Estimation of Vaccine Efficacy and Critical Vaccination Coverage in Partially Observed Outbreaks
Estimation of Vaccine Efficacy and Critical Vaccination Coverage in Partially Observed Outbreaks
Abstract
Classical approaches to estimate vaccine efficacy are based on the assumption that a person’s risk of infection does not
depend on the infection status of others. This assumption is untenable for infectious disease data where such dependencies
abound. We present a novel approach to estimating vaccine efficacy in a Bayesian framework using disease transmission
models. The methodology is applied to outbreaks of mumps in primary schools in the Netherlands. The total study
population consisted of 2,493 children in ten primary schools, of which 510 (20%) were known to have been infected, and
832 (33%) had unknown infection status. The apparent vaccination coverage ranged from 12% to 93%, and the apparent
infection attack rate varied from 1% to 76%. Our analyses show that vaccination reduces the probability of infection per
contact substantially but not perfectly (V E ^ S = 0.933; 95CrI: 0.908–0.954). Mumps virus appears to be moderately
transmissible in the school setting, with each case yielding an estimated 2.5 secondary cases in an unvaccinated population
^ 0 = 2.49; 95%CrI: 2.36–2.63), resulting in moderate estimates of the critical vaccination coverage (64.2%; 95%CrI: 61.7–
(R
66.7%). The indirect benefits of vaccination are highest in populations with vaccination coverage just below the critical
vaccination coverage. In these populations, it is estimated that almost two infections can be prevented per vaccination. We
discuss the implications for the optimal control of mumps in heterogeneously vaccinated populations.
Citation: van Boven M, Ruijs WLM, Wallinga J, O’Neill PD, Hahné S (2013) Estimation of Vaccine Efficacy and Critical Vaccination Coverage in Partially Observed
Outbreaks. PLoS Comput Biol 9(5): e1003061. doi:10.1371/journal.pcbi.1003061
Editor: Neil Ferguson, Imperial College London, United Kingdom
Received October 29, 2012; Accepted March 28, 2013; Published May 2, 2013
Copyright: ß 2013 van Boven et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Funding was provided by the Dutch Ministry of Health, Welfare and Sport, and the Netherlands Organisation for Health Research and Development
(grant 125050004). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: [email protected]
all schools 2493 510–1342 0.62* 0.68 (485/709) 0.03 (25/952) 0.31
school 1 432 205–369 0.12 0.86 (204/237) 0.03 (1/31) 0.76
school 2 338 135–289 0.13 0.82 (131/160) 0.17 (4/24) 0.73
school 3 259 68–159 0.42 0.72 (68/94) 0 (0/74) 0.40
school 4 184 40–70 0.54 0.53 (37/70) 0.04 (3/84) 0.26
school 5 130 13–33 0.75 0.46 (13/28) 0 (0/82) 0.12
school 6 263 28–171 0.76 0.70 (19/27) 0.10 (9/93) 0.23
school 7 194 6–43 0.78 0.19 (6/31) 0 (0/126) 0.04
school 8 227 3–27 0.79 0.05 (2/41) 0.01 (1/162) 0.01
school 9 258 6–119 0.93 0.18 (2/11) 0.03 (4/134) 0.04
school 10 208 6–62 0.93 0.30 (3/10) 0.02 (3/142) 0.04
The column ‘number infected’ shows the possible range of actual infections, ranging from the number known to be infected to the sum of this number and the number
of persons with unknown infection status. Vaccination coverages and attack rates are calculated using persons with known vaccination status (vaccination coverage),
and known vaccination and infection status (attack rates). See Tables S1, S2 for the complete data.
*: averaged over schools.
doi:10.1371/journal.pcbi.1003061.t001
Figure 1. Posterior distributions of the basic reproduction number and vaccine efficacy (A) and vaccination coverages and attack
rates (B) when assuming common parameters across schools (baseline scenario). The median of the basic reproduction number is 2.49
(95% CrI: 2.36–2.63) and the median of vaccine efficacy is 0.933 (95CrI: 0.908–0.954). The estimated critical vaccination coverage is 0.642 (95% CrI:
0.617–0.666).
doi:10.1371/journal.pcbi.1003061.g001
the theoretical relation between vaccination coverage and attack reproduction number range from well below 1 to more than 3,
rate in a large population, and simulations of a finite population. while vaccine efficacy estimates can range from less than 0.20
Overall, the correspondence between the observed and simulated (schools 8–10) to almost 1 (schools 7–10; Table 3, Figure 3).
