Loading [MathJax]/jax/output/SVG/jax.js
Research article

Possible counter-intuitive impact of local vaccine mandates for vaccine-preventable infectious diseases


  • Received: 29 February 2024 Revised: 17 June 2024 Accepted: 26 June 2024 Published: 08 July 2024
  • We modeled the impact of local vaccine mandates on the spread of vaccine-preventable infectious diseases, which in the absence of vaccines will mainly affect children. Examples of such diseases are measles, rubella, mumps, and pertussis. To model the spread of the pathogen, we used a stochastic SIR (susceptible, infectious, recovered) model with two levels of mixing in a closed population, often referred to as the household model. In this model, individuals make local contacts within a specific small subgroup of the population (e.g., within a household or a school class), while they also make global contacts with random people in the population at a much lower rate than the rate of local contacts. We considered what would happen if schools were given freedom to impose vaccine mandates on all of their pupils, except for the pupils that were exempt from vaccination because of medical reasons. We investigated first how such a mandate affected the probability of an outbreak of a disease. Furthermore, we focused on the probability that a pupil that was medically exempt from vaccination, would get infected during an outbreak. We showed that if the population vaccine coverage was close to the herd-immunity level, then both probabilities may increase if local vaccine mandates were implemented. This was caused by unvaccinated pupils possibly being moved to schools without mandates.

    Citation: Maddalena Donà, Pieter Trapman. Possible counter-intuitive impact of local vaccine mandates for vaccine-preventable infectious diseases[J]. Mathematical Biosciences and Engineering, 2024, 21(7): 6521-6538. doi: 10.3934/mbe.2024284

    Related Papers:

    [1] Qianqian Cui, Zhipeng Qiu, Ling Ding . An SIR epidemic model with vaccination in a patchy environment. Mathematical Biosciences and Engineering, 2017, 14(5&6): 1141-1157. doi: 10.3934/mbe.2017059
    [2] Roy Curtiss III . The impact of vaccines and vaccinations: Challenges and opportunities for modelers. Mathematical Biosciences and Engineering, 2011, 8(1): 77-93. doi: 10.3934/mbe.2011.8.77
    [3] Lili Han, Mingfeng He, Xiao He, Qiuhui Pan . Synergistic effects of vaccination and virus testing on the transmission of an infectious disease. Mathematical Biosciences and Engineering, 2023, 20(9): 16114-16130. doi: 10.3934/mbe.2023719
    [4] Hamed Karami, Pejman Sanaei, Alexandra Smirnova . Balancing mitigation strategies for viral outbreaks. Mathematical Biosciences and Engineering, 2024, 21(12): 7650-7687. doi: 10.3934/mbe.2024337
    [5] Anthony Morciglio, R. K. P. Zia, James M. Hyman, Yi Jiang . Understanding the oscillations of an epidemic due to vaccine hesitancy. Mathematical Biosciences and Engineering, 2024, 21(8): 6829-6846. doi: 10.3934/mbe.2024299
    [6] Siyu Liu, Mingwang Shen, Yingjie Bi . Global asymptotic behavior for mixed vaccination strategy in a delayed epidemic model with interim-immune. Mathematical Biosciences and Engineering, 2020, 17(4): 3601-3617. doi: 10.3934/mbe.2020203
    [7] Martin Luther Mann Manyombe, Joseph Mbang, Jean Lubuma, Berge Tsanou . Global dynamics of a vaccination model for infectious diseases with asymptomatic carriers. Mathematical Biosciences and Engineering, 2016, 13(4): 813-840. doi: 10.3934/mbe.2016019
    [8] Bruce Pell, Matthew D. Johnston, Patrick Nelson . A data-validated temporary immunity model of COVID-19 spread in Michigan. Mathematical Biosciences and Engineering, 2022, 19(10): 10122-10142. doi: 10.3934/mbe.2022474
    [9] Andras Balogh, Tamer Oraby . Stochastic games of parental vaccination decision making and bounded rationality. Mathematical Biosciences and Engineering, 2025, 22(2): 355-388. doi: 10.3934/mbe.2025014
    [10] Diana Bolatova, Shirali Kadyrov, Ardak Kashkynbayev . Mathematical modeling of infectious diseases and the impact of vaccination strategies. Mathematical Biosciences and Engineering, 2024, 21(9): 7103-7123. doi: 10.3934/mbe.2024314
  • We modeled the impact of local vaccine mandates on the spread of vaccine-preventable infectious diseases, which in the absence of vaccines will mainly affect children. Examples of such diseases are measles, rubella, mumps, and pertussis. To model the spread of the pathogen, we used a stochastic SIR (susceptible, infectious, recovered) model with two levels of mixing in a closed population, often referred to as the household model. In this model, individuals make local contacts within a specific small subgroup of the population (e.g., within a household or a school class), while they also make global contacts with random people in the population at a much lower rate than the rate of local contacts. We considered what would happen if schools were given freedom to impose vaccine mandates on all of their pupils, except for the pupils that were exempt from vaccination because of medical reasons. We investigated first how such a mandate affected the probability of an outbreak of a disease. Furthermore, we focused on the probability that a pupil that was medically exempt from vaccination, would get infected during an outbreak. We showed that if the population vaccine coverage was close to the herd-immunity level, then both probabilities may increase if local vaccine mandates were implemented. This was caused by unvaccinated pupils possibly being moved to schools without mandates.



