• Nenhum resultado encontrado

Expanding vaccine efficacy estimation with dynamic models fitted to cross-sectional prevalence data post-licensure

N/A
N/A
Protected

Academic year: 2021

Share "Expanding vaccine efficacy estimation with dynamic models fitted to cross-sectional prevalence data post-licensure"

Copied!
12
0
0

Texto

(1)

ContentslistsavailableatScienceDirect

Epidemics

jo u r n al h om ep ag e :w w w . e l s e v i e r . c o m / l o c a t e / e p i d e m i c s

Expanding

vaccine

efficacy

estimation

with

dynamic

models

fitted

to

cross-sectional

prevalence

data

post-licensure

Erida

Gjini

a,∗

,

M.

Gabriela

M.

Gomes

b,c,d

aInstitutoGulbenkiandeCiência,Apartado14,2781-901Oeiras,Portugal

bCIBIO-InBIO,CentrodeInvestigac¸ãoemBiodiversidadeeRecursosGenéticos,UniversidadedePorto,Portugal

cInstitutodeMatemáticaeEstatística,UniversidadedeSãoPaulo,Brazil

dLiverpoolSchoolofTropicalMedicine,Liverpool,UnitedKingdom

a

r

t

i

c

l

e

i

n

f

o

Articlehistory:

Received8May2015

Receivedinrevisedform2November2015

Accepted25November2015

Availableonline9December2015

Keywords:

Vaccinationmodel

Strainreplacement

Co-infection Competition

ODEparameterinference

a

b

s

t

r

a

c

t

The efficacy of vaccinesis typically estimated prior toimplementation, on the basisof

random-izedcontrolledtrials.Thisdoesnotpreclude,however,subsequentassessmentpost-licensure,while

mass-immunizationandnonlineartransmissionfeedbacksareinplace.Inthispaperweshowhow

cross-sectionalprevalencedatapost-vaccinationcanbeinterpretedintermsofpathogen

transmis-sionprocessesandvaccineparameters,usingadynamicepidemiologicalmodel.Weadvocatetheuseof

suchframeworksformodel-basedvaccineevaluationinthefield,fittingtrajectoriesofcross-sectional

prevalenceofpathogenstrainsbeforeandafterintervention.UsingSIandSISmodels,weillustratehow

prevalenceratiosinvaccinatedandnon-vaccinatedhostsdependontruevaccineefficacy,theabsolute

andrelativestrengthofcompetitionbetweentargetandnon-targetstrains,thetimepostfollow-up,

andtransmissionintensity.Wearguethatamechanisticapproachshouldbeaddedtovaccineefficacy

estimationagainstmulti-typepathogens,becauseitnaturallyaccountsforinter-straincompetitionand

indirecteffects,leadingtoarobustmeasureofindividualprotectionpercontact.Ourstudycallsfor

sys-tematicattentiontoepidemiologicalfeedbackswheninterpretingpopulationlevelimpact.Atabroader

level,ourparameterestimationprocedureprovidesapromisingproofofprincipleforageneralizable

frameworktoinfervaccineefficacypost-licensure.

©2015TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBY-NC-ND

license(http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction

Mathematical epidemiological models for the dynamics of microparasiteinfectionshavealonghistoryofdevelopmentand usein thedesign andoptimizationofinterventionprogrammes (Andersonetal.,1992).Yet,manychallengesremaininapplying suchmodelsretrospectivelytointerpret andquantify interven-tion effects in host–pathogen systems (Keeling, 2005; O’Hagan etal.,2014;Wikramaratnaetal.,2014;Goeyvaertsetal.,2015). It is of public interest to quantify the relative effectiveness of differentcontrolstrategies,assesstheongoingchangesin trans-missiondynamicsfollowingsuchinterventions,andoptimizetheir designthroughacost–benefitanalysisforthefuture.Inthispaper, ourfocusisonvaccinationasatransmission-reducing interven-tion,andmorespecifically,inthecontextofendemicpathogens. Althoughtheamountofdataavailablefromepidemiologicaltrials,

∗ Correspondingauthor.

E-mailaddress:egjini@igc.gulbenkian.pt(E.Gjini).

cross-sectionalandlongitudinalsurveysisvastandrapidly increas-ing, our understanding and interpretation of such data on the basisoftransmissionmechanismsandepidemiologicalfeedbacks is limited.Thisisapparentfor manypathogensystems, includ-ingStreptococcuspneumoniae bacteria,humanpapillomaviruses, dengue,malaria,influenzaandrotaviruses.Currentlyseveral vac-cinesarebeingusedorcontemplatedtocontrolthesepathogens aroundtheworld(Comanduccietal.,2002;Insingaetal.,2007;del AngelandReyes-delValle,2013;Sabchareonetal.,2012;Agnandji et al., 2011; Black et al., 2000), and assessing theirefficacy is crucial.

Conceptualmodelscanplayakeyroleinthisassessment,firstby clearlydefiningthemeasuresofinterest,secondly,by distinguish-ingindividualfrompopulationindicators,andthirdly,byenabling ustoanticipatefutureoutcomesofvaccinationprogrammes.An importantvaccineparameterisefficacyagainstpathogen acquisi-tion,definedasreductionintheprobabilityofinfectionpercontact ofeachvaccinatedindividual(Haberetal.,1991).Beforeavaccine isintroduced,vaccineefficacyestimationistypically performed throughrandomizedcontrolledtrials,involvingasubsetofagiven

http://dx.doi.org/10.1016/j.epidem.2015.11.001

1755-4365/©2015TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/4.

(2)

population(Halloranetal.,2010).Suchvaccineevaluationstudies use1−RR(1minusriskratio),asameasureofefficacy,whereRRis someestimateofrelativeriskinvaccinatedvs.non-vaccinated indi-viduals.Thistendstoignoreindirecteffects(Halloranetal.,1991), suchasthechangesintransmissionmediatedbytheintervention, whichwhileinthetimeandcoverageoftrialsareindeedexpected tobenegligible,arenotquitenegligiblewhenmass-immunization isinplace(ShimandGalvani,2012).

Theassessmentofvaccinespost-licensureisalsoofinterest,and hereiswheredynamicmathematicalmodelscanbeuseful, along-sidestatisticalapproaches(BiondiandWeiss,2015;Croweetal., 2014;Andrewsetal.,2014).Thereareseveralreasonsforwhysuch a-posterioriassessmentisimportant.First,onlyadynamicmodel canproperlylinkpre-licensurevaccineexpectationsandobserved outcomesinapopulationundergoingimmunization,thereby pro-vidingavaliditytestforthenumericalestimatesofvaccineefficacy obtainedfromtrials,andavaliditytestforthepublic-health pro-jectionsmadeaprioriregardingeffectiveness,orpopulationlevel impact.Second,onlyadynamicmodelcantakeintoaccountina mechanisticmannerthetimesincetheonsetofthevaccination programme,regardlessofequilibriumrequirements(Rinta-Kokko etal.,2009),andconsidertheactualvaccinecoverageinagiven setting.Third,inthecontextofmulti-strainpathogens,where mul-tivalentvaccinestargetasubsetofpathogentypes,onlyadynamic modelcanproperlyimplementthenonlinearinteractionsbetween pathogentypes(Lipsitch,1997; Martchevaet al.,2008), arising throughdirectcompetition,cross-immunityorasymmetricvaccine protection.

Although there has been recognition of the importance of dynamictransmissionmodelsforvaccineassessment(Shimand Galvani,2012),fewstudiessofarhaveattemptedtoinfervaccine efficacyfittingdynamicmodelstotemporalprevalence trajecto-riespost-vaccination(Choietal.,2011;Gjinietal.,2016).Other approaches havesuggested thatprevalenceodds ratiosmaybe moresuitablethanprevalenceratiostodeterminevaccineefficacy, andthatspecialattentionmustbegiventothetimeofsampling post-vaccination(Scottetal.,2014).AnotherstudybyOmorietal. (2012)hasuseddynamic models(SISand SIR)toillustratethe biasinodds-ratioestimatorsofvaccineefficacyfortwo compet-ingpathogentypes,buttheirestimationwasbasedonprevalences atendemicsteadystateonly,posinga strongrestrictiononthe method.A recent study by van Boven et al. (2013) dealswith vaccineefficacyestimationinanepidemicscenario,andapplies adynamicmodellingframeworktomumpsoutbreakdatainthe Netherlands.

Here, we advocate a similar dynamic spirit in the context of endemic diseases. We propose a novel approach to vaccine efficacy estimation using cross-sectional prevalence data inte-gratedwithindynamicmathematicalmodels.Thisenablesadeeper understandingofvaccineperformanceinthefield,asmediatedby transmissionintensity,competitionbetweenpathogensubtypes andhost factors.Whenvaccinecoverage is high, the transmis-sioncycleencompassesvaccinatedandnon-vaccinatedindividuals interactingthroughcontact,thus affectingand beingsubjectto adynamicforceofinfection.Withagraduallydiminishing expo-suretovaccinetypes,in polymorphicsystems, subtyperelative frequenciescanchangeinthepopulationfromthecombinedeffects of vaccination and interactions between target and non-target pathogentypes.Ifavaccineinducesareplacementphenomenon, asithasbeenarguedforpneumococcus(Weinbergeretal.,2011) andHPV(BiondiandWeiss,2015),vaccineefficacyagainsttargeted pathogenstrains,canbeestimatedwhilethesestrainsarestillin circulation,namelywhiletypereplacementisnotyetcomplete, andsufficientinformationcanbeextracted.Itispreciselyinthis intermediatedynamicphasethatmostvaccineobservational stud-iesareconducted,andwhereepidemiologicalfeedbacks,including

changesinexposureandinteractionbetweenmultiplestrains,are mostlikelytoplayarole.

Tocorrectlycapturealltheseprocesses,more refined math-ematical frameworks are needed. This requires going beyond direct statistical comparisons, based on static data, e.g. snap-shotprevalenceoddsratiosfromobservationalstudies(Thompson etal.,1998),ortheindirectcohortmethodforcase–controldata (Andrewsetal.,2011),whichneglectspathogensubtype interfer-encealtogether.Evenmoreimportantly,thecohortmethodfails toacknowledgethattheprobabilityofinfectionofanindividual dependsontheinfectionprevalenceinthepopulation,i.e.onthe infectionstatusofothers.

Withadynamicmodellingapproach,instead,theproblemof constanthazardratios(Hernán,2010)canbecircumvented,ascan limitationsoftheindirectcohortmethod(MoberleyandAndrews, 2014)forpurposes ofvaccine efficacyestimation.Furthermore, datacanbeinterpretedrelaxingthestationarityrequirementand accountingforpathogentypereplacement.Otherstatistical esti-mationmethodssuchasincidencedensitysampling(Richardson, 2004),mightalsonotrequiretheassumptionofstationarity,but theydonotdealwithcompetitioninmulti-strainpathogen sys-tems.

Thedefinitionofvaccineefficacythatweconsiderinthispaper hasa clear biological meaning: reduction of the probability of pathogen acquisition per contact, which enables extrapolation beyondasinglestudypopulation.Thiscontrastsclassicalestimates ofvaccineefficacythatarebasedoncomparingattackratesin vac-cinatedandunvaccinatedindividuals(thecohortmethod),orthose thatusethevaccinationstatusoftheinfectedindividualsrelativeto thepopulationvaccinationcoverage(thescreeningmethod).Such vaccineefficacyindicatorslackaclearbiologicalmeaning,which makesinterpretationproblematic,andpreventsanticipationofthe criticalvaccinationcoveragesneededtoreachcertaindesired out-comes.

Inthisstudy,wearguethattemporaleffectsofvaccination pro-grammescanbeaddressedthroughdynamicmathematicalmodels, where parameters of efficacyare explicitlydefined in terms of underlyingtransmission mechanisms,and where epidemiologi-calfeedbacksamongimmunizedandnon-immunizedindividuals, andbetweenpathogenstrainsarecorrectlyaccountedfor.Inthe interestofsimplicityandclarity,weonlyconsiderminimal epi-demiological modelsto illustrate vaccine effects on single and multipleinfectionwithdifferentpathogentypes,buttheuncovered trendsshouldapplyinsimilarveintomorecomplexvaccination scenarios (Halloran et al., 1991, 2010). We delineate a proof-of-concept inferenceprocedure, based onODEmodelfitting,to cross-sectionaldatacollectedoverdifferenttimepointsafter vac-cineimplementation.

2. Materialsandmethods

Tobuildintuitioninourreader,initiallywepresent susceptible-infected (SI) model frameworks accounting for one and two pathogen types, while the susceptible-infected-recovered (SIR) analogues are elaborated in the Supplementary Text S2. Then weproceedtosusceptible-infected-susceptible(SIS)modelswith many-typepathogens,groupedaccordingtowhethertheyare tar-getedbyapolyvalentvaccineornot.Wealwaysassumethatthe vaccineis effective againsttype 1 pathogen(SI/SIR models), or againstpathogensubtypesingroup1(SISsetting).Themodeof actionofthevaccineweconsiderisleaky(Halloranetal.,1991), andthevaccineefficacyisdefinedasthereductionin probabil-ityof infection/pathogenacquisition percontact. Noticethat in thispaper,wewillusetheterms‘infection’and‘carriage’ inter-changeably.Asthesourceofprevalencedata,weconsideractive

(3)

surveillanceprogrammespre-andpost-vaccinationina popula-tion,whereby thecarriage statusand pathogentype(s) of each screenedindividualaredetermined.Thebasicstructureofthe mod-elsintheabsenceofaninterventionisgiveninFig.1.

2.1. SImodel–1pathogentype(n=1)

The first model we consider for illustration is a simple susceptible-infectedmodelwithonepathogentypethatis directly-transmitted(SI-1).Withacontinuousvaccinationprogrammein place,theproportionsofhostsindifferentcompartmentsaregiven by:

Non-vaccinated hosts dS0 dt =(1−)−ˇS 0(I0+I1)S0 dI0 dt =ˇS 0(I0+I1)I0

Vaccinated hosts dS1 dt =−ˇwS 1(I0+I1)S1 dI1 dt =ˇwS 1(I0+I1)I1

wheresubscripts0and1indicatenon-vaccinatedandvaccinated hoststatus,respectively.Theparameterˇistheper-capita trans-missioncoefficientandthebirthrate(equaltothedeathrate). Hostsarebornwithalifeexpectancyof1/.Thevaccination cov-erageatbirthisandvaccineefficacyisgivenby1−w,assuming a homogeneouseffect(leaky vaccine). In theabsenceof a vac-cine,forsuchpathogentopersist,thebasicreproductionnumber (Heesterbeek,2000)R0=ˇ/mustexceed1,andthehigherR0is,

thehigherthepathogenprevalence.

2.2. SImodelwith2competingpathogentypes(n=2)

Extendingtheabovemodeltotwopathogentypes(SI-2),we have:

Non-vaccinated hosts dS0 dt =(1−)−(1+2)S 0S0 dI01 dt =1S 0I0 112−I01 dI02 dt =2S 0I0 221−I02 dI0 12 dt =12I 0 1+21I20−I120

Vaccinated hosts dS1 dt =−(w1+2)S 1S1 dI11 dt =w1S 1I1 112−I11 dI12 dt =2S 1I1 22w1−I12 dI1 12 dt =12I 1 1+2w1I21−I112,

where1=ˇ(I01+I012/2+I11+I121 /2)and2=ˇ(I20+I012/2+I21+

I1

12/2). The above equations track the proportions of

non-vaccinatedandvaccinatedhostsin4classes:susceptibles,S,hosts carryingpathogentype1,I1,hostscarrying pathogentype2,I2,

andco-infectedhostsI12,whereS0+



I0=1andS1+



I1=

respectively for non-vaccinated and vaccinated host compart-ments. Overall we have: S+



I=1. Upon primary pathogen exposure,asusceptiblehostcanacquirepathogentype1ortype 2.Theforcesofinfection(FOI)dependexplicitlyonprevalence: 1(t)fortype 1,and 2(t)fortype 2.Singlecarriers block

sub-sequentacquisitionof thesame typebut canacquiretheother pathogen type witha reduced rate i, this due to competition

betweentheresidentandthenewcomerstrain.Assumingno clear-ance,thereisanendemicpersistenceequilibriumwheneverˇ> intheabsenceofintervention(=0).Stablecoexistencebetween types requires further constraints on 1 and 2, as shown in

Fig.S1.

2.3. SISmodelfor2groupsofpathogentypes(n2)

For pathogens with larger antigenic diversity, many types can circulate simultaneously in a host population. The con-stituentpathogensubtypescanbeequivalentinmostlife-history traits,includingbasictransmissionandclearancepotential. How-ever,whenvaccinationwithpolyvalentvaccinesisconsidered,it becomespracticaltoaggregatethemintotypesofgroup1, (vac-cinetypes,i.e.thosethatwillbetargetedbyanintervention),and group2(remainingones,ornon-vaccinetypes).Typicallyahost mayacquireanypathogentypeofgroup1,orgroup2,orbea dou-ble carrieroftwotypes:fromgroup1,fromgroup2,oroneof each.

Acquisitionof a second pathogentype generally occursat a reducedratecomparedtosinglecarriage,duetodirectcompetition betweentheresidentandnewcomerpathogentypes.Competition betweenanytwopathogentypesisrepresentedbyparameter<1. Whentheybelongtothesamegroupweapplyanextrareduction factortoaccountforthefactthatthereisonelesstypeavailable forcolonizationwithinthesamegroup.Inthissenseshouldbe seenasafactorrepresentingdepletionofavailabletypes,which canexertasmallorlargeeffectoncoinfectionbytypeswithinthe samegroup,dependingonwhetherthegrouphasmanytypesor justafew.

Althoughinprincipleanyindividualcarriageepisodecaninduce sometype-specificimmunity,weconsiderthemagnitudeofsuch immunity negligible when assessed on the entire pool of all pathogentypesfromtheparentgroup.Thereis,however, noth-ingtoprecludetheinclusion ofcumulative immunityinmodel extensionsinformedbydata onspecific pathogens.Meanwhile, weareeffectivelyworkingwithSISepidemiologicaldynamicsat thelevel of groupsof subtypes.The systemwithvaccinationis givenby:

Non-vaccinated hosts dS0 dt =(1−)−(1+2+)S 0+(1S0) dI01 dt =1S 0I0 1(2+1)−(+)I10 dI20 dt =2S 0 I20(1+2)−(+)I20 dI011 dt =1I 0 1−(+)I 0 11 dI022 dt =2I 0 2−(+)I220 dI012 dt =(2I 0 1+1I20)−(+)I120

Vaccinated hosts dS1 dt =−(w1+2)S 1S1+(S1) dI11 dt =w1S 1 I11(2+w1)−(+)I11 dI12 dt =2S 1 I21(w1+2)−(+)I12 dI111 dt =w1I 1 1−(+)I111 dI122 dt =2I 1 2−(+)I221 dI112 dt =(2I 1 1+w1I12)−(+)I121 ,

(4)

Fig.1.Modeldiagramsintheabsenceofvaccination.Thedirectcompetitionphenomenonbetweenpathogensubtypesisrepresentedbythegreyarrows,modifyingthe

rateoftransitionfromsingletodualcolonization.

wheretheparameterdenotestheclearancerateofeachcarriage episode,assumedequalfor singleand dualcarriage,asinsome pneumococcusmodels(Colijnetal.,2009;Mitchelletal.,2015).The forces of infection are: 1=ˇ(I10+I110 +I120 /2+I11+I111 +I112/2)

and2=ˇ(I20+I220 +I120 /2+I21+I221 +I121 /2).Inthisformulation,

aswellasintheSI-2formulation,itisassumedthatanindividual carryingtwopathogensubtypesisonly50%infectiousforeither oneofthem,aspecialcaseofneutralnullmodels(Lipsitchetal., 2009).

Stabilityoftheendemicequilibriumintheabsenceof interven-tionrequiresR0=ˇ/(+)>1.Aconditionforstablecoexistenceof

thetwogroupsofpathogensubtypesintheabsenceofintervention isthat<1,whichisalwaystruegiventhemeaningof.Asinthe previousmodels,weassumethatafractionofthepopulationis vaccinated,withavaccinethatreducessusceptibilitytotargeted pathogentypes(heregroup1)byafactorw(0≤w≤1).Themodel canbegeneralizedtoincludefinerscalesofcompetitionbetween pathogentypes,ifimportantasymmetrieshappentobeimpliedby specificpathogendata.

2.4. Inferringvaccineefficacy

In all our transmission models, vaccineefficacy is given by VE=1−w, varying between 0 and 1, denoting the reduction in probability of pathogen acquisition per contact in vacci-natedindividuals relativetothosenon-vaccinated(Haberetal., 1991). We explore these models, with relation to retrospec-tiveanalysesofcross-sectionalprevalencedatapost-vaccination. Such data may be available through active sampling in obser-vational studies of populations subject to mass-immunization, and we assume they reflect pathogen carriage irrespective of symptoms.

We initially generate cross-sectional prevalence data post-vaccinationthroughnumericalsimulationofmodelsystemswith fixedparameters.Wegoontocomparetheratiooftargetpathogen prevalence(type1orgroup1)invaccinatedvs.non-vaccinated hosts,withthetruerelativeriskw,andwesystematicallyshow howthediscrepancybetweenthetwovarieswithtimeof obser-vation,withtransmissionintensityand competitionparameters betweenpathogensubtypes.ThisisperformedfortheSI-2and SIS-2models(butseeSupplementTextS2foranSIR-2formulation). Whatweproposeasasolutionistofitthefulldynamicmodelwith vaccinationtoprevalencetrajectories,andinferinthisway vac-cineefficacy,simultaneouslywithotherparameters.Noticethat thisapproachisdifferentfromtheonesuggestedbyScottetal. (2014),wheretheyadvocatethatsnapshotprevalenceratioitself, orprevalence-oddsratiobeused,onlyatappropriatetimes post-vaccination.Herewearenotproposingadirectuseofanystatic observationbutratheradynamicmodelfittomultipletemporal observationsofrelativeandabsoluteprevalenceofcarriage post-vaccination.

To motivate this approach, we illustrate the discrepancy betweenprevalenceratioandvaccineefficacy.Thus,webeginby providinganalyticalinsightusingthesimplerSI-1model,where itiseasytoderiveendemicprevalenceequilibriapre-and post-vaccination.Subsequentlyweexplorenumericallythemodelswith straincompetition.

3. Results

3.1. VaccineefficacyandendemicequilibriaoftheSI-1model Pathogencarriageprevalenceatthestableendemicequilibrium intheabsenceofvaccination(Section2.1,=0)isgivenbyIpre∗ =

1−1/R0,whereR0=ˇ/.Inthepresenceofvaccination(>0),by

settingthedifferentialequationstozero,wecancomputethenew endemicequilibriumpost-vaccine:

I∗=w(R0−1)−1+



4(1−)R0(1−w)w+[w(R0+1)−1]2

2R0w .

(1) Prevalencepost-vaccinebecomesclearlyanonlinearfunctionof basicreproductivenumberR0,vaccineefficacy(1−w),and

cov-erage  (Fig.2a).When thecoverage is perfect =1, we have I∗=1−(1/R0w).Onecanseefor examplethat toeliminatethe

pathogen,withperfectcoverage,wneedstobehigherthan1/R0,

orwithimperfectcoverage,thefractionvaccinatedandindividual vaccineprotectionwcanbejointlytraded-offagainstoneanother indifferentcriticalcombinations(i.e.thosethatsatisfyI*=0,inEq.

(1)).

These analytical expressions also illustratethat if we know theendemicprevalenceequilibriapre-andpost-vaccine,andthe vaccinationcoverage ,we caninfer simultaneouslyR0 and w,

hence vaccine efficacy (1−w), by comparingpre-vaccine with post-vaccine cross-sectional prevalence of carriage. The above expressioncanbeusedtoassesthe‘overall’effectivenessofa vac-cinationprogram,definedasthereductioninthetransmissionrate foranaverageindividualinapopulationwithavaccination pro-gramatagivenlevelofcoveragecomparedtoanaverageindividual inacomparablepopulationwithnovaccinationprogram(Halloran etal.,1991;Halloran,2006).

Similarly,ifweareinterestedintherelativeequilibrium preva-lence ratio (PR*) of infection in vaccinated and non-vaccinated

hosts,wecanalsoobtainitanalyticallyfromthemodel:

PR∗= I1∗/ I0∗/(1) = −1+w−R0w(1−2)+



4(1−)R0(1−w)w+[w(R0+1)−1]2 2R0w , (2)

(5)

0 5 10 15 20 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 R0 Equilibrium prevalence I *

A

0 5 10 15 20 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 w=0.5 R0

Relative prevalence ratio PR

*

B

w=0.1

w=0.5 w=0.1

Fig.2. Infectionprevalenceatpost-vaccinationequilibriumvs.transmissionintensity(SI-1model).(A)Absoluteprevalenceofcarriage.(B)Relativeprevalenceratioin

vaccinatedvs.non-vaccinatedhosts.Thedifferentlinesfromtoptobottomcorrespondtodifferentvaluesassumedforvaccineefficacy(VE=1−w=50%,90%)while

keepingfixedcoverage=0.5.Theratioofrelativeinfectionprevalenceinvaccinatedvs.non-vaccinatedindividualsatequilibriumdoesnotreflectthesamevalueofrelative

risk(w=1−VE)astransmissionintensityR0changes.Whileinthisfigure,allthepointsalongeachlinereflectscenarioswiththesamevaccineefficacy,theprevalenceratio

observedisdifferentforeachtransmissionintensityR0.Thisshowsthatonecannotsimplytranslate1−prevalenceratioatequilibriumtoadirectlyinterpretablemeasure

ofvaccineefficacy,asdefinedbythe1−wparameter,encapsulatingtheinstantaneousper-unitexposureprotectionfactorexperiencedbyeachvaccinatedindividual.We

omitanalysisofthecorrespondingSIS-1typesystem,astheresultsareidenticaltotheSI-1model.

whichrevealsthatthisquantityvariesnotonlywiththevaccine parametersandw,butalsowithtransmissionintensity,here rep-resentedbyR0 (Fig.2b).Thisanalyticalresultclearlystatesthat

prevalenceratio,evenatequilibrium,isunsuitableasadirect indi-catorofvaccineefficacy,aspreviouslynoted(Haberetal.,1991; ShimandGalvani,2012).Infact,PR*iscommonlygreaterthanw

exceptforspecialtripartitecombinationsof(R0,,w).

Nonethe-less,itispreciselysuchnonlinearrelationshipbetweenR0,vaccine

efficacy,andprevalence,thatliesattheheartoftypicalvaccination programmesagainstchildhooddiseases(typicallycharacterizedby SIRdynamics),wherethecriticalvaccinationcoverageneededto eliminateapathogenhasbeendeterminedthrough mathemati-calmodels,andappliedtocontrolmeasles,smallpox,mumps,and rubella.

Interestingly, when considering another ratio in our epi-demiologicalmodel, namely theprevalence odds ratio (POR),at equilibriumpost-vaccineweget:

POR∗= I1∗/S1∗

I0∗/S0∗ =w, (3)

confirmingtheclassicalresultthattheprevalence-odds-ratio,at leastinasimpleSI-1setting,isaperfectestimatoroftruerelative risk,providedweareatequilibrium.Thishasbeenshown previ-ouslybystudiescomparingprevalenceratiosandprevalence-odds ratios(Greenland,1987;Strömberg,1994;Thompsonetal.,1998). Next,webrieflyaddresshowinpolymorphicpathogensystems, therelationshipbetweenwandprevalenceratiodepends system-atically also onthe strength of competition betweenpathogen types,andfinalizewithourproposaltoinferwasafundamental parameterofadynamicmodelthatcanbefittedtocross-sectional datapre-andpost-vaccination.

3.2. Whenprevalenceratiopost-vaccinationismodulatedby straincompetitionandreplacement

Hereweconsidertheeffectsofsubtypecompetitioninpathogen systemswithmanystrains.Wesimulatehypotheticalvaccination scenariosusingthedynamicSI-2andSIS-2models(Sections 2.2 and2.3)withtwointeractingstrainsorgroupsofstrainsforTtime

unitspost-vaccination.Timeismeasuredinsameunitsashostlife expectancy1/.Focusingonthepathogentypestargetedbythe vaccine,andneglectingthecontributionofdualcarriersofvaccine andnon-vaccinetypes,theinstantaneousprevalenceratiooftype 1orgroup1pathogen,inthetwomodelsisdefinedas:

PR(t)=I 1 1(t) I0 1(t) ×1−  (SI-2model) PR(t)=I 1 1(t)+I111 (t) I0 1(t)+I110 (t) ×1−  (SIS-2model) (4)

IntheSI-2model,weconsidertwopossiblescenarios:(i)vaccine targetstype1,whentype1isdominant,and(ii)vaccinetargetstype 1,whentype2isdominantpriortointervention.Thesescenarios aredeterminedbytheratioofcompetitioncoefficientsı=1/2:

withtype 1dominanceifı<1andtype 2dominanceviceversa. When1=2,thetwopathogentypesstablycoexistatequal

abun-dances.Toexploretype1dominance,inFig.3,weset2=1and

consider1between0and1.Toexploretype2dominancebyan

equalamount,weset1=1andvary2.

InagreementwiththeSI-1analysis(Fig.2),theprevalenceratio fromtheSI-2model(Eq.(4))isalsomostcommonlyhigherthan thetruerelativeriskw(Fig.3).Inaddition,herewecanseethat prevalenceratioisverysensitivetotherelativemagnitudesof com-petitioncoefficients,especiallywhenthetargetstrainisdominant. Forvaccinesthattargetthedominanttypeweexpectwtobe over-estimated(vaccineefficacyunder-estimated),byalargeramount thanforvaccinesthattargetthenon-dominanttype.Thisindicates theimportanceofusingflexiblemodelformalismsthat accommo-datecompetitionparameterstobeestimatedsimultaneouslywith vaccineefficacy.Fromtheorderingofthecurvesinthisfigure,we canexpectthatusingamodelthatignorescompetition(1=2=1)

wouldgenerallyleadtoanoverestimationofw(underestimation ofvaccineefficacy)iftype1isdominantintherealsystem,andthe oppositeiftype2isdominant.

Intuitively,thiscanbeunderstoodbyseeingthevaccineand theintrinsiccompetitiveinteractionsinthesystemastwoforces thataffecttheconstituentstraindynamics.Whentype1is dom-inant,anda vaccinetargetstype1,thenaturaltendencyof the

(6)

0 20 40 0 0.2 0.4 0.6 0.8 1

Time post−vaccination (years)

PR(t), type 1 β =0.032 w 0 20 40 0 0.2 0.4 0.6 0.8 1

Time post−vaccination (years)

PR(t), type 1 β =0.32 w 0 20 40 0 0.2 0.4 0.6 0.8 1

Time post−vaccination (years)

PR(t), type 1 β =0.032 w 0 20 40 0 0.2 0.4 0.6 0.8 1

Time post−vaccination (years)

PR(t), type 1

β =0.32

w

Fig.3.Prevalenceratiooftargettypeinvaccinatedandnon-vaccinatedhostsvs.truerelativeriskintheSI-2model.Type1canbedominantpriortovaccination

(0.2≤1/2≤1),oralternatively,type2canbedominant(0.2≤2/1≤1).Thecolouredlinesfrombluetoredcorrespondtoincreasingvaluesofthecompetitionratio,

1/2(abovetheredcurve)and2/1from0.2to1(belowtheredcurve).Althoughastimeincreasestheprevalenceratiotendstowardstheparameterw,thevalueremains

biasedaboveduetocompetitionbetweentypesinbothcases,moresoiftype1isthebettercompetitor.Otherparametervalues:=0.0167,=0.5.Vaccinetargetstype

1.Differentvaluesofwareassumedintop/bottompanels:w=0.5andw=0.3respectively.Initialconditionsatendemicequilibrium.Timeisinunitsofyears,wherethe

averagelifeexpectancyisequalto60years.Thelowtransmissioncases(ˇ=0.032),correspondtoR0=1.9.Whilethehightransmissioncases(ˇ=0.32)correspondtoR0=19.

FortheanalogousfigurewithPR(t)takingintoaccountmixedcarriageI12seeFig.S4.(Forinterpretationofthereferencestocolourinthisfigurelegend,thereaderisreferred

tothewebversionofthisarticle.)

systemand thevaccinegoin oppositedirections, thusthe sys-temrespondsmoreslowly.Whentype2isdominantinstead,the vaccinetargeting1anddirectcompetitionactinthesame direc-tion,andweexpect fasterpropagationofvaccineeffects inthe entiresystem,whereby also thedifferencebetween vaccinated andnon-vaccinatedhostsemergesearlier(PR(t)→wfasterpost vaccination).

Inthismodel,competitionaffectsalsotheconvergenceofPOR (prevalenceoddsratio)oftype1pathogentow.Weexpectfromthe equilibriumanalysisoftheSI-1model,thatPOR(t)shouldbecloser tothetruerelativeriskwthanPR(t).Thisiswhatwefind.POR(t) tendsfastertowardwaftervaccinationisinplaceintheSI-2model. However,inthetwo-strainsystemwithdirectcompetition,POR(t) isalsoaffectedbyrelativestraindominance(Fig.S2),althoughto alesserextentthanPR(t),anddependingon1/2,italsomaynot

alwaysreachasymptoticallythetruew.

IntheSIS-2modelwith2groupsofpathogentypes,weuncover aregimewheretheprevalenceratioofaggregatedpathogentypes targetedbythevaccine(insingleandmultiplecarriage:I1+I11)

invaccinatedvs.non-vaccinatedhostsmaybebelowrelativerisk, andthusyieldanover-estimationofvaccineefficacyifitwereto beusedforthispurpose.Thisoccursforlargetransmission inten-sityˇ,andcloseto1(Fig.4).Whenissmallinstead,thepattern PR(t)>wcouldpersistindefinitely.Asincreases,thereisinitially adownwardbiasifobservationsaremadetoosoonafter interven-tiononset,andonlyaftersometime doesPR(t)tendtothetrue valueofrelativerisk.Dependingontheexactmagnitudeofˇand w,thedeviationofrelativeprevalenceratiofromwcouldpersistfor alongtimeafterthestartofvaccination(Fig.S3).Indeed,as trans-missionintensityincreases,theprevalenceofmultiplecarriagein thehostpopulationincreases,thusamplifyinganyindirecteffects ofcompetitionbetweentypes.

Noticethattheimpactofcompetitionhierarchiesonprevalence ratioPR(t)isverysensitivetohowthisprevalenceratioisdefined: whetherittakesintoaccountornot,multiplemixedcarriageof 1and2:I12.InFigs.S4andS5,weshowtheanalogousscenarios

ofFigs.3and4,foraprevalenceratiothattakesintoaccountthe globalFOIoftargettype(s),summingalsothecontributionI12/2of

themixedcarriagehostclass.ThesensitivityofPR(t)tocompetition hierarchies,inthiscasedecreasesintheSI-2model,andincreases drasticallyintheSIS-2model,especiallyforhighˇ,withparallel exacerbateddeviationfromw,indicatingthatmixedcarriageof targetandnon-targettypescontributesmoreconfoundingfrom indirectvaccineeffects.

3.3. Usingadynamicmodeltoinfervaccineefficacy

Asbrieflyillustratedabove,theuseofprevalenceratiostoassess vaccineefficacypresentsfourmainproblems:

(i)eveninthebestcaseofexplicitlymatchingpre-vaccineand post-vaccineequilibria(SI-1model),thusevenwhensatisfying thestationarityassumption,theindirecttransmissioneffects mediatedbythevaccinearetypicallyconfoundedinthe preva-lenceratiooftargettypesinvaccinatedandnon-vaccinated individuals(subjecttotheinterplaybetweenR0 andvaccine

coverage);

(ii)in generalfor observationalstudies,the prevalenceratiois dependentonthetimeofthesurveypost-vaccination,which makesitnon-robustandhardtocompareorinterpretacross settings,withoutmechanisticallyembeddingtimeinthe anal-ysis;

(iii)in polymorphicpathogensystems,prevalence-based indica-torscanbebiaseddependingonthemagnitudeanddirection

(7)

0 10 20 30 40 0.4 0.6 0.8 1 Time (months) PR(t), group 1 w σ =1 β =2 0 10 20 30 40 0.4 0.6 0.8 1 Time (months) PR(t), group 1 w σ =1 β =6 0 10 20 30 40 0.4 0.6 0.8 1 Time (months) PR(t), group 1 w σ =0.5 β =2 0 10 20 30 40 0.4 0.6 0.8 1 Time (months) PR(t), group 1 w σ =0.5 β =6

Fig.4. Deviationoftheprevalenceratiofromtruevaccine-inducedprotectionintheSIS-2modeldependsontransmissionrateandthemagnitudeofcompetition.Higher

transmissionintensity,leadstolargerdiscrepancyintheimmediatetimescaleaftervaccineintroduction.Fixedparametervalues=0.02,=0.5,=0.6andcompetition

coefficient=1and0.5respectivelyinthetop,andbottompanels.Initialconditionsatendemicequilibriumwherebothgroupsoftypescoexistatequalabundance.VE=50%

(dashedlinedepictsw=1−VE).Thedepletionfactoreffectingwithin-groupcoinfectionisvariedbetween0.1and0.9(colouredlinesfrombluetored).Increasingoverall

competitioninthesystemmakesindirecteffectsweaker,thusprevalenceratiosreflectmoreaccuratelytruerelativeriskw.Timeunitsaremonths,ashereweconsidera

childhooddisease,andthemeanageofhostsis50months.Bytype1werefertoallpathogentypesingroup1targetedbythepolyvalentvaccine.Fortheanalogousfigure

withPR(t)takingintoaccountmixedcarriageI12seeFig.S5.(Forinterpretationofthereferencestocolourinthisfigurelegend,thereaderisreferredtothewebversionof

thisarticle.)

of pre-existing competitiveinteractions between strainsor groupsofstrains,whosevaluesarehardtoknowandfactor outapriori.

(iv)in addition,inpolymorphicsystemsshapedbycompetition among strains, how the prevalenceratio of target types is definedintermsofsingleandmultiplecarriageofpathogentype combinationsisacriticaldeterminantofthediscrepancywith truerelativerisk,asitispreciselythedetailsofcompetition at co-colonizationthatdriveindirecteffects ofthevaccine, especiallyathightransmissionintensities.

Evenifweweretouseprevalence-odds-ratios(POR),instead ofprevalenceratios(PR),analysiscouldbeinaccuratewithregards toestimatingtruerelativerisk(andhencevaccineefficacy),asthe matchbetweenthesequantitiesgenerallyappliesonlyat equilib-riumpost-vaccine,anditisnotexemptfrombiasintroducedby samplingtimeandcompetitionhierarchiesbetweenstrains.

Individual vaccine protection, in the dynamic transmission models(0≤w≤1)multipliestransmissionrate(ˇ)andpathogen prevalence,thusitisnaturallydefinedperinfectiouscontact.In contrast, thesnapshot prevalence ratio, often reportedin field studies,missestheexposuredimension,beinganoutputofthe overalldynamicswithnonlinearandoftennon-trivialrelationto theoriginalinputparameter.Toresolvetheseproblems,a produc-tivealternativeistousethefulldynamicmodel,fitittoprevalence data before and after vaccination, as obtained through cross-sectionalobservationalstudies,thusestimatingtogetherseveral key epidemiologicalparameters. Based onpathogen prevalence observations, analysis of epidemiological studies post-licensure couldfactoroutsimultaneouslytheeffectsofmultipleparameters: ˇ,1,2(withintheSI-2model)andˇ,,(intheSIS-2model),

inordertoextractthetruevalueofvaccineefficacy(VE=1−w) againstthetargetedtypes.Frequently,epidemiologistswillhave enoughinformationaboutthesystemtodefinetheequationsthat governitsbehaviour,ortestsimultaneouslycompeting appropri-ateformulations.Inourcase,theparametersand(andinthe SIS-2model)areassumedknown.

3.4. NumericalprocedureforODEmodelfittingandparameter inference

Asaproofofconcept,weapplythefollowingprocedure.We generate hypothetical data performing model simulations with different parameter values, fixing initialconditions at the pre-vaccinationendemicstate.Observingthestateofthesystemat ti time units post-vaccination, subsequently we apply

nonlin-earleast-squaresoptimization(routinelsqnonlin inMATLAB) to model-generatedtrajectories,inordertorecovertheunderlying parameters.Similarmethodologicalapproachesexistandare rou-tinelyappliedtoepidemiological data(e.g.inthecontext ofR0

estimationCintrón-Ariasetal.(2009)).Theerrorfunction(objective functiontobeminimized)isgivenby:

Error=



ti



S,I1,I2...

(Theoretical model proportions(ti)−Data prevalences(ti))2,

(5) wheredatamayberealorsyntheticallygenerated,asinourcase here.Thisestimationdependsnotonlyonfittingtheprevalence of thepathogentypes targetedby thevaccine, butalso onthe prevalenceofsusceptibles,andofhostsinfectedwithtype2,at spe-cifictimepointspost-vaccinationinvaccinatedandnon-vaccinated

(8)

Fig.5.Biasinparameterestimationusingthefullepidemiologicalmodellingframework.Dynamicswith100randomparametercombinationsweresimulatedforeach

modelandthenonlinearleast-squaresoptimizationwasperformedonthereducedSI-2typeandSIS-2systemataspecifictimepost-vaccination.Fixedparametervalues

were:=0.5,=0.0167(SI-2)and=0.3,=0.02,=0.57(SIS-2).Initialconditionswerefixedatendemiccoexistenceequilibriumineachmodel.Assumingimperfect

observationsandafinitepopulationsizeN=1000,stochasticitywasimplementedintheobservationprocessbysamplinghostsfromamultinomialdistributionatt=0andat

3time-pointspost-vaccination(t1=12,t2=24,t3=36)whichintheSI-2modelreflect‘years’andintheSIS-2modelreflect‘months’.Themodelswerefittedtotheresulting

‘synthetic’sampleproportions.Inthebiasboxplots(bias=1−(ˆ)/),thecentralmarkisthemedian,theedgesoftheboxarethe25thand75thpercentiles,thewhiskers

extendtothemostextremebiaspointsnotconsideredoutliers,andoutliersareplottedindividually.ˇdenotesthetransmissionrateperunitoftime,1and2denotethe

directcompetitioncoefficientsintheSI-2model,whiledenotesthecompetitioncoefficientandthecompetitioncoefficientcorrectedforwithin-groupcoinfectionin

theSIS-2model.

hosts.Theerrorsareassumedtobenormallydistributedinthis framework,butthisassumptioncanbechangedinmore sophisti-catedparameterestimationprocedures.

Allowingforimperfectobservations(inthesyntheticdata), real-isticsamplingerrorcanbeaddedtodeterministicsimulations.For this,wedrawhostnumbersindifferentcompartmentsaccording toamultinomialprobabilitydistribution,withprobabilitiesfrom trajectoriesoftheODEmodel(SI-2,SIS-2),andagivensamplesize N.Byapplyingnonlinearleastsquaresoptimizationtointerpolate suchsyntheticsampleprevalencesgeneratedaftervaccination,we canrecoveralltheparametersofinterestwithverygoodaccuracy andinanunbiasedmanner(Fig.5).

Acritical requirementfor thesamplingof prevalences post-vaccinationisthatthesamplingtime-pointsaresufficientlyspread overtheinterval[0,1/],whichisintuitiveforavaccine admin-istered at birth, unless its implementation is preceded by a high-coveragevaccination campaign. In order to retrieve suffi-cientvaccineinformation,onehastofollowthedynamicsfromthe onsetofvaccinationatleastovertheentirelife-spanofatypical individual,i.e.overonegeneration.Providedthisminimalrange requirement,thebiasdecreasesfurtherwithsamplesizeandwith

thenumberoftime-pointsusedtosamplethepopulationinthe post-vaccineera(Fig.6).

Ingeneral,otheradvantagesofdeployingdynamicmodelsover moredirectstatisticaldescriptionsofdata,arethattheycanfillthe gapsforunobservedintervalsofsystembehaviour(Fig.7),generate newhypotheses,andyieldmoreaccuratepredictionsforthefuture. 3.5. Theroleoftype-specificimmunityandinferenceinaSIR model

Inthescenariosabove,werestrictedourattentiontoinfections withnoimmunity. Ina separatemodel,we extendedthebasic SI-2framework,toallowfortype-specificimmunityupon recov-ery,allelsekeptequal,includingvaccinationagainsttype1(see

SupplementaryTextS2).WhenchangingtoanSIR-2framework, thebaseline prevalenceof infection inthe pre-vaccine popula-tiondecreasesforthesameR0,andcompetitionbetweentypesis

reduced.Wenotethatprevalence-ratiodeviationsfromtrue vac-cineefficacypersistintheSIR-2model,andgetexacerbatedbythe oscillatorynatureofepidemiologicaldynamicsofinfectionswith immunity(Fig.S6).

(9)

2 3 4 5 6 −0.05 0 0.05 0.1 0.15 0.2

Number of time points post−vaccination

Mean bias over all parameters

N = 1000 0 500 1000 1500 2000 −0.05 0 0.05 0.1 0.15 0.2 Sample size N

Mean bias over all parameters

tpoints = 3

Fig.6.BiasdecreaseswithnumberofsamplingpointsandsamplesizeN.Here,weshowfortheSI-2model,howthemean(absolute)biasinparameterestimatesacross

differentvaluesfittedthroughthedynamicmodel,varieswithnumberofsamplingtimepointsandsamplesize.Thefirsttime-pointistakenat12timeunitspost-vaccination,

andthesubsequenttimepointsaretakeninstepsof12.Weassume=0.0167,correspondingtoalifeexpectancyof60yearsandforvaccinationcoverage=0.5.Thevalues

ofˇand1,2aredrawnrandomlyintherange[0.06,0.32]and[0,1],respectively,subjecttotheconstraintofstableendemiccoexistencepriortovaccination.Aslongas

thesamplingtimescovertheinterval[0,1/],informationaboutthevaccinewhichisadministeredatbirthcanbeextracted.

Fig.7. IllustrationofdynamicmodelfittingtoprevalencedataintheSI-2model.Aftervaccinationagainsttype1pathogen,cross-sectionalprevalencedata(circles)canbe

integratedwithinadynamicmathematicalmodel(lines)toestimateepidemiologicalparameters,suchasvaccineefficacyandcompetitionparametersbetweendifferent

pathogentypes.Hereonlyoverallprevalenceoftype1(blueline,filledcircles)andprevalenceoftype2(redline,emptycircles)areshown.Parametersusedinthissimulation:

ˇ=0.3,w=0.2,1=0.1,2=0.5and=0.5,=0.0167arefixed.N=1000in(a)andN=100in(b)hasbeenusedinthemultinomialsamplingschemetoaddsampling

errortosyntheticdata.Applyingthenonlinearleastsquaresroutine,intheseparticularinstances,wehaveobtainedthesepointestimates:(a)ˇ=0.297,w=0.215,1=

0.108,2=0.441and(b)ˇ=0.416,w=0.184,1=0.059,2=0.283.(Forinterpretationofthereferencestocolourinthisfigurelegend,thereaderisreferredtotheweb

versionofthisarticle.)

Whenapplyingdynamicmodelfitting,wewereabletoinfer theparametersofinterestˇ,w,1,2,alsofortheSIR-2model

fromprevalenceobservationspost-vaccination.Wegrouped‘data’ into:pathogen-freehosts(susceptibleandimmune),those carry-ingtype1,type2,andthosecarryingboth1and2.Underrandom samplingeffects(usingthemultinomialframework),eventhough transmissionrateandvaccineefficacycouldbeinferredaccurately, thequantificationofthedirectcompetitioncoefficients1and2

washarder(Fig.S7).Thisisunsurprising,giventhelowerpathogen prevalenceexpectedintheSIR-2model,asalargeproportionof hostseventuallybecomeimmunetobothtypesandcompetition informationislost.Infact,weexpectthatestimationofdirect com-petitionparametersinaSIRframeworkrequireslargersamplesizes andmoretime-points,oracompleteresolutionofODEvariables withregardstohostimmunestatusandserologicalhistory.

Taken together, our simulations convey that in principle, dynamicmodelfittingthroughinterpolatingpre-vaccineand post-vaccine pathogen prevalences can be used to retrospectively assessvaccineeffectsindifferenthostpopulations,under differ-entcompetitionscenariosbetweentargetandnon-targetstrains, accountingforessentialnonlinearitiesintransmission.

4. Discussion

In this study, we have suggested toclose the gap between mechanisticmodelsofvaccineeffectsandstatisticalapproaches for estimationof vaccineefficacy,addressing inaddition multi-typepathogens characterizedbydirectcompetition.Onereason for why mechanistic models have been somewhat neglected in vaccine assessment studies, is that for validation, dynamic

(10)

epidemiologicalmodelshavetorelyonspatiotemporallyresolved data,andtodate,epidemiologicaldatatendtobestatic,and there-foremoreamenabletothetoolsofstatistics.Yet,withincreasing infrastructuresandeffortstomonitorhealthinpopulationsover time,comesanopportunitytoapplymoredynamicframeworks in the future, and match them with dynamic vaccine study designs.

4.1. Theimportanceandchallengesofdynamicmodelling

Here,wedrawattentiononanewapplicationof mathemat-ical models: namely, in the inference of vaccine efficacy from cross-sectionalprevalencedata,accountingforothercritical deter-minantsofpathogendynamicssuchastransmissionrate,vaccine coverage,and competition betweenmultipletypes (Choiet al., 2011;Gjinietal.,2016).Thesedatacanbegatheredfromactive samplingand cross-sectional populationsurveyspre- and post-vaccination,nationalscreeningprogramsaftermassimmunization, orreportedpathogenprevalenceindifferentcountriesunder spe-cificvaccinationcoveragerates.

We focused on scenarios of treatment effect homogeneity (leakyvaccine)andbaselinehomogeneity(nostructureinthehost population),assumingrandomizationacrossvaccinatedand non-vaccinated hosts.Of coursethemodels canbeshaped soasto incorporatealternativescenarios,butthiswasbeyondourscope here.Depending on thetype of data that are available (e.g. if prevalenceinformationisstratifiedbyage,geographyetc.), epi-demiologicalmodelscanincludetransmissionbetweenagegroups ordifferentspatiallocations(e.g.asin(Goeyvaertsetal.,2015)).The spiritofinferenceremainsthesame;onlyinthosecasesonewould havetospecifycontactparametersandfeedbacksbetween differ-entsub-populations.In thecurrentpaper, themainstructuring dimensionwashostvaccinationstatus.

Theneed to modify current vaccine evaluation practices by including theexposure dimension in vaccine studies has been recently pointed out by Gomes et al. (2014), and additional computationalframeworksarealreadyemergingtoestimate per-exposurevaccineeffects at theindividuallevel (O’Haganet al., 2014).Interpretation of vaccineefficacyis inevitably entangled withcontext-specificepidemiologicalinteractionsandfeedbacks. Thus,accuratequantificationofsuchindividualprotection param-eter,whichisrobust,identifiableandcomparableacrosssettings, canbenefitfrominclusionofmoremathematicalapproaches in vaccinestudies.Anylengthofpost-vaccinefollow-upperiodcan beinterpretedwithindynamicmodels,aslongasthetime-scale ofobservationscoversthetypicalhostlife-spaninterval[0,1/], especiallyinthecaseofSIandSIRdynamics,andaslongas approx-imationsunderlyingmodelstructure(e.g.homogeneousmixing) arevalid.

Notice,thattheshortcomingofhavingtosampleoveran inter-valof1/time-unitstoextractepidemiologicalparametersand vaccineefficacy,ifnecessary,canbedealtwiththrough parame-terinferenceproceduresthatextractbaselineparameterssuchas transmissionrateandcompetitioncoefficientsfromfitting analyt-icalequilibriumexpressions toprevalencespre-vaccination,and thenusingthoseestimatestoprojectdynamicsforwardinthe vac-cineera.Inthatcase,withalreadyknowncoveragerateandknown baselineparameters,inferenceofvaccineefficacyalonethrough dynamicfittingofprevalenceobservationsovershortertime-scales shouldproveeasier.

Dependingonthetransmissionmodel,caremustbetakenwith regardstocorrelationbetweenparameters.Some epidemiologi-calquantities, e.g.total prevalenceoftarget types,maydisplay thesameglobalmagnitudeatagiventimepost-vaccination,for verydifferentparametercombinations(e.g.andVEintheSIS-2 model,showninFig.S8).However,iffiner-scaledataareavailable,

including stratification with respect to host vaccination status and single/multiplecarriage,theapparently correlated parame-terscanthenbeseparated.Thishighlightstheimportanceofhigh resolution and highquality epidemiological datafor parameter estimation.

Here we show that transmission rate and vaccine efficacy against targeted pathogen types can be robustly estimated by dynamicmodelfitting,eveninthepresenceofobservationerror. Yet,wealsoshowthatourpowertoidentifydirectcompetition coefficientsfrompost-vaccineprevalencedatamaydependonthe overalllevelofpathogenprevalenceinthepopulation.Infectious agentsthatdisplaylowerprevalencearelikelytorequirelarger samplesizesinprevalencestudies,foracorrectresolutionoftheir diversityandinferenceofunderlyingsubtypecompetition.Thisis especiallyrelevant insystemswhere hostsrecoverwith immu-nity.Lackofestimabilityofinteractioncoefficientshoweverdoes not necessarily meanthemodel assumingthose interactions is invalid;ratherit mayhint that thoseparticular parametersare notasimportantfortheoveralldynamicsatthepopulationscale considered(Wikramaratna etal.,2014), orthata better resolu-tionofhostimmunologicalstatus,besidessimplycarriagestatus,is needed.

4.2. Vaccineefficacyandeffectiveness:linkingtrialsand populationobservations

Noticethatwhileefficacyisadirectparameter,thatiseasily understoodand can be averagedacross settings,vaccine effec-tivenessis a somewhatmore subjective criterion: (i) it canbe definedoveraspecifictimepost-vaccination,e.g.reductionin over-allprevalenceinthe3yearsfollowingvaccination;(ii)inthecase ofmulti-typesystems,effectivenesscanbedefinedwithregards tocarriageofthetargetedpathogentypes,ortotalcarriage;(iii) itcanbedefinedwithregardstothevaccinatedsub-population only,orasanoverallmeasurefortheentirepopulation irrespec-tiveofvaccinationstatus;(iv)itcanreflectdiseasemanifestations ofthepathogenatthepopulationlevelorasymptomaticcarriage states.Allthese‘effectiveness’ indicatorsare importantoutputs oftheepidemiologicaldynamics,thatcanbecalculatedtomeet thedemands of various policy-makers,and while theymaybe different,theintrinsicvaccineprotectionVEperindividualisthe same.

When linking vaccine efficacy against acquisition obtained throughrandomizedcontrolledtrialspre-licensure,andthe vac-cineefficacyestimateddynamicallypost-licensure,weexpectin principlethetwoquantitiesshouldmatch.Thatiswhythe frame-work we propose couldalso serve as a validation test. Notice howeverthat oftenvaccinetrialsare conductedinone popula-tion,whilevaccinesareimplementedinanother.Thus,differences couldarise.For example,interferencewithmaternalantibodies againstlocalpathogensmayalterthevaccineprotectionagainst otherpathogensinanewvaccinationprogramme.Immunological responsesmayalsobeinfluencedbynutritionstatusandother fac-tors,thusthereisnothingtoprecludepost-licensureassessmentof VEinpopulationswheretrialswerenotconductedinthefirstplace. Alltheseissuescanbeexploredusingdynamicmodelsof transmis-siontointerpretvaccination-inducedchangesinpopulations,and extractthebasicprotectionparameter.Evenifthetwoquantitiesdo notmatch,onecouldarguethatstudyingtheirdiscrepancywould openthewayforbetterdatacollection,modelimprovement, fine-tuningofvaccinationcoverage,inclusionofhostheterogeneities, andultimatelyabetterunderstandingofthedynamics.Ideally,one shouldworktogetherwiththetwoapproaches:bothpre-licensure andpost-licensure,andtrytomatchpredictionswith retrospec-tiveanalyses.Althoughnotprimarilyintendedtoaddresswaningof immunity,adynamicframeworkisidealtoexplorealsotheissueof

(11)

thewaningofvaccine-mediatedprotection,especiallywhen inter-polatingprevalenceobservationsoverlongintervalssincetheonset ofavaccinationprogramme.

A dynamic model, byits natureof mechanistically connect-ingmulti-dimensionalepidemiologicalobservationsovertime,can explainandlinkalsomorekindsofdata,includingtheprevalenceof multiplecarriage,ofcompetingpathogentypes,andalterationsin prevalenceoftotalcarriageovertime.Allthesequantitiescould have implications for disease, for the evolutionarypotential of thepathogen,andinteractionswithotherpathogens.Adynamic modelthusoffersabroaderviewof‘effectiveness’.Predictionsfor pathogendynamicsinthepopulationcanbefurtherintegratedinto mechanisticmodelsforspecificsymptoms,asshownby Rodriguez-Barraqueretal.(2013).

4.3. Methodologicalprospects

Our aimsin this paperwereillustrative and conceptual.For thisreason,wedidnotelaborateextensivelyonmethodological aspects.ApplyingdynamicmodelfittingwithinaBayesian frame-work,toaccountforparameteruncertainty,isalsopossible.One coulduse thesame objective function asin Section 3.4,or an explicitmultinomiallikelihoodfunction,fortheexactnumbersof hostsobservedindifferentepidemiologicalclasses.Inthelatter case,thesamplesizeNandexpectedprobabilityvectorcoming fromnumericalintegrationoftheODEsystemwithitssetof param-eters,wouldenterexplicitlythelikelihoodterm.Moreover,prior informationcouldbeaddedaboutspecificparameters,ifavailable. OnesuchBayesianprocedureisexplainedindetail,andapplied toaspecificdatasetonpneumococcusvaccinationinPortugal,ina relatedpaperbytheauthors(Gjinietal.,2016),whereinadditionan analysisofcompetingmodelformulationsisundertaken.Another importantfactorisprocessnoise andstochasticity(Ellneretal., 1998;Shresthaetal.,2011).Parameterinferenceinmechanistic modelsbasedonsystemsofcoupledODEsisatimelyand compu-tationallychallengingproblem.Manyadvancesinthisrespectare beingmadeacrossfieldssuchasstatisticsandcomputing(Haario etal.,2006;Calderheadetal.,2009;GirolamiandCalderhead,2011; Dondelingeretal.,2013;Kingetal.,2014),andappliedsuccessfully acrossdomainsofscience,includingsystemsbiology,astrophysics andclimatestudies.Refinement ofsuchmethodologiesfor epi-demiologicalmodelswithvaccinationshouldbestraightforward giventhemultitudeoftoolsthatalreadyexistorarein develop-ment.Theconceptualbackboneofdynamicinferenceislikelyto applyacrosssystems,however,challengesinspecificdiseaseswill inevitablyinvolveadaptationofmodelstructureandrefinementof theinferenceprocedure.

Finally,therearemanywaysinwhichtwopathogenstrainscan competewitheachother,mostnotablyviacross-immunity,which wedidnotconsiderhere.Theexistenceofnaturalcross-immunity, suchasamongdengue(Adamsetal.,2006;WearingandRohani, 2006), orHuman Papilloma Virus strains(Elbasha and Galvani, 2005; Durham et al., 2012), poses the issueof eventual cross-immunityalsointhevaccine-inducedprotectionagainstparticular pathogen types.To account for immune-mediated interactions, modelsneedtospecifyhownaturallyacquiredtype-specific immu-nitymightinterferewithvaccine-inducedimmunity(Gomesetal., 2004)andvaccineeffectiveness(Omorietal.,2012).Robust infer-enceofvaccineprotectioninthesecasesremainsanopenareafor futureresearch,callingforsophisticatedvaccinestudydesigns,as dootherfactors,suchashostpopulationstructure,seculartrends andstochasticity.Anaccurateunderstandingofintervention effec-tivenessinpopulationsundermassimmunizationwillincreasingly requireintegratedmodellingframeworksforpre-andpost-vaccine dynamics,accountingfortheinterplayofallthesefactors.

AppendixA. Supplementarydata

Supplementarydataassociatedwiththisarticlecanbefound,in theonlineversion,athttp://dx.doi.org/10.1016/j.epidem.2015.11. 001.

References

Adams,B.,Holmes,E.,Zhang,C.,Mammen,M.,Nimmannitya,S.,Kalayanarooj,S.,

Boots,M.,2006.Cross-protectiveimmunitycanaccountforthealternating

epi-demicpatternofdenguevirusserotypescirculatinginBangkok.Proc.Natl.Acad. Sci.U.S.A.103(38),14234–14239.

Agnandji,S.T.,Lell,B.,Soulanoudjingar,S.S.,Fernandes,J.F.,Abossolo,B.P.,

Conzel-mann,C.,Methogo,B.,Doucka,Y.,Flamen,A.,Mordmüller,B.,etal.,2011.First

resultsofphase3trialofRTS,S/AS01malariavaccineinAfricanchildren.N.Engl. J.Med.365(20),1863–1875.

Anderson,R.M.,May,R.M.,Anderson,B.,1992.InfectiousDiseasesofHumans:

DynamicsandControl,vol.28.WileyOnlineLibrary.

Andrews,N.,Waight,P.a.,Borrow,R.,Ladhani,S.,George,R.C.,Slack,M.P.E.,Miller,

E.,2011.Usingtheindirectcohortdesigntoestimatetheeffectivenessofthe

sevenvalentpneumococcalconjugatevaccineinEnglandandWales.PLoSONE 6(12),e28435.

Andrews,N.J.,Waight,P.A.,Burbidge,P.,Pearce,E.,Roalfe,L.,Zancolli,M.,Slack,M.,

Ladhani,S.N.,Miller,E.,Goldblatt,D.,2014.Serotype-specificeffectivenessand

correlatesofprotectionforthe13-valentpneumococcalconjugatevaccine:a postlicensureindirectcohortstudy.LancetInfect.Dis.14(9),839–846.

Biondi,B.E.,Weiss,S.H.,2015.Impactofbivalenthumanpapillomavirus(HPV)

vac-cinationupontheriskofacquisitionofotherHPVtypes.Med.Res.Arch.1(1).

Black,S.,Shinefield,H.,Fireman,B.,Lewis,E.,Ray,P.,Hansen,J.R.,Elvin,L.,Ensor,

K.M.,Hackell,J.,Siber,G.,etal.,2000.Efficacy,safetyandimmunogenicityof

heptavalentpneumococcalconjugatevaccineinchildren.Pediatricinfect.Dis.J. 19(3),187–195.

Calderhead,B.,Girolami,M.,Lawrence,N.D.,2009.AcceleratingBayesianinference

overnonlineardifferentialequationswithGaussianprocesses.In:Advancesin NeuralInformationProcessingSystems,pp.217–224.

Choi,Y.H.,Jit,M.,Gay,N.,Andrews,N.,Waight,P.A.,Melegaro,A.,George,R.,Miller,

E.,2011.7-ValentpneumococcalconjugatevaccinationinEnglandandWales:

isitstillbeneficialdespitehighlevelsofserotypereplacement.PLoSONE6(10), e26190.

Cintrón-Arias,A.,Castillo-Chávez,C.,Bettencourt,L.M.,Lloyd,A.L.,Banks,H.,2009.

Theestimationoftheeffectivereproductivenumberfromdiseaseoutbreakdata. Math.Biosci.Eng.6(2),261–282.

Colijn,C.,Cohen,T.,Fraser,C.,Hanage,W.,Goldstein,E.,Givon-Lavi,N.,Dagan,R.,

Lipsitch,M.,2009.Whatisthemechanismforpersistentcoexistenceof

drug-susceptibleanddrug-resistantstrainsofstreptococcuspneumoniae?J.R.Soc. Interface,rsif20090400.

Comanducci,M.,Bambini,S.,Brunelli,B.,Adu-Bobie,J.,Aricò,B.,Capecchi,B.,Giuliani,

M.M.,Masignani,V.,Santini,L.,Savino,S.,etal.,2002.NadA,anovelvaccine

candidateofNeisseriameningitidis.J.Exp.Med.195(11),1445–1454.

Crowe,E.,Pandeya,N.,Brotherton,J.M.,Dobson,A.J.,Kisely,S.,Lambert,S.B.,

White-man,D.C.,etal.,2014.Effectivenessofquadrivalenthumanpapillomavirus

vaccineforthepreventionofcervicalabnormalities:case–controlstudynested withinapopulationbasedscreeningprogrammeinAustralia.BMJ348,g1458.

delAngel,R.M.,Reyes-delValle,J.,2013.Denguevaccines:stronglysoughtbutnot

arealityjustyet.PLoSPathog.9(10),e1003551.

Dondelinger,F.,Filippone,M.,Rogers,S.,Husmeier,D.,2013.ODEParameter

Infer-enceUsingAdaptiveGradientMatchingwithGaussianProcesses.

Durham, D.P., Poolman, E.M., Ibuka, Y., Townsend, J.P., Galvani, A.P., 2012.

Reevaluationofepidemiologicaldatademonstratesthatitisconsistentwith cross-immunityamonghumanpapillomavirustypes.J.Infect.Dis.206(8), 1291–1298.

Elbasha,E.H.,Galvani,A.P.,2005.VaccinationagainstmultipleHPVtypes.Math.

Biosci.197(1),88–117.

Ellner,S.,Bailey,B.,Bobashev,G.,Gallant,A.,Grenfell,B.,Nychka,D.,1998.Noise

andnonlinearityinmeaslesepidemics:combiningmechanisticandstatistical approachestopopulationmodeling.Am.Nat.151(5),425–440.

Girolami,M.,Calderhead,B.,2011.RiemannmanifoldLangevinandHamiltonian

MonteCarlomethods.J.R.Stat.Soc.:Ser.BStat.Methodol.73(2),123–214.

Gjini,E.,Valente,C.,Sa-Leao,R.,Gomes,M.,2016.Howdirectcompetitionshapes

coexistenceandvaccineeffectsinmulti-strainpathogensystems.J.Theoret. Biol.388,50–60.

Goeyvaerts,N.,Willem,L.,VanKerckhove,K.,Vandendijck,Y.,Hanquet,G.,

Beu-tels,P.,Hens,N.,2015.Estimatingdynamictransmissionmodelparametersfor

seasonalinfluenzabyfittingtoageandseason-specificinfluenza-likeillness incidence.Epidemics13,1–9.

Gomes,M.G.M., Lipsitch,M.,Wargo, A.R.,Kurath,G.,Rebelo, C.,Medley,G.F.,

Coutinho,A.,2014.Amissingdimensioninmeasuresofvaccinationimpacts.

PLoSPathog.10(3).

Gomes,M.G.M.,White,L.J.,Medley,G.F.,2004.Infection,reinfection,andvaccination

undersuboptimalimmuneprotection:epidemiologicalperspectives.J.Theoret. Biol.228(4),539–549.

Greenland,S.,1987.Interpretationandchoiceofeffectmeasuresinepidemiologic

(12)

Haario,H.,Laine,M.,Mira,A.,Saksman,E.,2006.DRAM:efficientadaptiveMCMC. Stat.Comput.16(4),339–354.

Haber,M.,Longini,I.M.,Halloran,M.E.,1991.Measuresoftheeffectsofvaccination

inarandomlymixingpopulation.Int.J.Epidemiol.20(1),300–310.

Halloran,M.,2006.Overviewofvaccinefieldstudies:typesofeffectsanddesigns.J.

Biopharmaceut.Stat.16(4),415–427.

Halloran,M.E.,Haber,M.,Longini, I.M.,Struchiner,C.J., 1991.Directand

indi-recteffectsinvaccineefficacyandeffectiveness.Am.J.Epidemiol.133(4), 323–331.

Halloran,M.E.,Longini,I.M.,Struchiner,C.J.,Longini,I.M.,Struchiner,C.J.,2010.

DesignandAnalysisofVaccineStudies.Springer.

Heesterbeek,J.,2000.MathematicalEpidemiologyofInfectiousDiseases:Model

Building,AnalysisandInterpretation,vol.5.Wiley.

Hernán,M.A.,2010.Thehazardsofhazardratios.Epidemiology(Cambridge,Mass.)

21(1),13.

Insinga,R.P.,Dasbach,E.J.,Elbasha,E.H.,Puig,A.,Reynales-Shigematsu,L.M.,2007.

Cost-effectivenessofquadrivalenthumanpapillomavirus(HPV)vaccination inMexico:atransmissiondynamicmodel-basedevaluation.Vaccine26(1), 128–139.

Keeling,M.J.,2005.Modelsoffoot-and-mouthdisease.Proc.R.Soc.Lond.B:Biol.Sci.

272(1569),1195–1202.

King,A.A.,Nguyen,D.,Ionides,E.L.,2014.Statisticalinferenceforpartiallyobserved

Markovprocessesviatherpackagepomp.J.Stat.Softw.55(2).

Lipsitch,M.,1997.Vaccinationagainstcolonizingbacteriawithmultipleserotypes.

Proc.Natl.Acad.Sci.U.S.A.94(12),6571–6576.

Lipsitch,M.,Colijn,C.,Cohen,T.,Hanage,W.P.,Fraser,C.,2009.Nocoexistencefor

free:neutralnullmodelsformultistrainpathogens.Epidemics1(1),2–13.

Martcheva,M.,Bolker,B.M., Holt,R.D.,2008.Vaccine-inducedpathogen strain

replacement:whatarethemechanisms?J.R.Soc.Interface5(18),3–13.

Mitchell,P.K.,Lipsitch,M.,Hanage,W.P.,2015.Carriageburden,multiple

coloniza-tionandantibioticpressurepromoteemergenceofresistantvaccineescape pneumococci.Philos.Trans.R.Soc.Lond.B:Biol.Sci.370(1670),20140342.

Moberley,S.,Andrews,R.,2014.Anevaluationoftheindirectcohortmethodto

esti-matetheeffectivenessofthepneumococcalpolysaccharidevaccine.J.Vaccines Immunol.2,4–6.

O’Hagan,J.J.,Lipsitch,M.,Hernán,M.A.,2014.Estimatingtheper-exposureeffectof

infectiousdiseaseinterventions.Epidemiology25(1),134–138.

Omori,R.,Cowling,B.J.,Nishiura,H.,2012.Howisvaccineeffectivenessscaledby

thetransmissiondynamicsofinteractingpathogenstrainswithcross-protective immunity?PLoSONE7(11),e50751.

Richardson,D.,2004.Anincidencedensitysamplingprogramfornestedcase-control

analyses.Occup.Environ.Med.61(12),e59.

Rinta-Kokko,H.,Dagan,R.,Givon-Lavi,N.,Auranen,A.,2009.Estimationof

vac-cineefficacyagainstacquisitionofpneumococcalcarriage.Vaccine27(29), 3831–3837.

Rodriguez-Barraquer,I.,Mier-yTeran-Romero,L.,Burke,D.S.,Cummings,D.A.,2013.

Challengesintheinterpretationofdenguevaccinetrialresults.PLoSNeglect. Trop.Dis.7(8),e2126.

Sabchareon,A.,Wallace,D.,Sirivichayakul,C.,Limkittikul,K.,Chanthavanich,P.,

Suvannadabba,S.,Jiwariyavej,V.,Dulyachai,W.,Pengsaa,K.,Wartel,T.A.,etal.,

2012.Protectiveefficacyoftherecombinant,live-attenuated,CYDtetravalent

denguevaccineinThaischoolchildren:arandomised,controlledphase2btrial. Lancet380(9853),1559–1567.

Scott,P.,Herzog,S.A.,Auranen,K.,Dagan,R.,Low,N.,Egger,M.,Heijne,J.C.,2014.

Tim-ingofbacterialcarriagesamplinginvaccinetrials:amodellingstudy.Epidemics 9,8–17.

Shim,E.,Galvani,A.P.,2012.Distinguishingvaccineefficacyandeffectiveness.

Vac-cine30(47),6700–6705.

Shrestha,S.,King,A.A.,Rohani,P.,2011.Statisticalinferenceformulti-pathogen

systems.PLoSComput.Biol.7(8),e1002135.

Strömberg,U.,1994.Prevalenceoddsratiovsprevalenceratio.Occup.Environ.Med.

51(2),143.

Thompson,M.L.,Myers,J.,Kriebel,D.,1998.Prevalenceoddsratioorprevalenceratio

intheanalysisofcrosssectionaldata:whatistobedone?Occup.Environ.Med. 55(4),272–277.

vanBoven,M.,Ruijs,W.L.,Wallinga,J.,O’Neill,P.D.,Hahne,S.,2013.Estimationof

VaccineEfficacyandCriticalVaccinationCoverageinPartiallyObserved Out-breaks.

Wearing,H.J.,Rohani,P., 2006.Ecologicalandimmunologicaldeterminantsof

dengueepidemics.Proc.Natl.Acad.Sci.U.S.A.103(31),11802–11807.

Weinberger,D.M.,Malley,R.,Lipsitch,M.,2011.Serotypereplacementindisease

afterpneumococcalvaccination.Lancet378(9807),1962–1973.

Wikramaratna,P.S.,Kucharski,A.,Gupta,S.,Andreasen,V.,McLean,A.R.,Gog,J.R.,

Referências

Documentos relacionados

corrente de injeção para os modelos de SOAs simulados de comprimento de 0,65 mm (a) e 2 mm (b), com respeito à variação da densidade de portadores de referência.. 3.3.3 Coeficiente

Para entender o processo de consolidação dos jornais mineiros também foi preciso se buscar nos estudos da história da imprensa brasileira definições das fases da imprensa no século

As doenças mais frequentes com localização na cabeça dos coelhos são a doença dentária adquirida, os abcessos dentários mandibulares ou maxilares, a otite interna e o empiema da

Daí a pertinência de averiguar as perceções/representações e a autoestima profissional dos professores do ensino básico face à aprendizagem na sala de aula dos

O Mestrado Profissional em Ensino de Biologia em Rede Nacional – PROFBIO visa, como proposta auxiliar, ao processo de ensino-aprendizagem de biologia no ensino médio,

Although the prevalence levels were the same in the wild animals of the Javaés River and the specimens of the breeding facility of the Praia Alta Ranch, the mean level of parasitemia

OBJECTIVES: A cross-sectional study was conducted to measure the prevalence of work-related injuries among physiotherapists in Malaysia and to explore the influence of factors such

A importância dessa intervenção para o município de Ubá é incorporar um novo pensar e agir, do usuário como protagonista de seu processo saúde- doença tornando