data is excellent for schools with low vaccination coverage and Further, the analyses show that in schools with high attack rates
high attack rates, while there is a tendency for higher attack rates (schools 1–4) the parameter estimates are quite close to those of the
than expected in schools with high vaccination coverage and a baseline scenario, indicating that estimates of transmissibility and
small number of infections. vaccine efficacy in the baseline scenario are dominated by schools
To investigate the information contained in the data by school with large numbers of infections and low vaccination coverages.
we perform analyses in which each school is equipped with its own
transmissibility and vaccine efficacy. It appears that precise Estimation of vaccine efficacy by the cohort method
estimates of transmissibility and vaccine efficacy can be obtained In comparison with our estimates of vaccine efficacy as the
in schools with high attack rates (schools 1–4), but not in schools reduction in the probability of infection (Table 3), estimates of
with only a handful of infections (schools 7–10). In fact, in schools vaccine efficacy by the cohort method tend to be somewhat lower
with less than 10 confirmed infections credible intervals of the in schools with low vaccination coverage and high infection attack
Table 2. Estimates of vaccination coverage and attack rate per school when assuming common epidemiological parameters
across schools (baseline scenario).
school 1 382 (372–389) 327 (315–337) 50 (43–59) 3 (1–6) 0.12 (0.10–0.14) 0.77 (0.74–0.79)
school 2 293 (286–301) 243 (232–256) 45 (37–52) 6 (4–11) 0.13 (0.11–0.15) 0.74 (0.70–0.78)
school 3 150 (141–158) 110 (101–119) 109 (101–117) 3 (0–8) 0.42 (0.39–0.45) 0.44 (0.41–0.47)
school 4 84 (79–90) 45 (41–50) 100 (94–104) 4 (3–6) 0.53 (0.51–0.56) 0.27 (0.24–0.30)
school 5 33 (30–37) 15 (13–18) 97 (93–100) 0 (0–2) 0.75 (0.72–0.77) 0.12 (0.10–0.15)
school 6 67 (59–76) 50 (42–59) 196 (187–204) 20 (14–27) 0.75 (0.71–0.78) 0.27 (0.23–0.30)
school 7 42 (38–46) 10 (7–13) 152 (148–156) 0 (0–3) 0.78 (0.76–0.80) 0.05 (0.04–0.07)
school 8 48 (45–52) 3 (2–5) 179 (175–182) 1 (1–2) 0.79 (0.77–0.80) 0.02 (0.01–0.03)
school 9 18 (13–25) 6 (3–11) 240 (233–245) 11 (7–16) 0.93 (0.90–0.95) 0.07 (0.05–0.09)
school 10 13 (10–17) 4 (3–7) 195 (191–198) 5 (3–8) 0.94 (0.92–0.95) 0.05 (0.03–0.06)
^ ^ ^
basic reproduction number (R 0) vaccine efficacy (VE S) critical vaccination coverage (q C)
The table shows the estimates of the basic reproduction number, vaccine efficacy, and critical vaccination coverage. Estimates are represented by posterior medians
with 95% credible intervals.
doi:10.1371/journal.pcbi.1003061.t003
Figure 3. Estimated reproduction numbers (top, blue dots) and vaccine efficacies (bottom, gray dots) per school, with associated
95% credible intervals (cf. Table 3). Note that estimated vaccine efficacy is consistently high in schools with high exposure (schools 1–6), but
cannot be estimated with any precision in schools with low exposure (schools 8–10).
doi:10.1371/journal.pcbi.1003061.g003
words, it is conceivable that the inclusion process systematically methodology provides a natural weighting of schools, in which
favours inclusion of schools with uncharacteristically high attack schools with small number of infections have lower weight than
rates, thereby leading to selection bias. For schools with low schools with high number of infections. If specific details were
vaccination coverage (and high attack rates) this is arguably not a available on the inclusion process, one could envisage extension of
problem as variation in outbreak sizes is expected to be minor, the analyses in which the selection process is modelled explicitly.
given the sizes of the schools included (Figure 2). For schools with This, however, would introduce more model options, additional
high vaccination coverage, however, selection bias may well have parameters to be estimated, and would certainly lead to a more
played a role, and may explain the relatively high attack rates in complicated analysis.