    Measles is a viral disease that yearly kills over a hundred thousand people worldwide [1]. It is one of the most contagious diseases known, with a basic reproduction number R0 estimated to be between 12 and 18 [2] (see Section 1.2 and [3] for a definition of R0). Because measles is extremely contagious and immunity after infection is typically life-long, in the absence of vaccination, most people will be infected before reaching adulthood, hence measles is often called a childhood disease.

    In some countries, measles and other vaccine-preventable infectious diseases, such as rubella, mumps, and pertussis, are still endemic. That is, they are always present and cause regular peaks of prevalence. In many other countries, measles is largely absent because of successful vaccination campaigns [1,4], which caused herd-immunity in the population [3, p69]. We focus on those countries. In several of the countries in which there have not been considerable outbreaks of measles for decades, the vaccine coverage has been declining, and there is fear that the "pool of susceptibles" will grow big enough to lose herd-immunity and allow for new waves of infection. Governments and public health authorities try to tackle this with campaigns to promote vaccine uptake and sometimes by legislation, which makes some vaccines compulsory for children, e.g., in Italy [5], or considering giving schools/daycare centers the freedom to use a vaccine mandate, as was considered in the Netherlands [6]. The ethical debate on whether childhood vaccination should be mandatory or not is a complex and contested topic. See, e.g., [7] for a review of the ethical literature.

    In this paper, we consider the case in which vaccination is not mandatory for everybody and the decision of requiring the vaccine is left to the schools, as was proposed in the Netherlands. In this situation, children who cannot get a vaccine for medical reasons can freely choose where to attend school, whether it has a mandate or not. We refer to those children as medically exempt children. So, in this situation, unvaccinated children that are not medically exempt all go to schools without a mandate, and in those schools the fraction of unvaccinated children may be higher than if vaccine mandates are not allowed, which (as we show) may lead to higher risk of outbreaks of the vaccine-preventable diseases.

    Skepticism and opposition toward vaccination represent a serious matter, to the point that the World Health Organization (WHO) has listed vaccine hesitancy as one of the ten threats to global health in 2019 [8]. Fear of possible side effects, religious convictions, and spread of misinformation are some of the causes of decline in vaccination coverage [9,10]. It is reasonable to argue that, if given the choice, parents who are reluctant toward vaccines are more likely to enroll their children in schools that do not require vaccination. On the other hand, it is likely that medically exempt children go to schools that require vaccination, in order to obtain a safer environment. We are interested in the question whether and under which conditions local vaccine mandates will lead to increased risk of infection for unvaccinated children that are medically exempt.

    In this paper, we use a mathematical model to analyze the spread of measles among elementary school-aged children, where we take into account that most spread of the pathogen takes place within classes. Although we take measles as the pathogen of interest, the analysis applies equally to other highly infectious vaccine preventable diseases. We use a stochastic SIR (susceptible, infectious, recovered) epidemic model in which the relevant population is closed and partitioned in school classes. So, the total number N of children is fixed: We do not consider enrollments nor graduations during the time of study, which correspond to births and deaths, respectively, in the usual SIR model terminology, and we assume that the time scale of an outbreak is such that we may ignore children moving to different classes. We define nc as the number of children per class, which we, for convenience of exposition, consider fixed, and mc as the number of classes. So, N=ncmc. In Section 4.1, we show that the mathematical analysis is easily expanded to allow for variation in the class sizes. We remark that, strictly speaking, we analyze our model for a sequence of populations, where N.

    In SIR models, each individual can be susceptible to the disease, infectious or recovered and immune. A contact trough which the virus is spread is called an infectious contact. An infectious individual remains so for a random length of time that is called the infectious period, during which they are able to infect other individuals in the population. In our analysis we only focus on the probability that an introduction of a virus may lead to a large outbreak and on the probability that a given child gets infected. We do not consider the dynamics of such an outbreak in real time. Because of this, possible exposed/latent periods between a child getting infected and becoming infectious have no impact on our quantities of interest [11].

