Professional Documents
Culture Documents
Jurnal Vaksin Dan Herd Immunity
Jurnal Vaksin Dan Herd Immunity
a r t i c l e i n f o a b s t r a c t
Article history: We study the influence of population heterogeneity on herd immunity level and on individual’s vaccina-
Received 13 December 2020 tion decision making process. We first formulate the mathematical model in a population with two sub-
Revised 6 May 2021 groups, based on different activity levels or different susceptibilities. The herd immunity threshold is
Accepted 30 May 2021
derived and discussed. It is calculated that the required vaccine coverage level for herd immunity in a
Available online 5 June 2021
heterogeneous mixing population can be varied significantly. The required vaccine coverage level is lower
than the classical herd immunity level, if the vaccine coverage level in the more active group or more sus-
Keywords:
ceptible group is higher than the other subgroup. It is suggested that the classical herd immunity levels
Epidemiology
Population heterogeneity
can be misleading in the process of planning mass vaccination programs. The analysis is further extended
Herd immunity to study the population with more subgroups. We then study the formal vaccination games to simulate
Optimal vaccination strategy the process of vaccination decision making, in either homogeneous or heterogeneous mixing populations.
Game theory It is proved that the Nash equilibrium in the vaccination game is not unique if population heterogeneity is
considered. Moreover, herd immunity is not achieved if individuals are solely driven by self-interests.
Ó 2021 Elsevier Ltd. All rights reserved.
1. Introduction is inferred from early observed exponential growth rate for an epi-
demic, indicated by daily data of incidence cases (Brauer, 2018b;
1.1. Population heterogeneity, herd immunity and vaccine rollout Wallinga and Lipsitch, 2006).
strategies It was argued in another important paper (Fox et al., 1971) that
the simple herd immunity threshold might not be appropriate for
Herd immunity is the indirect protection against an infectious public health. Some possible reasons are imperfect immunity,
disease. With the presence and proximity of immune individuals, heterogeneous mixing population, nonrandom vaccination and
the risk of infection among susceptible individuals is reduced the so-called ‘‘freeloaders” (Fine et al., 2011). Recently, the pan-
(Fine, 1993; Fine et al., 2011). It was initially stated in (Smith, demic COVID-19 has lead scientists to argue whether population
1970) that, given the basic reproduction number R0 , the infection heterogeneity can reduce the required herd immunity threshold
incidence will decline if the proportion of immune individuals to some degree (Britton et al., 2020; Omer et al., 2020; Fontanet
exceeded the herd immunity threshold: and Cauchemez, 2020; Anderson et al., 2020). In Britton et al.
(2020), the authors analyzed the levels of disease-induced immu-
hom 1 nity in several regions and examined the influence of heterogene-
hc ¼1 : ð1Þ
R0 ity on herd immunity. With fixed value of basic reproduction
A thorough introduction of development for the study of herd number R0 ¼ 2:5, several mathematical models with different
immunity can be found in Fine (1993), Fine et al. (2011) and structures were considered and corresponding numerical simula-
Jones and Helmreich (2020). Mathematically speaking, if we tions were performed to find the threshold value such that there
assume the proportion of immune individuals is p, then the is no second wave of epidemic. It was shown that, by considering
resulted control reproduction number is given by Rc ¼ ð1 pÞR0 age groups and different activity levels, the actual herd immunity
(Macdonald, 1957; Anderson and May, 1991). Thus, in order for level may be substantially lower than the classical herd immunity
hom level in Eq. (1). The discrepancy between two levels is greater if the
Rc 6 1, it is necessary to have p P hc in (1). The estimation of
hom
social heterogeneity is more obvious. It was also pointed out in
herd immunity threshold hc requires the quantity of R0 which Britton et al. (2020) that, the social activity heterogeneity plays a
greater role in reduced the required herd immunity than heteroge-
E-mail address: bai@hcm.uni-bonn.de
https://doi.org/10.1016/j.jtbi.2021.110795
0022-5193/Ó 2021 Elsevier Ltd. All rights reserved.
F. Bai Journal of Theoretical Biology 526 (2021) 110795
neous age-group mixing. Mainly inspired by papers (Fine et al., 2. Herd immunity in the heterogeneous mixing population
2011; Britton et al., 2020), we developed several mathematical
models to perform quantitative analysis on significant effect of We first present the simple SEIR epidemic model for homoge-
heterogeneity on herd immunity. Some statements made in neous mixing population,
Britton et al. (2020) are consistent with the analytical results in 8 0
this paper. Furthermore, it is shown that the actual required vac- >
> S ¼ bS NI
> 0
<
cine coverage for herd immunity can be varied, due to different E ¼ bS NI rE
ð2Þ
vaccination strategies. Priority vaccinations for more active group >
>
> I0 ¼ rE cI
: 0
(s) or more susceptible group(s) yield that the herd immunity is R ¼ cI:
much easier to achieve, so is the goal of disease eradication. It is
also worthwhile to emphasize that, vaccine rollout strategies are The population is divided into susceptible members, exposed
far more complicated than the estimates of herd immunity levels. members, infectious members and recovered members. The aver-
Prioritization of vaccinations should take many factors into age contact rate in the population is assumed to be b, the mean
account, such as the activity levels and the mortality rates in differ- exposed period is r1 and the mean infectious period is 1c. The basic
ent age groups and the characteristics of vaccines. The optimal vac- reproduction number for model (2) is R0 ¼ bc. The herd immunity
cination strategies are contextualized and need to be adapted over threshold is estimated as in Eq. (1). In the remainder of the paper,
time, especially when the supplies of vaccines are initially limited we assume that R0 is identical for models with heterogeneity
for some emerging epidemics, such as current COVID-19 (European considered.
Centre for Disease Prevention and Control, 2020).
2.1. The population with 2 subgroups
1.2. Vaccination games The population under study is divided into two subgroups
based on different activity levels. The sizes of two subgroups are
Vaccination is one of the most effective methods for prevention N 1 and N 2 , respectively (see Brauer, 2008; Bai, 2020). We denote
of the spread of infectious diseases. All individuals in the popula-
k ¼ NN21 as the ratio of sizes of two subgroups. Individuals in sub-
tion are supposed to make rational decisions on whether or not
group i make bi contacts in unit time. pij is the probability that
to take vaccination. Ultimately, a certain percentage of individuals
determines to take vaccination, while the rest determine to be sus- one contact made by an individual of group i that is with an indi-
ceptible. We consider the process of decision making is one type of vidual of group j, for i; j ¼ 1; 2. We have
evolution but with a shorter time scale. Thus, evolutionary game p11 þ p12 ¼ p21 þ p22 ¼ 1;
theory is appropriate to be applied to simulate the process. The
first paper to discuss game theory and epidemiology is written and the balance relation equation (Brauer, 2008; Bai, 2020)
by Fine and Clarkson in 1986 (see Fine and Clarkson, 1986). Subse- p21 b2 N2 ¼ p12 b1 N 1 : ð3Þ
quently, several papers discussed vaccination games in infinite
population, e.g., Bauch and Earn (2004), Tassier (2013), Bai It is noted that the recovery rates for both subgroups are equal,
(2016b, 2019), Galvani et al. (2007), Reluga et al. (2006), and which suggests that the only source of heterogeneity is the differ-
Shim et al. (2012a,b). In this paper, we assume that the occurrence ent activity levels. Similar with assumptions made in Brauer (2008,
of decision making process is prior to the outbreak of the disease. 2018a,b) and Bai (2020), we set b1 > b2 , which indicates that sub-
Information about possible outcomes of infection is provided to all group 1 is more active than subgroup 2. For some study cases, sub-
individuals, and all individuals use same information to assess out- group 1 can be referred as the group of superspreaders who are
comes (Bai, 2016a; Bauch and Earn, 2004). The impact of transmis- responsible for the majority of transmissions. Different values of
sion of infection is ignored, the approach of differential games was b1 and b2 can also be interpreted as different susceptibilities. For
discuss in Reluga (2010) and many other papers. example, if subgroup 2 represents the cluster of individuals who
follow some certain public orders, such as social distancing, then
b1 > b2 still holds. It is also applicable for the cases of age hetero-
1.3. Outline geneity or combinations of different factors (see Britton et al., 2020
for full epidemic model).
The paper consists of two major parts. We first discuss the rela- The epidemic model with two subgroups is
tion between herd immunity and population heterogeneity. In this 8
>
> X2
section, the population under study is considered to have 2 sub- >
> S 0
¼ S b
I
pij Njj
>
> i i i
groups with different activity levels. With suitable setups, the herd >
> j¼1
>
>
immunity threshold relation is generated and compared with the < X 2
E0i ¼ Si bi pij Nj rEi ð4Þ
I
classical herd immunity level. We further provide the conditions >
>
>
j
for which the herd immunity threshold level is lower by consider- >
>
j¼1
>
>
ing heterogeneity. The optimal vaccination strategy is then sug- >
> I0i ¼ rEi cIi
>
: 0
gested. To complement the analysis, we extend the theoretical Ri ¼ cIi ;
results to the case of n subgroups, with n > 2. The second part of
the study is to formulate vaccination games. We first briefly review for i ¼ 1; 2. We apply the next generation matrix approach (van den
the results of vaccination game in the homogeneous mixing popu- Driessche and Watmough, 2002) and calculate the basic reproduc-
lation. The emphasis is however on the second vaccination game in tion number
the heterogeneous mixing population. It is proved that, Nash equi- 0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1
libria exist under certain conditions and are not unique. More 1 @p11 b1 þ p22 b2 ðp11 b1 p22 b2 Þ2 þ 4p12 p21 b1 b2
R0 ¼ þ A:
importantly, we conclude that the optimal vaccine coverage level 2 c c
(herd immunity level) can never be reached if all individuals are
solely driven by self-interest. To summarize the paper, we present As discussed in Brauer (2008) and Bai (2020), it is necessary to
some limitations of current research and some possible extensions make some assumptions about the nature of mixing between two
as well. subgroups in order to obtain more useful expression. One reason-
2
F. Bai Journal of Theoretical Biology 526 (2021) 110795
able assumption is the proportionate mixing, which suggests that It is also observed that if the value of R0 is larger, higher vaccine
mixing is random but constrained by the activity level. Thus, we coverage levels in both subgroups are generally required. Our next
assume step is to substitute the formulas (8) in herd immunity threshold
relation (12) to obtain
bj Nj
pij ¼ ; i; j ¼ 1; 2: ð5Þ c2 b c2 kb
b1 N1 þ b2 N2 ð1 h1 Þ ðc1 þc1 2 kÞc þ ð1 h2 Þ ðc1 þc
2
¼1
2 kÞc
ð13Þ
Then we have p11 ¼ p21 ¼: p1 and p12 ¼ p22 ¼: p2 . Furthermore, c2
ð1 h1 Þ ðc1 þc1 2 kÞ þ ð1 h2 Þ ðc1 þc
2
c2 k
¼ R10 ¼ 1 hc
hom
;
2 kÞ
the basic reproduction number R0 is
and replace the denominator ðc1 þ c2 kÞ by ðc21 þ c22 kÞ,
p1 b1 þ p2 b2
R0 ¼ : ð6Þ
c c21 c2 k hom
1 h1 2 2 2 h2 ¼ 1 hc : ð14Þ
For the rest of this paper, we use the notations p1 and p2 to c21 þ c2 k
2
c1 þ c2 k
replace the parameters p11 ; p12 ; p21 and p22 in model (4). As men- It leads to the following equation,
tioned in Brauer (2018a,b) and Bai (2020), in order to guarantee
that the basic reproduction number R0 in (6) for model (4) has hom c21 c2 k
hc ¼ h1 þ 2 2 2 h2 : ð15Þ
the same value as the model (2) with homogeneous mixing, we c21 þ c22 k c1 þ c2 k
have The classical herd immunity level can be written as the linear
b ¼ p1 b1 þ p2 b2 : ð7Þ combination of vaccine coverage levels in both subgroups. We
het
denote hc as the vaccine coverage level in the whole population
The average contact rate b is weighted by b1 and b2 . We further
to reach the herd immunity and have
assume b1 ¼ c1 b and b2 ¼ c2 b. Without loss of generality, we have
c1 > 1 and c2 < 1. Thurs, the formulas of proportionate mixing in het N1 h1 þ N2 h2 h1 þ kh2
hc ¼ ¼ : ð16Þ
Eq. (5) can be written as N1 þ N2 1þk
p1 ¼ b1 Nb11þb
N1
2 N2
c1
¼ c1 þc 2k
; hom het
Both hc and hc are in the form of linear combinations of h1
ð8Þ and h2 . We use the following theorem to summarize the compar-
p2 ¼ b1 Nb12þb
N2
2 N2
¼ c1cþc
2k
2k
:
hom het
ison between hc and hc .
We further obtain
Theorem 2.1. For a population consists of two subgroups with
c1 c2 k
b ¼ c1 b þ c2 b ; ð9Þ different activity levels, the required vaccine coverage level hc in the
het
c1 þ c2 k c1 þ c2 k
whole population to reach herd immunity is lower than the classic
which leads to the relation between heterogeneity parameters, hom
herd immunity level hc :¼ 1 R10 , if and only if the vaccine coverage
c1 þ c2 k ¼ c21 þ c22 k: ð10Þ in the more active subgroup is higher than the other subgroup. The
het hom
Once the population structure is appropriately set, we are able difference between hc and hc increases if the population hetero-
to further analyze the relation between herd immunity level and geneity is more obvious.
population heterogeneity. We assume that the vaccine coverage
levels in subgroup 1 and subgroup 2 are h1 and h2 , respectively.
hom het
If the following inequality Proof. The comparison between hc and hc depends on the rela-
c21
ð1 h1 Þp1 b1 þ ð1 h2 Þp2 b2 tion of coefficients c22 k
and 1k. We have:
61 ð11Þ
c
hom het c21 c22 k
hc hc ¼ h 1 þ h2
satisfies, the herd immunity can be achieved in the population as a c21 þ c22 k c21 þ c22 k
whole. We define the herd immunity threshold relation in a two-
1 k
subgroup population as h1 þ h2
1þk 1þk
ð1 h1 Þp1 b1 þ ð1 h2 Þp2 b2 c21 1 c2 k k
¼ 1: ð12Þ ¼ 2 h1 þ 2 2 2 h2
c c1 þ c2 k 1 þ k
2
c1 þ c2 k 1 þ k
The total infections contributed by subgroup 1 and subgroup 2 c2 1
¼ 2 1 2 ðh1 h2 Þ: ð17Þ
are measured by p1cb1 and p2cb2 , respectively. Three possibilities with c1 þ c2 k 1 þ k
respect to the infection contributions made by two subgroups are:
c21 c2 c2
Because c22 k
= 1k ¼ c12 > 1, we obtain c2 þc1 2 k 1þk
1
> 0. Thus, there are
2 1 2
pi bi
1. If c > 1 for i ¼ 1; 2. Both subgroups have to be partially vacci- three possibilities:
nated, because both subgroups contribute to the total infections
hom het
significantly. 1. If h1 ¼ h2 ; hc ¼ hc ,
2. If p1cb1 > 1 and p2cb2 6 1, subgroup 1 has to be partially vaccinated. 2. if h1 >
hom
h2 ; hc >
het
hc ,
hom het
Otherwise, herd immunity can never be reached. Similar argu- 3. if h1 < h2 ; hc < hc .
ment can be made for the case of p1cb1 6 1 and p2cb2 > 1.
3. If pi bi
6 1 for i ¼ 1; 2. Partially vaccinations in either subgroup 1 When the vaccine coverage level h1 in subgroup 1 is higher than h2 ,
c
the actual vaccine doses needed for herd immunity are fewer. Eq.
or subgroup 2 or both groups possibly reach herd immunity. c1 het
This is the case typically requires the lowest vaccine coverage (17) indicates that if c2
increases, the difference between hc and
hom
level in the population. hc also increases. As desired. h
3
F. Bai Journal of Theoretical Biology 526 (2021) 110795
4
F. Bai Journal of Theoretical Biology 526 (2021) 110795
Fig. 1. The feasible region in blue represents the sufficient vaccine coverage levels to eliminate the epidemic in a population of two subgroups, the basic reproduction number
R0 is 2:5. Two axes represent the vaccine coverage levels in two subgroups. The solid line in red represents the herd immunity threshold relation. Sub-figure ðaÞ indicates that
both subgroups are major infection contributors with p1cb1 > 1 and p2cc2 > 1. Sub-figure ðbÞ indicates that one subgroup is major infection contributor while the other subgroup
is minor infection contributor.
Fig. 2. R0 ¼ 1:5. Two axes represent the vaccine coverage levels in two subgroups. We further assume that a population with size N can be divided
The solid line in red represents the herd immunity threshold relation. into n > 2 subgroups, based on different activity levels or different
Fig. 3. The overall required vaccine coverage to reach herd immunity in the population with two subgroups, which depends on the vaccine coverage levels h1 and h2 in
respective subgroups. The dashed black line represents the overall required vaccine coverage when h1 and h2 satisfy the herd immunity threshold relation in Eq. (12). Its
projection on x-y plane is exactly the herd immunity threshold relation, which is represented by the purple solid line. When h1 ¼ 75% and h2 ¼ 0, the required vaccine
coverage can be as low as 15%. Under this circumstances, herd immunity is much easier to achieve than the classical herd immunity level indicated in Eq. (.1).
5
F. Bai Journal of Theoretical Biology 526 (2021) 110795
p¼ ðc1 k1 ; c2 k2 ; ; cn kn Þ; c2i ki
ck i¼1
The basic reproduction number R0 for model (21) can be calcu- Although we can only guarantee that vk11 > 1, it is observed that
lated by the next generation matrix approach (see Diekmann et al.,
the sequence ðvk i Þ is decreasing. The implication is that impact
2009; van den Driessche and Watmough, 2002). The next genera- i iP1
tion matrix is from the most active subgroup 1 is most significant in the compar-
2 p1 b1 p2 b1 N 1 p3 b1 N 1 pn b1 N 1
3 2p 3 ison between hc
hom het
and hc . The analysis leads to the following
1 b1 p1 b2 p1 b3 p1 bn
c c N2 c N3 c Nn c c c c
6p b N 7 6p b 7 conjecture,
6 1 2 2 p2 b2 p3 b2 N 2
pn b2 N 2 7 6 2 1 p2 b2 p2 b3
p2 bn 7
6 c N1 c c N3 c Nn 7 6 7
FV 1
¼6 7¼6 c c c c
7;
6 . .. .. .. 7 6 . .. .. .. 7 Conjecture 2.1. For a heterogeneous mixing population with n
6 .. . . . 7 6 .. . . . 7 (n > 2) subgroups which are categorized by different activity levels,
4 5 4 5
p1 bn N n p2 bn N n p3 bn N n pn bn pn b1 pn b2 pn b3 pn bn
c N1 c N2 c N3 c c c c c
if the vaccine coverage levels in more active subgroups are higher than
in less active subgroups, the required overall vaccine coverage level in
ð22Þ
the whole population to reach herd immunity is lower than the one
where we have applied the following balance relation equation predicted by assuming homogeneity of population.
pi bj Nj ¼ pj bi Ni : ð23Þ
The next generation matrix in (22) has the zero determinant, Remark. To study a heterogeneous mixing population, the most
and the largest non-zero eigenvalue is the trace of the matrix. Thus, popular method is to divide the population into subgroups based
the basic reproduction number is on age (Mossong et al., 2008). It is assumed that the members in
pb the same subgroup have similar behavioral pattern, such as the
R0 ¼ : ð24Þ
c total number of contacts per day and the preference of contacts
in other subgroups. It requires the statistical analysis of demo-
Because the basic reproduction numbers for models (2) and (21) graphic data and certain types of social survey studies to extract
are identical, we derive the following relation between hetero- the values of parameter vectors k and c. The more accurate formula
geneity parameters, of herd immunity threshold relation (27) can be further obtained.
X
n X
n
cj kj ¼ c2j kj : ð25Þ
j¼1 j¼1 3. Vaccination game in the heterogeneous mixing populations
We assume that the vector of vaccine coverage levels in each
3.1. Review of vaccination game in the homogeneous mixing
subgroup is h ¼ ðh1 ; h2 ; ; hn Þ, then if the following inequality
population
holds,
X
n
In a homogeneous mixing population, the vaccination game (a
ð1 hi Þpi bi two-player game) is studied by setting the elements of the follow-
i¼1
6 1; ð26Þ ing payoff matrix in Table 1, based on the fact that, prior to the
c
occurrence of vaccination, there are exact two strategies for indi-
the epidemic can be eliminated. Thus, the herd immunity threshold viduals to choose (Bai, 2016a, 2019; Bauch and Earn, 2004;
relation can be expressed as Tassier, 2013; Shim et al., 2012b). Fitness of different strategies
6
F. Bai Journal of Theoretical Biology 526 (2021) 110795
Table 1 x_ v ¼ xv ðf v f Þ;
ð32Þ
x_ s ¼ xs ðf s f Þ;
Payoff matrix of the vaccination game in a homogeneous mixing population. Each
individual has to choose either strategy V (‘‘to be vaccinated”) or strategy S (‘‘not to be
vaccinated”). The payoffs in the payoff matrix are assumed to be non-positive. It is
noticed that all individuals’ decisions on whether to be vaccinated is completely where f denotes the average fitness of both strategies,
determined by payoffs.
f ¼ xv f þ xs f :
Vaccination Game in a homogeneous mixing population v s
cination game is determined by the relative cost C r . If C r > p0 ; S is the relative vaccination cost.
the dominant strategy of the game. No matter how other individ-
uals play this game, choosing strategy S always has a higher payoff.
This eventually leads to the vaccine coverage level p ¼ 0 in the
population. This scenario is possibly due to the factors such as
the vaccine is too expensive or has serious side effects, or the dis-
ease is considered less serious. On the other hand, if C r < p0 , the
vaccination game is a coexistence game. There exists a fixed pH
which represents the proportion of the population who adopts
the vaccination. The estimation of pH and its comparison with
the herd immunity threshold are our primary interests.
We apply evolutionary game theory to formulate the following
replicator Eqs. (32) to simulate the decision making process.
Assume the proportion of vaccinated individuals is xv and the pro-
portion of unvaccinated individuals is xs ; fxv ; xs g ¼ fp; 1 pg (see
Bai, 2016b, 2019; Bauch and Earn, 2004; Tassier, 2013). The fitness
of two strategies can be expressed as:
f v ¼ C v xv C v xs ¼ C v ;
ð31Þ
f s ¼ pp C i xs ; Fig. 4. The red dashed line represents the piecewise function ð1 pÞpp about
vaccine coverage level p. When p P 0:6, herd immunity is achieved. As a result, the
where f v is the fitness of individuals with strategy V and f s is the
function value is 0 for p P 0:6. For 0 6 p 6 0:6, the function is decreasing
fitness of individuals with strategy S, respectively. It is noted that monotonically. If we set the relative vaccination cost C r ¼ 0:3 (the green solid
fitness is equal to payoff in our settings of vaccination game. Thus, horizontal line), there exists a unique solution pH 43%. It is also observed that if
the replicator equations for vaccination game (1) can be expressed C r is smaller, or even very close to 0; pH will increase towards the herd immunity
as threshold 60%, but never reach it.
7
F. Bai Journal of Theoretical Biology 526 (2021) 110795
Table 2
Payoff matrix of the vaccination game in the heterogeneous mixing population with two subgroups. Subscripts 1; 2 represent two subgroups. It is noted that all individuals can
not choose which subgroup they belong to, but can only choose between the adoptions of strategy V or strategy S.
We formulate the replicator equations to simulate the process log 11p1 ¼ ð1pÞp
c
1 b1
p1 þ ð1qÞp
c
2 b1
p2
of decision making regarding whether or not to be vaccinated. ð39Þ
The proportions of individuals who adopt strategy V in subgroup
log 11p2 ¼ ð1pÞp
c
1 b2
p1 þ ð1qÞp
c
2 b2
p2 :
1 and subgroup 2 are xv 1 and xv 2 , respectively. The proportions of It is not necessary to explicitly solve for p1 and p2 , we only need
individuals who adopt strategy S in subgroup 1 and subgroup 2 to guarantee that both p1 and p2 decrease strictly with the vaccine
are xs1 and xs2 , respectively. Thus, the replicator equations are coverage levels p and q, which is
xv_ 1 ¼ xv 1 ðf v 1 f1 Þ;
@ p1 @ p1 @ p2 @ p2
x_s1 ¼ xs1 ðf s1 f1 Þ; < 0; < 0; < 0; < 0: ð40Þ
ð34Þ @p @q @p @q
xv_ 2 ¼ xv 2 ðf f Þ;
v2 2
x_s2 ¼ xs2 ðf s2 f2 Þ; This will be confirmed and explained in detail in the following
subsection of numerical simulations.
with fi denotes the average fitness/payoffs for individuals in sub-
group i. The payoffs can be calculated accordingly,
3.4. Numerical simulations
f v 1 ¼ C v ;
f s1 ¼ p1 ðp; qÞC i p1 ð1 pÞ p2 ðp; qÞC i p2 ð1 qÞ; We perform the numerical simulations to show the existence of
ð35Þ
f v 2 ¼ C v ; ðpH ; qH Þ for vaccination game (2), in Eq. (36). The parameters for
f s2 ¼ p1 ðp; qÞC i p1 ð1 pÞ p2 ðp; qÞC i p2 ð1 qÞ: the simulations are 1c ¼ 5; b ¼ 0:5; c1 ¼ 2; c2 ¼ 12 and k ¼ 8. It is cal-
p1 and p2 were defined previously in Eq. (8), to represent the culated that p1 ¼ 13 and p2 ¼ 23. Furthermore, we have
probabilities that the contact is made with an individual in sub-
groups 1 and 2, respectively. Solving replicator Eqs. (34), the final p1 b1 5 p2 b1 10 p1 b2 5 p2 b2 5
¼ ; ¼ ; ¼ ; ¼ ;
proportions pH and qH are implicitly expressed in the following c 3 c 3 c 12 c 6
equation,
and the basic reproduction number is R0 ¼ 2:5.
p1 ðpH ; qH Þp1 ð1 pH Þ þ p2 ðpH ; qH Þp2 ð1 qH Þ ¼ C r : ð36Þ p1 ðp; qÞp1 ð1 pÞ þ p2 ðp; qÞp2 ð1 qÞ is a function of p and q. We vary
We are interested in two following questions: whether ðpH ; qH Þ the values of pair ðp; qÞ 2 ½0; 12 to draw the surface and observe
is unique? Whether ðpH ; qH Þ is in the feasible region which results how this function depends on different values of p and q. It is shown
in herd immunity? To answer these questions, we ought to analyze in Fig. 5 that the function value monotonically decreases as p or q
the properties of infection probabilities p1 and p2 first. increasing, which confirms the inequalities in Eq. (40). The function
value becomes 0 once p and q satisfy the herd immunity threshold
Remark. As expressed in Eq. (35), the payoffs/fitness for unvac- relation. The maximum of function value occurs when p ¼ q ¼ 0,
cinated or vaccinated individuals in both subgroups are identical. and the maximum is approximately 0:7. Thus, if the relative cost
Theoretically, the vaccination game (2) can be reduced to a C r is greater than 0:7, all individuals are tend to be unvaccinated
classical 2-player game. in order to gain higher personal payoffs. If and only if C r < 0:7, there
exists the final fractions ðpH ; qH Þ. However, such fractions form a
curve on the 2-dimensional plane and the uniqueness of Nash equi-
3.3. Incorporation of the epidemic model librium does not hold anymore.
If we pick different values of C r 2 ð0; 0:7Þ, different curves for
To precisely calculate infection probabilities p1 and p2 , we predicted vaccine coverage levels can obtained, which are pre-
derive final size relations based on the epidemic model (4) with sented in Fig. 6. However, the herd immunity level curve can never
two subgroups. The corresponding final size relations are be reached. This indicates that, in a heterogeneous mixing popula-
h i h i tion, herd immunity can not be achieved if individuals are solely
ð0Þ
log SS11ð1Þ ¼ p1cb1 S1 ð0ÞS1 ð1Þ
N1
þ p2cb1 S2 ð0ÞS2 ð1Þ
N2 driven by self-interest.
h i h i ð37Þ We are also interested in how individual’s behavior changes if
ð0Þ S1 ð0ÞS1 ð1Þ S2 ð0ÞS2 ð1Þ
log SS22ð1Þ ¼ p1cb2 N1
þ p2cb2 N2
: the disease is more contagious, for instance, when there exists
the new mutated form of coronavirus in current COVID-19 pan-
If we explicitly involve the vaccine coverage levels p and q, the
demic (Bai and Brauer, 2021; Hou et al., 2020). We assume
above relations lead to
h i h i b ¼ 0:85 to account for higher infectivity and have R0 ¼ 4:25.
ð0Þ
log SS11ð1Þ ¼ ð1pÞp
c
1 b1 S1 ð0ÞS1 ð1Þ
S1 ð0Þ
þ ð1qÞp
c
2 b1 S2 ð0ÞS2 ð1Þ
S2 ð0Þ
The predicted vaccine coverage levels in both subgroups are pre-
h i h i ð38Þ sented in Fig. 7. Interestingly, the upper limit of tolerance regards
ð0Þ
log SS22ð1Þ ¼ ð1pÞp
c
1 b2 S1 ð0ÞS1 ð1Þ
S1 ð0Þ
þ ð1qÞp
c
2 b2 S2 ð0ÞS2 ð1Þ
S2 ð0Þ
: relative cost of vaccine increases above 0:8. For same value of C r ,
vaccination intents in both subgroups are increasing when people
Because p1 ¼ S1 ð0ÞS 1 ð1Þ
S1 ð0Þ
and p2 ¼ S2 ð0ÞS 2 ð1Þ
S2 ð0Þ
, we further obtain are facing more contagious diseases.
8
F. Bai Journal of Theoretical Biology 526 (2021) 110795
Acknowledgements
Fig. 7. R0 ¼ 4:25 with a larger value of b ¼ 0:85. The vaccine coverage levels in
both subgroups are associated with different values of relative vaccination cost C r . The work is funded by the Post-doctoral fellowship offered by
These expected coverage levels are also sub-optimal. Hausdorff Center for Mathematics and the University of Bonn.
9
F. Bai Journal of Theoretical Biology 526 (2021) 110795
The author thanks two reviewers for very constructive comments Fox, J.P., Elveback, L., Scott, W., Gatewood, L., Ackerman, E., 1971. Herd immunity:
basic concept and relevance to public health immunization practice. Am. J.
to improve this work.
Epidemiol. 94 (3), 179–189.
Galvani, A.P., May, R.M., 2005. Dimensions of superspreading. Nature 438 (7066),
293–295.
References Galvani, A.P., Reluga, T.C., Chapman, G.B., 2007. Long-standing influenza vaccination
policy is in accord with individual self-interest but not with the utilitarian
Anderson, R.M., May, R.M., 1991. Infectious diseases of humans: dynamics and optimum. Proc. Nat. Acad. Sci. 104 (13), 5692–5697.
control. Oxford University Press, Oxford. Hou, Y.J., Chiba, S., Halfmann, P., Ehre, C., Kuroda, M., Dinnon, K.H., Leist, S.R.,
Anderson, R.M., Vegvari, C., Truscott, J., Collyer, B.S., 2020. Challenges in creating Schäfer, A., Nakajima, N., Takahashi, K., Lee, R.E., Mascenik, T.M., Graham, R.,
herd immunity to SARS-CoV-2 infection by mass vaccination. Lancet 396 Edwards, C.E., Tse, L.V., Okuda, K., Markmann, A.J., Bartelt, L., de Silva, A.,
(10263), 1614–1616. Margolis, D.M., Boucher, R.C., Randell, S.H., Suzuki, T., Gralinski, L.E., Kawaoka,
Bai, F., 2016a. Uniqueness of Nash equilibrium in vaccination games. J. Biol. Dyn. 10 Y., and Baric, R.S. 2020. SARS-CoV-2 d614g variant exhibits efficient replication
(1), 395–415. ex vivo and transmission in vivo. Science, eabe8499..
Bai, F. 2016b. Vaccination models in infectious diseases. PhD thesis, University of Jones, D., Helmreich, S., 2020. A history of herd immunity. Lancet 396 (10254), 810–
British Columbia.. 811.
Bai, F., 2019. Modeling vaccination decision making process in a finite population. Lloyd-Smith, J.O., Schreiber, S.J., Kopp, P.E., Getz, W.M., 2005. Superspreading and
Math. Biosci. 311, 82–90. the effect of individual variation on disease emergence. Nature 438 (7066),
Bai, F., 2020. Evaluating different epidemiological models with the identical basic 355–359.
reproduction number R0 . J. Biol. Dyn. 14 (1), 849–870. Macdonald, G., 1957. The epidemiology and control of malaria. Oxford University
Bai, F., Brauer, F., 2021. The effect of face mask use on COVID-19 models. Press, London.
Epidemiologia 2 (1), 75–83. Monod, M., Blenkinsop, A., Xi, X., Hebert, D., Bershan, S., Tietze, S., Baguelin, M.,
Bauch, C.T. and Earn, D.J.D. 2004. Vaccination and the theory of games. Proc. Natl. Bradley, V.C., Chen, Y., Coupland, H., Filippi, S., Ish-Horowicz, J., McManus, M.,
Acad. Sci., 101(36), 13391–13394.. Mellan, T., Gandy, A., Hutchinson, M., Unwin, H.J.T., van Elsland, S.L., Vollmer, M.
Brauer, F., 2005. The kermack–McKendrick epidemic model revisited. Math. Biosci. A.C., Weber, S., Zhu, H., Bezancon, A., Ferguson, N.M., Mishra, S., Flaxman, S.,
198 (2), 119–131. Bhatt, S., and Ratmann, O. 2021. Age groups that sustain resurging COVID-19
Brauer, F., 2008. Epidemic models with heterogeneous mixing and treatment. Bull. epidemics in the united states. Science, page eabe8372..
Math. Biol. 70 (7), 1869–1885. Mossong, J., Hens, N., Jit, M., Beutels, P., Auranen, K., Mikolajczyk, R., Massari, M.,
Brauer, F. 2018a. Early estimates of epidemic final sizes. J. Biol. Dyn., 1–8.. Salmaso, S., Tomba, G.S., Wallinga, J., Heijne, J., Sadkowska-Todys, M., Rosinska,
Brauer, F., 2018b. The final size of a serious epidemic. Bull. Math. Biol. 81 (3), 869– M., Edmunds, W.J., 2008. Social contacts and mixing patterns relevant to the
877. spread of infectious diseases. PLoS Med. 5, (3) e74.
Brauer, F., Castillo-Chavez, C., 2012. Mathematical Models in Population Biology and Omer, S.B., Yildirim, I., Forman, H.P., 2020. Herd immunity and implications for
Epidemiology. Springer, New York. SARS-CoV-2 control. JAMA 324 (20), 2095.
Britton, T., Ball, F., Trapman, P., 2020. A mathematical model reveals the influence of Reluga, T.C., 2010. Game theory of social distancing in response to an epidemic.
population heterogeneity on herd immunity to SARS-CoV-2. Science 369 PLoS Comput. Biol. 6, (5) e1000793.
(6505), 846–849. Reluga, T.C., Bauch, C.T., Galvani, A.P., 2006. Evolving public perceptions and
Davies, N.G., Klepac, P., Liu, Y., Prem, K., Jit, M., Eggo, R.M., 2020. Age-dependent stability in vaccine uptake. Math. Biosci. 204 (2), 185–198.
effects in the transmission and control of COVID-19 epidemics. Nat. Med. 26 (8), Shim, E., Chapman, G.B., Townsend, J.P., Galvani, A.P., 2012a. The influence of
1205–1211. altruism on influenza vaccination decisions. J. R. Soc. Interface 9 (74), 2234–
Diekmann, O., Heesterbeek, J.A.P., Roberts, M.G., 2009. The construction of next- 2243.
generation matrices for compartmental epidemic models. J. R. Soc. Interface 7 Shim, E., Grefenstette, J.J., Albert, S.M., Cakouros, B.E., Burke, D.S., 2012b. A game
(47), 873–885. dynamic model for vaccine skeptics and vaccine believers: Measles as an
European Centre for Disease Prevention and Control. 2020. COVID-19 vaccination example. J. Theor. Biol. 295, 194–203.
and prioritisation strategies in the EU/EEA. 22 December 2020. ECDC: Smith, C.E.G., 1970. Prospects for the control of infectious disease. Proc. R. Soc. Med.
Stockholm.. 63(11P2):1181–1190..
European Centre for Disease Prevention and Control. 2021. Overview of the Tassier, T., 2013. The economics of epidemiology. Springer, Heidelberg.
implementation of COVID-19 vaccination strategies and vaccine deployment van den Driessche, P., Watmough, James, 2002. Reproduction numbers and sub-
plans in the EU/EEA. 29 March 2021. ECDC: Stockholm.. threshold endemic equilibria for compartmental models of disease
Fine, P.E.M., 1993. Herd immunity: History, theory, practice. Epidemiol. Rev. 15 (2), transmission. Math. Biosci. 180 (1–2), 29–48.
265–302. Viner, R.M., Mytton, O.T., Bonell, C., Melendez-Torres, G.J., Ward, J., Hudson, L.,
Fine, P.E.M., Clarkson, J.A., 1986. Individual versus public priorities in the Waddington, C., Thomas, J., Russell, S., van der Klis, F., Koirala, A., Ladhani, S.,
determination of optimal vaccination policies. Am. J. Epidemiol. 124 (6), Panovska-Griffiths, J., Davies, N.G., Booy, R., Eggo, R.M., 2021. Susceptibility to
1012–1020. SARS-CoV-2 infection among children and adolescents compared with adults.
Fine, P., Eames, K., Heymann, D.L., 2011. herd immunity: A rough guide. Clin. Infect. JAMA Pediatrics 175 (2), 143.
Dis. 52 (7), 911–916. Wallinga, J., Lipsitch, M., 2006. How generation intervals shape the relationship
Fontanet, A., Cauchemez, S., 2020. COVID-19 herd immunity: where are we? Nat. between growth rates and reproductive numbers. Proc. R. Soc. B: Biol. Sci. 274
Rev. Immunol. 20 (10), 583–584. (1609), 599–604.
10