some of these schools (school 6 and to a lesser extend schools 9–10) We have assumed throughout that infections outside the school
(Figure 2). Fortunately, one could argue that our statistical played a marginal role. Again, this assumption is probably less
problematic in schools with low vaccination coverage and high
infection attack rates than in schools with high vaccination
Table 4. Estimates of vaccine efficacy by the cohort method, coverage and lower attack rates, as variation in the expected
i.e. as 1 minus the relative risk of infection in vaccinated number of infections is expected to be small in schools with low
versus unvaccinated persons (1{RR). vaccination coverage. Moreover, there was no sustained commu-
nity transmission during the study period, suggesting that the
^ impact of infection outside the schools may have been small.
vaccine efficacy (12R R )
Nevertheless, it would be interesting to extend the current
school 1 0.93 (0.81–0.99) analyses, e.g., along the lines of [21,22] by inclusion of other
school 2 0.76 (0.56–0.92) major transmission settings.
school 3 0.98 (0.93–1.0)
Classical estimates of mumps transmissibility have been based
on the mean age at infection in the pre-vaccination era ([23] and
school 4 0.91 (0.80–0.98)
references therein), or on seroprevalence data from the pre-
school 5 0.97 (0.90–1.0) vaccination era [24,25]. These analyses yielded estimates of the
school 6 0.84 (0.73–0.93) basic reproduction number in fully unvaccinated populations that
school 7 0.96 (0.83–1.0) are substantially higher (,7–20) than our estimates (,2–3). It
school 8 0.75 (0.0–0.98) should be noted that these population-based estimates cannot
school 9 0.78 (0.25–0.96)
directly be translated to our school-based estimates. Still, should
those early estimates be indicative of the current transmissibility of
school 10 0.90 (0.68–0.98)
mumps at the population level, then not only are schools an
Estimates are represented by posterior medians with 95% credible intervals. See important transmission route but other settings also have the
text for details. potential to contribute significantly to overall transmission. Again,
doi:10.1371/journal.pcbi.1003061.t004 to assess the contribution of different settings to the overall
transmission dynamics, it would be desirable to extend the current Table 4 and Table S3). In schools with high vaccination
studies beyond the school setting, by including household coverage and small numbers of infections the reverse tends to be
information and, in the specific case of this study, information true, and estimates of vaccine efficacy generally are both higher
on the churches attended by the participants [9]. This, however, is and more precise when using the cohort method. For instance,
only possible if detailed information were available on these in school 10 there are 6 confirmed infections, and vaccine
settings, not only with respect to their composition but also with efficacy is poorly estimated in our analysis (95%CrI: 0.16–0.96)
respect to vaccination and infection status of a sizeable part of the but with fair precision by the cohort method (95%CrI: 0.68–
population. 0.98). This is arguably an artefact of the latter method’s
Vaccine efficacy and transmissibility together determine the assumption that all 139 uninfected vaccinated persons have
critical vaccination coverage needed to prevent epidemic out- been exposed to an infected person, thereby artificially
breaks. In our study, estimates of the critical vaccination coverage increasing the precision of the estimates of vaccine efficacy.
are 64% (95%CrI: 62%–67%) in the baseline scenario, and range Our results point to strategies to efficiently allocate catch-up
from 63% (95%CrI: 58%–68%; school 1) to 76% (95%CrI: 66%– vaccination efforts in heterogeneously vaccinated populations. No
83%; school 6) in schools with more than 10 confirmed infections additional vaccination is needed in schools with high vaccination
(schools 1–6). This indicates that the critical vaccination coverage coverage (.75%, say) as these are already protected against
does not need to be as high as suggested by early population-based epidemic outbreaks affecting a large fraction of students. Similarly,
estimates, which are in the range of 86%–95%. allocating vaccines to schools with low vaccination coverage
In none of the analyses presented here have we made a (,50%, say) is inefficient as it does not markedly reduce the
distinction between children who had been vaccinated once and probability of infection for those who are not vaccinated, i.e. the
those that had been vaccinated twice. This was done because indirect benefits of vaccination are small in these populations. Our
preliminary analyses and previous results [9,11] could not find any analyses suggest that vaccination of populations in the range
evidence for differences in vaccine efficacy between the two between these two extremes is most efficient, and that in these
groups. In view of the data this is not unexpected, as the total populations a single vaccination can potentially prevent almost
number of infections in vaccinated children was small, and as two infections. Of course, in practice other considerations, for
attack rates in the two subpopulations were identical (15 infections instance on ethical issues, communication, and cost-effectiveness
among the 582 children who had been vaccinated once; 10 would also come into play.