    In many epidemic models, it is assumed that the population is homogeneous, which means that the individuals mix uniformly. However, in real life this is rarely the case. Here, we introduce heterogeneity by allowing individuals to have local (within class) and global contacts (between classes) with different probabilities. We assume that, conditioned on an individual being infected, the events that whether or not infectious contacts are made with different individuals are independent. So, we implicitly assume that the infectious period is not random [12]. Obviously, the infectious period for measles is random. However, the variance of time between the start of the infectious period and the onset of rash (which is close to the period that a child stays socially active and is likely to infect other children) is very small [13]. This makes that we can simplify the analysis and assume that contacts between individuals occur independently [12]. In our model, an infectious individual makes an infectious contact with each classmate with probability pL, which for the moment we assume to be equal to 1, and with other individuals in the population with probability pG. This model is also referred to as the household model, see e.g., [14] or [11, Section 6.3], since people are divided into small groups (such as workplaces, households, or classrooms) where transmission of the virus is more frequent. The global transmission probability is typically much smaller than the local probability, and in our case we assume that pG is asymptotically inversely proportional with the population size. That is, if we consider a sequence of epidemic models indexed by the population size N, then there exists a constant λG(0,) such that limNNpG=λG. This assumption implies that in the model, the rate of global contacts made by each child is asymptotically independent of the total population size [3, page 4].

    The basic reproduction number R0 is the expected number of infections caused by a "typical" infected individual infected during the initial stages of an epidemic in a population that is mostly susceptible [3] (so, R0 refers to a situation in which there is no vaccination.) In homogeneously mixing populations, defining R0 is straightforward. Although there is some issue with deciding what a typical infected individual is, it is also possible to define R0 and other reproduction numbers for household models [15,16]. In Section 2, we deduce the model parameter λG from estimates of R0 from literature.

    For measles, the basic reproduction number R0 is estimated to be between 12 and 18 [2,13], which makes the virus one of the most contagious viruses in the world. Assuming pL=1 means that if one of the unvaccinated children in a class gets the disease, then they will infect all of their unvaccinated classmates. Setting pL<1 would complicate the exposition considerably, while the qualitative behavior of the model stays similar, as long as pL is large enough (say, larger than 4 divided by the class size, when the probability that the entire class gets infected is roughly 98%); see Section 4.2.

    Because pL=1, a child and their non-vaccinated classmates together can be seen as a "super individual", of which the susceptibility and infectivity is proportional to the number of unvaccinated children in the class. We assume that the latent period of measles is relatively long, so that we can safely assume that all secondary infections in a class are directly caused by the initially infected child in the class, and not by another secondary infected in the population. Lastly, we define a time unit as the fixed infectious period [11, p4].

    To analyze the impact of vaccination (or the lack thereof), we assume that the vaccine against measles is perfect, i.e., once vaccinated, a person can neither get infected nor infect other people. So, vaccinated people are effectively omitted from the model. A fraction v of the population is vaccinated; we call Nu(v)=(1v)N the number of unvaccinated children, which are initially the susceptibles in our model. Another way to introduce vaccines is to say that each child is independently vaccinated with probability v. The two assumptions lead to slightly different models, but since they are asymptotically equivalent, we use one or the other depending on what is more convenient in the situation. For example, when it is useful to think about Nu as a fixed number, we will use the first assumption; if we want the number of initially susceptible children in different classes to be independent, then the second model is better.

    In order to protect children at a school, the school may (if allowed by law) institute a local vaccine mandate, which requires all pupils that are not medically exempt to be vaccinated. In our model, we denote the fraction of classes with such a mandate by π. We require πv for reasons of consistency; see Section 2.

    To evaluate the global infection rate λG for given R0, we consider a situation with no vaccines nor mandates, i.e., both v and π are set to zero. Recall that classes are all of fixed size nc. We distinguish between children that are infected through a global contact and children that are infected within a class through a local contact. The first group of children are the primary cases or type-one individuals, which are the first individuals of their classes to get infected; each primary case will then infect all the nc1 other (unvaccinated) children in their class, which we refer to as secondary cases or type-two individuals. We assume that both types make global contacts (and cause primary cases) at rate λG, but only primary cases cause secondary cases. For this household model, it is known that the epidemic threshold parameter R0 is the dominant eigenvalue of the next generation matrix

    A=[λGnc1λG0] (2.1)

    where element Ai,j is given by the mean number of type-j individuals that a type-i individual infects during their infectious period ([14,17]). So, R0 solves

    (λGR0)R0λG(nc1)=0.

    So,

    λG=R20R0+nc1 (2.2)

    If R0=15 and nc=25, this leads to λG5.8.

    Note that in [17], the dimension of the next generation matrix is nH×nH, where nH is the maximal household size. In our case, it is enough to consider a 2×2 matrix because we only need to divide individuals into primary and secondary cases. In Section 4.2, when dealing with pL<1, we need higher dimensional matrices.

    The dominant eigenvalue of the matrix A is sometimes called the individual reproduction number and is denoted by RI [16]. Assuming pL=1, the reproduction numbers R0 and RI are the same because all secondary cases are infected by primary ones only.

    As mentioned before, there are in total mc classes, each with nc students. So, N=mcnc. The number of classes without mandate is mc(1π), so the maximum possible number of unvaccinated children is mcnc(1π)=N(1π). Therefore, the number of unvaccinated children N(1v) has to be less or equal than N(1π). So, for the model to be consistent, we require πv.

    In a class without a vaccine mandate, the distribution of the number X of unvaccinated children is Bin(nc,1v1π) and we write πk=P(X=k). In what follows, we use

    u:=1v1π.

