrEP4y SQ+Test2y+EarlyART1mo SQ+Test2y+EarlyART2.5y+PrEP4.4y SQ+Test2y+EarlyART1mo+ImmediatePrEP SQ+Test2y+EarlyART2.5y+PrEP6mo SQ+Test2y+EarlyART1mo+PrEP1mo SQ+Test2y+EarlyART3mo SQ+Test2y+EarlyART1mo+PrEP1.2y SQ+Test2y+EarlyART3mo+ImmediatePrEP SQ+Test2y+EarlyART1mo+PrEP2y SQ+Test2y+EarlyART3mo+PrEP1mo SQ+Test2y+EarlyART1mo+PrEP3mo SQ+Test2y+EarlyART3mo+PrEP1.2y Continued...Note:Strategiesinboldfontarethoseonthefrontier,orassessedinthebasecaseanalysis. 301 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies SQ+Test2y+EarlyART1mo+PrEP3y SQ+Test2y+EarlyART3mo+PrEP2y SQ+Test2y+EarlyART1mo+PrEP4y SQ+Test2y+EarlyART3mo+PrEP3mo SQ+Test2y+EarlyART1mo+PrEP4.4y SQ+Test2y+EarlyART3mo+PrEP3y SQ+Test2y+EarlyART1mo+PrEP6mo SQ+Test2y+EarlyART3mo+PrEP4y SQ+Test2y+EarlyART1y SQ+Test2y+EarlyART3mo+PrEP4.4y SQ+Test2y+EarlyART1y+ImmediatePrEP SQ+Test2y+EarlyART3mo+PrEP6mo SQ+Test2y+EarlyART1y+PrEP1mo SQ+Test2y+EarlyART3y SQ+Test2y+EarlyART1y+PrEP1.2y SQ+Test2y+EarlyART3y+ImmediatePrEP SQ+Test2y+EarlyART1y+PrEP2y SQ+Test2y+EarlyART3y+PrEP1mo SQ+Test2y+EarlyART1y+PrEP3mo SQ+Test2y+EarlyART3y+PrEP1.2y SQ+Test2y+EarlyART1y+PrEP3y SQ+Test2y+EarlyART3y+PrEP2y SQ+Test2y+EarlyART1y+PrEP4y SQ+Test2y+EarlyART3y+PrEP3mo SQ+Test2y+EarlyART1y+PrEP4.4y SQ+Test2y+EarlyART3y+PrEP3y SQ+Test2y+EarlyART1y+PrEP6mo SQ+Test2y+EarlyART3y+PrEP4y SQ+Test2y+EarlyART2y SQ+Test2y+EarlyART3y+PrEP4.4y SQ+Test2y+EarlyART2y+ImmediatePrEP SQ+Test2y+EarlyART3y+PrEP6mo SQ+Test2y+EarlyART2y+PrEP1mo SQ+Test2y+EarlyART6mo SQ+Test2y+EarlyART2y+PrEP1.2y SQ+Test2y+EarlyART6mo+ImmediatePrEP SQ+Test2y+EarlyART2y+PrEP2y SQ+Test2y+EarlyART6mo+PrEP1mo SQ+Test2y+EarlyART2y+PrEP3mo SQ+Test2y+EarlyART6mo+PrEP1.2y SQ+Test2y+EarlyART2y+PrEP3y SQ+Test2y+EarlyART6mo+PrEP2y SQ+Test2y+EarlyART2y+PrEP4y SQ+Test2y+EarlyART6mo+PrEP3mo SQ+Test2y+EarlyART2y+PrEP4.4y SQ+Test2y+EarlyART6mo+PrEP3y SQ+Test2y+EarlyART2y+PrEP6mo SQ+Test2y+EarlyART6mo+PrEP4y SQ+Test2y+EarlyART6mo+PrEP4.4y SQ+Test3mo+EarlyART2y+PrEP4y SQ+Test2y+EarlyART6mo+PrEP6mo SQ+Test3mo+EarlyART2y+PrEP4.4y SQ+Test2y+ImmediatePrEP SQ+Test3mo+EarlyART2y+PrEP6mo SQ+Test2y+PrEP1mo SQ+Test3mo+EarlyART2.5y SQ+Test2y+PrEP1.2y SQ+Test3mo+EarlyART2.5y+ImmediatePrEP SQ+Test2y+PrEP2y SQ+Test3mo+EarlyART2.5y+PrEP1mo Continued...Note:Strategiesinboldfontarethoseonthefrontier,orassessedinthebasecaseanalysis. 302 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies SQ+Test2y+PrEP3mo SQ+Test3mo+EarlyART2.5y+PrEP1.2y SQ+Test2y+PrEP3y SQ+Test3mo+EarlyART2.5y+PrEP2y SQ+Test2y+PrEP4.4y SQ+Test3mo+EarlyART2.5y+PrEP3mo SQ+Test2y+PrEP6mo SQ+Test3mo+EarlyART2.5y+PrEP3y SQ+Test3mo SQ+Test3mo+EarlyART2.5y+PrEP4y SQ+Test3mo+EarlyART1mo SQ+Test3mo+EarlyART2.5y+PrEP4.4y SQ+Test3mo+EarlyART1mo+ImmediatePrEP SQ+Test3mo+EarlyART2.5y+PrEP6mo SQ+Test3mo+EarlyART1mo+PrEP1mo SQ+Test3mo+EarlyART3mo SQ+Test3mo+EarlyART1mo+PrEP1.2y SQ+Test3mo+EarlyART3mo+ImmediatePrEP SQ+Test3mo+EarlyART1mo+PrEP2y SQ+Test3mo+EarlyART3mo+PrEP1mo SQ+Test3mo+EarlyART1mo+PrEP3mo SQ+Test3mo+EarlyART3mo+PrEP1.2y SQ+Test3mo+EarlyART1mo+PrEP3y SQ+Test3mo+EarlyART3mo+PrEP2y SQ+Test3mo+EarlyART1mo+PrEP4y SQ+Test3mo+EarlyART3mo+PrEP3mo SQ+Test3mo+EarlyART1mo+PrEP4.4y SQ+Test3mo+EarlyART3mo+PrEP3y SQ+Test3mo+EarlyART1mo+PrEP6mo SQ+Test3mo+EarlyART3mo+PrEP4y SQ+Test3mo+EarlyART1y SQ+Test3mo+EarlyART3mo+PrEP4.4y SQ+Test3mo+EarlyART1y+ImmediatePrEP SQ+Test3mo+EarlyART3mo+PrEP6mo SQ+Test3mo+EarlyART1y+PrEP1mo SQ+Test3mo+EarlyART3y SQ+Test3mo+EarlyART1y+PrEP1.2y SQ+Test3mo+EarlyART3y+ImmediatePrEP SQ+Test3mo+EarlyART1y+PrEP2y SQ+Test3mo+EarlyART3y+PrEP1mo SQ+Test3mo+EarlyART1y+PrEP3mo SQ+Test3mo+EarlyART3y+PrEP1.2y SQ+Test3mo+EarlyART1y+PrEP3y SQ+Test3mo+EarlyART3y+PrEP2y SQ+Test3mo+EarlyART1y+PrEP4y SQ+Test3mo+EarlyART3y+PrEP3mo SQ+Test3mo+EarlyART1y+PrEP4.4y SQ+Test3mo+EarlyART3y+PrEP3y SQ+Test3mo+EarlyART1y+PrEP6mo SQ+Test3mo+EarlyART3y+PrEP4y SQ+Test3mo+EarlyART2y SQ+Test3mo+EarlyART3y+PrEP4.4y SQ+Test3mo+EarlyART2y+ImmediatePrEP SQ+Test3mo+EarlyART3y+PrEP6mo SQ+Test3mo+EarlyART2y+PrEP1mo SQ+Test3mo+EarlyART6mo SQ+Test3mo+EarlyART2y+PrEP1.2y SQ+Test3mo+EarlyART6mo+ImmediatePrEP SQ+Test3mo+EarlyART2y+PrEP2y SQ+Test3mo+EarlyART6mo+PrEP1mo Continued...Note:Strategiesinboldfontarethoseonthefrontier,orassessedinthebasecaseanalysis. 303 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies SQ+Test3mo+EarlyART2y+PrEP3mo SQ+Test3mo+EarlyART6mo+PrEP1.2y SQ+Test3mo+EarlyART2y+PrEP3y SQ+Test3mo+EarlyART6mo+PrEP2y SQ+Test3mo+EarlyART6mo+PrEP3mo SQ+Test3y+EarlyART2y+PrEP2y SQ+Test3mo+EarlyART6mo+PrEP3y SQ+Test3y+EarlyART2y+PrEP3mo SQ+Test3mo+EarlyART6mo+PrEP4y SQ+Test3y+EarlyART2y+PrEP3y SQ+Test3mo+EarlyART6mo+PrEP4.4y SQ+Test3y+EarlyART2y+PrEP4y SQ+Test3mo+EarlyART6mo+PrEP6mo SQ+Test3y+EarlyART2y+PrEP4.4y SQ+Test3mo+ImmediatePrEP SQ+Test3y+EarlyART2y+PrEP6mo SQ+Test3mo+PrEP1mo SQ+Test3y+EarlyART2.5y SQ+Test3mo+PrEP1.2y SQ+Test3y+EarlyART2.5y+ImmediatePrEP SQ+Test3mo+PrEP2y SQ+Test3y+EarlyART2.5y+PrEP1mo SQ+Test3mo+PrEP3mo SQ+Test3y+EarlyART2.5y+PrEP1.2y SQ+Test3mo+PrEP3y SQ+Test3y+EarlyART2.5y+PrEP2y SQ+Test3mo+PrEP4.4y SQ+Test3y+EarlyART2.5y+PrEP3mo SQ+Test3mo+PrEP6mo SQ+Test3y+EarlyART2.5y+PrEP3y SQ+Test3y SQ+Test3y+EarlyART2.5y+PrEP4y SQ+Test3y+EarlyART1mo SQ+Test3y+EarlyART2.5y+PrEP4.4y SQ+Test3y+EarlyART1mo+ImmediatePrEP SQ+Test3y+EarlyART2.5y+PrEP6mo SQ+Test3y+EarlyART1mo+PrEP1mo SQ+Test3y+EarlyART3mo SQ+Test3y+EarlyART1mo+PrEP1.2y SQ+Test3y+EarlyART3mo+ImmediatePrEP SQ+Test3y+EarlyART1mo+PrEP2y SQ+Test3y+EarlyART3mo+PrEP1mo SQ+Test3y+EarlyART1mo+PrEP3mo SQ+Test3y+EarlyART3mo+PrEP1.2y SQ+Test3y+EarlyART1mo+PrEP3y SQ+Test3y+EarlyART3mo+PrEP2y SQ+Test3y+EarlyART1mo+PrEP4y SQ+Test3y+EarlyART3mo+PrEP3mo SQ+Test3y+EarlyART1mo+PrEP4.4y SQ+Test3y+EarlyART3mo+PrEP3y SQ+Test3y+EarlyART1mo+PrEP6mo SQ+Test3y+EarlyART3mo+PrEP4y SQ+Test3y+EarlyART1y SQ+Test3y+EarlyART3mo+PrEP4.4y SQ+Test3y+EarlyART1y+ImmediatePrEP SQ+Test3y+EarlyART3mo+PrEP6mo SQ+Test3y+EarlyART1y+PrEP1mo SQ+Test3y+EarlyART3y SQ+Test3y+EarlyART1y+PrEP1.2y SQ+Test3y+EarlyART3y+ImmediatePrEP Continued...Note:Strategiesinboldfontarethoseonthefrontier,orassessedinthebasecaseanalysis. 304 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies SQ+Test3y+EarlyART1y+PrEP2y SQ+Test3y+EarlyART3y+PrEP1mo SQ+Test3y+EarlyART1y+PrEP3mo SQ+Test3y+EarlyART3y+PrEP1.2y SQ+Test3y+EarlyART1y+PrEP3y SQ+Test3y+EarlyART3y+PrEP2y SQ+Test3y+EarlyART1y+PrEP4y SQ+Test3y+EarlyART3y+PrEP3mo SQ+Test3y+EarlyART1y+PrEP4.4y SQ+Test3y+EarlyART3y+PrEP3y SQ+Test3y+EarlyART1y+PrEP6mo SQ+Test3y+EarlyART3y+PrEP4y SQ+Test3y+EarlyART2y SQ+Test3y+EarlyART3y+PrEP4.4y SQ+Test3y+EarlyART2y+ImmediatePrEP SQ+Test3y+EarlyART3y+PrEP6mo SQ+Test3y+EarlyART2y+PrEP1mo SQ+Test3y+EarlyART6mo SQ+Test3y+EarlyART2y+PrEP1.2y SQ+Test3y+EarlyART6mo+ImmediatePrEP SQ+Test3y+EarlyART6mo+PrEP1mo SQ+Test4y+EarlyART2y+ImmediatePrEP SQ+Test3y+EarlyART6mo+PrEP1.2y SQ+Test4y+EarlyART2y+PrEP1mo SQ+Test3y+EarlyART6mo+PrEP2y SQ+Test4y+EarlyART2y+PrEP1.2y SQ+Test3y+EarlyART6mo+PrEP3mo SQ+Test4y+EarlyART2y+PrEP2y SQ+Test3y+EarlyART6mo+PrEP3y SQ+Test4y+EarlyART2y+PrEP3mo SQ+Test3y+EarlyART6mo+PrEP4y SQ+Test4y+EarlyART2y+PrEP3y SQ+Test3y+EarlyART6mo+PrEP4.4y SQ+Test4y+EarlyART2y+PrEP4y SQ+Test3y+EarlyART6mo+PrEP6mo SQ+Test4y+EarlyART2y+PrEP4.4y SQ+Test3y+ImmediatePrEP SQ+Test4y+EarlyART2y+PrEP6mo SQ+Test3y+PrEP1mo SQ+Test4y+EarlyART2.5y SQ+Test3y+PrEP1.2y SQ+Test4y+EarlyART2.5y+ImmediatePrEP SQ+Test3y+PrEP2y SQ+Test4y+EarlyART2.5y+PrEP1mo SQ+Test3y+PrEP3mo SQ+Test4y+EarlyART2.5y+PrEP1.2y SQ+Test3y+PrEP3y SQ+Test4y+EarlyART2.5y+PrEP2y SQ+Test3y+PrEP4.4y SQ+Test4y+EarlyART2.5y+PrEP3mo SQ+Test3y+PrEP6mo SQ+Test4y+EarlyART2.5y+PrEP3y SQ+Test4y SQ+Test4y+EarlyART2.5y+PrEP4y SQ+Test4y+EarlyART1mo SQ+Test4y+EarlyART2.5y+PrEP4.4y SQ+Test4y+EarlyART1mo+ImmediatePrEP SQ+Test4y+EarlyART2.5y+PrEP6mo SQ+Test4y+EarlyART1mo+PrEP1mo SQ+Test4y+EarlyART3mo Continued...Note:Strategiesinboldfontarethoseonthefrontier,orassessedinthebasecaseanalysis. 305 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies SQ+Test4y+EarlyART1mo+PrEP1.2y SQ+Test4y+EarlyART3mo+ImmediatePrEP SQ+Test4y+EarlyART1mo+PrEP2y SQ+Test4y+EarlyART3mo+PrEP1mo SQ+Test4y+EarlyART1mo+PrEP3mo SQ+Test4y+EarlyART3mo+PrEP1.2y SQ+Test4y+EarlyART1mo+PrEP3y SQ+Test4y+EarlyART3mo+PrEP2y SQ+Test4y+EarlyART1mo+PrEP4y SQ+Test4y+EarlyART3mo+PrEP2y SQ+Test4y+EarlyART1mo+PrEP4.4y SQ+Test4y+EarlyART3mo+PrEP3mo SQ+Test4y+EarlyART1mo+PrEP6mo SQ+Test4y+EarlyART3mo+PrEP3y SQ+Test4y+EarlyART1y SQ+Test4y+EarlyART3mo+PrEP4y SQ+Test4y+EarlyART1y+ImmediatePrEP SQ+Test4y+EarlyART3mo+PrEP4.4y SQ+Test4y+EarlyART1y+PrEP1mo SQ+Test4y+EarlyART3mo+PrEP6mo SQ+Test4y+EarlyART1y+PrEP1.2y SQ+Test4y+EarlyART3y SQ+Test4y+EarlyART1y+PrEP2y SQ+Test4y+EarlyART3y+ImmediatePrEP SQ+Test4y+EarlyART1y+PrEP3mo SQ+Test4y+EarlyART3y+PrEP1mo SQ+Test4y+EarlyART1y+PrEP3y SQ+Test4y+EarlyART3y+PrEP1.2y SQ+Test4y+EarlyART1y+PrEP4y SQ+Test4y+EarlyART3y+PrEP3mo SQ+Test4y+EarlyART1y+PrEP4.4y SQ+Test4y+EarlyART3y+PrEP3y SQ+Test4y+EarlyART1y+PrEP6mo SQ+Test4y+EarlyART3y+PrEP4y SQ+Test4y+EarlyART2y SQ+Test4y+EarlyART3y+PrEP4.4y SQ+Test4y+EarlyART3y+PrEP6mo SQ+Test6mo+EarlyART1y+PrEP6mo SQ+Test4y+EarlyART6mo SQ+Test6mo+EarlyART2y SQ+Test4y+EarlyART6mo+ImmediatePrEP SQ+Test6mo+EarlyART2y+ImmediatePrEP SQ+Test4y+EarlyART6mo+PrEP1mo SQ+Test6mo+EarlyART2y+PrEP1mo SQ+Test4y+EarlyART6mo+PrEP1.2y SQ+Test6mo+EarlyART2y+PrEP1.2y SQ+Test4y+EarlyART6mo+PrEP2y SQ+Test6mo+EarlyART2y+PrEP2y SQ+Test4y+EarlyART6mo+PrEP3mo SQ+Test6mo+EarlyART2y+PrEP3mo SQ+Test4y+EarlyART6mo+PrEP3y SQ+Test6mo+EarlyART2y+PrEP3y SQ+Test4y+EarlyART6mo+PrEP4y SQ+Test6mo+EarlyART2y+PrEP4y SQ+Test4y+EarlyART6mo+PrEP4.4y SQ+Test6mo+EarlyART2y+PrEP4.4y SQ+Test4y+EarlyART6mo+PrEP6mo SQ+Test6mo+EarlyART2y+PrEP6mo SQ+Test4y+ImmediatePrEP SQ+Test6mo+EarlyART2.5y Continued...Note:Strategiesinboldfontarethoseonthefrontier,orassessedinthebasecaseanalysis. 306 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies SQ+Test4y+PrEP1mo SQ+Test6mo+EarlyART2.5y+ImmediatePrEP SQ+Test4y+PrEP1.2y SQ+Test6mo+EarlyART2.5y+PrEP1mo SQ+Test4y+PrEP2y SQ+Test6mo+EarlyART2.5y+PrEP1.2y SQ+Test4y+PrEP3mo SQ+Test6mo+EarlyART2.5y+PrEP2y SQ+Test4y+PrEP3y SQ+Test6mo+EarlyART2.5y+PrEP3mo SQ+Test4y+PrEP6mo SQ+Test6mo+EarlyART2.5y+PrEP3y SQ+Test6mo SQ+Test6mo+EarlyART2.5y+PrEP4y SQ+Test6mo+EarlyART1mo SQ+Test6mo+EarlyART2.5y+PrEP4.4y SQ+Test6mo+EarlyART1mo+ImmediatePrEP SQ+Test6mo+EarlyART2.5y+PrEP6mo SQ+Test6mo+EarlyART1mo+PrEP1mo SQ+Test6mo+EarlyART3mo SQ+Test6mo+EarlyART1mo+PrEP1.2y SQ+Test6mo+EarlyART3mo+ImmediatePrEP SQ+Test6mo+EarlyART1mo+PrEP2y SQ+Test6mo+EarlyART3mo+PrEP1mo SQ+Test6mo+EarlyART1mo+PrEP3mo SQ+Test6mo+EarlyART3mo+PrEP1.2y SQ+Test6mo+EarlyART1mo+PrEP3y SQ+Test6mo+EarlyART3mo+PrEP2y SQ+Test6mo+EarlyART1mo+PrEP4y SQ+Test6mo+EarlyART3mo+PrEP3mo SQ+Test6mo+EarlyART1mo+PrEP4.4y SQ+Test6mo+EarlyART3mo+PrEP3y SQ+Test6mo+EarlyART1mo+PrEP6mo SQ+Test6mo+EarlyART3mo+PrEP4y SQ+Test6mo+EarlyART1y SQ+Test6mo+EarlyART3mo+PrEP4.4y SQ+Test6mo+EarlyART1y+ImmediatePrEP SQ+Test6mo+EarlyART3mo+PrEP6mo SQ+Test6mo+EarlyART1y+PrEP1mo SQ+Test6mo+EarlyART3y SQ+Test6mo+EarlyART1y+PrEP1.2y SQ+Test6mo+EarlyART3y+ImmediatePrEP SQ+Test6mo+EarlyART1y+PrEP2y SQ+Test6mo+EarlyART3y+PrEP1mo SQ+Test6mo+EarlyART1y+PrEP3mo SQ+Test6mo+EarlyART3y+PrEP1.2y SQ+Test6mo+EarlyART1y+PrEP3y SQ+Test6mo+EarlyART3y+PrEP2y SQ+Test6mo+EarlyART1y+PrEP4y SQ+Test6mo+EarlyART3y+PrEP3mo SQ+Test6mo+EarlyART1y+PrEP4.4y SQ+Test6mo+EarlyART3y+PrEP3y SQ+Test6mo+EarlyART3y+PrEP4y EnhancedTT(TT+Test1y+PrEP4y) SQ+Test6mo+EarlyART3y+PrEP4.4y EnhancedTT(TT+Test1y+PrEP4.4y) SQ+Test6mo+EarlyART3y+PrEP6mo EnhancedTT(TT+Test1y+PrEP6mo) SQ+Test6mo+EarlyART6mo EnhancedTT(TT+Test1y) Continued...Note:Strategiesinboldfontarethoseonthefrontier,orassessedinthebasecaseanalysis. 307 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies SQ+Test6mo+EarlyART6mo+ImmediatePrEP EnhancedTT(TT+Test2y+ImmediatePrEP) SQ+Test6mo+EarlyART6mo+PrEP1mo EnhancedTT(TT+Test2y+PrEP1mo) SQ+Test6mo+EarlyART6mo+PrEP1.2y EnhancedTT(TT+Test2y+PrEP1.2y) SQ+Test6mo+EarlyART6mo+PrEP2y EnhancedTT(TT+Test2y+PrEP2y) SQ+Test6mo+EarlyART6mo+PrEP3mo EnhancedTT(TT+Test2y+PrEP3mo) SQ+Test6mo+EarlyART6mo+PrEP3y EnhancedTT(TT+Test2y+PrEP3y) SQ+Test6mo+EarlyART6mo+PrEP4y EnhancedTT(TT+Test2y+PrEP4y) SQ+Test6mo+EarlyART6mo+PrEP4.4y EnhancedTT(TT+Test2y+PrEP4.4y) SQ+Test6mo+EarlyART6mo+PrEP6mo EnhancedTT(TT+Test2y+PrEP6mo) SQ+Test6mo+ImmediatePrEP EnhancedTT(TT+Test2y) SQ+Test6mo+PrEP1mo EnhancedTT(TT+Test3mo+PrEP4.4y) SQ+Test6mo+PrEP1.2y EnhancedTT(TT+Test3mo) SQ+Test6mo+PrEP2y EnhancedTT(TT+Test3y+ImmediatePrEP) SQ+Test6mo+PrEP3mo EnhancedTT(TT+Test3y+PrEP1.2y) SQ+Test6mo+PrEP3y EnhancedTT(TT+Test3y+PrEP2y) SQ+Test6mo+PrEP4.4y EnhancedTT(TT+Test3y+PrEP3mo) SQ+Test6mo+PrEP6mo EnhancedTT(TT+Test3y+PrEP3y) Test-and-Treat,TT(SQ+ImmediateEarlyARTatCD4>500) EnhancedTT(TT+Test3y+PrEP4y) EnhancedTT(TT+Test3y+PrEP1mo) EnhancedTT(TT+Test3y+PrEP4.4y) EnhancedTT(TT+ImmediatePrEP) EnhancedTT(TT+Test3y+PrEP6mo) EnhancedTT(TT+PrEP1mo) EnhancedTT(TT+Test3y) EnhancedTT(TT+PrEP1.2y) EnhancedTT(TT+Test4y+ImmediatePrEP) EnhancedTT(TT+PrEP2y) EnhancedTT(TT+Test4y+PrEP1mo) EnhancedTT(TT+PrEP3mo) EnhancedTT(TT+Test4y+PrEP1.2y) EnhancedTT(TT+PrEP3y) EnhancedTT(TT+Test4y+PrEP2y) EnhancedTT(TT+PrEP4y) EnhancedTT(TT+Test4y+PrEP3mo) EnhancedTT(TT+PrEP4.4y) EnhancedTT(TT+Test4y+PrEP3y) EnhancedTT(TT+PrEP6mo) EnhancedTT(TT+Test4y+PrEP4y) EnhancedTT(TT+Test1y+ImmediatePrEP) EnhancedTT(TT+Test4y+PrEP4.4y) EnhancedTT(TT+Test1y+PrEP1mo) EnhancedTT(TT+Test4y+PrEP6mo) Continued...Note:Strategiesinboldfontarethoseonthefrontier,orassessedinthebasecaseanalysis. 308 TableC.19:Completelistofallstrategiessimulatedintheanalysis. Strategies EnhancedTT(TT+Test1y+PrEP1.2y) EnhancedTT(TT+Test4y) EnhancedTT(TT+Test1y+PrEP2y) EnhancedTT(TT+Test6mo+PrEP4.4y) EnhancedTT(TT+Test1y+PrEP3mo) EnhancedTT(TT+Test6mo) EnhancedTT(TT+Test1y+PrEP3y) PrEP(TT+Test6mo+PrEP4y) EnhancedPrEP(PrEP+Test3mo+PrEP1mo) EnhancedPrEP(PrEP+PrEP1.2y) EnhancedPrEP(PrEP+Test3mo+ImmediatePrEP) EnhancedPrEP(PrEP+ImmediatePrEP) EnhancedPrEP(PrEP+Test3mo+PrEP1.2y) EnhancedPrEP(PrEP+PrEP1mo) EnhancedPrEP(PrEP+Test3mo+PrEP2y) EnhancedPrEP(PrEP+PrEP2y) EnhancedPrEP(PrEP+Test3mo+PrEP3mo) EnhancedPrEP(PrEP+PrEP3mo) EnhancedPrEP(PrEP+Test3mo+PrEP3y) EnhancedPrEP(PrEP+PrEP3y) EnhancedPrEP(PrEP+Test3mo+PrEP4y) EnhancedPrEP(PrEP+PrEP6mo) EnhancedPrEP(PrEP+Test3mo+PrEP6mo) 309 C.4.1 Thebasecaseanalysis Ourbasecaseanalysisresultssuggestthatrelativetothestatusquostrategy,andatthecurrent US average willingness to pay (AWTP) threshold of $150,000/QALY gained, based on the WHO’ssocietalwillingnesstopay[99],Test-and-Treat,followedbyPrEP,arethemostcost- efectivestrategies(TableC.20).Thetest-and-treatstrategyconsistingofHIVtestingevery4.4 yearsfollowedbyimmediateARTstart(SQ+ImmediateEarlyART;strategy2onFigure4.2) reducesHIVinfectionsby4.3%andcosts$21,053/QALY.Incomparison,thePrEPstrategy consistingofHIVtestingevery6months,immediateARTstart,andPrEPstartevery4years (TT+Test6mo+PrEP4y;strategy7onFigure4.2)yieldsthelargestreduction(59.0%)in HIVincidenceandishighlycost-efectiverelativetothestatusquo($26,006/QALYgained). The testing strategy consisting of annual HIV testing and ART start every 2.5 years by the treatment-eligibleindividualswouldcost$27,268/QALYgained,andreducenewinfections by 21.4% (Table C.20 and Figure C.3). When the choice set consists only of these base case strategies,thetestingstrategyisweaklydominatedbythePrEPstrategy,andcosts$32,440per QALYgained,relativetotest-and-treat[21]. Relativetothetest-and-treatstrategy,thePrEP strategyishighlycost-efective($42,511/QALYgained). C.4.2 Resultsfromthefullanalysis Combinationsofstrategieswerealsosimulatedinthisstudy.Overall,theyindicatethatmore aggressivetest-and-treatandPrEPstrategiesarecostefectiveintheLACMSMpopulation. TableC.20-TableC.22summarizeourmainresults,whicharefurtherdiscussedinthemain paper. 310 TableC.20:Basecaseanalysis–benetsandcostsofthetesting,test-and-treat,andPrEPstrategies. Scenario HIVInfections a TotalCost, 2013$B a Total QALYs,M a ICER,$/QALY NewCases, n PreventedCases,n (%) RelativetoStatus Quo RelativetoPriorBest Strategy StatusQuo(SQ) c 99,874 - 43.58 2.96 - - Test-and-treat,TT (SQ+ImmediateEarlyART) d 95,542 4,332(4.3) 45.18 3.05 21,053 21,053 Testing e 78,517 21,357(21.4) 49.32 3.17 27,268 32,440(WD) f PrEP (TT+Test6mo+PrEP4y) 40,992 58,881(59.0) 58.03 3.48 26,006 42,511 Benetsandcostsrepresentlifetimeestimatesduring20yearsprogramdurationintheLACMSMpopulation. a CumulativeHIVincidencefortheprogramdurationof20y; b Discountedat3%discountrate; c Test4.4y+ART2.5yatCD4≤ 500; d EarlyARTdenedasARTstartatCD4>500; e Testevery1y+ARTstartevery2.5yatCD4≤ 500. f WD=Weaklydominated;TestingisweaklydominatedbyPrEPwhenthealternativesetisreducedtothe4basecasestrategies. Itishoweverstronglydominatedwhenall623strategies(TableC.19)areconsidered. 311 TableC.21: CumulativeHIVinfectionsavertedduring20yearsintheLACMSMpopulationunderthestatusquoandrationaltest- and-treatandPrEPstrategiesontheecientfrontier. RationalDecision (OntheEcientFrontier) New HIV Cases a ,n HIVCasesAverted,n(%) RelativetoPrEP RelativetoPrior RationalStrategy Relativeto StatusQuo Relativeto Test-and-treat Status-quo(SQ) b 99,874 - - - - Test-and-treatTT(SQ+ImmediateEarlyART) c 95,542 4,332(4.3) - - 4,332(4.3) EnhancedTT(TT+Test3y) 76,387 23,487(23.5) 19,155(20.0) - 19,155(20.0) EnhancedTT(TT+Test2y) 67,762 32,111(32.2) 27,779(29.1) - 8,624(11.3) EnhancedTT(TT+Test1y) 57,826 42,048(42.1) 37,716(39.5) - 9,936(14.7) EnhancedTT(TT+Test6mo) 52,115 47,759(47.8) 43,427(45.5) - 5,711(9.9) PrEP(TT+Test6mo+PrEP4y) 40,992 58,881(59.0) 54549(57.1) - 11,123(21.3) EnhancedPrEP(PrEP+PrEP3y) 39,023 60,851(60.9) 56,519(59.2) 1,970(2.1) 1,970(4.8) EnhancedPrEP(PrEP+PrEP2y) 36,205 63,668(63.7) 59,336(62.1) 4,787(5.0) 2,817(7.2) EnhancedPrEP(PrEP+PrEP1.2y) 33,088 66,505(66.6) 62,173(65.1) 7,623(8.0) 2,836(7.8) EnhancedPrEP(PrEP+Test3mo+PrEP2y) 33,369 66,786(66.9) 62,454(65.4) 7,904(8.3) 281(0.8) EnhancedPrEP(PrEP+Test3mo+PrEP1.2y) 30,484 69,390(69.5) 65,058(68.1) 10,509(11.0) 2,604(7.9) EnhancedPrEP(PrEP+Test3mo+ImmediatePrEP) 22,573 77,301(77.4) 72,969(76.4) 18,420(19.3) 7,911(26.0) a CumulativeHIVincidencefortheprogramdurationof20y; b Test4.4y+ART2.5yatCD4≤ 500. c EarlyARTdenedasARTstartatCD4> 500. 312 TableC.22:Mostcost-efectivetest-and-treatandPrEPstrategies(expanded). RationalDecision TotalDiscounted Incremental,WRTSQ Incremental,WRTPRD ExtendedlyDominatedStrategies Costs QALYs Costs QALYs ICER Costs QALYs ICER Status-quo(SQ) 43.58 2.96 - - - - - - - Test-and-treatTT (SQ+Immediate EarlyART) 45.18 3.05 1.6 82,915 19,302 1.6 82,915 19,302 SQ+Test4y SQ+EarlyART3y SQ+EarlyART2.5y SQ+EarlyART2y SQ+EarlyART1y SQ+EarlyART6mo SQ+EarlyART3mo SQ+EarlyART1mo SQ+Test4y+EarlyART3y SQ+Test4y+EarlyART2.5y SQ+Test4y+EarlyART2y SQ+Test4y+EarlyART6mo EnhancedTT (TT+Test3y) 48.97 3.23 5.39 268,436 20,096 3.79 185,522 20,451 SQ+Test2y SQ+Test2y+EarlyART3y SQ+Test2y+EarlyART2y SQ+Test2y+EarlyART1y SQ+Test3y+EarlyART3mo SQ+Test4y+EarlyART1mo SQ+Test3y+EarlyART1mo EnhancedTT(TT+Test4y) EnhancedTT(TT+Test2y) PrEP(SQ+EarlyART1y+PrEP4y) PrEP(SQ+EarlyART1mo+PrEP3y) PrEP(TT+Test4y+PrEP3y) EnhancedTT (TT+Test2y) 50.85 3.31 7.27 345,318 21,053 1.88 76,882 24,394 SQ+Test2y+EarlyART3mo SQ+Test2y+EarlyART1mo Continued... 313 TableC.22:Mostcost-efectivetest-and-treatandPrEPstrategies(expanded). RationalDecision TotalDiscounted Incremental,WRTSQ Incremental,WRTPRD ExtendedlyDominatedStrategies Costs QALYs Costs QALYs ICER Costs QALYs ICER EnhancedTT (TT+Test1y) 53.32 3.39 9.74 424,845 22,921 2.47 79,527 31,036 SQ+Test1y+EarlyART6mo SQ+Test1y+EarlyART3mo SQ+Test1mo+EarlyART3mo PrEP(TT+Test3y+PrEP2y) EnhancedTT (TT+Test6mo) 55.22 3.44 11.64 474,260 24,544 1.9 49,415 38,492 SQ+Test6mo+EarlyART3mo SQ+Test6mo+EarlyART1mo TT+Test2y+PrEP2y PrEP (TT+Test6mo+ PrEP4y) 58.03 3.48 14.45 518,716 27,863 2.81 44,457 63,269 EnhancedTT(TT+Test3mo) TT+Test1y+PrEP4y SQ+Test6mo+EarlyART3mo+PrEP4y SQ+Test6mo+EarlyART1mo+PrEP4y TT+Test1y+PrEP3y SQ+Test1y+EarlyART1mo+PrEP3y TT+Test1y+PrEP2y EnhancedPrEP (PrEP+PrEP3y) 58.55 3.49 14.97 524,827 28,529 0.52 6,111 85,117 SQ+Test3mo+EarlyART1mo+PrEP3y EnhancedPrEP (PrEP+PrEP2y) 59.36 3.5 15.78 532,534 29,633 0.81 7,707 104,788 SQ+Test6mo+EarlyART1mo+PrEP2y EnhancedPrEP (PrEP+PrEP1.2y) 60.33 3.5 16.75 539,479 31,045 0.97 6,945 139,346 EnhancedPrEP(PrEP+Test3mo+PrEP4y) EnhancedPrEP(PrEP+Test3mo+PrEP3y) EnhancedPrEP (PrEP+Test3mo+ PrEP2y) 61.01 3.51 17.43 544,155 32,033 0.68 4,676 145,956 EnhancedPrEP(PrEP+PrEP1.2y) Continued... 314 TableC.22:Mostcost-efectivetest-and-treatandPrEPstrategies(expanded). RationalDecision TotalDiscounted Incremental,WRTSQ Incremental,WRTPRD ExtendedlyDominatedStrategies Costs QALYs Costs QALYs ICER Costs QALYs ICER EnhancedPrEP (PrEP+Test3mo+ PrEP1.2y) 61.93 3.51 18.35 549,048 33,429 0.92 4,892 188,714 EnhancedPrEP(PrEP+PrEP6mo) EnhancedPrEP (PrEP+Test3mo+ ImmediatePrEP) 64.38 3.52 20.8 559,477 37,181 2.45 10,429 234,726 EnhancedPrEP(PrEP+Test3mo+PrEP1.2y) EnhancedPrEP(PrEP+Test3mo+PrEP3mo) EnhancedPrEP(PrEP+Test3mo+PrEP1mo) EnhancedPrEP(PrEP+PrEP1mo) EnhancedPrEP(PrEP+ImmediatePrEP) 315 C.5 Sensitivityanalyses TotesttherobustnessofourICERestimatesandtestthesensitivityoftheseestimatestothe uncertaintyinherentinthemodelweconductedaseriesof1-wayandprobabilisticsensitivity analysesthatincorporatebootstrapping. C.5.1 One-waysensitivityanalyses Wesensitizedeachinputparametervalueinthemodelwithinitsuncertaintyrangeoneata time, whileholdingconstantallotherparametervaluesinthemodeltotheirnonsensitized values. To facilitate presenting the results, we rst conducted this analysis on the epidemic parameters, then on the cost and efectiveness parameters. For each type of parameter, we assessedtheefectofthevariationsintheparametervaluesontheICERsforallourbasecase strategies and the strategies on the ecient frontier. We present in this appendix only the resultsforthebasecasescenarios,andthemostefectivestrategyonthefrontier(i.e. strategy 13onFigure4.2). Theresultsfromallotherstrategiesonthefrontierdisplaysimilarpatterns astheonespresentedinthisappendix. C.5.1.1 One-waysensitivityanalysesontheepidemicparameters Ingeneral,theresultssuggestthattheICERestimatesaremostsensitivetovariationsinthe sexual mixing and transmission parameters, as well as the frequency of testing, the rates of treatment initiation, and adherence to treatment. The sensitivity of antibody testing also appearstoafecttheICERestimates. However,alltheseefectsarenotsubstantialtochange theefectivenessproleofanystrategy,ascanbeobservedfromFiguresC.3-FigureC.14. 316 Figure C.3: Sensitivity of the ICERs to variations in the epidemic parameters – TT (SQ + ImmediateEarlyART)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ART discontinuation rate (AIDS) Antibody test specificity (post−seroconversion) Nucleic Acid Amplification test specificity Population inflow rate multiplier Reduction in sexual behavior due to testing and counseling Antibody test sensitivity (pre−seroconversion) Rate of acquired MDR Antibody test specificity (pre−seroconversion) Progression rate to AIDS (ART−treated, drug−resistant) Antibody test sensitivity (post−seroconversion) HIV transmission probability (primary stage; unaware HIV+, drug−sensitive) HIV−related death rate (untreated AIDS, drug−resistant) Progression rate to AIDS (treatment−eligible, drug−resistant, CD4<=500) ART discontinuation rate (symptomatic HIV) Average duration of identification status since last HIV test (uninfected) Progression rate to symptomatic HIV (drug−sensitive, untreated, unaware) Nucleic Acid Amplification test sensitivity Reduction in sexual infectivity due to ART HIV transmission probability (AIDS stage; aware untreated HIV+, drug−sensitive) Natural death rate HIV testing rate Progression rate to asymptomatic HIV HIV transmission probability (asymptomatic stage; aware HIV+, drug−sensitive) HIV−related death rate (ART−treated AIDS, drug−sensitive) HIV transmission probability (symptomatic stage; aware HIV+, drug−sensitive) HIV−related death rate (ART−treated AIDS, drug−resistant) Progression rate to symptomatic HIV (ART−treated) Progression rate to AIDS (ART−treated, drug−sensitive) Mutation rate from drug−resistant to drug−sensitive strain Progression rate to AIDS (treatment−eligible, drug−sensitive, CD4<=500) HIV−related death rate (untreated AIDS, drug−sensitive) ART initiation rate (AIDS) HIV transmission probability (asymptomatic stage; unaware HIV+, drug−sensitive) Sexual mixing rate ART initiation rate (symptomatic HIV) HIV transmission probability (asymptomatic stage; ART−treated, drug−sensitive) HIV transmission probability (AIDS stage; ART−treated, drug−sensitive) 10,000 12,000 14,000 16,000 18,000 20,000 22,000 24,000 26,000 28,000 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Epidemic Parameters Note: ThehorizontalbarsrepresenttheICERsrangesforTest-and-Treat,TT(SQ+ImmediateEarlyARTat CD4 > 500;strategy2inTable4.3andonFigure4.2)relativetoStatusQuo(SQ;strategy1onFigure4.2),as theparametervaluesarechangedtotheboundsoftheiruncertaintyranges. Theredlineandblackdotsatthe centeroftheplotdenotethenonsensitizedICERestimateof$19,302/QALYgainedunderTT,relativetoSQ. 