infections among the 370 who had been vaccinated twice). The
fact that attack rates were identical is somewhat surprising, as one Methods
could have expected more infections in the group that had been
vaccinated only once, more than five years ago. For completeness, Study design and data collection
we have presented the full data in Table S2. In the Netherlands, several large outbreaks of mumps virus
Further, in our analyses we assume that the vaccine works by (genotype D) occurred in 2007–2009. We collected data from
reducing the probability of transmission (i.e. we assume a leaky children attending primary schools with evidence of mumps virus
vaccine), rather than by providing all-or-nothing immunity. This transmission (report of at least one laboratory confirmed mumps
was done for simplicity, and since the current data do not allow us case or more than one clinical mumps case) [9,11]. Children’s
to distinguish between the different workings of the vaccine. If parents were asked to fill out a questionnaire asking for
additional data was available, e.g., on the pre-outbreak antibody information on the child’s vaccination status and occurrence of
titres, one could consider extension of the method by using pre- mumps. Individual data on vaccination status were also retrieved
outbreak antibody titres as an indicator for the ‘level of immunity’, from the national Dutch vaccination register. When these were not
and use this indicator to estimate how the level of pre-existing available, we used the self-reported vaccination status (vaccinated/
immunity relates to the probability of infection. In most situations, unvaccinated). Children who were vaccinated more than twice
however, such information will be hard to get, as this would (one case), and who were reported to have had mumps before
necessitate a large prospective study. September 2007 (three cases) were excluded. The study was
Our definition of vaccine efficacy has a clear-cut biological approved by the medical ethics committee of the University
interpretation (reduction of the probability of infection per Medical Centre Utrecht and the Radboud University Nijmegen
contact). This makes it possible to meaningfully average over Medical Centre. The data are presented in Table 1 and Tables S1,
populations with varying vaccination coverages and exposure S2.
levels, and also to extrapolate beyond the study population. This
contrasts with traditional estimates of vaccine efficacy that are Estimation of vaccine efficacy in a disease transmission
based on a comparison of attack rates in vaccinated and framework
unvaccinated individuals (the cohort method), or that simply use Model structure. The analyses are based on the distribution
the vaccination status of the infected individuals together with the of the number of persons infected in an outbreak [16–18].
population vaccination coverage (the screening method) [14,26]. Specifically, we use so-called final size distributions of a two-type
Vaccine efficacy estimated by these methods lack a clear biological SEIR (susceptible-exposed-infectious-recovered) model in which
interpretation, and in essence assumes that a person’s risk of the two types represent unvaccinated and vaccinated persons.
infection is independent of whether or not others in the population In SEIR models, each individual in the population can either be
are infected. This makes interpretation of the estimates problem- susceptible (i.e. healthy), exposed (infected but not able to infect
atic, and forbids estimation of the critical vaccination coverage others), infective (infected and able to infect others) or recovered (not
[15,27–29]. infectious and now immune). We assume that infectious contacts are
Even though our definition of vaccine efficacy differs funda- made at the level of the school, and not at other organizational levels
mentally from vaccine efficacy measured by the cohort method, (e.g., class, household, community). These assumptions seem
the results are quantitatively in fair agreement with traditional reasonable since there was no evidence of sustained community
estimates, especially in populations with low vaccination transmission during the study period, while only limited information
coverage and large number of infections (Table 3 versus was available on class structure within schools.
We focus on estimation of two key epidemiological quantities, likelihood pðdjpÞ can be extremely complicated, even if no
the basic reproduction number R0 which quantifies the transmis- infection or vaccination information is missing. We therefore
sibility of the pathogen, and vaccine efficacy VES which adopt an alternative approach in which attention is shifted to the
determines the reduction in the probability of infection for those joint posterior density of the parameters of interest and random
who have been vaccinated. We use a Bayesian inferential directed graphs (digraphs) G which describe the potential
framework in which these parameters are estimated and the pathways of infection. Specifically, a given graph Gdescribes all
missing information is imputed. directed contacts made between individuals, some of which may
Throughout we assume that a pair of individuals makes contacts correspond to actual infections (e.g. a link from an infective to a
at a rate that is inversely proportional to the school size N, thus susceptible) while others may not (e.g. a link from a susceptible to
ensuring that each person makes an identical expected number an infective). Knowledge of G, and the identity of the initially
of contacts per unit of time. Specifically, whilst infective an infective individual(s), determines which individuals in the
unvaccinated person makes infectious contacts with each population ultimately become infected, i.e. the final outcome. Full
unvaccinated individual according to a Poisson process of details of this method are given in [18,19] and so we now recall the
rateNl , and with each vaccinated individual according to a salient points in our setting.
Poisson process of rateð1{VES Þ Nl , where VES represents The augmented posterior density is given by
vaccine efficacy for susceptibility [1]. All Poisson processes are
assumed mutually independent. Notice that vaccine efficacy as pðp,GjdÞ!pðdjp,GÞpðGjpÞpðpÞ,
defined here can be interpreted as the reduction in the
probability of infection for a contact that would have resulted where pðdjp,GÞpðGjpÞ is the augmented likelihood, and pðpÞ is
in infection if the contacted person was unvaccinated. Hence, the prior density of the parameters [18]. The (augmented)
we have 0ƒVES ƒ1 by definition. likelihood contains two factors, of which the first indicates whether
Final size data alone do not allow us to estimate parameters with a certain combination of parameters and digraphs is compatible
respect to calendar time, but only relative to other model with the data, i.e. pðdjp,GÞ~1 if the parameters and digraphs are
parameters. To set a time-scale, and for simplicity, we therefore compatible with the data, and pðdjp,GÞ~0 otherwise. The second
assume that the infectious period (i.e. the time that an individual is factor gives the likelihood of a certain infection graph conditional
in the infective state) is fixed at length 1 time unit and set the basic on the values of the transmission parameters. For a single school
reproduction number R0 equal to the contact rate parameter l. An the likelihood of a digraph G given parameters p, pðG jpÞ, is given
alternative possibility, which could also be incorporated into our by the product of the likelihoods of all infectious links and non-
modelling and inference framework, is to assume that infectious links (i.e. the absence of an infectious link) in the set of infected
periods are exponentially distributed with mean 1, which yields a and potentially infected persons, times the product of all non-
geometric distribution for the number of contacts made by an links from infected persons to persons who were known to be
individual whilst they remain infective [30,31]. In practice, this uninfected. Hence, in a school with n1 infected persons, n2
alternative choice of infectious period distribution rarely makes potentially infected persons, and n3 uninfected persons, pðG jpÞ
any material difference to the results [18]. is given by the product of the probabilities of all
Our model contains two epidemiological parameters, namely ðn1 zn2 Þðn1 zn2 {1Þ links and non-links in the set of infected
the basic reproduction number R0 determining overall trans- and potentially infected persons, times the product of the
missibility, and vaccine efficacyVES which quantifies the extent probabilities of the non-links from persons that are infected in
to which vaccination reduces the probability of becoming the digraph to persons who are known to be uninfected. Since
infected by a single contact. In a large population with there are at least n1 infected persons and n3 uninfected persons
vaccination coverage q the reproduction number in the early the number of non-links to uninfected persons is at least n1 n3 . It
stages of an epidemic, Rq , takes a simple form, namely can be higher if some of the persons with unknown infection
Rq ~R0 ð1{q VES Þ [32]. The critical vaccination coverage qc status are infected, and reaches a maximum of ðn1 zn2 Þn3 if all
which makes major outbreaks highly improbable is found by persons with unknown infection status are infected.
solving the above equation for Rq ~1, yielding The above description can be made mathematically precise
[18,19]. If we denote by I the (unobserved) set of infected
qc ~VES{1 1{R{1
0 : individuals (of which there are at least n1 and at most n1 zn2 ), by
I the set of individuals who are known to be infected together
This equation is used to estimate the critical vaccination with the individuals who may or may not have been infected, by J
coverage directly from estimates of the basic reproduction the set of individuals who are known to be uninfected, by U (for
number and vaccine efficacy. unvaccinated) and V (for vaccinated) the possible person types,
The parameters R0 , VES , and q also determine outbreak size in by tðiÞ[fU,V g the type of an individual with label i, by ptðiÞt the
a large population through the final size equations probability that there is a link from the individual with label i to
zU ~1{expð{ð1{qÞRUU zU {qRVU zV Þand zV ~1{expð{ð1 an individual of type t, by nit the number of links from
{qÞRUV zU {qRVV zV Þ, where zU and zV denote the fractions individual i to persons
P of type t, by Nt the number of individuals
infected in the unvaccinated and vaccinated groups, and Rij of type t in I ( Nt ~n1 zn2 ), and by Mt the number of
(i,j[fU,V g) represent the type-reproduction numbers [32]. In our t[fU, V g
P
model we have R.U ~R0 and R.V ~ð1{VES ÞR0 . individuals of type t in J ( Mt ~n3 ), then the likelihood
t[fU, V g
The likelihood function. In a Bayesian framework the key
object of interest is the posterior density of the model parameters p pðG jpÞ is given by
given data d, pðpjdÞ. Using Bayes’ rule the posterior density can be
expressed as pðpjdÞ!pðdjpÞpðpÞ, where pðdjpÞ is the likelihood pðGjpÞ~
and pðpÞ the prior density of the parameters. In practical
n N {1 {n M ð1Þ
applications, this formulation is of limited use because the P P ptitðiÞt 1{ptðiÞt t ftðiÞ~tg it x P P 1{ptðiÞt t :
i[I t[fU,V g i[I t[fU,V g
Recall that we assume that the infectious periods are of fixed persons with unknown infection status cannot easily be removed.
duration. Together with our earlier assumption that contacts are Notice that this operation leaves the topology of the graph intact
made according to mutually independent Poisson processes with (distribution of links and types), and thus does not affect the
rates that are inversely proportional to school size, the infection likelihood.
probabilities pst (s,t[fU,V g) are given bypst ~1{exp { RNst , After running a number of exploratory analyses output is
where Rst are the type-specific reproduction numbers, Our generated for a single chain of length 30,000–50,000, of which the
parameterization implies R.U ~R0 and R.V ~ð1{VES ÞR0 . last 20,000–25,000 cycles are used to obtain a thinned sample of
Scenarios and estimation. The parameter vector p contains size 5,000 or 10,000. Inspection of convergence of the chain is
the epidemiological parameters R0 and VES , and the unknown performed visually. Run times are approximately 7–10 days on a
vaccination statuses. Throughout, the basic reproduction number 3.2Ghz eight-core workstation.
and vaccine efficacy are assigned uninformative uniform prior
distributions (0ƒVES v1 and R0 w0). We further assume that the Simulated outbreaks
probability that a person with unknown vaccination status is To explore the correspondence between the parameter
vaccinated is given by the observed vaccination coverage of the estimates with the data, we simulated outbreaks in schools of size
school in which the person resides (Table 2). We consider two 200 using the digraph construction described above. To prevent
scenarios: One in which both R0 and VES are identical across early extinction we introduced three infectious persons with
schools, and the other in which R0 and VES are estimated for each random vaccination status in each simulation. For each vaccina-
school separately. tion composition, we generated 5,000 random digraphs with the
The posterior density is explored using a Markov chain Monte values of the basic reproduction number and vaccine efficacy
Carlo method, whereby the missing vaccination statuses are sampled without replacement from the posterior distribution.
included as latent parameters [18–20]. Digraphs are updated by Subsequently, for each graph we calculated the attack rate among
adding and deleting edges at random from I. Specifically, we use those that were initially susceptible, and present the median and
a birth-death construction for updating, in which infectious 2.5% and 97.5% percentiles of the resulting distributions (the
contacts between any two persons that are (potentially) infected black line and grey area in Figure 2).
are chosen uniformly at random [18–20]. The Metropolis-
Hastings acceptance probability of adding an infectious link to a Estimation of vaccine efficacy by the cohort method
0
digraph G resulting in a digraph G 0 , is given by ppððGGDpDpÞÞ ‘max {‘G
‘G z1 ,
To compare our results with estimates of vaccine efficacy using
where ‘max ~ðn1 zn2 Þðn1 zn2P {1Þ P is the maximum number of the cohort method [14], we have calculated vaccine efficacy as
infectious contacts, and ‘G ~ nit the number of infec- 1 minus the relative risk of infection in vaccinated versus
i[I t[fU,V g unvaccinated persons. In these analyses only information of
tious contacts in I. Likewise the acceptance probability of an persons with known vaccination and infection status was taken into
0
attempt to delete an edge is ppððGGDpDpÞÞ ‘max {‘
‘G
, where it is understood account (Table S1). As in the above we employ a Bayesian
G z1 framework in which the probabilities in the unvaccinated and
that digraphs that are not compatible with the data have zero vaccinated groups are assigned uniform prior distributions,
likelihood. Notice that, in contrast to earlier studies, the likelihood yielding beta-binomial posterior distributions for the infection
ratio of two graphs differing by one link in general does not reduce probabilities. Estimates are obtained using Markov chain Monte
to the likelihood ratio of having an infectious contact at a Carlo (MCMC) methods, specifically by taking a thinned sample
particular position versus not having an infectious contact at that of 10,000 from a converged chain of length 500,000. Table S3
position [18–20]. In particular, it is possible that by adding or reports classical (frequentist) estimates of vaccine efficacy using the
deleting an infectious contact the number of infected persons cohort method [14].
increases or decreases by more than one, because an addition or
deletion of an infectious contact could result in the addition or
Supporting Information
deletion of a number of vertices to the connected component
which determines the final size. Hence, updating of the graphs Table S1 Overview of the outbreaks of mumps in Dutch
requires calculation of the full likelihood of the proposed graph, primary schools. See Ruijs et al. (2011) (ref [9]) and Snijders et al.
which is a computationally expensive operation, resulting in long (2012)(ref [11]) for details.
runtimes if the number of infected persons is large. (DOC)
The basic reproduction number and vaccine efficacy are
Table S2 Summary statistics of the study population, distin-
updated with a random-walk Metropolis algorithm using Gaussian
guishing between one and two vaccinations (cf. Table 1). The
proposal distributions with standard deviations of 0.2–2 and 0.02–
column ‘number infected’ shows the possible range of actual
0.2, respectively. Vaccination statuses are updated by flipping the
infections, ranging from the number known to be infected to the
vaccination status of a randomly selected person with unknown
sum of this number and the number of persons with unknown
vaccination status [20]. The index case is assumed to be an
infection status. Vaccination coverage and attack rates are
unvaccinated person.
calculated using persons with known vaccination status (vaccina-
Updating is performed in blocks, in the order 1) update the
tion coverage), and known vaccination and infection status (attack
value of the reproduction number, 2) update the value of vaccine
rates).
efficacy, 3) for each school update the vaccination status of a
(DOC)
randomly chosen person with missing vaccination information, 4)
for each school attempt to add an infectious contact, and 5) for Table S3 Classical estimates of vaccine efficacy by the cohort
each school attempt to delete an infectious contact. Each cycle of method, i.e. as 1 minus the relative risk of infection in vaccinated
the chain thus contains 32 updating events. To improve mixing versus unvaccinated persons (Orenstein et al. 1985) (ref [14]).
every 50th cycle the positions in the digraph of infected persons Approximate 95% confidence intervals of the parameter estimates
(both vaccinated and vaccinated) are randomly permuted so that are given between brackets.
the chain does not get stuck in topologies from which links to (DOC)
References
1. de Melker HE, Schellekens JF, Neppelenbroek SE, Mooi FR, Rümke HC, et al. 16. O’Neill PD (2010) Introduction and snapshot review: Relating infectious disease
(2000) Reemergence of pertussis in the highly vaccinated population of the transmission models to data. Statistics in Medicine 29: 2069–2077
Netherlands: observations on surveillance data. Emerging Infectious Diseases 6: 17. Cauchemez S, Ferguson NM (2012) Methods to infer transmission risk factors
348–357 in complex outbreak data. Proceedings of the Royal Society Interface 9: 456–
2. Wichmann O, Hellenbrand W, Sagebiel D, Santibanez S, Ahlemeyer G, et al. 569
(2006) Large measles outbreak at a German public school, 2006. Pediatric 18. Demiris N, O’Neill PD (2005) Bayesian inference for stochastic multitype
Infectious Disease Journal 26: 782–786 epidemics in structured populations via random graphs. Journal of the Royal
3. Quinlisk MP (2010) Mumps control today. Journal of Infectious Diseases 202: Statistical Society B 67: 731–745
655–656 19. O’Neill PD (2009) Bayesian inference for stochastic multitype epidemics in
4. Greenland K, Whelan J, Fanoy E, Borgert M, Hulshof K, et al. (2012) Mumps structured populations using sample data. Biostatistics 10: 779–791
outbreak among vaccinated university students associated with a large party, the 20. van Boven M, Kretzschmar M, Wallinga J, O’Neill PD, Wichmann O (2010)
Netherlands, 2010. Vaccine 30: 4676–4680 Estimation of measles vaccine efficacy and critical vaccination coverage in a
5. Cohen C, White JM, Savage EJ, Glynn JR, Choi Y, et al. (2007) Vaccine highly vaccinated population. Journal of the Royal Society Interface 7: 1537–
effectiveness estimates, 2004–2005 mumps outbreak, England. Emerging 1544
Infectious Diseases 13: 12–17 21. Neal PJ, Roberts GO. (2004) Statistical inference and model selection for the
6. Dayan GH, Quinlisk MP, Parker AA, Barskey AE, Harris ML, et al. (2008) 1861Hagelloch measles epidemic. Biostatistics. 5: 249–61
Recent resurgence of mumps in the United States. New England Journal of 22. Groendyke C, Welch D, Hunter DR (2012) A network-based analysis of the
Medicine 358:1580–1589 1861 Hagelloch measles data. Biometrics 68: 755–765
7. Muhsen K, Shohat T, Aboudy Y, Mendelson E, Algor N, et al. (2011)
23. Anderson RM, May RM (1991) Infectious Diseases of Humans. Dynamics and
Seroprevalence of mumps-antibodies in subpopulations subsequently affected by
Control. Oxford: Oxford University Press.
a large scale mumps epidemic in Israel. Vaccine 29: 3878–3882
24. Kanaan MN, Farrington CP (2005) Matrix models for childhood infections: a
8. Rubin SA, Link MA, Sauder CJ, Zhang C, Ngo L, et al. (2012) Recent mumps
Bayesian approach with applications to rubella and mumps. Epidemiology and
outbreaks in vaccinated populations: no evidence of immune escape. Journal of
Infection 133: 1009–1021
Virology 86: 615–620
9. Ruijs WL, Hautvast JL, Akkermans RP, Hulscher ME, van der Velden K (2011) 25. Wallinga J, Teunis P, Kretzschmar M (2006) Using data on social contacts to
The role of schools in the spread of mumps among unvaccinated children: a estimate age-specific transmission parameters for respiratory-spread infectious
retrospective cohort study. BMC Infectious Diseases 11: 227 agents. American Journal of Epidemiology 164: 936–944
10. Wielders CC, van Binnendijk RS, Snijders BE, Tipples GA, Cremer J, et al. 26. Farrington CP (1993) Estimation of vaccine effectiveness using the screening
(2011) Mumps epidemic in orthodox religious low-vaccination communities in method. International Journal of Epidemiology 22: 742–746
the Netherlands and Canada, 2007 to 2009. Euro Surveillance 16: pii 19989 27. Becker NG, Britton T, O’Neill PD (2003) Estimating vaccine effects on
11. Snijders BE, Lier AV, van de Kassteele J, Fanoy EB, Ruijs WL, et al. (2012) transmission of infection from household outbreak data. Biometrics 59: 467–475
Mumps vaccine effectiveness in primary schools and households, the Nether- 28. Hudgens MG, Halloran ME (2008) Toward causal inference with interference.
lands, 2008. Vaccine 30: 2999–3002 Journal of the American Statistical Association 103: 832–842
12. Ruijs WL, Hautvast JL, van Ansem WJ, Akkermans RP, van’t Spijker K, et al. 29. VanderWeele TJ, Vandenbroucke JP, Tchetgen EJ, Robins JM (2012) A
(2012) Measuring vaccination coverage in a hard to reach minority. European mapping between interactions and interference: implications for vaccine trials.
Journal of Public Health 22: 359–364 Epidemiology 23: 285–292
13. van Lier A, Oomen P, de Hoogh P, Drijfhout I, Elsinghorst B, et al. (2012) 30. Lloyd-Smith JO, Schreiber SJ, Kopp PE, Getz WM (2005) Superspreading and
Præventis, the immunization register of the Netherlands: a tool to evaluate the the effect of individual variation on disease emergence. Nature 438: 355–359
National Immunization Programme. Euro Surveillance 17: pii = 20153 31. van Boven M, Koopmans M, Du Ry van Beest Holle M, Meijer A, Klinkenberg
14. Orenstein WA, Bernier RH, Dondero TJ, et al. (1985) Field evaluation of D, et al. (2007) Detecting emerging transmissibility of avian influenza virus in
vaccine efficacy. Bulletin of the World Health Organization 63: 1055–1068 human households. PLoS Computational Biology 3: e145
15. Halloran ME, Longini IM, Struchiner CJ (2010) Design and Analysis of Vaccine 32. Diekmann O, Heesterbeek JAP, Britton T (2012) Mathematical Epidemiology of
Studies. New York: Springer. Infectious Diseases. Model Building, Analysis and Inference. New York: Wiley.