    The expected number of unvaccinated children per class is given by μc=ncu=nc1v1π. We observe that the probability that a child, chosen uniformly at random among the unvaccinated children in the population, belongs to a class with k unvaccinated children is given by the size-biased distribution determined through

    ˉπk=kπkE[X]=k(nck)(u)k(1u)nckncu=(nc1k1)uk1(1u)nc1k+1. (2.3)

    We let ˉX be a random variable with probability masses defined through the ˉπk and we notice that ˉX is distributed as Y+1, where Y=Y(v) is a random variable with a binomial distribution with parameters nc1 and u.

    We are interested in how the reproduction number R0,e=R0,e(v,π) (where the e stands for effective) changes once the vaccine is introduced in the population and depends on v and π. To analyze this, we compute the dominant eigenvalue of the next generation matrix

    A=A(v,π)=[λG(1v)(nc1)(1v)(1π)λG(1v)0],

    where λG(1v) is the infection rate of global contacts between the unvaccinated. The results are depicted in Figure 1.

    Figure 1.  R0,e as a function of v, if both vaccines and mandates are introduced.

    Using λG from (2.2), we also compute R,e=R,e(v,π), which is the household reproduction number after vaccination for the household model [14,16] and represents the mean number of clumps (or in our case, classes) that are infected by a typical infectious clump/class. It is defined to be R,e=λG,vμ, where λG,v is the expected number of global infections caused by a typical infected individual, and μ=μ(v) is the mean number of infected children in an infected class. λG,v is given by λG(1v), while

    μ=E[ˉX]=nck=1kˉπk=1+(nc1)u.

    The latter follows from the size biased distribution in (2.3). Substituting this in the expression for R,e, we obtain

    R,e=[1+(nc1)u](1v)λG. (2.4)

    The household reproduction number R,e, as well as most other reproduction numbers defined in literature, including R0,e, is useful because an epidemic goes extinct with probability 1 if and only if R,e1 [16]. The minimal value of v for which R,e1 is the critical vaccination coverage vcr.

    R,e>1(1v)2(1π)λG(nc1)>1(1v)λG.

    Since nc represents the number of children in a class, the left-hand side is always nonnegative. So, if the expected number of contacts with unvaccinated children (i.e., (1v)λG) is greater than 1, i.e., 1v>1λG, then the epidemic is above threshold. Assume instead that both terms are positive; then this inequality gives us an upper bound for the fraction 1π of schools without mandate:

    1π<(1v)2λG(nc1)1(1v)λG.

    So, if the number of schools without mandate is small enough and the unvaccinated children are concentrated in those schools, the value of R,e is above 1 (see Figure 2).

    Figure 2.  Left: plot of R,e as a function of v[0.8,1] when R0=15, nc=25, and π=0. The red asterisk denotes when R,e=1 and we call vcr the correspondent vaccination coverage. Right: plot of R,e as a function of π when v=vcr, again for R0=15 and nc=25.

    Assume that the epidemic starts with one randomly chosen unvaccinated child in the population and say that the class of the child has non-vaccinated children. Then, the probability that a class with k unvaccinated children gets infected by the class of the initially infected child is given by the following expression:

    πk[1(1pG)k]nck=0πk[1(1pG)k]. (2.5)

    When pG<<1 (as we assume), the expression 1(1pG)k is well approximated by kpG, and we obtain that (2.5) is approximated by

    πkpGknck=0πkpGk=πkknck=0πkk=ˉπk.

    If the number of classes is large, also the population size N and the number of unvaccinated children Nu are large. So, we can use a branching process approximation to describe the initial stages of the epidemic: In those early stages, infectious individuals contact only individuals that – with large probability – are still susceptible, so the number of new infections caused by households are asymptotically independent and global contacts lead to a tree-structured epidemic generated graph, which has the law of a multitype branching process, in which individuals correspond to classes in the epidemic process. A class of size k (i.e., a class with k unvaccinated children) corresponds to a type-k individual of the Galton-Watson approximating tree. Since the infection rate for global contacts between unvaccinated is λG(1v), the offspring of a type-k individual is Poiss(λG(1v)k) [18].

    More formally, by a birthday problem approach [21, p.24], we can argue that, with high probability, the first k global contacts, k=o(Nu), are all with different households. Furthermore, it is well-known (e.g., [11, Chapter 3]) that if we denote the number of infected children in an SIR epidemic by |ENu| and the size of the approximating branching process (i.e., a Galton-Watson process with offspring distribution described above) by |B|, then as Nu,

    P(|ENu|>logNu)P(|B|=).

    If |ENu|>logNu, we talk about a major outbreak of the epidemic.

    If the first infectious child belongs to a class of size k (i.e., a class with k unvaccinated children), then the probability that the process dies out is equal to the probability that the infectious offspring of each student in the first infectious class dies out. This leads to the following expressions for ˉq, the extinction probability of the approximating branching process for an epidemic initiated by a uniformly chosen unvaccinated individual from the population.

    ˉq=nck=1P(ˉX=k)=0[λG(1v)k]!eλG(1v)kˉql=nck=1P(ˉX=k)eλG(1v)k(1ˉq)=nc1j=0P(ˉX1=j)eλG(1v)(j+1)(1ˉq)=nc1j=0(nc1j)uj(1u)nc1jeλG(1v)(j+1)(1ˉq)=(1u+ueλG(1v)(1ˉq))nc1eλG(1v)(1ˉq) (2.6)

    which means that ˉq is a solution of the fixed point equation

    ˉq=(1u+ueλG(1v)(1ˉq))nc1eλG(1v)(1ˉq).

    From the theory of branching processes [19], we know that ˉq is the smallest solution of this equation, which is strictly less than 1 if and only if R,e>1.

    By assumption, a medically exempt child is enrolled in a school that has a vaccine mandate. This means that they belong to a class of effective size 1. By the epidemic generated graph representation of an epidemic [18], we know that pme, the probability that a medically exempt child avoids being infected if a large outbreak occurs, is equal to the probability that a large outbreak occurs if the outbreak starts with a single infected child in a class of effective size 1. This result can also be obtained by using susceptibility sets [20] and see that the so-called forward and backward branching process have the same offspring distribution. Using arguments as in (2.6), we obtain that

    pme=pme(v,π)=eλG(1v)(1ˉq). (2.7)

    If there is no vaccine mandate, a medically exempt child is just like any other unvaccinated child and the probability of avoiding infection is ˉq for the case that π=0. We denote this probability by pwm=pwm(v), where wm stands for "without mandate".

    Considering the entire population, we may expect that mandates make the global situation worse, because unvaccinated children have, on average, more other unvaccinated children in their class, and children in the same class contact each other at a higher rate than other pairs of children. However, in this situation, a medically exempt child can only be infected by global contacts. In the case without mandate, even if the global situation is more controlled, such a child could be enrolled in a class with other unvaccinated children, and the probability pme for a given medically exempt child to avoid infection in a situation with vaccine mandates may be larger than that probability in a situation without mandates pwm. In this section, we analyze the model to justify this intuition.

    In Figure 3 we consider the case where π=1/2 and v is varying. Here, pme is higher than pwm up to vvs93%, and we have the opposite inequality for v>vs, where the s stands for switching level. The graph tells us that if v is low, a medically exempt child is profiting from vaccine mandates at schools, since they will for sure be enrolled in a class with no other unvaccinated children. However, when the vaccine percentage is sufficiently high, the probability of having only one unvaccinated child per class is already high and mandates worsen the global situation for everyone. Note that pwm reaches 1 at a lower v value than pme. This confirms the intuition that without mandates, a lower vaccine coverage is sufficient to guarantee herd-immunity.

    Figure 3.  Plot of the probability pme that a medically exempt child avoids infection as a function of v with π set to 0.5, and of the probability pwm that a random unvaccinated child avoids infection in the case with no mandates (π=0). We see that the two lines cross at vs0.925. The minimal value of v for which pwm=1 is at v0.933, while the minimal value of v for which pme=1 is at v0.95.

    In Figure 4, we again consider pme, the probability that a medically exempt individual avoids infection, but now as a function of π, when v=0.93 and v=0.90. For v=0.93, we can see that, for low values of π (lower than approximately 0.20), the medically exempt are more protected in the model with mandates. But if π increases, then the mandates are actually disadvantageous to the medically exempt children, even if they are the only unvaccinated children in their class. If v=0.90, already without any vaccine mandates, the global spread of the disease is so easy that the global spread does not increase much by introducing vaccine mandates, while the relative protection of having no other unvaccinated children in the class is considerable and vaccine mandates are always beneficial for medically exempt children.

    Figure 4.  Plot of pme, the probability that a medically exempt child avoids infection as a function of π, when v=0.93 (left) and when v=0.90 (right). We compare it with the probability pwm that a random unvaccinated child avoids infection in the case with no mandates (π=0).

    Up to now, we have assumed that the introduction of mandates in a society will not change the vaccine percentage. This is not realistic, since some parents might vaccinate their children in order to let them go to the school of their preference. We call those parents "converted". We may think of v as the sum of two components: vi is the initial proportion of vaccinated individuals and vc is the additional portion of "converted". Note that vc1vi. The graph in Figure 5 shows a plot of R,e as a function of vc, when π=0.5, vi=0.9, and nc=25. As expected, R,e is strictly decreasing in vc: If more people get "converted" and, therefore, more children get vaccinated, the overall public health situation improves.

    Figure 5.  Left: R,e as a function of vc when π=0.5, vi=0.9, and nc=25. The red asterisk is the intersection point with y=1. (Right: Plot of pme and pwm as a function of vc when π=0.5, vi=0.9, and nc=25. Note that this figure may be deduced from Figure 3 with v restricted to [0.9,1]. This is because we fix the initial fraction of vaccinated vi and then define v=vi+vc and compute pme and pwm with such combined v.).

    In this section, we present some extensions to the basic model. In Section 4.1, we look at the case in which the number of children per class is random. In Section 4.2, we study the case of nontrivial local infection probabilities, with pL<1. Finally, in Section 4.3, we analyze the spread of other infectious diseases with basic reproduction number R0 lower than the one of measles.

    Our analysis is easily extended to random class sizes, in which the number of children in a class is distributed as a random variable Nc. Call pk=P(Nc=k); then the probability that a newly infected class is of size k is given by the size-biased version ¯Nc of Nc and we define ˉpk=P(¯Nc=k). To evaluate the expected number of type-2 individuals that a type-1 individual infects (which is element (1,2) of matrix A in (2.1)), we compute

    E[¯Nc1]=k=1(k1)ˉpk=k=1k(k1)pkE[Nc]=E[Nc(Nc1)]E[Nc].

    For example, if NcPoiss(m), say, with m=25, then E[Nc]=m and, with the same procedure as in Section 2, we obtain λG5.6.

    We denote the random number of unvaccinated per class after both vaccines and mandates are introduced in the model again by X. Then, X conditioned on Nc=k is Bin(k,u) distributed and

    P(X=l)=kP(X=l|Nc=k)P(Nc=k)

    while

    P(ˉX=l)=lP(X=l)E[X]=kP(ˉX=l|Nc=k)P(Nc=k). (4.1)

    We observe that

    P(ˉX=l|Nc=k)=lP(X=l|Nc=k)E[X|Nc=k]

    which implies, after some computations, that ˉX conditioned on Nc=k is distributed as 1+Bin(k1,u), and then allows us to compute the unconditioned distribution of ˉX using (4.1).

    Throughout all our analysis above, we have considered the case in which the local infection probability pL=1, which simplifies the model as explained in Section 1.2. Here, we want to study nontrivial cases with pL<1 that correspond to a different contact pattern.

    Assume that an infectious individual infects their classmates with probability pL=1/2 per classmate during their infectious period. Setting both v and π to zero, we compute the global infection rate λG as in Section 2. We consider a class of size 25 in which one student gets infected from a global contact. The probability that a given one of their classmates avoids infection within two steps is 12(114)23 and the probability that at least one of the classmates avoids infection is at least 24×12(114)230. This means that all the pupils of that class will have contracted the virus within two generations of infection with very high probability. A primary case infects, on average, (nc1)/2 classmates, which will, in turn, infect as many remaining susceptibles. The next generation matrix (2.1) then becomes

    A=[λGnc120λG01λG00]

    and R0 solves

    R20(λGR0)+R0λGnc12+λGnc12=0.

    If R0=15 and nc=25, then

    λG=R30R20+nc12+R0nc128.1.

    Recall that if pL=1, then λG5.8.

    Now, introduce both vaccines and mandates in the model. We consider a class with n+1 unvaccinated children, one of which is infected through a global contact. The epidemic that starts spreading via local contacts in the class is described by – to use the notation of Andersson and Britton [11, Chapter 2] – the standard SIR epidemic model En,1(λL,1), where λL indicates the local infectious rate. To compute the value of λL, we link it to pL by observing that an infectious individual makes contacts with a given classmate at the time steps of a Poisson(λL/n) process; therefore, the probability 1pL of having no contacts with a given individual is given by eλL/n.

    We indicate by Pnj the probability that j of the n initial susceptibles get infected during the (local) epidemic. The distribution of these probabilities is obtained by solving a lower triangular linear system given in [11, Chapter 2, Theorem 2.2]. Then, in this case, the probability of extinction ˉq in (2.6) will be

    ˉq=nck=1P(ˉX=k)k1j=0Pk1jl=0[λG(1v)(j+1)]ll!eλG(1v)(j+1)ˉql=nck=1P(ˉX=k)k1j=0Pk1jeλG(1v)(j+1)(1ˉq). (4.2)

    We observe that, keeping pL fixed and increasing the number of susceptibles (of the local epidemic) n, the probability Pnn that everyone gets infected gets closer and closer to 1, as seen in Figure 6.

    Figure 6.  Value of Pnn plotted as a function of n.

    In Figure 7 we see, on the left, a plot of the probabilities pme and pwm, that a medically exempt child avoids infection in the case with and without class mandates, respectively, when π=1/2, as functions of v. On the right, we can see a comparison between the cases pL=1 and pL=1/2. We observe that even if in the latter case the local contact probability is halved, the probability pme that a medically exempt child avoids infection is higher in the first case. This is caused by the dependence on the value of the global infection rate λG, which we remember increases from 5.8 in the first case to 8.1 in the second: If R0 stays the same but pL decreases, then pG increases, which means that the medically exempts are more exposed to the virus. For the probability pwm that a random unvaccinated avoids infection in the case without mandates, we conversely see almost no difference in the two cases.

    Figure 7.  Left: Plot of the probability pme that a medically exempt child avoids infection as a function of v with π set to 0.5, and of the probability pwm that a random unvaccinated child avoids infection in the case with no mandates (π=0), when the local probability pL is set to 1/2. The two lines cross at vs0.925, as in Figure 3. (Right: Contrast between the cases pL=1 (see Figure 3) and pL=1/2.).

    Similarly, we can analyze the cases pL=1/3 and pL=1/5. For the same reason as above, we can assume each individual in a class to be infected within a maximum of two generations of infectives. If one child is infected by a global contact, it will infect, on average, (nc1)/3 and (nc1)/5 classmates, respectively, which will, in turn, infect the remaining 2(nc1)/3 and 4(nc1)/5 children. The next generation matrices become

    A1/3=[λGnc130λG02λG00]A1/5=[λGnc150λG04λG00]

    which lead to λG9.3 and 10.7, respectively. Then, using the same expression for ˉq found in (4.2), we obtain similar graphs for pme and pwm, as seen in Figure 8.

    Figure 8.  Plot of the probability pme that a medically exempt child avoids infection as a function of v with π set to 0.5, and of the probability pwm that a random unvaccinated child avoids infection in the case with no mandates (π=0), when the local probability pL is set to 1/3 (left) and 1/5 (right).

    Finally, in Figure 9 we can see a comparison between the plots for pme (on the left) and pwm (on the right) for the different pL values we have previously considered. What we observe is that, while pme tends to decrease when pL decreases, pwm has the opposite behavior. When mandates are not present and unvaccinated are uniformly spread in the classrooms, having a lower pL value slightly increases the probability pwm of avoiding infection, as the clusters of infectives forming during the spread of the virus are smaller. When mandates are present, the probability pme that a medically exempt child avoids infection decreases with pL, because the global infection rate λG reaches higher values.

    Figure 9.  Comparison of the plots of the probability pme that a medically exempt child avoids infection as a function of v when π=0.5 (left), and of the probability pwm that a random unvaccinated child avoids infection in the case of no vaccine mandates (right), for the different pL values considered above.

    We are now interested in considering the case of a less infectious disease than measles, such as mumps or rubella; as an example, we take the basic reproduction number R0=6.

    Comparing Figures 3 and 10, we see that in the first case, the functions describing pme and pwm increase much faster than in the case R0=6. This means that with a less infectious virus, the difference between the cases with and without mandates is more pronounced. In fact, as shown in Figure 10, a medically exempt child escaping infection is significantly more probable in the case with vaccine mandates, up to the switching point v0.82. However, there is a wider range of vaccine coverage v, for which medically exempt children are worse off with than without vaccine mandates.

    Figure 10.  Plot of the probability pme that a medically exempt child avoids infection as a function of v with π set to 0.5, and of the probability pwm that a random unvaccinated child avoids infection in the case with no mandates (π=0). We see that the two lines cross at vs0.82. The minimal value of v for which pwm=1 is at v0.83, while the minimal value of v for which pme=1 is at v0.88. To compare it with the case R0=15, see Figure 3.

    In Figure 11, we see how pme and pwm vary as functions of π when the value of v is fixed and close to the herd immunity level; the behavior is the same as the one observed in Figure 4.

    Figure 11.  Plot of pme, the probability that a medically exempt child avoids infection as a function of π, when v=0.83 (left) and when v=0.80 (right). We compare it with the probability pwm that a random unvaccinated child avoids infection in the case with no mandates (π=0). See Figure 4 for comparison with the case R0=15.

    In this paper, we study how vaccine mandates might affect the spread of an infectious disease in elementary schools, and especially how the probability of being infected changes for medically exempt children. This effect may be counterintuitive and local mandates might worsen the risks for medically exempt children. This work was inspired by a law that was proposed by the Dutch House of Representatives and was later rejected by the Senate [6]. In this measure, childcare owners could freely decide whether or not joining the public vaccination program was compulsory for children to enroll at the daycare.

    In our study, we do not distinguish between child-daycare and elementary schools, and we only allow the disease to spread within schools, while the population structure and contact patterns are actually much more complex and the infection could take place in households, sport centers, and so on. Moreover, we do not consider multiple waves of infection nor the possibility of becoming susceptible after being recovered, which indeed is not that relevant for measles, but might be for other diseases.

    Despite the limitations due to the model hypotheses, the analysis describes the qualitative different behaviors of the two situations (with and without vaccine mandates) and can be important for further ethical and political decisions made on vaccines.

    The analysis can easily be extended to diseases with lower reproduction numbers than measles; see Section 4.3. It depends on the relative spread within and between households whether this will lead to less or more pronounced possible undesirable effects of vaccine mandates.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    The authors declare there is no conflict of interest.



    [1] World Health Organisation, Measles, 2023. Available from: https://www.who.int/news-room/fact-sheets/detail/measles.
    [2] S. Thompson, J. C. Meyer, R. J. Burnett, S. M. Campbell, Mitigating vaccine hesitancy and building trust to prevent future measles outbreaks in England, Vaccines, 11 (2023), 288. https://doi.org/10.3390/vaccines11020288 doi: 10.3390/vaccines11020288
    [3] O. Diekmann, H. Heesterbeek, T. Britton, Mathematical Tools for Understanding Infectious Disease Dynamics, Princeton University Press, 2013.
    [4] World Health Organisation, History of Measles Vaccine, Available from: https://www.who.int/news-room/spotlight/history-of-vaccination/history-of-measles-vaccination.
    [5] Ministero della Salute (Italian Ministery of Health), Legge Vaccini, 2024. Available from: https://www.salute.gov.it/portale/vaccinazioni/dettaglioContenutiVaccinazioni.jsp?id = 4824 & area = vaccinazioni & menu = vuoto.
    [6] Eerste Kamer (Dutch Senate), Initiatiefvoorstel-Raemakers en Van Meenen, 2022. Available from: https://www.eerstekamer.nl/wetsvoorstel/35049_initiatiefvoorstel_raemakers.
    [7] K. Wiley, M. Christou-Ergos, C. Degeling, R. McDougall, P. Robinson, K. Attwell, et al., Childhood vaccine refusal and what to do about it: a systematic review of the ethical literature, BMC Med. Ethics, 24 (2023), 96. https://doi.org/10.1186/s12910-023-00978-x doi: 10.1186/s12910-023-00978-x
    [8] World Health Organisation, Ten Threats to Global Health in 2019. Available from: https://www.who.int/news-room/spotlight/ten-threats-to-global-health-in-2019.
    [9] World Health Organisation, Infodemics and Misinformation Negatively Affect People's Health Behaviours, New Who Review Finds, 2022. Available from: https://www.who.int/europe/news/item/01-09-2022-infodemics-and-misinformation-negatively-affect-people-s-health-behaviours–new-who-review-finds.
    [10] W. L. M. Ruijs, J. L. A. Hautvast, G. van IJzendoorn, W. J. C. van Ansem, K. van der Velden, M. E. Hulscher, How orthodox protestant parents decide on the vaccination of their children: a qualitative study, BMC Public Health, 12 (2012), 408. https://doi.org/10.1186/1471-2458-12-408 doi: 10.1186/1471-2458-12-408
    [11] H. Andersson, T. Britton, Stochastic Epidemic Models and Their Statistical Analysis, Springer Science & Business Media, 2000.
    [12] K. Kuulasmaa, The spatial general epidemic and locally dependent random graphs, J. Appl. Probab., 19 (1982), 745–758. https://doi.org/10.2307/3213827 doi: 10.2307/3213827
    [13] J. Hübschen, I. Gouandjika-Vasilache, J. Dina, Measles, Lancet, 399 (2022), 678–690. https://doi.org/10.1016/S0140-6736(21)02004-3 doi: 10.1016/S0140-6736(21)02004-3
    [14] F. Ball, D. Mollison, G. Scalia-Tomba, Epidemics with two levels of mixing, Ann. Appl. Probab., 7 (1997), 46–89. https://doi.org/10.1214/aoap/1034625252 doi: 10.1214/aoap/1034625252
    [15] L. Pellis, F. Ball, P. Trapman, Reproduction numbers for epidemic models with households and other social structures, Ⅰ. definition and calculation of R0, Math. Biosci., 235 (2012), 85–97. https://doi.org/10.1016/j.mbs.2011.10.009 doi: 10.1016/j.mbs.2011.10.009
    [16] F. Ball, L. Pellis, P. Trapman, Reproduction numbers for epidemic models with households and other social structures Ⅱ: Comparisons and implications for vaccination, Math. Biosci., 274 (2013), 108–139. https://doi.org/10.1016/j.mbs.2016.01.006 doi: 10.1016/j.mbs.2016.01.006
    [17] N. G. Becker, K. Dietz, The effect of household distribution on transmission and control of highly infectious diseases, Math. Biosci., 127 (1995), 207–219. https://doi.org/10.1016/0025-5564(94)00055-5 doi: 10.1016/0025-5564(94)00055-5
    [18] T. Britton, K. Y. Leung, P. Trapman, Who is the infector? general multi-type epidemics and real-time susceptibility processes, Adv. Appl. Probab., 51 (2019), 606–631. https://doi.org/10.1017/apr.2019.25 doi: 10.1017/apr.2019.25
    [19] P. Jagers, Branching Processes with Biological Applications, Wiley, 1975.
    [20] F. Ball, O. D. Lyne, Stochastic multi-type SIR epidemics among a population partitioned into households, Adv. Appl. Probab., 33 (2001), 99–123. https://doi.org/10.1017/S000186780001065X doi: 10.1017/S000186780001065X
    [21] G. Grimmett, D. Stirzaker, Probability and Random Processes, 3rd edition, Oxford University Press, 2002.
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1508) PDF downloads(78) Cited by(0)

Figures and Tables

Figures(11)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog