Overview of a “neglected” infectious disease:...

Overview of a “neglected” infectious disease: dengue

Outline

• Currentepidemiologicaltrends• Mechanismsunderlyingdenguetransmission• Denguecontrol• Overviewofrecentmodelingefforts

• Driversofserotypeinteractions• Implicationsforvaccines

• Discussion

Dengue in a nutshell

• Estimated400millionannualinfectionsworldwide– broadclinicalspectrumofinfection:silent→DF→DHF/DHSS

• DENVsingle-strandedRNAvirus(Flavivirus)• 4antigenically-distinct

serotypes• permanentimmunity

serotype-specific(?)• severediseaseinresponse

tosequentialinfection

Principally transmitted by peridomestic mosquito Aedesaegypti(alsoAedesalbopictus)

BloodMeals:•femalestakemultiplebloodmeals•feedasearlyas21hours•linkedtoa4daygonotrophiccycle•temperaturedependence: flightlimitedbelow21°C conflictingbitingresults

Oviposition:•laidinstandingwaterbutfoodavailabilityisimportant•eggscansurvivegreatextremes:(8.9°C-43.3°C)•activationafter1weekat18.3°Cormore

AdultMortality:•maximuminlab:225days;maximuminfield:42days•averageinfield:15days•temperaturedependent–15minsat48.9°C->100%mortality -extendedperiodsat40.6°C

Shifting epidemiology Increasingcasereports…

Shifting epidemiology

Increasingseverity…

DHFincidenceinAmericas,WHO/PAHO/CDC

Dengue in 2010

TheglobaldistributionandburdenofdengueBhattetal.,Nature2013

“Dengue Re-emerges in U.S., Spurring Race for Vaccine”, NY Times

“WilldenguefeverspreadinU.S.?Toosoontotell,expertssay”,USAToday

Dengue in U.S.? – Headlines from 2010

“MiamiHasFirstDengueFeverCasein50Years”,USNews

PublicHealthReports,Oct261934

… and 2013

Christoffersonetal.,Parasites&Vectors2014

Phylogeny of dengue

viruses

Twiddyetal.,Mol.Biol.Evol.2003

Strainreplacement–whatdeterminesspreadofnovelgenotypes?

Why is dengue re-emerging?

• Increasedhumanmovement(transportingvirus&initiallymosquitoes)

• Rapidurbanization,poorlivingconditions• Nosustainedeffectivevectorcontrol• Absenceofvaccine(until2016…butseelater)

What dengue case reports don’t reveal

• Tipoftheiceberg–howmanyinfectionsareasymptomatic?

• Serotype?Genotype?

• Individualhistoryofinfection?– 4!=24differentserotypesequences

Vector-borne Disease

Focusofpathogentransmission

Arboviruses

Arthropod-borneviruses

anthroponosis zoonosis

Dengue Transmission Cycleinfectivityofhoststovectors

infectivityofvectorstohosts

#bitespervector

probabilitythatexposedvectorsurvivesEIP

hostbitingrate

durationofhostinfection

Dengue Transmission Cycle

c

b a/g

e-gn

ma

1/r

Vectorial Capacity & Basic Reproduction Number

V =ma2e�gnc

g=) R0 =

b

rV

ratioofvectortohostpopulationdensity

vectorbitingrate

extrinsicincubationperiod

dailymortality

infectionprobability

recoveryrate

transmissionprobability

What processes determine maintenance and expansion of dengue?

• Human-virusinteractions– Humansusceptibilitytoinfection&diseasedependsonsequenceofinfection,age,viralstrain,etc.

• Mosquito-virusinteractions– Viralfitnesswithinmosquitodependsonviralstrain,temperature,etc.

• Non-viralfactors(shapingmosquito-humancontact)– Temporal&spatialvariation(atdifferentscales)inclimate,habitat,behavior,movement

Potential mechanisms underlying dengue dynamics

Environmental:Vectorbiologyandvirustransmissioninfluencedbyseasonality(andperhapsinter-annualvariation)inclimaticvariables

Iquitos study

Cases captured by clinic-based surveillance system in Iquitos, Peru between 2000–2010

Seasonaltimingofepidemics–interplayofclimate,humanimmunity,vectorcontrol&stochasticity

Dengue Control

• Vector-targetedcontrol– Reducemosquitopopulations– Reduceinfectiousmosquitopopulations

• Shortenfemalelifespan• Introducemosquitopopulationswithreducedcapacitytotransmitdengue

• Human-targetedcontrol– Vaccines(severalinvariousstagesoftesting)

Dengue vaccine

• OnevaccineDengvaxia(CYD-TDV)developedbySanofiPasteur,licensed• liveattenuatedtetravalentchimericvaccinewithyellowfeverbackbone

• Approximatelyfiveothervaccinecandidatesinclinicaldevelopment

• two(developedbyButantanandTakeda)inPhaseIIItrials

Seriousconcernsaboutwhethervaccinesmaycausemoreharmthangood

WHY?

Dengue replicationhttps://youtu.be/3LhWuaTRCME

Serotype Specific Data: Mexico

What mechanisms determine these patterns of interaction?


p Antibody response to infection – antibody-dependent enhancement (ADE): wane to

sub-neutralising levels, second episode of infection with heterotypic serotype may lead to enhanced viral replication (Halstead, 1970)

Ferguson et al. (1999; PNAS)

Studied two-strain model: direct transmission, no seasonality

Concluded that antibody-dependent enhancement can 1. generate persistent (sometimes

chaotic) serotype dynamics, qualitatively consistent with data

2. Facilitate coexistence of serotypes

Thai dengue data

l Serotype-specific clinical data

p Aggregated DF & DHF case reports

Nisalak et al. (2003; Am J Trop Med Hyg)

Sabin’s “experiments”

RESEARCH ON DENGUE DURING WORLD WAR 111

ALBERT B. SABIN'

Army Epidemiological Board, Preventive Medicine Division, Office of the Surgeon General

STATUS OF PROBLEM PRIOR TO WORLD WAR II

Most of the basic and significant contributions to our knowledge of dengueprior to World War II were made by honored members of the medical department of the U. S. Army. Ashburn and Craig (1) provided the evidence for theviral etiology of the disease. Siler, Hall and Hitchens (2) clearly established theperiod of infectivity of dengue patients for Aedes aegypti mosquitoes, the periodrequired for the development of the virus in these mosquitoes before they couldtransmit the infection, as well as the very long period during which these mosquitoes were capable of transmitting dengue. Simmons, St. John and Reynolds (3) established (a) the role of Aedes albopictus in the transmission of dengue,(b) the occurrence of inapparent infection in certain monkeys under experimental and possibly also natural conditions, thus suggesting the existence of aâ€œjungleâ€•type of dengue fever exclusive of the human cycle, (c) the persistenceof immunity to the homologous strain of virus for 13 months in human volunteers residing in an endemic region, and (d) many of the properties of the virus.it is necessary to recall, however, that the latter investigators completed theirstudies in 1930, before most of the important, newer virological techniques andprocedures had been developed. In 1934: Snijders, Postmus and SchÃ¼ffner(4)reported some immunity experiments on human beings in Holland with twodifferent strains of virus which left the subject of immunity to dengue in a ratherunsettled state. In 1936, Shortt, Rao and Swaminath (5) reported the successfulcultivation of dengue virus on the chorioallantoic membrane of chick embryos,but their conclusions were not based on tests on human beings. Otherwise, littleor no work was done on dengue during the period of 1930 to 1940.Thus, while a good deal of fundamental information about dengue was avail

able at the beginning of World War II, it was also apparent that most of theelementary requirements which would permit one to carry out systematic studieswith the virus of dengue fever were lacking. No strains of the virus were anywhere available; there was neither a suitable laboratory animal for experimentalwork nor an established method of in vitro cultivation, and almost nothing wasknown regarding some of the basic physical and biological properties of the virus.

1 This article was prepared as part of the medical history program of the Army, and,with certain minor modifications, will appear in the forthcoming HISTORY OF PREVENTIVE MEDICINE, U. S. ARMY MEDICAL DEPARTMENT, WORLD WAR II.It is an honor and privilege to have this article published in a volume dedicated to ColonelCharlesFranklinCraig,who, togetherwithAshburn,demonstratedtheviraletiologyofdengueand supplieda greatdealofadditionalimportantfundamentalinformationregardingthisdisease.

2 Lieutenant Colonel, Medical Corps, Army of the United States. Present address: TheChildren'sHospitalResearchFoundation,UniversityofCincinnatiCollegeofMedicine,Cincinnati29,Ohio.

30


p Antibody response to infection – antibody-dependent

enhancement (ADE): wane to sub-neutralising levels, second episode of infection with heterotypic serotype may lead to enhanced viral replication (Halstead, 1970)

– transient cross-immunity: cross-reactive antibody levels elevated for 2-9 months following infection (Sabin, 1952)

time since exposure

Antibody levels within an individual

cross-immunity

ADE

Typical model framework

susceptible

exposed

infectious

cross-immune

susceptible

exposed

infectious

recovered

susceptible

exposed

infectious

susceptible

exposed

infectious

Serotype1 Serotype2mortality

ADE: increased susceptibility

temporary cross-immunity (weeks)

stre

ngth

of A

DE,

incr

ease

d su

scep

tibili

ty

Wearing & Rohani (2006; PNAS)

Dominant inter-epidemic period of each serotype

Dominant inter-epidemic period of aggregate cases

Correlation between serotypes

Critical community size

Data from http://www.jhsph.edu/cir/dengue.html

Thai DHF data

means of 50 realisations with standard errorbars

Wearing&RohaniPNAS2006

Interrogating data

2.2. Adapted time-series susceptible-infectedrecovered model

We have adapted the time-series susceptible-infected recovered(TSIR) model, developed by Finkenstadt & Grenfell [33]. Weassume a transmission model of

It;i ¼ rt " Ia1t#1;i " S

a2t#1;i " et;i; ð2:1Þ

where It,i is the count of infecteds at time t for serotype i, St,i is thecount of susceptibles and rt is a transmission parameter that istime-varying with period of 1 year. The a are mixing parameters,which, if both equal 1, define a population with homogeneousmixing, whereas values not equal to 1 have been used to describedepartures from mass-action mixing [34] and to account for dis-cretization of continuous-time transmission processes [35]. It isassumed that the error term e t,i has the following properties:E½log et;i' ¼ 0 and Varðlog et;iÞ ¼ s2

e . We used a time-step of twoweeks (a ‘bi-week’) which roughly coincides with the generationtime of dengue fever. Additionally, we accounted for multi-strainsusceptible dynamics according to the following equation:

St;i ¼ Bt#d þ St#1;i # It;i # dQt;L;#i; ð2:2Þ

where Bt – d is the number of births entering the susceptible cohortat time t, after receiving an assumed d ¼ 8 bi-weeks of maternalantibody protection. Also, Qt,L, – i is a term which accounts for thetransitions between susceptible and convalescent states for aparticular strain i. This term depends on the parameters, thespecified model of cross-protection (k or l, for fixed durationand exponential models, respectively) as well as the maximumpossible length of cross-protection (k or L). More details aboutthe susceptible accounting can be found in the electronic sup-plementary material, §1. The parameter d is the fraction ofinfected individuals that gain transient immunity to all otherserotypes—i.e. are removed from the susceptible population forother serotypes—for some period of time.

The model represents the possible trajectories of individualsthrough particular states regarding their infection with each sero-type (presented in electronic supplementary material, figure S6).Individuals begin life susceptible to all serotypes and becomeinfected with each serotype over the course of their lifetime.

Mosquitos are represented implicitly in our model. The potentiallytime-varying transmission parameter, rt, describes the rateat which the multiple processes involved in transmission (i.e. mos-quito feeding, viral growth in mosquitos, mosquito survival) givesrise to new infections in humans. Upon infection, individualsbecome temporarily immune to infection by the other serotypes.That is, those infected with strain j are removed for a period oftime from the susceptible class for strain i, where j = i, to a conva-lescent, cross-protected class CP. This state could representprotection from infection or protection from symptomatic, hospi-tal-attended clinical disease. Importantly, we assume that peoplecan be infected after leaving this temporarily cross-protectedstate. Thus, if only cross-protected from clinical disease, weassume that a subclinical illness does not elicit a protectiveimmune response to the infecting serotype. We are interested inestimating parameters that govern the length and duration ofthis state of cross-protection.

2.3. Parametric forms of cross-protectionWe assumed two parametric forms for the period of cross-protec-tion. We refer to the first parametrization as a ‘fixed duration’model. In these models, we assume that some fraction d of allinfected individuals experience a period of cross-protection with afixed length, k. Fixed duration models experienced some problemswith convergence for some of the large values of k included inthe analysis. For this reason, the range of k was truncated to be100 biweeks, which included the 90% and 95% confidence regionsfor k and d. We refer to the second parametrization as ‘exponential’models. In these models, we assume that infected individuals leavethe cross-protected class according to a random exponential survi-val distribution with mean l. The exponential model could alsobe described as having protection in individuals wane overtime. We truncate this distribution, assuming that the maximumamount of time a person can be cross-protected is equal to L,which was fixed at 10 years and five months. This length wasdecided upon because it was the 75th percentile of the exponentialdistribution with a mean of 7.5 years—the highest l considered. It isnecessary to truncate the distribution because this determines howmuch data are needed to ‘seed’ the initial time-step of the model.

1975

case

cou

nts

1980 1985 1990 1995 2000 2005 2010

80604020

0

sero

type

1se

roty

pe 2

sero

type

3se

roty

pe 4

80604020

0

80604020

0

80604020

0

Figure 1. The time series of monthly serotype-specific case counts of dengue from Queen Sirikit National Institute of Child Health in Bangkok, Thailand. This facilityis a paediatric healthcare facility which serves as a reference hospital for dengue in Bangkok. The counts shown here represent the total number of cases, bothprimary and secondary infections.

rsif.royalsocietypublishing.orgJR

SocInterface10:20130414

3

Reichetal.(2013;JRSocInterface)

DengueserotypesinThailand

St,i = Bt�d + St�1,i � It,i � �Qt,j 6=i

It,i = rt · It�1,i · St�1,i · ✏t,i

Transmissionmodel

Monthlyserotype-specificcasecountsofdenguefromQueenSirikitNaqonalInsqtuteofChildHealthinBangkok,Thailand

Transmissionrateinmontht

Birthsaccountingfor4monthsofmaternalimmunity

LossfromSusceptiblestoserotypeiwheninfectionwithserotypej

Interrogating data• Usedstatisticalmethods—basedonmaximumlikelihood—todistinguishamongdifferentmodels

reported) to test the robustness of our estimates to onlyobserving a fraction of all cases. In 869 of 1000 datasetswith no cross-protection where 10 per cent of all cases wereassumed to be reported, the TSIR model correctly identifiedthat no cross-protection was present. When 1 per cent ofcases were assumed reported, the TSIR model correctly ident-ified no cross-protection in 667 of 1000 simulated datasets.Detailed methodology and results from the simulation arepresented in table S3, electronic supplementary material,figures S3 and S4 and §2.

3.5. Consideration of immune enhancementModels Fa, Fb, Fc and Fd restricted the parameter d to liebetween 0 and 1. In this case, d can be interpreted as the frac-tion of infected individuals who experience some level ofcross-protection. The models we present in this manuscriptfocus on these cross-protection interpretations of d. How-ever, allowing d to assume negative values admits anotherinterpretation, namely that d describes the relative contri-bution of those individuals infected with one serotype to apopulation’s susceptibility to a different serotype. In thisscenario, negative values of d are consistent with immuneenhancement of severity of or susceptibility to infectionsamong those previously infected. In a simple sensitivityanalysis that extended our primary models, we exploredthe range of d values from 0 to 21 in model Fc. These resultsare presented in electronic supplementary material, figure S5.

For periods of 0–4 years, we did not find any support forincreased contribution of recently infected individuals tothe pool of susceptible individuals.

4. DiscussionIt has been proposed that the phylogenetic structure ofdengue viruses reflects competing processes. On the onehand, cross-protective immunity may increase the geneticdistance between viruses within different serotypes by select-ing for genetic variants that escape cross-protective immunity[22]. On the other hand, antibody-dependent enhancementmay only occur in viruses that are antigenically similar [41].Here, we have provided strong evidence that at least one ofthese processes (cross-protection) plays an important role inthe transmission dynamics of each dengue serotype.

We find strong evidence for short-term cross-protection ofa duration of approximately 2 years. Our results were robust todifferent assumptions of the distribution of cross-protection,seasonality of transmission and heterogeneity in transmissionbetween serotypes. These estimates provide the first quantitat-ive estimate of the duration of short-term cross-protectionbetween dengue serotypes since Albert Sabin’s experimentaldata collected in the 1940s.

The data that we have analysed do not allow us todistinguish between cross-protection against infection andcross-protection against clinically apparent disease. More-over, the medically attended dengue incidence consideredhere may be strongly correlated with the predominant sero-types circulating in the population or it may have a morecomplicated relationship with serotype-specific incidence ofinfection. Heterogeneity in severity by serotype or infectionsequence (e.g. secondary infections of type j following type i)may threaten our ability to estimate durations of cross-protection and other transmission parameters, because itreduces the correlation between observed clinical case dataand the temporal incidence of infection. Longitudinal obser-vations of the incidence of infection are needed in order todetermine the causal mechanisms underlying our obser-vations. Though longitudinal cohort data would be ideal forresolving specific mechanisms of interaction, the expense ofcollecting this data means that long durations (on the orderof 40 years) have not and will not be performed. Our datahave allowed us to model and estimate non-specific serotypeinteractions over four decades.

Given that secondary infections with dengue are morelikely to result in severe illness than primary infections, anyhospital-based case data (such as the data used in this analy-sis) will have an over-representation of secondary infections.Temporal differences in primary and secondary infectiondynamics could impact our ability to accurately estimatethe duration of cross-protection. This is a limitation of thework presented here. However, research that has shownthat primary dynamics are tightly coupled to secondarydynamics across a wide range of theoretical dengue models[19,20] suggests that bias introduced by unrepresentativeprimary and secondary case sampling may be minimal.

We assume that individuals can be infected with onlyone serotype at a time. Co-infections of multiple dengueserotypes have been observed to occur [42,43], though onestudy has suggested that competitive interactions betweenthe serotypes within mosquitoes may limit the transmission

–810(a)

(b)

log-

likel

ihoo

d

log-lik

–812

–814

–816

0

0.2

d0.4

0.6

0.8

1.0

0 1average length of cross-protection (years)

k, the duration of cross-protection (years)

2

0 1 2 3 4

3 4

–807–809.6–810.3–812–814–816–818–820n.a.

5 6 7

Figure 3. Profile likelihood surfaces for the best exponential models (a) andfixed duration models (b). In panel (a), the maximum-likelihood and 95% CIfor l are shown by the dashed vertical lines. In (b), the axes index the twoparameters of a fixed duration distribution of cross-protection. The fraction ofthe population that experiences cross-protection is d and k is the duration ofcross-protection. The two lightest regions represent the 90% and 95% like-lihood confidence regions. The confidence region for the average duration(which is calculated as k . d) represents the range of average durationscontained in the respective confidence region. (Online version in colour.)



6

Reichetal.(2013;JRSocInterface)

the duration of cross-protection. Figure 4 shows the patternof time-varying or seasonal transmission coefficients fitfor each serotype and aggregated across all serotypes forthe exponential cross-protection models. Seasonal trans-mission coefficients varied from 84% of the mean valueat its peak to 116% of its mean value at its trough. Thereis general correspondence between peaks in temperature,rainfall and the high values of the transmission coefficient,but several peaks occur during dry and cool periods aswell (figure 4).

3.3. Model generates 3 – 5 year multi-annual patternsTo investigate whether our estimated models could generatemulti-annual patterns displayed in the observed incidencedata, we simulated serotype-specific incidence over thetime-period of the observed data. Using Fourier transforms,we estimated the frequency of multi-annual oscillations inthe simulated data and compared this with empirical data.Details on these simulations are included in the electronicsupplementary material, §1. Simulated datasets from systemswith a fixed duration of cross-protection showed strong evi-dence of annual cycles as well as 3–5 year serotype-specificmulti-annual cycles. These frequencies roughly align withperiodicities observed in the actual data, although they failto capture some of the observed longer term dynamics.In contrast, not including cross-protection resulted in no

consistent multi-annual oscillations. Electronic supplemen-tary material, figure S2 shows the serotype-specific spectrafor the observed data as well as for the simulated datasets.

3.4. Multi-strain model framework is sensitive tocross-protection

To assess our estimation procedure, we fit our models to datagenerated from simulations in which the true parameterswere known. Using the multi-strain TSIR framework, we ana-lysed 12 000 simulated datasets from a continuous-timemulti-strain state-space model that incorporates exponentiallydistributed cross-protection and susceptibility enhancement(i.e. individuals with immunity to one serotype are morelikely to acquire a second) [39]. For each simulated dataset,the 13 candidate models were fitted and the best-fittingmodel was chosen as the one with the lowest AIC. TheTSIR model was both sensitive and specific in identifyingthe presence or the absence of cross-protection in simulateddatasets, based on the results from the best-fitting models.The TSIR model had 77 per cent specificity—correctly identi-fying that no cross-protection was present in 77 per centof the datasets that had none. The TSIR model also had100 per cent sensitivity—identifying a significant level ofcross-protection in all datasets that came from a systemwith some level of cross-protection. We analysed simulateddata with varying reporting rates (10 and 1% of all cases

average duration (years)

1.93

2.00

2.13

2.27

1.88

2.23

expo

nent

ial (

l)fi

xed

dura

tion

(k d

)

% protected (d ) r*i r(t)*

100

100

100

84 (39–100)

76 (32–100)

80 (33–100)

r(t) = seasonal transmission parameters includedri = serotype-specific transmission parameters included

log-lik = log-likelihood for the given modeld.f. = degrees of freedom of the modelDAIC = change in Akaike Information Criterion over null model

log-lik d.f.

–873.9

–810.0

–773.5

–870.5

–771.0

–807.3

20

42

120

21

121

43

0.7716–943.3001

0.48 100 (30–100)

DAICaverage duration of cross protection

Ea

Eb

Ec

Ed

Fa

Fb

Fc

Fd

0 1 2 3 4 5 6 7

Na

Nb

Nc

Nd

N

no c

ross

-pro

tect

ion

0

0

0

0

0

–882.9

–817.2

–781.1

19

41

119

–946.7

–902.4 3

15

–23.1

–106.9

–23.9

–28.0

–26.8

–110.2

107.7

106.7

0

–7.0

–94.4

–10.7

112.6

–941.8 17

Figure 2. Estimated parameters and model characteristics for exponential and fixed duration models of cross-protection. Model results shown in bold indicate thatthe model showed a statistically significant improvement over a null model that included no form of cross-protection and no serotype-specific or seasonal trans-mission parameters. Based on AIC and likelihood ratio tests, models Ec and Fc showed the most improvement over the null model and also showed significantlybetter fit to the data than a model that did not include cross-protection but did include seasonal transmission. The point estimates of the average duration ofcross-protection are shown by the vertical tick mark and the 95% CIs are shown with horizontal lines.



5

Overall Modeling Conclusions

Strongestresult:modelsneedtoincludetemporarycross-immunitytogeneratebestmatchwithThaidata(seealsoAdamsetal.2006PNAS)

SuggestsweaktransmissionconsequencesofADE

However,doesnotexcludeasignificantroleforADEindenguepathogenesis

Consequences for vaccination

3

1. Materials and Methods 1.1 Model of vaccine action Figure S1 illustrates the default model of vaccine action described in detail in section 2 below. As discussed in the main text, the model assumes that the immunological effect of vaccination is akin to a (silent) natural infection.

Figure S1: Mechanism of action of the vaccine assumed in the default model.

Unvaccinated individuals (top row of Figure S1) experience a moderate severity primary infection, a more severe secondary infection, then mild tertiary and quaternary infections. Seronegative individuals who are vaccinated while still fully susceptible to dengue (middle row) are transiently protected against the four dengue serotypes as is generally observed after the first natural infection(22, 37). As antibody titers decay, they stop being protective and become enhancing, thus increasing the probability of symptomatic and severe disease upon a primary breakthrough infection(4, 5). Thus, the model assumes that in individuals who are vaccinated while still seronegative, a first breakthrough infection will cause symptomatic or hospitalized disease with the same (high) probability as a secondary natural infection in unvaccinated individuals. Conversely, vaccination of individuals who have experienced one or more dengue infections (bottom row of Figure S1) boosts their immunity to levels comparable to those of individuals who have experienced two or more infections. Thus, a secondary infection in an individual vaccinated after their primary infection will cause symptomatic or severe disease with the same (low) probability as a tertiary infection in unvaccinated individuals. Table S1 shows the key temporal trends in vaccine efficacy seen in the phase 3 trials, shown as the relative risk of hospitalized dengue comparing the vaccine group with the control group. The relative risks vary by age (but not significantly by trial within the same age group) and over time, in both studies and in both the under 9 and over 9 age groups. However, all relative risks increased approximately 3 fold between the active phase and the first year of long term follow-up. 3



Unvaccinated individuals (top row of Figure S1) experience a moderate severity primary infection, a more severe secondary infection, then mild tertiary and quaternary infections. Seronegative individuals who are vaccinated while still fully susceptible to dengue (middle row) are transiently protected against the four dengue serotypes as is generally observed after the first natural infection(22, 37). As antibody titers decay, they stop being protective and become enhancing, thus increasing the probability of symptomatic and severe disease upon a primary breakthrough infection(4, 5). Thus, the model assumes that in individuals who are vaccinated while still seronegative, a first breakthrough infection will cause symptomatic or hospitalized disease with the same (high) probability as a secondary natural infection in unvaccinated individuals. Conversely, vaccination of individuals who have experienced one or more dengue infections (bottom row of Figure S1) boosts their immunity to levels comparable to those of individuals who have experienced two or more infections. Thus, a secondary infection in an individual vaccinated after their primary infection will cause symptomatic or severe disease with the same (low) probability as a tertiary infection in unvaccinated individuals. Table S1 shows the key temporal trends in vaccine efficacy seen in the phase 3 trials, shown as the relative risk of hospitalized dengue comparing the vaccine group with the control group. The relative risks vary by age (but not significantly by trial within the same age group) and over time, in both studies and in both the under 9 and over 9 age groups. However, all relative risks increased approximately 3 fold between the active phase and the first year of long term follow-up.

3




3





• Assessingdenguevaccinepotentialusingmodeling

DENGUE VACCINE

Benefits and risks of theSanofi-Pasteur dengue vaccine:Modeling optimal deploymentNeil M. Ferguson,1*† Isabel Rodríguez-Barraquer,2* Ilaria Dorigatti,1

Luis Mier-y-Teran-Romero,2 Daniel J. Laydon, Derek A. T. Cummings2,3

The first approved dengue vaccine has now been licensed in six countries. We propose thatthis live attenuated vaccine acts like a silent natural infection in priming or boosting hostimmunity. A transmission dynamic model incorporating this hypothesis fits recent clinical trialdata well and predicts that vaccine effectiveness depends strongly on the age groupvaccinated and local transmission intensity. Vaccination in low-transmission settings mayincrease the incidence of more severe “secondary-like” infection and, thus, the numbershospitalized for dengue. In moderate transmission settings, we predict positive impactsoverall but increased risks of hospitalization with dengue disease for individuals who arevaccinated when seronegative. However, in high-transmission settings, vaccination benefitsboth the whole population and seronegative recipients. Our analysis can helpinform policy-makers evaluating this and other candidate dengue vaccines.

The first dengue vaccine, the product of a 20-year development process by Sanofi PasteurLtd., has now been approved for use in sixcountries. Its development was considera-bly more challenging than for other flavi-

virus infections because of the immunologicalinteractionsbetween the fourdengue virus (DENV)serotypes and the risk of immune-mediated en-hancement of disease (1–3). Individuals experi-encing their second natural DENV infection havea higher risk, by more than sixfold, of severe dis-ease compared with those experiencing primaryinfection (4, 5), which is hypothesized to be dueto heterotypic antibody-dependent enhancement(4). If future trials are to avoid similar conse-quences, the ideal DENV vaccine should gener-ate a balanced protective response against eachof the four serotypes (1).The Sanofi-Pasteur vaccine, Dengvaxia, a re-

combinant chimeric live attenuated DENV vaccinebased on a yellow fever 17D vaccine backbone,was evaluated in two large multicenter phase 3trials. One trial was conducted in Southeast Asia(6), among ~10,000 children aged 2 to 14 years,and the other in Latin America (7), among ~21,000children aged 9 to 16 years. Both trials reportedefficacy of ~60% against virologically confirmedsymptomatic dengue disease (the primary out-come), as well as higher efficacy against severedengue and variation in efficacy by serotype(6–8). The trials also revealed high efficacy in re-cipients who were seropositive to DENV at the

time of vaccination, but much lower (and statis-tically insignificant) efficacy in those who wereseronegative at the time of vaccination. Both trialsalso found lower vaccine efficacies in younger agegroups—a pattern consistent with reduced effi-cacy in individualswho have not lived long enoughto experience a natural infection.Reduced efficacy in seronegative recipients ini-

tially indicates that it would be beneficial, but notessential, to optimize the target age group whendeveloping vaccination programs. However, inJuly 2015, long-term follow-up results for the thirdyear of the trial showed that vaccinees in theyoungest age group (2- to 5-year-olds) of the Asiantrial had a substantially and significantly higherrisk of hospitalization for virologically confirmeddengue disease than controls had (9). In other agegroups (in both trials), the vaccine was still pro-tective against hospitalization, albeit efficacy waslower than that seen in the active phase of thetrial [see supplementary materials (10)]. Im-munogenicity data (11–18) have shown that sero-positive vaccine recipients attainhighandsustainedantibody levels after the first dose of vaccine,whereas peak antibody levels in seronegative re-cipients are on average a factor of 10 lower andshow rapid decay, apparent even between vaccinedoses (18). Serological data were only collectedfrom a subset of participants in each phase 3trial, so it is not possible to determine whetherthe risk excess seen in the 2- to 5-year-old agegroup is driven by the effect of vaccine in the largeproportion of seronegative recipients in this agegroup, but at present, this appears to be the mostplausible explanation (19).These trial results pose challenges in consid-

ering how best to use the vaccine. The hetero-geneities in the efficacy profile—combined withthe uncertainties regarding the vaccine’s mech-anism of action (20) and the underlying com-plexity of DENV epidemiology and transmission

dynamics—make it far from simple to extrapolatefrom the trial results to predict the potential im-pact of wide-scale use of this vaccine.We therefore developed mathematical models

of DENV transmission (10) to explore hypothesesabout vaccine action and to examine the poten-tial consequences for the impact of routine use ofthis vaccine. Given the trial results (see table S1),any model needs to incorporate waning of effi-cacy over time. Hence, we fitted a “simple”modelto the publicly available trial data (6–8), whereefficacy was allowed to decay from an initial highvalue to some lower long-term value, with theseefficacy values assumed to be different for sero-positive and seronegative vaccine recipients. Theresulting parameter estimates and poor overallfit (table S5 and fig. S5) led us to propose a morebiologically motivated model, in which the im-munological effect of vaccination is comparableto a silent natural infection (fig. S1). Seronegativerecipients gain transient protective cross-reactiveimmunity akin to that observed for natural infec-tion (21–23). After this protection decays, lowerconcentrations of heterotypic antibodies increasethe risk of severe disease upon a breakthroughprimary infection to the same level seen for sec-ondary infections in nonvaccinees (4, 5). Con-versely, vaccination of recipients who have alreadyhad one DENV infection results in a boosting ofimmunity to levels comparable with someonewho has had two natural infections, and theirnext infection will not have the higher severityassociated with natural secondary infections, butrather, the much lower risk of severe disease as-sociated with tertiary and quaternary (post-secondary) infections (24).This model fitted well the patterns seen in both

the active and long-term follow-up phases of thephase 3 clinical trial, including the variation invaccine efficacy by age, serostatus at the time ofvaccination, and time since vaccination (Fig. 1).The poorest aspect ofmodel fit is to the sevenfoldgreater incidence of hospitalization with dengueseen in 2- to 5-year-old vaccine recipients com-pared with controls in the first year of the long-term follow-up in the Asian trial. However, modelpredictions lie within the confidence bounds ofthe data, and the model successfully reproducesa relative risk >1 for vaccine recipients comparedwith controls in that age group. Indeed, had thelong-term follow-up data on the effects of vacci-nation in the 2- to 5-year-old age group not beenincluded, our model would still have predicted arelative risk >1 in that age group, based on trendsseen in the other age groups and the results ofthe active phase (table S4).Consistent with prior knowledge (5), our pa-

rameter estimates indicated that secondary infec-tions are about twice as likely to cause symptomaticinfection as either primary or postsecondaryinfections (table S3). In addition, we estimatedthat the vaccine initially induces near-perfectheterologous protection in seronegative recip-ients but that this decays rapidly, with a meanduration of 7 months [95% credible interval (CI)of 4 to 11 months]. Our analysis did not resolvethe extent to which such transient heterologous

SCIENCE sciencemag.org 2 SEPTEMBER 2016 • VOL 353 ISSUE 6303 1033

1MRC Centre for Outbreak Analysis and Modelling, School ofPublic Health, Imperial College London, Norfolk Place,London W2 1PG, UK. 2Department of Epidemiology, JohnsHopkins Bloomberg School of Public Health, 615 North WolfeStreet, Baltimore, MD 21205, USA. 3Department of Biologyand Emerging Pathogens Institute, University of Florida, PostOffice Box 100009, Gainesville, FL 32610, USA.*These authors contributed equally to this work. †Correspondingauthor. Email: [email protected]

RESEARCH | REPORTS

on

Sept

embe

r 7, 2

016

http

://sc

ienc

e.sc

ienc

emag

.org

/D

ownl

oade

d fr

om

protection is induced in seropositive recipi-ents; themodal posterior estimate of the efficacyof such protection is 0 but the 95% CI spans 0to 100%.To predict the implications of our model of

vaccine responses on the effectiveness of immu-nizationpolicies,we simulated the effect of routinevaccination at 80% coverage, and explored theeffect of varying the age at vaccination between 2and 18 years of age. We deliberately examinedages below the 9-year minimum age approvedby regulators to give greater insight into the in-teraction between age, transmission intensity,seroprevalence, and the impact of vaccinationon dengue disease. Owing to the dependence ofefficacy on serostatus at the time of vaccination,the impact of the vaccine critically depends onthe proportion of the target age group who haveexperienced 0, 1, or more natural DENV infec-tions before vaccination. Therefore, we quantifytransmission intensity as the long-term averageof the proportion of 9-year-olds who are sero-positive. Thismetricmapsmonotonically onto themore commonly used metric of the basic repro-duction number, R0 (fig. S3), but has the advan-tages of being directly related to the key driver ofvaccine efficacy (i.e., serostatus), which is readilymeasurable and interpretable and not depen-dent on specific model assumptions (25).The predicted mean population impact of rou-

tine vaccination on symptomatic dengue diseaseand case incidence of hospitalization with den-gue over 10- and 30-year periods is shown in Fig. 2.In high-transmission settings, vaccination is asso-ciated with modest (20 to 30%) reductions inboth symptomatic disease and hospitalization.For a specific level of transmission, there is anoptimal age of vaccination that decreases as trans-mission intensity increases. Although short-term(10-year) impacts are generally positive, overlonger periods of time (30 years), vaccinationmay have positive or negative impacts on theincidence of symptomatic dengue disease andhospitalized dengue. This is particularly true inlow-transmission settings. Vaccination is morelikely to have a negative outcome for hospitalizeddengue than symptomatic dengue as secondaryor secondary-like infections (i.e., primary infec-tions in vaccine recipients) have an approximatelyeightfold higher risk of hospitalization than pri-mary infections but only a twofold higher riskof uncomplicated symptomatic disease (10, 26).The population-level impacts of vaccination hide

enormous heterogeneity in benefits and risks atthe level of the individual recipient (Fig. 3, A andB): Seropositive recipients always gain a sub-stantial benefit from vaccination (>90% reduc-tion in the risk of being hospitalizedwith dengue),whereas seronegative recipients experience anincreased risk of being hospitalized with dengue.This is true both in the short-term (see supplemen-tarymaterials) and in the long-term and raises fun-damental issues about individual versus populationbenefits of vaccination. The increase in risk isgreatest for low-transmission settings, where asubstantial fractionof seronegative recipientswouldnot normally experience a natural secondary in-

fection. Conversely, in the highest-transmissionsettings, the main effect of vaccination on sero-negative individuals is to bring forward in timethemore severe secondary-like infection that theywould have eventually naturally experienced. This,combined with a small indirect effect of vaccina-tion on reducing transmission, leads to a smalloverall positive benefit to all recipients in high-transmission settings. Restricting the minimumlicensed age of use of the vaccine to 9 years mit-igates, but does not remove, the risk of negativepopulation-level impacts in low-transmission set-tings where the majority of 9-year-olds are stillseronegative. Conversely, in high-transmission set-tings, the optimal age to target for vaccinationcan be below 9 years.The vaccination policies that risk producing

adverse outcomes can therefore be defined. Theminimum average prevaccination seroprevalencerequired to avoid negative impacts is shown inFigure 3C from both the individual and popula-tion perspectives. An overall negative impact onthe entire population can be avoided by choos-ing a target age for vaccination in which averageseroprevalence exceeds ~35%. In contrast, it isharder to avoid an increased risk of hospitaliza-tion for individuals who are seronegative whenvaccinated. Doing so requires that the indirecteffects of vaccination in reducing overall denguetransmission exceed the increased risk of diseasewhich vaccination causes in seronegative indi-viduals via immune priming. Over a period of 30years, this is only possible in high–transmissionintensity settings when R0 > 3 or seropreva-lence in 9-year-olds exceeds ~70%. Only for theyoungest age of vaccination considered (2 years,below the licensed minimum age) do population

and individual thresholds converge. In part basedon the modeling presented here, the WorldHealth Organization’s Strategic Group of Expertson Immunization has recently recommendedpopulation serological surveys be undertakenin populations where the vaccine is being con-sidered for use and that vaccination is only recom-mended where seroprevalence in the targeted agegroup exceeds 50% (and preferably 70%) (27).Serological testing of individuals offers an alter-

native solution to mitigate the potential risks andto maximize the benefits of dengue vaccination;rapid diagnostic tests could be used to screen po-tential vaccine recipients, with only seropositiveindividuals being vaccinated. Indeed, data fromimmunogenicity studies suggest that a single-dosevaccination schedulemight be enough to achieveprotective immunity in seropositive individuals.Such a policy could result in up to a 30% reductionin the incidence of hospitalization for dengueand a much-reduced risk of negative outcomes(Fig. 4A) after vaccinating only a fraction (thosetesting seropositive) of the target age group (Fig.4B). Although such a policy would be logisticallychallenging in the context of mass vaccinationcampaigns, it should not be ruled out—if the costof testing can be reduced to a level comparablewith the cost of buying and delivering a singlevaccine dose, such a strategy is likely to havesubstantially greater cost-effectiveness than thecurrent three-dose strategy without testing. Usingserological testing to inform vaccination decisionsis not an entirely novel concept, as it has beenrecommended for pregnant women in relation torubella and hepatitis B vaccination (28, 29).Because vaccination only transiently reduces the

risk of infection and themain effect of vaccination

1034 2 SEPTEMBER 2016 • VOL 353 ISSUE 6303 sciencemag.org SCIENCE

Fig. 1. Model fit to publicly available data from the Asian phase 3 clinical trial. See (6). Modal (best-fit) estimate and 95% credible intervals for four conditions. (A) Proportion of participants of the immu-nological subset of the trial who were seronegative at the time of receiving their first dose, by age. (B) Attackrate of virologically confirmed symptomatic dengue in immunological subset in first 2 years after dose 1 by trialarm and baseline serostatus. (C) Attack rate of virologically confirmed symptomatic dengue in all trialparticipants in first 2 years after dose 1 by trial armandage group. (D) Attack rate of virologically confirmedhospitalization with dengue disease in all trial participants in third year after dose 1 (first year of long-termfollow-up) by trial arm and age group. Fit to Latin American trial shown in the supplementarymaterials (10).

RESEARCH | REPORTS

on

Sept

embe

r 7, 2

016

http

://sc

ienc

e.sc

ienc

emag

.org

/D

ownl

oade

d fr

om

Contrastmodelfitstodata

Best-fittingmodel:(i)secondinfectionstwiceaslikelytoleadtosymptomaticdengueasprimaryand(ii)~7moperiodofheterologousprotectioninseronegativevaccinerecipients

is to modify the risk of disease, our findingspredict that the indirect effect of vaccination onDENV transmissionwill be limited. This explainswhy we found that the predicted impacts of rou-tine vaccination (whether positive or negative)scale almost linearly with vaccine coverage. Ourdefault assumption was that symptomatic infec-tions are twice as infectious as asymptomaticinfections, which leads to vaccination slightly

reducing transmission in high–transmission in-tensity settingsbut slightly increasing transmissionin low-transmission settings. Making the alter-native assumption that all infections are equallyinfectious reduces the chance and magnitude ofnegative impacts of vaccine for low-transmissionintensities but also reduces the positive impactsof vaccination when transmission intensity is high(figs. S9 and S10).

Our results also show that the effectiveness ofvaccination would be expected to vary over time(figs. S6 to S8). In low-transmission settings, theintroduction of vaccination could perturb trans-mission dynamics and lead to transient reductionsin dengue disease incidence for 5 to 10 years. Onlywhen the transmission dynamics reequilibrate arethe long-term impacts seen. From the individualperspective, it is also important to consider theeffect of vaccination on the cumulative lifetimerisk of dengue disease andhospitalization. Amongseronegative recipients, reductions in risk result-ing from short-term vaccine-induced protectionmight exceed later increases in risk resulting fromvaccine-induced immunological priming. This isparticularly true in high-transmission settingswhere, in the absence of vaccination, nearly every-one experiences secondary infection with dengueat some point in their lives. Special considerationshould be given to the policy and ethical consid-erations of shifting infections and/or symptomaticepisodes among individuals to different times intheir lives.Our analysis has several limitations. We were

not able to estimate serotype-specific efficacy pa-rameters. Owing to cross-reactive immunity, inany one year, DENV incidence in single popula-tions tends to be dominated by a single serotype,which is reproduced by our transmission model.However, the phase 3 trials showed substantialattack rates from all four serotypes, but under-pinning this was much greater heterogeneity inserotype-specific attack rates between the coun-tries contributing to the trial. To capture observedserotype-specific attack rates it is necessary to fitcountry- and serotype-specific trial data, whichare not currently publicly available (30). How-ever, in the supplementarymaterials (10), we showhow the apparent serotype-specific efficacies seenmay reflect differences between serotypes in thepropensity to cause disease in primary, secondary,and postsecondary infection rather than actualdifferences in (serostatus-specific) efficacy (fig.S12). Including such asymmetry does not qual-itatively affect model predictions (figs. S13 andS14).We also do not consider persistent variationin exposure to DENV at the individual or neigh-borhood level; if substantial proportions of thepopulation consistently experience lower orhigher levels of exposure than the average through-out their lives, then both the risks (to the low-exposure group) andbenefits (to the high-exposuregroup) of vaccination may be larger than we es-timate here. Although characterizing real-worldlevels of exposure heterogeneity is difficult, thisissue should be a priority for future work.All efficacy outcomesmeasured in the trial were

based on clinically apparent disease, so we are cur-rently unable to resolve whether the vaccineprotects against infection or just against disease(20, 31). Our baseline model assumes a combina-tion of both—short-lived protection against in-fection, followed by a long-lived modificationof future disease risk. We are also unable to assessthe impact of breakthrough infections on vaccine-acquired immunity. If vaccination truly acts as asilent infection, then breakthrough infections in

SCIENCE sciencemag.org 2 SEPTEMBER 2016 • VOL 353 ISSUE 6303 1035

Fig. 2. Predicted population effects of vaccination on dengue disease for a range of transmissionintensities (x axes) and ages of vaccination ( y axes).Color scale indicates proportion of cases avertedin the whole population (A) over 10 years, for all symptomatic dengue; (B) over 10 years, for participantshospitalized with dengue; (C) over 30 years, for all symptomatic dengue; and (D) over 30 years, forhospitalization with dengue. Negative proportions of cases averted indicate vaccination increases risk.Solid contours indicate the optimal age of vaccination for each transmission intensity. Dashed contoursindicate the youngest age group that may be targeted to avoid negative effects at the population level.

Fig. 3. Predicted individual effects of vaccination over 30 years. Proportion of hospitalized casesaverted among individual vaccine recipients who are vaccinated: (A) when seronegative and (B) whenseropositive. Dashed contour indicates the youngest age group that may be targeted to avoid negativeeffects at the individual level. (C) Minimum proportion of the age group (1-year age bands) targeted forroutine vaccination that should be seropositive at the time of vaccine introduction to avert negativeimpacts (over a 30-year time frame) at the population (red) and individual (blue) level.

RESEARCH | REPORTS

on

Sept

embe

r 7, 2

016

http

://sc

ienc

e.sc

ienc

emag

.org

/D

ownl

oade

d fr

om


Proportioncasesaverted

Reflectstransmissionsetting

Allsymptomaticdengue Hospitalizations

10yrhorizon

30yrhorizon

1.Low-transmission settings: vaccination may increase incidence of more severe “secondary-like” infection plus numbers hospitalized

2.Moderate transmission settings: positive impacts overall, but increased risks of hospitalization with dengue disease for individuals who are vaccinated when seronegative.

3.High-transmission settings: vaccination benefits both whole population and seronegative recipients

Discussion

• Dengueincidenceincreasing– Underlyingcausescomplex,likelyrelatedtochangingclimate,land-useandmobility

• Epidemiologyofdenguecomplexbecauseofassociationsbetween– Serotypes(viaimmunememory)– Sequenceofinfection(primary/secondary/tertiary)– Seasonalityinvectorbiology

• Importantconsequencesforvaccination– may,perversely,leadtoincreaseddiseaseinsomesettings

Date post:	21-Apr-2020
Category:	Documents
Upload:	others
View:	2 times
Download:	0 times

Overview of a “neglected” infectious disease:...

Documents