317 FigureC.4:SensitivityoftheICERstovariationsintheepidemicparameters–EnhancedTT (TT+Test6mo)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Natural death rate Antibody test specificity (pre−seroconversion) Nucleic Acid Amplification test sensitivity HIV testing rate ART initiation rate (symptomatic HIV) Rate of acquired MDR ART discontinuation rate (AIDS) ART discontinuation rate (symptomatic HIV) Progression rate to asymptomatic HIV Antibody test specificity (post−seroconversion) Mutation rate from drug−resistant to drug−sensitive strain HIV−related death rate (untreated AIDS, drug−resistant) HIV−related death rate (ART−treated AIDS, drug−sensitive) Progression rate to AIDS (ART−treated, drug−resistant) Progression rate to AIDS (treatment−eligible, drug−resistant, CD4<=500) Population inflow rate multiplier HIV−related death rate (ART−treated AIDS, drug−resistant) Reduction in sexual behavior due to testing and counseling Antibody test sensitivity (post−seroconversion) Nucleic Acid Amplification test specificity Reduction in sexual infectivity due to ART Progression rate to symptomatic HIV (drug−sensitive, untreated, unaware) HIV transmission probability (asymptomatic stage; aware HIV+, drug−sensitive) HIV transmission probability (primary stage; unaware HIV+, drug−sensitive) Antibody test sensitivity (pre−seroconversion) ART initiation rate (AIDS) Progression rate to AIDS (treatment−eligible, drug−sensitive, CD4<=500) HIV transmission probability (asymptomatic stage; ART−treated, drug−sensitive) HIV−related death rate (untreated AIDS, drug−sensitive) Average duration of identification status since last HIV test (uninfected) Progression rate to AIDS (ART−treated, drug−sensitive) HIV transmission probability (symptomatic stage; aware HIV+, drug−sensitive) HIV transmission probability (AIDS stage; ART−treated, drug−sensitive) Progression rate to symptomatic HIV (ART−treated) HIV transmission probability (AIDS stage; aware untreated HIV+, drug−sensitive) HIV transmission probability (asymptomatic stage; unaware HIV+, drug−sensitive) Sexual mixing rate 24,000 28,000 32,000 36,000 40,000 44,000 48,000 52,000 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Epidemic Parameters Note:ThehorizontalbarsrepresenttheICERsrangesforEnhancedTT(TT+Test6mo;strategy6inTable4.3 andonFigure4.2)relativetoEnhancedTT(TT+Test1y;strategy5onFigure4.2),astheparametervaluesare changedtotheboundsoftheiruncertaintyranges. Theredlineandblackdotsatthecenteroftheplotdenote thenonsensitizedICERestimateof$38,492/QALYgainedunderEnhancedTT(TT+Test6mo),relativeto EnhancedTT(TT+Test1y). 318 FigureC.5: SensitivityoftheICERstovariationsintheepidemicparameters–PrEP(TT+ Test6mo+PrEP4y)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● HIV−related death rate (untreated AIDS, drug−resistant) Nucleic Acid Amplification test specificity Progression rate to AIDS (treatment−eligible, drug−resistant, CD4<=500) Natural death rate Antibody test specificity (post−seroconversion) Nucleic Acid Amplification test sensitivity Mutation rate from drug−resistant to drug−sensitive strain HIV−related death rate (untreated AIDS, drug−sensitive) Population inflow rate multiplier ART discontinuation rate (AIDS) Reduction in sexual behavior due to testing and counseling Antibody test sensitivity (post−seroconversion) Progression rate to AIDS (ART−treated, drug−resistant) Antibody test sensitivity (pre−seroconversion) Progression rate to asymptomatic HIV Antibody test specificity (pre−seroconversion) HIV−related death rate (ART−treated AIDS, drug−resistant) ART initiation rate (AIDS) HIV testing rate Progression rate to symptomatic HIV (drug−sensitive, untreated, unaware) Reduction in sexual infectivity due to ART HIV transmission probability (asymptomatic stage; aware HIV+, drug−sensitive) Progression rate to AIDS (treatment−eligible, drug−sensitive, CD4<=500) HIV transmission probability (AIDS stage; aware untreated HIV+, drug−sensitive) Average duration of identification status since last HIV test (uninfected) Rate of acquired MDR ART initiation rate (symptomatic HIV) HIV transmission probability (symptomatic stage; aware HIV+, drug−sensitive) PrEP adherence rate HIV−related death rate (ART−treated AIDS, drug−sensitive) Sexual mixing rate HIV transmission probability (AIDS stage; ART−treated, drug−sensitive) HIV transmission probability (primary stage; unaware HIV+, drug−sensitive) HIV transmission probability (asymptomatic stage; ART−treated, drug−sensitive) Progression rate to AIDS (ART−treated, drug−sensitive) ART discontinuation rate (symptomatic HIV) Progression rate to symptomatic HIV (ART−treated) HIV transmission probability (asymptomatic stage; unaware HIV+, drug−sensitive) 40,000 45,000 50,000 55,000 60,000 65,000 70,000 75,000 80,000 85,000 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Epidemic Parameters Note: The horizontal bars represent the ICERs ranges for PrEP (TT + Test 6 mo + PrEP 4 y; strategy 7 in Table 4.3 and on Figure 4.2) relative to Enhanced TT (TT + Test 6 mo; strategy 6 on Figure 4.2), as the parametervaluesarechangedtotheboundsoftheiruncertaintyranges.Theredlineandblackdotsatthecenter oftheplotdenotethenonsensitizedICERestimateof$63,269/QALYgainedunderPrEP(TT+Test6mo+ PrEP4y),relativetoEnhancedTT(TT+Test6mo). 319 Figure C.6: Sensitivity of the ICERs to variations in the epidemic parameters – Enhanced PrEP(PrEP+Test3mo+ImmediatePrEP)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Nucleic Acid Amplification test sensitivity Natural death rate Nucleic Acid Amplification test specificity Reduction in sexual behavior due to testing and counseling ART discontinuation rate (AIDS) Mutation rate from drug−resistant to drug−sensitive strain Population inflow rate multiplier HIV−related death rate (untreated AIDS, drug−resistant) Antibody test specificity (post−seroconversion) Progression rate to AIDS (treatment−eligible, drug−resistant, CD4<=500) Antibody test sensitivity (pre−seroconversion) HIV−related death rate (untreated AIDS, drug−sensitive) Progression rate to asymptomatic HIV Antibody test sensitivity (post−seroconversion) Progression rate to AIDS (ART−treated, drug−resistant) HIV transmission probability (AIDS stage; ART−treated, drug−sensitive) Progression rate to symptomatic HIV (drug−sensitive, untreated, unaware) Progression rate to AIDS (treatment−eligible, drug−sensitive, CD4<=500) HIV transmission probability (symptomatic stage; aware HIV+, drug−sensitive) Progression rate to AIDS (ART−treated, drug−sensitive) HIV transmission probability (asymptomatic stage; aware HIV+, drug−sensitive) HIV transmission probability (asymptomatic stage; unaware HIV+, drug−sensitive) HIV testing rate Progression rate to symptomatic HIV (ART−treated) HIV−related death rate (ART−treated AIDS, drug−resistant) Antibody test specificity (pre−seroconversion) Rate of acquired MDR ART initiation rate (symptomatic HIV) HIV−related death rate (ART−treated AIDS, drug−sensitive) ART initiation rate (AIDS) Reduction in sexual infectivity due to ART PrEP adherence rate HIV transmission probability (asymptomatic stage; ART−treated, drug−sensitive) HIV transmission probability (primary stage; unaware HIV+, drug−sensitive) Average duration of identification status since last HIV test (uninfected) Sexual mixing rate ART discontinuation rate (symptomatic HIV) HIV transmission probability (AIDS stage; aware untreated HIV+, drug−sensitive) 210,000 220,000 230,000 240,000 250,000 260,000 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Epidemic Parameters Note: The horizontal bars represent the ICERs ranges for Enhanced PrEP (PrEP + Test 3 mo + Immediate PrEP; strategy 13 in Table 4.3 and on Figure 4.2) relative to Enhanced PrEP (PrEP + Test 3 mo + PrEP 1.2 y; strategy12onFigure4.2),astheparametervaluesarechangedtotheboundsoftheiruncertaintyranges. The red line and black dots at the center of the plot denote the nonsensitized ICER estimate of $234,726/QALY gainedunderEnhancedPrEP(PrEP+Test3mo+ImmediatePrEP),relativetoEnhancedPrEP(PrEP+Test3 mo+PrEP1.2y). C.5.1.2 One-waysensitivityanalysesonthecostandefectivenessparameters The results from the sensitivity analysis with the cost and efectiveness parameters indicate thattheICERsarethemostsensitivetothecostofARTandthehealthcarecostsofmanaging uninfectedHIV. 320 FigureC.7: SensitivityoftheICERstovariationsinthecostparameters–TT(SQ+Imme- diateEarlyART)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Cost per antibody test Annual health care costs of acute HIV Annual health care costs of symptomatic HIV (treated with ART) Cost of posttest counseling + linkage to care (HIV+) Cost of posttest counseling (HIV−) Cost of HIV diagnosis Cost per CD4 count test Cost of pretest counseling Cost per HIV RNA test Annual health care costs of symptomatic HIV (untreated) Annual health care costs of AIDS (treated with ART) Annual health care costs of asymptomatic HIV (untreated) Cost per physician visit Cost per HIV genotype test Annual cost of ART Annual health care costs of AIDS (untreated) 17,000 17,500 18,000 18,500 19,000 19,500 20,000 20,500 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Cost Parameters Note: ThehorizontalbarsrepresenttheICERsrangesforTest-and-treat,TT(SQ+ImmediateEarlyARTat CD4 > 500;strategy2inTable4.3andonFigure4.2)relativetoStatusQuo(SQ;strategy1onFigure4.2),as theparametervaluesarechangedtotheboundsoftheiruncertaintyranges. Theredlineandblackdotsatthe centeroftheplotdenotethenonsensitizedICERestimateof$19,302/QALYgainedunderTT,relativetoSQ. 321 FigureC.8:SensitivityoftheICERstovariationsinthecostparameters–EnhancedTT(TT +Test6mo)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Annual health care costs of acute HIV Annual health care costs of symptomatic HIV (treated with ART) Cost of posttest counseling + linkage to care (HIV+) Cost per antibody test Cost per CD4 count test Cost per physician visit Cost per HIV RNA test Cost per HIV genotype test Cost of pretest counseling Cost of posttest counseling (HIV−) Annual health care costs of symptomatic HIV (untreated) Annual health care costs of AIDS (treated with ART) Annual cost of ART Cost of HIV diagnosis Annual health care costs of asymptomatic HIV (untreated) Annual health care costs of AIDS (untreated) 36,000 36,500 37,000 37,500 38,000 38,500 39,000 39,500 40,000 40,500 41,000 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Cost Parameters Note:ThehorizontalbarsrepresenttheICERsrangesforEnhancedTT(TT+Test6mo;strategy6inTable4.3 andonFigure4.2)relativetoEnhancedTT(TT+Test1y;strategy5onFigure4.2),astheparametervaluesare changedtotheboundsoftheiruncertaintyranges. Theredlineandblackdotsatthecenteroftheplotdenote thenonsensitizedICERestimateof$38,492/QALYgainedunderEnhancedTT(TT+Test6mo),relativeto EnhancedTT(TT+Test1y). 322 FigureC.9:SensitivityoftheICERstovariationsinthecostparameters–PrEP(TT+Test6 mo+PrEP4y)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Cost of posttest counseling + linkage to care (HIV+) Cost of posttest counseling (HIV−) Cost per HIV genotype test Cost per CD4 count test Cost per HIV RNA test Monthly cost of PrEP Cost of pretest counseling Cost per test (serum BUN and creatinine levels) Annual health care costs of acute HIV Annual health care costs of symptomatic HIV (untreated) Annual health care costs of symptomatic HIV (treated with ART) Cost per antibody test Cost per test (other sexuallty−transmitted infections) Annual cost of ART Cost of HIV diagnosis Annual health care costs of AIDS (treated with ART) Annual health care costs of asymptomatic HIV (untreated) Cost per physician visit Annual health care costs of AIDS (untreated) 45,000 50,000 55,000 60,000 65,000 70,000 75,000 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Cost Parameters Note: The horizontal bars represent the ICERs ranges for PrEP (TT + Test 6 mo + PrEP 4 y; strategy 7 in Table 4.3 and on Figure 4.2) relative to Enhanced TT (TT + Test 6 mo; strategy 6 on Figure 4.2), as the parametervaluesarechangedtotheboundsoftheiruncertaintyranges.Theredlineandblackdotsatthecenter oftheplotdenotethenonsensitizedICERestimateof$63,269/QALYgainedunderPrEP(TT+Test6mo+ PrEP4y),relativetoEnhancedTT(TT+Test6mo). 323 FigureC.10: SensitivityoftheICERstovariationsinthecostparameters–EnhancedPrEP (PrEP+Test3mo+ImmediatePrEP)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Cost of pretest counseling Cost per CD4 count test Cost of posttest counseling + linkage to care (HIV+) Annual health care costs of acute HIV Cost per test (serum BUN and creatinine levels) Cost per antibody test Monthly cost of PrEP Cost per HIV genotype test Cost per HIV RNA test Cost per test (other sexuallty−transmitted infections) Cost of posttest counseling (HIV−) Annual health care costs of AIDS (treated with ART) Cost per physician visit Annual health care costs of symptomatic HIV (treated with ART) Annual cost of ART Annual health care costs of symptomatic HIV (untreated) Annual health care costs of asymptomatic HIV (untreated) Annual health care costs of AIDS (untreated) Cost of HIV diagnosis 215,000 220,000 225,000 230,000 235,000 240,000 245,000 250,000 255,000 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Cost Parameters Note: The horizontal bars represent the ICERs ranges for Enhanced PrEP (PrEP + Test 3 mo + Immediate PrEP; strategy 13 in Table 4.3 and on Figure 4.2) relative to Enhanced PrEP (PrEP + Test 3 mo + PrEP 1.2 y; strategy12onFigure4.2),astheparametervaluesarechangedtotheboundsoftheiruncertaintyranges. The red line and black dots at the center of the plot denote the nonsensitized ICER estimate of $23,4726/QALY gainedunderEnhancedPrEP(PrEP+Test3mo+ImmediatePrEP),relativetoEnhancedPrEP(PrEP+Test3 mo+PrEP1.2y). 324 FigureC.11:SensitivityoftheICERstovariationsintheefectivenessparameters–TT(SQ+ ImmediateEarlyART)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Reduction in QOL due to false−positive test (multiplier) QOL: Acute HIV (treated with ART) QOL: Uninfected QOL: AIDS (aware, treated with ART) QOL: Symptomatic HIV (aware, treated with ART) Reduction in QOL due to ART side effects (multiplier) QOL: Asymptomatic HIV (aware, treated with ART) QOL: Asymptomatic HIV (unaware) QOL: Asymptomatic HIV, Y ear 1 (aware, untreated) QOL: Acute HIV (aware) Reduction in QOL due to drug−resistance QOL: Symptomatic HIV (unaware) QOL: AIDS (unaware) QOL: AIDS (aware, untreated) QOL: Symptomatic HIV (aware, untreated) QOL: Asymptomatic HIV, Y ear 2+ (aware, untreated) QOL: Acute HIV (unaware) 18,900 19,000 19,100 19,200 19,300 19,400 19,500 19,600 19,700 19,800 19,900 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Effectiveness Parameters Note: ThehorizontalbarsrepresenttheICERsrangesforTest-and-Treat,TT(SQ+ImmediateEarlyARTat CD4>500; strategy 2 in Table 4.3 and on Figure 4.2) relative to Status Quo (SQ; strategy 1 on Figure 4.2), as theparametervaluesarechangedtotheboundsoftheiruncertaintyranges. Theredlineandblackdotsatthe centeroftheplotdenotethenonsensitizedICERestimateof$19,302/QALYgainedunderTT,relativetoSQ. 325 FigureC.12:SensitivityoftheICERstovariationsintheefectivenessparameters–Enhanced TT(TT+Test6mo)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● QOL: Symptomatic HIV (aware, untreated) Reduction in QOL due to drug−resistance QOL: Symptomatic HIV (aware, treated with ART) QOL: AIDS (aware, treated with ART) Reduction in QOL due to ART side effects (multiplier) QOL: Acute HIV (unaware) QOL: Asymptomatic HIV, Y ear 2+ (aware, untreated) QOL: Acute HIV (treated with ART) Reduction in QOL due to false−positive test (multiplier) QOL: Acute HIV (aware) QOL: Asymptomatic HIV (unaware) QOL: AIDS (unaware) QOL: Symptomatic HIV (unaware) QOL: Asymptomatic HIV (aware, treated with ART) QOL: Uninfected QOL: Asymptomatic HIV, Y ear 1 (aware, untreated) QOL: AIDS (aware, untreated) 37,400 37,600 37,800 38,000 38,200 38,400 38,600 38,800 39,000 39,200 39,400 39,600 39,800 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Effectiveness Parameters Note:ThehorizontalbarsrepresenttheICERsrangesforEnhancedTT(TT+Test6mo;strategy6inTable4.3 andonFigure4.2)relativetoEnhancedTT(TT+Test1y;strategy5onFigure4.2),astheparametervaluesare changedtotheboundsoftheiruncertaintyranges. Theredlineandblackdotsatthecenteroftheplotdenote thenonsensitizedICERestimateof$38,492/QALYgainedunderEnhancedTT(TT+Test6mo),relativeto EnhancedTT(TT+Test1y). 326 FigureC.13:SensitivityoftheICERstovariationsintheefectivenessparameters–PrEP(TT +Test6mo+PrEP4y)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● QOL: Acute HIV (unaware) QOL: Acute HIV (treated with ART) QOL: Asymptomatic HIV, Y ear 1 (aware, untreated) Reduction in QOL due to ART side effects (multiplier) QOL: Asymptomatic HIV, Y ear 2+ (aware, untreated) QOL: AIDS (aware, treated with ART) QOL: Asymptomatic HIV (aware, treated with ART) QOL: Acute HIV (aware) Reduction in QOL due to drug−resistance QOL: Asymptomatic HIV (unaware) QOL: Symptomatic HIV (unaware) QOL: Symptomatic HIV (aware, treated with ART) QOL: AIDS (aware, untreated) Reduction in QOL due to false−positive test (multiplier) QOL: Symptomatic HIV (aware, untreated) QOL: AIDS (unaware) QOL: Uninfected 54,000 56,000 58,000 60,000 62,000 64,000 66,000 68,000 70,000 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Effectiveness Parameters Note: The horizontal bars represent the ICERs ranges for PrEP (TT + Test 6 mo + PrEP 4 y; strategy 7 in Table 4.3 and on Figure 4.2) relative to Enhanced TT (TT + Test 6 mo; strategy 6 on Figure 4.2), as the parametervaluesarechangedtotheboundsoftheiruncertaintyranges.Theredlineandblackdotsatthecenter oftheplotdenotethenonsensitizedICERestimateof$63,269/QALYgainedunderPrEP(TT+Test6mo+ PrEP4y),relativetoEnhancedTT(TT+Test6mo). 327 FigureC.14:SensitivityoftheICERstovariationsintheefectivenessparameters–Enhanced PrEP(PrEP+Test3mo+ImmediatePrEP)strategy. ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● QOL: AIDS (aware, treated with ART) Reduction in QOL due to ART side effects (multiplier) QOL: Asymptomatic HIV, Y ear 1 (aware, untreated) QOL: Acute HIV (treated with ART) QOL: Asymptomatic HIV (aware, treated with ART) Reduction in QOL due to drug−resistance QOL: Acute HIV (aware) QOL: Acute HIV (unaware) QOL: Symptomatic HIV (aware, untreated) QOL: Asymptomatic HIV (unaware) QOL: AIDS (aware, untreated) QOL: Symptomatic HIV (aware, treated with ART) QOL: Asymptomatic HIV, Y ear 2+ (aware, untreated) QOL: AIDS (unaware) QOL: Symptomatic HIV (unaware) QOL: Uninfected Reduction in QOL due to false−positive test (multiplier) 215,000 225,000 235,000 245,000 255,000 Incremental costeffectiveness ratios (ICERs), 2013 US$ per QALY gained Effectiveness Parameters Note: The horizontal bars represent the ICERs ranges for Enhanced PrEP (PrEP + Test 3 mo + Immediate PrEP;strategy13onFigure4.2)relativetoEnhancedPrEP(PrEP+Test3mo+PrEP1.2y;strategy12inTable4.3 andonFigure4.2),astheparametervaluesarechangedtotheboundsoftheiruncertaintyranges. Theredline and black dots at the center of the plot denote the nonsensitized ICER estimate of $234,726/QALY gained underEnhancedPrEP(PrEP+Test3mo+ImmediatePrEP)relativetoEnhancedPrEP(PrEP+Test3mo+ PrEP1.2y). C.5.2 Bootstrappingprobabilisticsensitivityanalysis Weassignedeachmodelinputparametervalueaprobabilitydistributionandanuncertainty range derived from the published literature. The uncertainty ranges associated with each modelvariableareprovidedinTableC.5-TableC.16. 328 Fourtypesofprobabilitydistributionswereusedinthisanalysis:theprogramevaluationand reviewtechnique(PERT),lognormalnormal,anduniformdistributions. ThePERTdistri- butionisavariantofthebetadistribution,andthetwoarerelatedbythefollowingrelation- ship[2,11,41,96] PERT(a,b,c) = Beta(α 1 ,α 2 )(c−a) +a, (C.83) whereaandcaretheminimumandmaximumvaluesoftheparameteruncertaintyrangeand bisthemostlikelyvalue(mode).Further,theparametersα 1 andα 2 arecalculatedasfollows: α 1 = (μ−a)(2b−a−c) (b−μ)(c−a) (C.84) α 2 = α 1 (c−μ) (μ−a) , (C.85) where μ = (a + 4b +c) 6 . (C.86) The equation for the meanμ allows for the determination of the values ofα 1 andα 2 . It also helps demonstrate how the mean for the PERT distribution is 4 times more sensitive tothemostlikelyvaluethanitistotheminimumandmaximumvalues,unlikethetriangle distribution in which the mean is equally sensitive to each parameter. Because of this, the PERTdistributiondoesnotsuferfromthepotentialsystematicbiasproblemsofthetriangle distributioninoverestimatingthevalueofthemeanoftherobustnessanalysisresultsinwhich themaximumforthedistributionisverylarge. Further, thisdistributionisrelativelyeasily denedandhasaexibleshape,anditisthereforewellsuitedformodelingexpertestimates andpolicyvariables.[82,96] 329 Kenney and Keeping [46] show that there is an empirical relationship between the sample meanmedianandmode: x−θ ≈ 3(x− ˜ x), (C.87) wherex,θand ˜ xdenoterespectivelythemeanmodeandmedianvalues.Further,Hozoetal. [38]showthatthemeanandstandarddeviationsofaparametercanbeestimatedfromthe medianandminimumandmaximumasfollows: x ≈ (x min + 2˜ x +x max ) 4 , and (C.88) S ≈ v u u u t 1 12 [x min − 2˜ x +x max ] 2 4 + (x max −x min ) 2 ! , (C.89) wherex min ,x max ,and ˜ xarerespectivelytheminimum,maximum,andmedianvalues. We sample the life expectancy parameters from a PERT distribution. We sampled the epi- demic,policyandefectivenessparametersaccordingtothePERTdistribution,whereasthe costparametersweresampledfollowingalognormaldistributiontoaccountfortheskewed andfat-tailnatureofcostdata[88]. Becausewehadonlytherangesandmediansofthecost parameters,weadoptedtheapproachinHozoetal.[38]toestimatethemeanandstandard deviations.Thelocationandscaleparametersforthelognormaldistributionwerethencalcu- latedasfollows: μ = ln x 2 √ S 2 +x 2 ! and (C.90) σ = v u u u u u t ln     1 + S 2 x 2     . (C.91) 330 Usingtheseparametervalues,wethensampledthecostparametersaccordingtothelognormal distribution. Other parameters were sampled following a normal or uniform distribution. Theprobabilitydistributionswereassumedtobeindependentfromeachother. AMonteCarlosimulationwasthenconducted,wherebytheresultsofthemodelwerecalcu- latedasmanytimesforeachpolicyscenario,eachtimerepresentingamodelsimulation.Fora singleiterationofthesimulation,asinglevaluewassampledfromeachparameteruncertainty rangefollowingthepre-specieddistributiontoobtainafullsetofvaluesforallmodelvari- ables. Thisprocesswasiterated2,000timedinthisstudy. Wethenusedthe2,000sampled parametersetstotestthesensitivityofourresultstodiferentassumptionsaboutparameter values.ThesesimulatedvalueswerealsousedtocalculatethesimulationaverageICERvalue andtoconstructthebootstrapping95%simulationintervalsaroundtheseaverageICERval- ues,undereachpolicyscenario. Althoughtheapproachdescribedaboveismoredicultboththeoreticallyandcomputation- ally,ithastheadvantageofyieldingmuchmorerobustandusefulresultsthanthetraditional one-wayordeterministicsensitivityanalyses,andithelpstoavoidsomeofthetheoreticalchal- lengesassociatedwiththeoreticaldistributions. C.5.3 Averagesocietalwillingnesstopaythresholds. Using the approach described above, we rst estimated the share of ICERs that remained cost-efectiveatvariousaveragesocietalwillingness-to-paythresholds,andasthemodelinput parametervalueswerevariedwithintheiruncertaintyranges.TheresultfortheICERsrelative tostatusquoaresummarizedinFigureC.15whichsuggeststhatallICERsestimatedfromour bootstrappingprocessremaincostefectiveattheaveragesocietalwillingnesstopaythreshold of$150,000/QALYgained. 331 FigureC.15: Sensitivityanalysis: Shareofcost-efectiveiterationsasafunctionofthesocietal averagewillingnesstopay. 0.0 0.2 0.4 0.6 0.8 1.0 Average Willingness to Pay (2013 US$, '000) Share of Cost−Effective Iterations 0 10 20 30 40 50 60 70 150 160 Test−and−treat, TT (SQ + Immediate Early ART at CD4>500; strategy 2) Enhanced TT (TT + Test 6 mo; strategy 6) PrEP (TT + Test 6 mo + PrEP 4 y; strategy 7) Enhanced PrEP (PrEP + Test 3 mo + Immediate PrEP; strategy 13) Average U.S. willingness−to−pay threshold ($150,000/QAL Y gained) Note:AllICERsarecalculatedrelativetoStatusQuo.Simulationsarebasedon2,000iterations;all parametervaluesaredrawnfromtheiruncertaintyrange,followingaPERT,lognormal,normal,or uniformdistribution. Wealsopresenttheresultsforthesharesofcost-efectiveiterationswhentheICERsarecom- puted relative to the prior strategy on the frontier (Figure C.16). They suggest that at the societalwillingnesstopaythresholdof$150,000/QALYgained,thetest-and-treat,enhanced Test-and-Treat, and PrEP strategies remain cost-efective, when compared to the prior e- cient strategy on the frontier. The most aggressive enhanced PrEP strategy (PrEP + Test 3 mo+ImmediatePrEP;strategy13onFigure4.2)ishowevercost-inefective,relativetoPrEP 332 enhancedwithHIVtestingevery3months,andPrEPstartevery1.2years(PrEP+Test3mo +PrEP1.2y;strategy12onFigure4.2). FigureC.16:Sensitivityanalysisrelativetopriorecientstrategy:Shareofcost-efectiveitera- tionsasafunctionofthesocietalaveragewillingnesstopay. 0 50 100 150 200 250 300 350 0.0 0.2 0.4 0.6 0.8 1.0 Average Willingness to Pay (2013 US$, '000) Share of Cost−Effective Iterations Test−and−treat, TT (SQ + Immediate Early ART at CD4>500) relative to Status−quo (SQ) Enhanced TT (TT + Test 6 mo) relative to Enhanced TT (TT + Test 1 y) PrEP (TT + Test 6 mo + PrEP 4 y) relative to Enhanced TT (TT + Test 6 mo) Enhanced PrEP (PrEP + Test 3 mo + Immediate PrEP) relative to Enhanced PrEP (PrEP + Test 3 mo + PrEP 1.2 y) Average U.S. willingness−to−pay threshold ($150,000/QAL Y gained) Note:ICERsforeachstrategyarecalculatedrelativetothepriorecientstrategyontheecientfron- tier.Simulationsarebasedon2,000iterations;allparametervaluesaredrawnfromtheiruncertainty range,followingaPERT,lognormal,normal,oruniformdistribution. C.5.4 Sensitivityonpricereductionfollowinggenericentry InordertoaccountfortheefectofreductionsinARTandPrEPcostscausedbygenericentry followingpatentexpiration,wealsoconductedasensitivityanalysisofICERstovariouslevels 333 ofgenericpricediscountasapercentageofbrandprices.Priorstudieshavesuggestedthatsuch pricediscountscouldbeupto70%ofthebrandprice[71,73]. WerstinvestigatedtheefectsofthesepricereductionsontheICERsfortheTest-and-Treat (SQ+ImmediateEarlyART;strategy2onFigure4.2),Test-and-TreatenhancedwithHIV testingevery6months(TT+Test6mo;strategy6onFigure4.2),PrEP(TT+Test6mo+ PrEP4y;strategy7onFigure4.2),andPrEPenhancedwithimmediatePrEPstartandHIV testingevery3months(PrEP+Test3mo+ImmediatePrEP;strategy13onFigure4.2),relative tothestatusquostrategy.AllICERsdeclinewithincreasedgenericcompetition,indicatingan improvementinthecost-efectivenessprolesofthestrategies(FigureC.17).FigureC.17also suggeststhattheICERestimatesremainrobusttoperturbationsinthemodelinputvaluesat allgenericpricediscountlevels,asevidencedbythealignmentofthesimulationmeanvalues withthebasecaseICERestimates. Further, allbasecaseICERestimatesliewithinthe95% simulation intervals for all strategies and at all generic price discount levels, although these intervalsarewide. WealsoinvestigatedtheefectsofpricereductionsontheICERsforeachstrategy,relativeto itsprecedingstrategyontheecientfrontier.TheresultsarealsodepictedinFigureC.18and suggest similar trend as explained above. The ICERs continue to remain robust to generic pricechanges. 334 FigureC.17:RobustnessofICERstochangesinthemodelinputparametervaluesatvarious genericpricediscount(ICERsrelativetoStatusQuo). Test−and−treat, TT (SQ + Immediate Early ART at CD4>500; strategy 2) Enhanced TT (TT + Test 6 mo; strategy 6) PrEP (TT + Test 6 mo + PrEP 4 y; strategy 7) Enhanced PrEP (PrEP + Test 3 mo + Immediate PrEP; strategy 13) 16000 17000 18000 19000 20000 21000 22000 23000 24000 25000 24000 25000 26000 27000 28000 33000 34000 35000 36000 37000 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Generic Price Discount as % of Brand Price ICER Relative to Status Quo, 2013 US$/QALY Base case Simulation mean 95% simulation interval bounds Note: The generic price discounts presented as percent of brand price. ICERS are calculated relative to the Status Quo strategy. Simulations are based on 2,000 iterations; all parameter values are drawn from their uncertaintyranges,followingaPERT,lognormal,normal,oruniformdistribution. 335 FigureC.18:RobustnessofICERstochangesinthemodelinputparametervaluesatvarious genericpricediscount(relativetopriorecientstrategy). Test−and−treat, TT (SQ + Immediate Early ART at CD4>500; strategy 2) relative to Status−quo (SQ; strategy 1) Enhanced TT (TT + Test 6 mo; strategy 6) relative to Enhanced TT (TT + Test 1 y; strategy 5) PrEP (TT + Test 6 mo + PrEP 4 y; strategy 7) relative to Enhanced TT (TT + Test 6 mo; strategy 6) Enhanced PrEP (PrEP + Test 3 mo + Immediate PrEP; strategy 13) relative to Enhanced PrEP (PrEP + Test 3 mo + PrEP 1.2 y; strategy 12) 16000 17000 18000 19000 34000 35000 36000 37000 38000 39000 61000 62000 63000 233000 233500 234000 234500 235000 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Generic Price Discount as % of Brand Price ICER Relative to Status Quo, 2013 US$/QALY Base case Simulation mean 95% simulation interval bounds Note: The generic price discounts presented as percent of brand price. ICERS are calculated relative to the priorecientstrategyontheecientfrontier(seeTable4.3andFigure4.2). Simulationsarebasedon2,000 iterations;allparametervaluesaredrawnfromtheiruncertaintyranges,followingaPERT,lognormal,normal, oruniformdistribution. 336 C.6 Systematicreviewoftheliterature Wedevelopedinclusionandexclusioncriteriaandasearchstrategytohelpasystematicscreen- ingofstudies. Wesearchedpredeneddatabasesforsearchtermscoveringthereviewtopic andscreenedthestudiesgatheredfromthesesearchesforcontentandquality. Theincluded studieswerethencodedandcategorizedfollowingpredenedinclusionandexclusioncrite- ria. Wedidnotconductaquantitativemeta-analysisthattestsforrelativediferencesacross interventions,becausesuchanexercisegoesbeyondthescopeofthepresentstudy.However, weextractedoutcomesacrossincludedstudiestoenablecomparativequalitativeanalysis. C.6.1 Inclusionandexclusioncriteria This review included studies that assessed the cost-efectiveness of treating or screening HIV/AIDSamongtheMSMpopulationintheUS.Weonlyincludedpeer-reviewedjournals articlesintheEnglishlanguage,publishedafterDecember31,2002. Astudywasincludedas relevanttothisreviewifitmetalloftheinclusioncriteriaoutlinedinTableC.23.Studieswere excludediftheymetanyoftheexclusioncriteria. 337 TableC.23:Inclusionandexclusioncriteriafortheliteraturereview. Aspect InclusionCriteria ExclusionCriteria Language English OtherthanEnglish Setting UnitedStates Othercountry StudyDesign Cost-efectivenessanalysis,mathematicalmodeloftheHIV/AIDSepidemic, experimental or quasi-experimental studies of HIV/AIDS screening, treat- ment,andprophylaxisinterventions Other than experimental and quasi- experimental Population IncludestheMSMpopulation,aged13-65y DoesnotincludetheMSMpopulation TopicandContextRelevance Cost-efectivenessofincreasedHIV/AIDSscreeningandtreatmentamong MSMpopulationinUnitedStates;CostofHIV/AIDSscreeningandtreat- ment;EfectofHIVdiagnosisortreatmentonhealth-relatedQOL Topicfocusandcontextdonotapplytothe reviewquestion Outcomes At least 1 of the following outcomes: infections averted, new infec- tions, quality-adjusted life years (QALY), QOL, costs, incremental cost- efectiveness,incrementalcost-efectivenessratio(ICER) Difer from infections averted, new infec- tions,QALY,costs,ICER Results ImputabletotheMSMpopulationintheUnitedStates Do not present results imputable to the MSMpopulationintheUnitedStates 338 C.6.2 Searchstrategy WesearchedtheCochrane,GoogleScholar,JSTOR,PubMed,WebofScience,andTuftsUni- versityCost-efectivenessAnalysisRegistrydatabasesforcost-efectivenessstudiesofscreen- ingandtreatmentstrategiesofHIV/AIDSintheMSMpopulationintheUnitedStates. We denedasearchstrategybyconstructingsearchtermsthatmimictheinclusioncriteriasum- marizedinTableC.24.Thesesearchtermswerechoseninamannerthatensuredoverinclusion tominimizetheriskofmissingrelevantarticles. 339 TableC.24:Searchcodes Database SearchCode Cochrane (costORefectiveness)AND(HIVORAIDS)AND(USAORUnitedStates)AND(MSMORmenwhohave sexwithmen)AND(test*ORtreat*ORscreen*ORprophylaxis) GoogleScholar Cost+efectiveness+HIV/AIDS+test+treat+screen+prophylaxis+MSM+menwhohavesexwithmen+ USA+UnitedStatesrestrictedtoyears2003-2013andexcludingpatentsandcitations. JSTOR (costORefectiveness)AND(HIVORAIDS)AND(USAORUnitedStates)AND(MSMORmenwhohave sexwithmen)AND(test*ORtreat*ORscreen*ORprophylaxis)AND(cty:(journal)ANDty:(aORbrv)AND (year:[2003 TO 2013]) AND la:(eng) AND disc:(economics-discipline OR health-discipline OR healthsciences- discipline) PubMed (cost[All Fields] OR efectiveness[All Fields]) AND (HIV[All Fields] OR AIDS[All Fields]) AND (test*[All Fields] OR treat*[All Fields] OR screen*[All Fields] OR prophylaxis[All Fields]) AND (MSM[All Fields] OR men who have sex with men[All Fields]) AND (USA[All Fields] OR United States[All Fields]) AND 2003/04/05[PDAT]:2013/04/01[PDAT]ANDhumans[MeSHTerms]) WebofScience Topic=(costORefectiveness)ANDTopic=(HIVORAIDS)ANDTopic=(test*ORtreat*ORscreen*ORpro- phylaxis)ANDTopic=(MSMORmenwhohavesexwithmen)ANDTopic=(USAORUnitedStates)ANDYear Published=(2003-2013) TuftsCEARegistry Searchedmanually. 340 C.6.3 Publicationscreening We rst screened all the titles in the reference universe and excluded all titles that evidently metanyoftheexclusioncriteria. Thetitlesgeneratedbythesearchstrategythatappearedto fulllourinclusioncriteriaandthosethatdidnotprovideenoughinformationtoascertain suitabilityforinclusionwereselectedforabstractreview.Consequently,75titleswereincluded forabstractsscreening. Second,andfollowingthetitlescreeningapproach,wescreenedthe abstracts of remaining articles, following the inclusion/exclusion criteria. The articles that werenotexcluded(iethosethatappearedtofulllourinclusioncriteriaandthosethatdidnot provideenoughinformationtoascertainsuitabilityforinclusion)wereretainedforretrievalof fulltexts.Followingthisstage,weretained55studiesforfurtherscreening.Finally,wescreened theintroductionsandthefull-bodyofthe55selectedstudiesfromtheabstractscreeningstage, followingthepredenedinclusion/exclusioncriteria. Analtotalof15studieswereselected forourreviewandincluded5systematicreviewstudies. C.6.4 Results Using the search strategy, on April 1, 2013, we searched the Cochrane, PubMed, Google Scholar, JSTOR, Web of Science, and the Tufts University Cost-efectiveness Registry databases for articles treating the review subject. These searches yielded 463 publications, whichwerecompiledasreferencesinEndNote.Afterremovingduplicatestudiesthedatabase searchyielded419uniquepotentialreferences. TableC.25summarizesthesearchresultsfor each of these sources. After the screenings of titles, abstracts, and full-texts, 15 studies were retainedforinclusioninthisreview.TableC.25representstheowchartofthescreeningpro- cess. 341 TableC.25:Summarystatisticsofthedatabasesearches. Database ArticleCount Cochrane 25 GoogleScholar 91 JSTOR 209 PubMed 39 TuftsUniversityCEARegistry 42 WebofScience 60 (LessDuplicates) -43 Total 419 FigureC.19:Flowdiagramoftheliteraturescreening. Ourreviewofthe15retainedstudiesindicatedasubstantialgapintheliteratureonthepoten- tialtrade-ofsofearlierandexpandedHIVtreatmentorincreasedfrequenciesofHIVscreen- ing compared with PrEP use in concentrated epidemics settings, such as that of the MSM populationintheUnitedStates.Therefore,weconcludedthatmorecost-efectivenessstudies thatgathernewevidenceonthisquestionareneeded.Thereviewofthesestudiesalsohelped 342 usidentifytheepidemic,cost,efectiveness,andpolicyparameterestimatesneededtopopu- lateandsimulateourmodel. Thefulltextoftheliteraturereviewandthelistofallreviewed studiesareavailablefromtheauthorsuponrequest. References [1] Abbas,U.L.,Anderson,R.M.andMellors,J.W. Potentialimpactofantiretroviralther- apyonHIV-1transmissionandAIDSmortalityinresource-limitedsettings. Journal of Acquired Immune Deficiency Syndromes,41(5):632–41,Apr152006. [2] Abramowitz,M.andStegun,I.A. Handbook of mathematical functions with formulas, graphs, and mathematical tables. NewYork,NY:Wiley,1972. [3] Baeten,J.M.etal. AntiretroviralprophylaxisforHIVpreventioninheterosexualmen andwomen. New England Journal of Medicine,367(5):399–410,Aug022012. [4] Baeten,J.M.etal. PreexposureprophylaxisforHIVprevention: wherehavewebeen andwherearewegoing? JournalofAcquiredImmuneDeficiencySyndromes,63(Suppl 2):S122–9,Jul2013. [5] Barnett, P.G. et al. Determinants of the cost of health services used by veterans with HIV. Medical Care,49(9):848–56,Sep2011. [6] Bollinger,R.C.etal. RiskfactorsandclinicalpresentationofacuteprimaryHIVinfec- tion in India. Journal of the American Medical Association, 278(23):2085–9, Dec 17 1997. [7] Bozzette, S.A. et al. Expenditures for the care of HIV-infected patients in the era of highlyactiveantiretroviraltherapy. New England Journal of Medicine,344(11):817–23, Mar152001. 343 [8] Brauer,F. Compartmentalmodelsinepidemiology. InBauer,F.,vandenDriessche,P. andWu,J.,editors, Mathematical Epidemiology,volume1945,chapter2,pages19–79. Oxford,UK:Springer,2008. [9] BureauofLaborStatistics. NationalCompensationSurvey(NCS). Availableathttp: //www.bls.gov/ncs/data.htm,Apr282015. [10] CenterforDiseaseControlandPrevention.HIVAmongGayandBisexualMen.Avail- ableathttp://www.cdc.gov/hiv/group/msm/,2015. AccessedDec15,2015. [11] Cheng,R.C.H. Generatingbetavariateswithnonintegralshapeparameters. Commu- nications of the ACM,21(4):317–22,1978. [12] Choopanya,K.etal.AntiretroviralprophylaxisforHIVinfectionininjectingdrugusers in Bangkok, Thailand (the Bangkok Tenofovir Study): a randomised, double-blind, placebo-controlledphase3trial. Lancet,381(9883):2083–90,Jun152013. [13] CMS. 2013ClinicalDiagnosticLaboratoryFeeSchedule. Availableathttp://www.cms. gov/apps/ama/license.asp?le=/ClinicalLabFeeSched/downloads/13CLAB.ZIP, 2013. AccessedJun05,2013. [14] Cohen,M.S.etal. Narrativereview:Antiretroviraltherapytopreventthesexualtrans- missionofHIV-1. Annals of Internal Medicine,146(8):591–601,Apr172007. [15] Cohen,M.S.etal. PreventionofHIV-1infectionwithearlyantiretroviraltherapy. New England Journal of Medicine,365(6):493–505,Aug112011. [16] Cohen,S.E.etal. ImplementationofPrEPinSTDClinicsandaCommunityHealth Center: HighUptakeandDrugLevelsamongMSMintheDemoProject. In CROI. Mar 3-6; Boston, MA,2014. 344 [17] Cohen,S.E.etal. Highinterestinpreexposureprophylaxisamongmenwhohavesex with men at risk for HIV infection: baseline data from the US PrEP demonstration project. Journal of Acquired Immune Deficiency Syndromes,68(4):439–48,Apr12015. [18] Cragin, L. et al. Cost-efectiveness of a fourth-generation combination immunoassay forhumanimmunodeciencyvirus(HIV)antibodyandp24antigenforthedetection ofHIVinfectionsintheUnitedStates. HIV Clinical Trials,13(1):11–22,Jan–Feb2012. [19] Daar, E.S. et al. Diagnosis of primary HIV-1 infection. Los Angeles County Primary HIVInfectionRecruitmentNetwork. Annals of Internal Medicine,134(1):25–9,Jan2 2001. [20] Donnell, D. et al. HIV protective ecacy and correlates of tenofovir blood concen- trations in a clinical trial of PrEP for HIV prevention. Journal of Acquired Immune Deficiency Syndromes,66(3):340–8,Jul12014. [21] Drummond,M.etal.MethodsfortheEconomicEvaluationofHealthCareProgrammes. Oxford,UK:OxfordUniversityPress,2005. [22] Eakle, R., Venter, W.D. and Rees, H. Pre-exposure prophylaxis for HIV prevention: ReadyforprimetimeinSouthAfrica? South African Medical Journal,103(8):515–6, Aug2013. [23] Ettner,S.L.andWeissman,J. UtilizationofformalandinformalhomecarebyAIDS patients in Boston: A comparison of intravenous drug users and homosexual males. Medical Care,32(5):459–70,May1994. [24] Ettner, S.L. etal. Triply-diagnosed patientsintheHIV/AIDSTreatment Adherence, HealthOutcomesandCostStudy:patternsofhomecareuse. AIDSCare,20(10):1177– 89,Nov2008. 345 [25] Farnham,P.G.etal. ComparingthecostsofHIVscreeningstrategiesandtechnologies inhealth-caresettings. Public Health Reports,123(Suppl3):51–62,Nov–Dec2008. [26] Gebo, K.A. et al. Costs of HIV medical care in the era of highly active antiretroviral therapy. AIDS,13(8):963–9,May281999. [27] Gold,M.R.etal. Cost-eectiveness in health and medicine: Report of the panel on cost- eectiveness in health and medicine. NewYork,NY:OxfordUniversityPress,1996. [28] Golub,S.A.etal.Fromecacytoefectiveness:FacilitatorsandbarrierstoPrEPaccept- ability and motivations for adherence among MSM and transgender women in New YorkCity. AIDS Patient Care and STDs,27(4):248–54,Apr2013. [29] Grant,R.M.etal.PreexposurechemoprophylaxisforHIVpreventioninmenwhohave sexwithmen. New England Journal of Medicine,363(27):2587–99,Dec302010. [30] Grant,R.M.etal.Scale-upofpre-exposureprophylaxisinSanFranciscotoimpactHIV incidence. InCROI. Feb 23-26. Seattle, W A.,2015. [31] Grijsen,M.etal. TemporaryantiretroviraltreatmentduringprimaryHIV-1infection hasapositiveimpactonhealth-relatedqualityoflife:DatafromthePrimo-SHMcohort study. HIV Medicine,13(10):630–5,Nov2012. [32] Hanmer, J., Lawrence, W.F. and Anderson, J.P. Report of nationally representative valuesforthenoninstitutionalizedUSadultpopulationfor7health-relatedquality-of- lifescores. Medical Decision Making,26(4):391–400,Jul–Aug2006. [33] Hecht,F.M.etal.Identifyingtheearlypost-HIVantibodyseroconversionperiod.Jour- nal of Infectious Diseases,204(4):526–33,Aug152011. [34] Hendrix,C.W. ExploringconcentrationresponseinHIVpre-exposureprophylaxisto optimizeclinicalcareandtrialdesign. Cell,155(3):515–8,Oct242013. 346 [35] Hogg,R.etal. Lifeexpectancyofindividualsoncombinationantiretroviraltherapyin high-incomecountries:Acollaborativeanalysisof14cohortstudies. Lancet,372(9635): 293–9,Jul262008. [36] Holtgrave,D.R.andPinkerton,S.D. Updatesofcostofillnessandqualityoflifeesti- matesforuseineconomicevaluationsofHIVpreventionprograms.JournalofAcquired Immune Deficiency Syndromes and Human Retrovirology,16(1):54–62,Sep11997. [37] Honiden,S.etal. TheefectofdiagnosiswithHIVinfectiononhealth-relatedquality ofLife. Qualilty of Life Research,15(1):69–82,Feb2006. [38] Hozo, S.P., Djulbegovic, B.andHozo, I. Estimatingthemeanandvariancefromthe median, range, and the size of a sample. BMC Medical Research Methodology, 5:13, 2005. [39] Hsu,H.andWalensky,R.P. Cost-efectivenessanalysisandHIVscreening: Theemer- gencymedicineperspective. Annals of Emergency Medicine,51(1Suppl1):S145–50,Jul 2011. [40] Hutchinson,A.B.etal. TheeconomicburdenofHIVintheUnitedStatesintheeraof highlyactiveantiretroviraltherapy:Evidenceofcontinuingracialandethnicdiferences. Journal of Acquired Immune Deficiency Syndromes,43(4):451–7,Dec12006. [41] Johnson,N.L.,Kotz,S.andBalakrishnan,N.Continuousunivariatedistributions.Wiley SeriesinProbabilityandMathematicalStatistics.NewYork,NY:Wiley&Sons,2edi- tion,1994. [42] Juusola,J.L.,Brandeau,M.L.andLong,E.F. Thecost-efectivenessofsymptom-based testingandroutinescreeningforacuteHIVinfectioninmenwhohavesexwithmenin theUnitedStates. AIDS,25(14):1779–87,Sep102011. 347 [43] Juusola,J.L.etal.Thecost-efectivenessofpreexposureprophylaxisforHIVprevention intheUnitedStatesinmenwhohavesexwithmen. Annals of Internal Medicine,156 (8):541–50,Apr172012. [44] Kamb,M.L.etal. Ecacyofrisk-reductioncounselingtopreventhumanimmunode- ciencyvirusandsexuallytransmitteddiseases: Arandomizedcontrolledtrial.Project RESPECTStudyGroup. JournaloftheAmericanMedicalAssociation,280(13):1161–7, Oct71998. [45] Karim,S.S. HIVpre-exposureprophylaxisininjectingdrugusers. Lancet,381(9883): 2060–2,Jun152013. [46] Kenney,J.F.andKeeping,E.S. Mathematicsofstatistics. NewYork,NY:VanNostrand company,3edition,1954. [47] Khazeni,N.etal. Efectivenessandcost-efectivenessofvaccinationagainstpandemic inuenza(H1N1)2009. Annals of Internal Medicine,151(12):829–39,Dec152009. [48] Koblin, B.A. et al. High-risk behaviors among men who have sex with men in 6 US cities:baselinedatafromtheEXPLOREStudy. AmericanJournalofPublicHealth,93 (6):926–32,Jun2003. [49] Koblin,B.A.etal. RiskfactorsforHIVinfectionamongmenwhohavesexwithmen. AIDS,20(5):731–9,Mar212006. [50] Koppenhaver,R.T.andSorensen,S.W. Thecost-efectivenessofpre-exposureprophy- laxisinmenwhohavesexwithmenintheUnitedStates:Anepidemicmodel. Journal of Acquired Immune Deficiency Syndromes,58(2):e51–2,Oct12011. 348 [51] Leonardi,M.,Lee,E.andTan,D.H.Awarenessof,usageofandwillingnesstouseHIV pre-exposure prophylaxis among men in downtown Toronto, Canada. International Journal of STD & AIDS,22(12):738–41,Dec2011. [52] Liu,A.etal. Earlyexperiencesimplementingpre-exposureprophylaxis(PrEP)forHIV preventioninSanFrancisco. PLoS Medicine,11(3):e1001613,Mar2014. [53] Liu,A.Y.etal. LimitedknowledgeanduseofHIVpost-andpre-exposureprophylaxis amonggayandbisexualmen. Journal of Acquired Immune Deficiency Syndromes,47 (2):241–7,Feb12008. [54] London,A.S.etal. UseofunpaidandpaidhomecareservicesamongpeoplewithHIV infectionintheUSA. AIDS Care,13(1):99–121,Feb2001. [55] Long, E.F. and Brandeau, M.L. The cost-efectiveness and population outcomes of expandedHIVscreeningandantiretroviraltreatmentintheUnitedStates. Annals of Internal Medicine,153(12):778–89,Dec212010. [56] Long, E.F., Brandeau, M.L.andOwens, D.K. Potentialpopulationhealthoutcomes and expenditures of HIV vaccination strategies in the United States. Vaccine, 27(39): 5402–10,Aug272009. ISSN1873-2518(Electronic)0264-410X(Linking). URLhttp:// www.ncbi.nlm.nih.gov/pubmed/19591796.Long,ElisaFBrandeau,MargaretLOwens, DouglasKengR-01-DA-15612/DA/NIDANIHHHS/R01DA015612/DA/NIDANIH HHS/R01DA015612-08/DA/NIDANIHHHS/ResearchSupport,N.I.H.,Extramu- ralResearchSupport,U.S.Gov’t,Non-P.H.S.Netherlands2009/07/1409:00Vaccine. 2009Aug27;27(39):5402-10.doi:10.1016/j.vaccine.2009.06.063.Epub2009Jul8. [57] Marks,G.etal.Meta-analysisofhigh-risksexualbehaviorinpersonsawareandunaware theyareinfectedwithHIVintheUnitedStates:implicationsforHIVpreventionpro- grams. Journal of Acquired Immune Deficiency Syndromes,39(4):446–53,Aug12005. 349 [58] Mayer, K. et al. Controlling the HIV epidemic with antiretrovirals: IAPAC consen- susstatementontreatmentaspreventionandpreexposureprophylaxis. Journal of the International Association of Providers of AIDS Care,12(3):208–16,May-Jun2013. [59] McCormick,A.W.andWalensky,R.P. Theefectofantiretroviraltherapyonsecondary transmissionofHIVamongmenwhohavesexwithmen. Clinical Infectious Diseases, 44(8):1115–22,Apr152007. [60] McNeil, D.G.J. Many Doctors Remain Unaware of Drug That Helps Pre- vent HIV Transmission. The New York Times, page A26, Nov 26 2015. Available at http://www.nytimes.com/2015/11/26/health/many-doctors-unaware-of- truvada-drug-for-preventing-hiv.html. [61] Meara,E.,White,C.andCutler,D.M. Trendsinmedicalspendingbyage,1963-2000. Health Aairs,23(4):176–83,Jul–Aug2004. [62] Mimiaga,M.J.etal.Preexposureantiretroviralprophylaxisattitudesinhigh-riskBoston area men who report having sex with men: Limited knowledge and experience but potentialforincreasedutilizationaftereducation. Journal of Acquired Immune Defi- ciency Syndromes,50(1):77–83,Jan12009. [63] Neuner, J.M. et al. Diagnosis and management of adults with pharyngitis. A cost- efectivenessanalysis. Annals of Internal Medicine,139(2):113–22,Jul152003. [64] O’Brien,B.J.etal. Oseltamivirfortreatmentofinuenzainhealthyadults:Pooledtrial evidenceandcost-efectivenessmodelforCanada. Value Health,6(2):116–25,Mar–Apr 2003. [65] Osmond,D.H.etal. ChangesinprevalenceofHIVinfectionandsexualriskbehavior inmenwhohavesexwithmeninSanFrancisco:1997-2002.AmericanJournalofPublic Health,97(9):1677–83,Sep2007. 350 [66] Owen,S.M.etal. Alternativealgorithmsforhumanimmunodeciencyvirusinfection diagnosisusingteststhatarelicensedintheUnitedStates. Journal of Clinical Microbi- ology,46(5):1588–95,May2008. [67] Paltiel,A.D.etal. ExpandedHIVscreeningintheUnitedStates:Efectonclinicalout- comes, HIV transmission, and costs. Annals of Internal Medicine, 145(11):797–806, Dec52006. [68] Paltiel, A.D., Freedberg, K.A. and Scott, C.A. HIV preexposure prophylaxis in the United States: Impact on lifetime infection risk, clinical outcomes, and cost- efectiveness. Clinical Infectious Diseases,48(6):806–15,Mar152009. [69] Prachakul,W.andGrant,J.S. InformalcaregiversofpersonswithHIV/AIDS:Areview andanalysis. Journal of the Association of Nurses in AIDS Care,14(3):55–71,May–Jun 2003. [70] Quinn, T.C. et al. Viral load and heterosexual transmission of human immunode- ciencyvirustype1.RakaiProjectStudyGroup. NewEnglandJournalofMedicine,342 (13):921–9,Mar302000. [71] Richardson,E.etal. HealthPolicyBrief:Biosimilars. Health Aairs,pages1–5,Oct10 2013. [72] Rose,C.D.etal.HIVinterventionforprovidersstudy:Arandomizedcontrolledtrialof aclinician-deliveredHIVrisk-reductioninterventionforHIV-positivepeople. Journal of Acquired Immune Deficiency Syndromes,55(5):572–81,Dec152010. [73] Saha,A.etal. GenericcompetitionintheUSpharmaceuticalindustry. International Journal of the Economics of Business,13(1):15–38,Feb222006. 351 [74] Sanders,G.D.etal. Cost-efectivenessofscreeningforHIVintheeraofhighlyactive antiretroviraltherapy. New England Journal of Medicine,352(6):570–85,Feb102005. [75] Schacker,T.etal.ClinicalandepidemiologicfeaturesofprimaryHIVinfection.Annals of Internal Medicine,125(4):257–64,Aug151996. [76] Schackman,B.R.etal.Thelifetimecostofcurrenthumanimmunodeciencyviruscare intheUnitedStates. Medical Care,44(11):990–7,Nov2006. [77] Smith, R.J.etal. EvolutionarydynamicsofcomplexnetworksofHIVdrug-resistant strains:ThecaseofSanFrancisco. Science,327(5966):697–701,Feb52010. [78] Smith,R.J.etal. SupplementaryMaterialto“Evolutionarydynamicsofcomplexnet- works of HIV drug-resistant strains: The case of San Francisco”. Science, 327(5966): 697–701,Feb52010. [79] Sonnenberg,F.A.andBeck,J.R.Markovmodelsinmedicaldecisionmaking:Apractical guide. Medical Decision Making,13(4):322–38,Oct–Dec1993. [80] Sood,N.etal. Test-and-treatinLosAngeles: Amathematicalmodeloftheefectsof test-and-treatforthepopulationofmenwhohavesexwithmeninLosAngelesCounty. Clinical Infectious Diseases,56(12):1789–96,Mar132013. [81] Sood,N.etal.MathematicalDetailsS1:HIVModelApproach;SupplementaryDatato “Test-and-TreatinLosAngeles:AMathematicalModeloftheEfectsofTest-and-Treat forthePopulationofMenWhoHaveSexWithMeninLosAngelesCounty”. Clinical Infectious Diseases,56(12):1789–96,2013. [82] Sood,N.etal. ReplytoGonzalez-Sernaetal. Clinical Infectious Diseases,57(3):478–9, Apr262013. 352 [83] Steward, W.T. et al. Behavior change following diagnosis with acute/early HIV infection–amovetoserosortingwithotherHIV-infectedindividuals.TheNIMHMul- tisiteAcuteHIVInfectionStudy:III. AIDS and Behavior,13(6):1054–60,Dec2009. [84] Tengs,T.O.andLin,T.H. Ameta-analysisofutilityestimatesforHIV/AIDS. Medical Decision Making,22(6):475–81,Nov–Dec2002. [85] The National Institute of Mental Health (NIMH) Multisite HIV Prevention Trial Group. TheNIMHMultisiteHIVPreventionTrial:ReducingHIVsexualriskbehav- ior. Science,280(5371):1889–94,Jun191998. [86] Thigpen,M.C.etal.AntiretroviralpreexposureprophylaxisforheterosexualHIVtrans- missioninBotswana. New England Journal of Medicine,367(5):423–34,Aug22012. [87] Thompson,M.A.etal. AntiretroviraltreatmentofadultHIVinfection: 2010recom- mendations of the International AIDS Society-USA panel. Journal of the American Medical Association,304(3),Jul212010. [88] Thompson, S.G. and Nixon, R.M. How sensitive are cost-efectiveness analyses to choiceofparametricdistributions? Medical Decision Making,25(4):416–23,Jul–Aug 2005. [89] Tieu,H.V.etal. SexualpartneringandHIVriskamongblackmenwhohavesexwith men:NewYorkCity. Journal of Urban Health,87(1):113–21,Jan2010. [90] Turner,D.A.etal. Thecost-efectivenessofinuenzavaccinationofhealthyadults50- 64yearsofage. Vaccine,24(7):1035–43,Feb132006. [91] US Census Bureau. S0101. Age and Sex: 2006-2008 American Community Survey 3-YearEstimates. Availableathttp://factnder.census.gov/servlet/STTable?_bm=y&- 353 geo_id=01000US&-qr_name=ACS_2008_3YR_G00_S0101&-ds_name=ACS_2008_ 3YR_G00_,2008. [92] USDepartmentofVeteransAfairs.FSS,BIG4andNationalContractPrices.Available athttp://www.pbm.va.gov/PharmaceuticalPrices.asp,Jun52013. [93] Van Damme, L. et al. The FEM-PrEP trial of emtricitabine/tenofovir disoproxil fumarate (Truvada) among African women. In 19th Conference on Retroviruses and Opportunistic Infections. Mar 5-8, Seattle, W A. Abstract 32LB,2012. [94] Vergis, E.N.andMellors, J.W. NaturalhistoryofHIV-1infection. Infectious Disease Clinics of North America,14(4):809–25,Dec2000. [95] Vernazza,P.etal. HIVtransmissionunderhighlyactiveantiretroviraltherapy. Lancet, 372(9652):1806–7,Nov222008. [96] Vose,D.Quantitativeriskanalysis: AguidetoMonteCarlosimulationmodelling.John Wiley&Sons,3edition,Sep2009. [97] Wilson,D.P. ModellingbasedonAustralianHIVnoticationsdatasuggestshomosex- ual age mixing is primarily assortative. Journal of Acquired Immune Deficiency Syn- dromes,51(3):356–60,Jul12009. [98] Wilson,D.P.etal. RelationbetweenHIVviralloadandinfectiousness:Amodel-based analysis. Lancet,372(9635):314–20,Jul262008. [99] WorldHealthOrganization. Thewordhealthreport2002:Reducingrisks,promoting healthylife. Technicalreport,WorldHealthOrganization,2002. 354

Abstract (if available)

Abstract Infectious diseases are increasingly posing a threat to public health and economies in both developing and developed countries. The US is not shielded from these threats, as evidenced by recent epidemic outbreaks of SARS, swine flu, Ebola and Zika virus. Vaccination is a proven cost-effective strategy to mitigate these threats. However, both children and adult vaccination rates in the US remain below the Healthy People 2020 target levels. Identifying strategies to increase vaccine uptake is therefore an important public health goal. In order to help shed some light on this issue, we use insights from behavioral economics to develop and conducted an experiment aimed at determining which of a “no-fault” insurance scheme against the risk of vaccine side effects vs a monetarily equivalent direct subsidy scheme (i.e. expected value of insurance) is most effective at boosting vaccine uptake. We present results showing that a “no-fault” insurance scheme can modestly increase vaccine uptake, compared to a subsidy scheme of an equivalent monetary value. This result leads us to conclude that it might be possible to leverage on the current vaccine injury compensation program in the US to boost vaccine uptake. ❧ Next, we also exploit techniques from economic epidemiology to investigate the impacts of different HIV/AIDS prevent strategies on the HIV/AIDS epidemic among men who have sex with men (MSM) in Los Angeles County. First, we develop a compartmental epidemiological model of HIV/AIDS transmission to assess the effect of the “test-and-treat” policy on reducing HIV transmission. We also assess the impact of this policy on the prevalence of multi-drug resistance. Second, we expand the policy set to include pre-exposure prophylaxis (PrEP) and develop an economic model to assess the trade-offs between the costs and benefits of these policies. This approach allows us to identify the efficient frontier for optimal allocation of resources. ❧ Collectively, these studies offer powerful tools for prioritizing and optimally allocating resources in order to effectively combat infectious diseases and avert their burden. We conclude our analysis by outlining some areas for future research.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays in pharmaceutical and health economics

PDF

Essays on the equitable distribution of healthcare

PDF

Essays on the use of microsimulation for health and economic policy analysis

PDF

Evaluating treatment options for metastatic, castration-resistant prostate cancer: a comprehensive value assessment

PDF

Economic aspects of obesity

PDF

Disease modifying treatments for Alzeimer's disease: modeling, value assessment and eligible patients

PDF

The impact of Patient-Centered Medical Home on a managed Medicaid plan

PDF

The causal-effect of childhood obesity on asthma in young and adolescent children

PDF

Three essays on the identification and estimation of structural economic models

PDF

Long-term impacts of childhood adversity on health and human capital

PDF

Three essays on estimating the effects of government programs and policies on health care among disadvantaged population

PDF

The effects of interlocal collaboration on local economic performance: investigation of Korean cases

PDF

Three essays on economics of early life health in developing countries

PDF

Essays on understanding consumer contribution behaviors in the context of crowdfunding

Asset Metadata

Creator Drabo, Emmanuel Fulgence (author)

Core Title Essays on the economics of infectious diseases

School School of Pharmacy

Degree Doctor of Philosophy

Degree Program Pharmaceutical Economics and Policy

Publication Date 07/13/2016

Defense Date 04/05/2016

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag antiretroviral therapy,cash incentive,cost-effectiveness,discrete choice experiment,drug resistance,expected utility theory,HIV/AIDS,immunization,Koszegi-Rabin utility theory,loss aversion,mathematical model,no-fault insurance,OAI-PMH Harvest,pre-exposure prophylaxis,PrEP,prospect theory,subsidy,test and treat,testing,vaccination,VICP

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Sood, Neeraj (committee chair), Doctor, Jason N. (committee member), Hay, Joel W. (committee member), Lakdawalla, Darius (committee member), Vardavas, Raffaele (committee member)

Creator Email drabo@usc.edu,edrabo@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-266887

Unique identifier UC11280251

Identifier etd-DraboEmman-4481.pdf (filename),usctheses-c40-266887 (legacy record id)

Legacy Identifier etd-DraboEmman-4481.pdf

Dmrecord 266887

Document Type Dissertation

Format application/pdf (imt)

Rights Drabo, Emmanuel Fulgence

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA