key: cord-0004237-k6ad4qgu
authors: Kabir, K. M. Ariful; Tanimoto, Jun
title: Modelling and analysing the coexistence of dual dilemmas in the proactive vaccination game and retroactive treatment game in epidemic viral dynamics
date: 2019-12-04
journal: Proc Math Phys Eng Sci
DOI: 10.1098/rspa.2019.0484
sha: 3ebd75fd49d3a3c223e75574a2846b492b8f26e6
doc_id: 4237
cord_uid: k6ad4qgu

The dynamics of a spreadable disease are largely governed by four factors: proactive vaccination, retroactive treatment, individual decisions, and the prescribing behaviour of physicians. Under the imposed vaccination policy and antiviral treatment in society, complex factors (costs and expected effects of the vaccines and treatments, and fear of being infected) trigger an emulous situation in which individuals avoid infection by the pre-emptive or ex post provision. Aside from the established voluntary vaccination game, we propose a treatment game model associated with the resistance evolution of antiviral/antibiotic overuse. Moreover, the imperfectness of vaccinations has inevitably led to anti-vaccine behaviour, necessitating a proactive treatment policy. However, under the excessively heavy implementation of treatments such as antiviral medicine, resistant strains emerge. The model explicitly exhibits a dual social dilemma situation, in which the treatment behaviour changes on a local time scale, and the vaccination uptake later evolves on a global time scale. The impact of resistance evolution and the coexistence of dual dilemmas are investigated by the control reproduction number and the social efficiency deficit, respectively. Our investigation might elucidate the substantial impacts of both vaccination and treatment in the framework of epidemic dynamics, and hence suggest the appropriate use of antiviral treatment.

The dynamics of a spreadable disease are largely governed by four factors: proactive vaccination, retroactive treatment, individual decisions, and the prescribing behaviour of physicians. Under the imposed vaccination policy and antiviral treatment in society, complex factors (costs and expected effects of the vaccines and treatments, and fear of being infected) trigger an emulous situation in which individuals avoid infection by the pre-emptive or ex post provision. Aside from the established voluntary vaccination game, we propose a treatment game model associated with the resistance evolution of antiviral/antibiotic overuse. Moreover, the imperfectness of vaccinations has inevitably led to anti-vaccine behaviour, necessitating a proactive treatment policy. However, under the excessively heavy implementation of treatments such as antiviral medicine, resistant strains emerge. The model explicitly exhibits a dual social dilemma situation, in which the treatment behaviour changes on a local time scale, and the vaccination uptake later evolves on a global time scale. The impact of resistance evolution and the coexistence of dual dilemmas are investigated by the control reproduction number and the social efficiency deficit, respectively. 

The appearance of epidemiological dynamics in the mechanism of pre-emptive voluntary vaccination has been studied in various contexts [1] , such as perfect and imperfect vaccination [2, 3] , dynamical behaviour of vaccination [4] , vaccination with information spreading [5] , metapopulation migration modelling [6] and heterogeneous networks [7] . Furthermore, Chen & Fu [8] studied an effective antiviral treatment with prescribing behaviour and resistance evolution. Remarkably, influenza-like illnesses, Oseltamivir (Tamiflu) [9] is a widely used ex post treatment originally administered against influenza A and B viruses. However, over the years, the societal benefits of antiviral treatment have lessened with overuse, leading to treatment resistance. These trends are evidenced by the interplay between prescription behaviour and resistance evolution. Here, the theoretical studies of vaccination and treatment strategies have considered different effectiveness, associated costs, payoff structures and time scales.

Previously, compartment models with the mean-field approximation, such as the SI [10] , SIS [11] , SIR [12] , SEIR [13] and SEIQR [14] models, are exhibited by dividing the population into several distinct groups. In these designations, S, I, R, E and Q represent the proportions of susceptible, infected, recovered, exposed and quarantined individuals, respectively. Recently, Kabir et al. extended the simple SIR model by introducing an awareness effect on epidemic spreading and implemented a two-layer SIR-UA model on well-mixed [15] and heterogeneous networks [16] . Additionally, treatment is an important compartmented state that reduces the disease after infection. Treatments such as antibiotics, quarantine and isolation have been theoretically investigated by many researchers [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] . The consequences of vaccination and treatment on an epidemic model were investigated in an influenza model with age structure by Qiu & Feng [27, 28] and Feng et al. [29] , an SIVS model with vaccination age by Li et al. [30] , an SIR epidemic model with optimal control theory by Zaman et al. [31] and a pandemic influenza model by Towers et al. [32] . All these works presumed that vaccination, quarantine or treatment would reduce epidemic infection in a simple dynamical situation on local time scales and with no game aspect. By contrast, the present study aims to establish a theoretical epidemic model encompassing both vaccination and treatment as an evolutionary game approach.

The human decision-making process is affected by the cost and risk of the vaccine, selfopinion, networks and neighbours' decisions; therefore, how vaccine acquiescence is influenced by various factors must be investigated [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] . According to prior studies, a game approach to epidemiological vaccination can fairly predict the infection risk in both vaccinated and nonvaccinated individuals [49] [50] [51] [52] . Such voluntary vaccination game approaches have been studied theoretically and in multi-agent simulations (MAS). To elucidate the mechanism of infectiousdisease control, these approaches incorporate a two-layer time scale: a local time scale (epidemic season) of epidemic diffusion and a global time scale on which the strategy updates at the end of the season (at local equilibrium), followed by repeated seasons. Kuga & Tanimoto [53] developed a theoretical model of imperfect vaccination on local and global time scales and validated it by MAS. However, Kabir & Tanimoto [54] claimed that an individual's decision to take a vaccination after social learning (dynamical behaviour) also occurs on local time scales, so this strategy should be updated instantly. Accordingly, it seems that the voluntary vaccination game approach can be implemented into the local time scale while maintaining the framework on the global time scale (strategy update at the end of each season). In the same context, antiviral treatment depends on the local time scale, antiviral resistance and prescribing behaviour.

To shed light on this complex phenomenon, we newly propose the dual-dilemma game structure that considers the roles of both the proactive vaccination and retroactive treatment games. This approach admits different strategy update rules and different time scales of the two provisions. In most of the previous studies [5, 6, [39] [40] [41] [49] [50] [51] [52] [53] , the proactive vaccination permits an individual to accept or decline a vaccination at the end of every epidemic season. This repeated choice is made on global time scales. On the other hand, the retroactive treatment prescribes the behaviour and the antiviral resistance of a certain individual only when s/he is actually infected at a certain time in an epidemic season, which occurs on local time scales. An excessive antiviral treatment may also trigger another viral resistant strain; this behaviour is an expected social problem concerning seasonal influenza in Japan [55] . In this case, the so-called vaccination dilemma modelled by the vaccination game is joined by another dilemma, whereby an individual seeking to use public goods (i.e. accepting the antiviral) induces another viral strain with devastating consequences for others. Using our novel idea backed by the theoretical game approach, we quantify the impact of the pre-emptive vaccination game (before the disease spreading) and the treatment game (after the infection), which includes the prescribing behaviour as an ex post provision. Both games are influenced by vaccine effectiveness, treatment efficacy, treatment cost and vaccination cost. Such a double social dilemma situation, perhaps quite ubiquitous in the real world, has not been considered in related previous studies. The vaccination game in the double-dilemma scenario occurs on a larger time scale (over repeated seasons) than the treatment game, which occurs on a day-by-day basis. Therefore, a coevolutionary process can be plausibly modelled in the present study. We employ a pre-emptive control measure that prevents the breakout of infection at an early stage based on an individual's decision. Meanwhile, the ex post treatment can be regarded as a fail-safe provision implemented after infection. Owing to the imperfections of vaccines and the unwillingness to take vaccines as a pre-emptive provision, people probably consider a retroactive treatment as the 'ultimate weapon' against disease dispersion. However, overuse of antiviral treatment and prescribing behaviour can trigger the emergence of resistant strains, encouraging more ex post provision activity by individuals. To handle these two provisions working on different time scales, our model implants the second social dilemma incurred by the antiviral treatment rather than the so-called vaccination dilemma acquired by the proactive provision. To our knowledge, no previous theoretical analysis has considered two provisions in the same context of the evolutionary game process. We also develop another new concept called the social efficiency deficit (SED). Such a framework can explicitly elucidate the social dual-dilemma on both global (vaccination) and local (treatment) time scales.

In our model, the retroactive antiviral treatment targets the individuals harbouring a sensitive or resistant strain that is controlled by the treatment (antiviral/antibiotic) potency (efficacy). The resistant strain incurs a high medical cost, mortality and the risk of antiviral/antibiotic use, which increases when the demand for antiviral use is driven by the individual's self-interest and overprescribing. Here, we emphasize the social learning behaviour for prescription of antiviral treatment under the evolutionary dynamics of resistance that can uphold the optimal use of the treatment. To explore this evolutionary process of vaccination and treatment, we impose three strategy update rules: individual-based risk assessment (IB-RA), society-based risk assessment (SB-RA) and direct commitment (DC). These rules govern the individual's connection with society. Moreover, we derive the control reproduction numbers of the sensitive and resistant strains, and hence analyse the disease-free equilibria situations in an epidemic.

The remainder of this paper is organized as follows. The 'Methods and model' section introduces the new model of epidemic vaccination with the antiviral treatment model and demonstrates it schematically. The 'Results and discussion' section validates the proposed model in numerical simulations. Finally, the 'Conclusion' section summarizes and further discusses our findings.

To model the social dual-dilemma as a two-stage game, the pre-emptive vaccination and ex post treatment are developed in the framework of SIR epidemic dynamics in a well-mixed population (figure 1). In stage 1, the individuals make their vaccination decisions (yes or no) that will control their infection risk during the pandemic season. In stage 2, the infected people with either the sensitive strain or resistant strain decide their treatment provision (treated or untreated) on the local time scale. The antiviral (antibiotic) treatment case incorporates a feedback loop between the prescription behaviour and resistance evolution. To model the disease diffusion in a single season, the initially susceptible people are compartmentalized into vaccinated and nonvaccinated groups. The individuals in the susceptible state can be infected with either sensitive or resistant strains, then seek treatment at an overall treatment rate (treatment probability) denoted as ωP , where P is the prescribing rate and ω determines the probability at which infected people accept treatment from the prescribing individuals. Consequently, the individuals can recover in two ways: natural recovery with no antibiotic/antiviral treatment or recovery after treatment. The epidemiological dynamics are described by the following system of ordinary differential equationsṠ Here, the fractions of vaccinated plus non-vaccinated individuals infected with the sensitive and resistant strains are denoted by I u sen and I u res , respectively. T denotes the fraction of infected individuals receiving treatment and R represents the fraction of individuals who have recovered from infection by a sensitive or resistant strain. In addition, β s and β r (γ s and γ r ) are the disease transmission rates (infection recovery rates) for the sensitive and resistant strains, respectively and γ T is the recovery rate of infected individuals receiving treatment. Finally, μ s and μ r are the mutation rates of the sensitive and resistant strains, respectively. Without mutation, the coexistence of sensitive and resistant strains is forbidden by the competitive exclusion principle. The portion of vaccinated individuals is separated into perfectly immune and nonimmune individuals, distinguished by the vaccine's effectiveness e(0 ≤ e ≤ 1). The treatment efficacy ε controls the treatment efficiency of the sensitive and resistant strains. When ε = 0, the treatment is far less beneficial against the resistant strain than against the sensitive strain. On the other hand, when ε = 1, the higher number of people in the resistant state is now taking the highest benefit of treatment and is ineffective against the sensitive strain.

The basic reproduction number (ratio) R 0 is the estimated number of infected individuals instigated by a susceptible individual (R 0 = β/γ ). In particular, if R 0 < 1, the disease will eventually die out, whereas if R 0 > 1, the disease will spread through the population. In this case, we presume separate control reproduction numbers R s and R r for the sensitive and resistant strains, respectively. To evaluate the control reproduction numbers, we initially set

Let us express the model dynamics as

we can write

Similarly, we have

As mentioned above, the control reproduction number must reflect the stability of the diseasefree equilibrium state. According to equations (2.7) and (2.8), the control reproduction numbers of both sensitive (R s ) and resistant (R r ) strains act like a decreasing function of P and x; both ranging from 0 to 1 displayed in figure In panels (a-ii) and (a-iii), we can find the critical treatment probability P c at which the treatment probabilities of the sensitive and resistant strains are equal. When 0 < P < P c , the sensitive strain is more prevalent than the resistant strain (i.e. R s > R r ). However, when P > P c , the resistant strain will outperform against the sensitive strain. At the social optimum, the critical treatment probability P c specifies the maximum treatment control under which resistant strains will not emerge. Assuming R s = R r in equations (2.7) and (2.8), P c is calculated as

. 

We incorporate two-game aspects (treatment and vaccination) in a single epidemiological game model. This model reproduces the coevolutions of accepting a vaccination at the beginning of every season and receiving treatment after becoming infected in a season. In the treatment game, the individual decision to receive or decline treatment against the infectious sensitive and resistant strains occurs on the local time scale. In the vaccination game, the individuals can consent to alter their strategy (accept or decline vaccination) based on the progress of the last pandemic season, which occurs on the global time scale.

Based on a feedback loop between the resistance evolution and prescription norm, the game approach establishes a social learning dynamical process that somehow controls the optimum use of the antiviral treatment. To quantify the evolutionary decision dynamics of treatment versus non-treatment (prescribing versus non-prescribing), we specify the relative treatment cost C T visà-vis the infection cost C i = 1. We also introduce the benefit B T of treating the sensitive strain (the resistant strain is excluded, because it is much more difficult to treat than the sensitive strain, so B T is always positive). This idea is formulated as a two-strategy game in table 1. The fractions of individuals infected with the sensitive and resistant strains are, respectively, given by

The expected payoffs of the treated and untreated individuals are, respectively, given by

and

To model the two-strategy game, we presume the DC strategy update rule presented in Iwamura & Tanimoto [51] . This rule is designed by comparing the expected payoffs of the treated π T and untreated π U individuals. In the present study, the strategy updates (in both the treatment and vaccination games (see later)) apply the mean-field approximation. The modified Fermi function of DC is given by

where π i and π j are the mean payoffs of the focal portion of individuals and the opponent strategy (fraction), respectively. Here, we consider pairwise comparison between two groups; which depends on the payoff difference π j − π i . Because, the pairwise fermi has been wellaccepted strategy-updating procedure that stochastically comparable to the real-world human decision-making process. The probabilities of the population transiting from untreated to treated and from treated to untreated are, respectively, calculated as

and

Consequently, the treatment game is expressed by the following DC dynamics: Considering the defined payoff structure and the portion of individuals presented in table 2, the social average payoff π , expected value of vaccinators π C and expected value of nonvaccinators π D are, respectively, given by

To formulate the evolutionary process, we consider two types of strategy adaptation procedures [53] ; IB-RA and SB-RA. In the case of IB-RA, an individual can update strategy by observing a neighbour's strategy. The update is governed by the transition probability Prob(s i ← s j ) taken from the pairwise Fermi function [1] . Alternatively, in an SB-RA, an individual relies on the mean payoff of all opposite neighbours [2] . We apply the mean-field approximation to formulate the adaptation dynamics in both the IB-RA and SB-RA rules. Here, we replace the first row by the actual transition probabilities in the second row of table 3.

To establish the dynamical system at the end of each epidemic season, we formulate IB-RA and SB-RA as mathematical models that modify the fraction of vaccinators x. The evolutionary dynamics of the IB-RA and SB-RA are, respectively, given by Individual-based risk assessment (IB-RA)

Society-based risk assessment (SB-RA)

We have now established all mathematical frameworks in both the vaccination and treatment cases. The above set of equations is numerically solved by the explicit finite difference method. The calculation results affected by the two-stage process (the SITR/V dynamical model and treatment update) are together obtained in a single season (local update) at equilibrium, and the vaccination strategy is adopted at the end of every season (global update). The initial values were set as V(

The results are presented in two-dimensional (2D) phase diagrams. The strategy update rules of the vaccination game (IB-RA or SB-RA) and the DC were applied on the global and local time scales, respectively (where the DC depicts the prescribing behaviour of the treatment policy). The coalescing impact of proactive vaccination and the retroactive treatment policy based on human behaviour was formulated by the conventional mean-field approximation. The simultaneous changes of two coevolutionary decision-making processes were globally demonstrated in two cases. We first explored the phase diagram of the final epidemic size (FES), vaccination coverage (VC), fraction of treated people (FTR) and the average social payoff (ASP) while varying two parameters: the vaccination effectiveness e and the treatment efficacy ε, maintaining sensible fixed values of the other parameters. In the second case, we introduce SED that explicitly reveals the As shown in figure 3 , the FES increased (higher infection region) with increasing vaccination cost C r and treatment cost C T . Reducing the cost of both vaccination (C r = 0.1) and treatment (C T = 0.1) improved the FES (lowered the infection region) ( figure 3(a-i) , both panels). Furthermore, reducing the costs (C r , C T < 0.5) more effectively benefitted the FES (in terms of the critical borderline between infection breakout and diminution) in IB-RA than in SB-RA (cf. panel sets A and B); the reverse tendency was found at higher costs (C r , C T ) as analogously reported by Fukuda et al. [2] . When the costs are relatively low, vaccination is more encouraged in SB-RA than in IB-RA, which hampers the reduction in the infected number of individuals in the early stage of each season. Thus, based on the human decision-making of whether to accept or decline both vaccination and treatment, the changing propensity of the FES can be significantly enhanced by the vaccination effectiveness, treatment efficacy and their corresponding costs.

As indicated in figure 4 , lowering the vaccination cost and increasing the reliability (effectiveness) of the vaccine enticed the individuals to accept more vaccines. This tendency was more marked in the SB-RA than in the IB-RA, and enhanced the vaccination acceptance when the post-infection treatment cost was high (C T = 0.9) on the local time scale. However, the portion of individuals making treatment provision (FTR) diminished at higher treatment costs (C T = 0.9) in both schemes (panels (*-iii) in figure 5 ). A small treatment cost attracts individuals to the treatment provision, whereas a higher cost hampers the treatment-seeking behaviour (lowers the FTR). Therefore, either lowering the vaccination cost or improving the vaccination effectiveness 

To explore the dual-dilemma structure on an epidemiological model, we considered the joint impact of vaccination and treatment games in the same context. A typical scenario is demonstrated in figure 7 . Our idea was motivated by the endorsement of social dilemma situations in the strategies of evolutionary game theory, in which the players are all individuals in a well-mixed population. In a coevolutionary adaptation process, it is important to know whether the social dual-dilemma exists under certain combinations of the model parameters, such as the vaccination effectiveness, treatment efficiency and their associated costs. Unlike simple 2 by 2 games in which the so-called dilemma strength (DS) can be explicitly defined [1] , a real social dilemma typically observed in the vaccination game [2, [5] [6] [7] [36] [37] [38] [39] [40] [41] [42] [43] [44] , traffic flow [46] [47] [48] [49] [50] [51] [52] and others may have a time-variable game structure. In the vaccination and traffic games, this structure is mainly influenced by the disease-spreading and traffic flow dynamics, respectively. The timevariable game structure is too complex to represent by the payoff matrix in a 2 × 2 game or the time-constant payoff structure function in the 2-strategy and n-player game. Thus, the DS is difficult to determine in advance even when the mathematical model is well defined. Our new SED concept (defined above and mathematically formulated blow) was inspired by the seminal idea of traffic flow analysis (e.g. [56] [57] [58] [59] [60] [61] [62] ). The DS indicates the existence or absence of a social dilemma (behaving as a prognostic index), whereas the SED provides an 'ex post' or diagnostic index. Here, let us define SED as the gap between the result of the evolutionary trail (which can be evaluated by the Nash equilibrium (NE)) and the optimum solution (ideal result from a social-welfare standpoint [63] ). The payoff at the NE can always be observed by taking an evolutionary game approach, whereas the optimal social payoff is observable in a model of any complexity. Therefore, one can evaluate the SED in any context, and hence predict the occurrence of a social dilemma (if the SED is positive, the gap exists; if it is zero, the evolutionary trail matches the optimum). Thus, the SED indicates that the payoff can be improved from that at the NE. Mathematically, the SED is given by SED = (social optimal payoff) − (payoff at Nash equilibrium) ( Again, let us reiterate that SED = 0 implies no social dilemma, while any social dilemma causes a positive SED. According to the abovementioned conceptual definition, SED in the current model (which deals with both vaccination and treatment games) is given by

Meanwhile, the ASP is the quantity of payoff. The superscript 'Opt' and subscript 'social' together indicate the social optimal. The C i was taken as the standardizing denominator as previously defined as 1. The SEDs in the vaccination and treatment game of the present model are, respectively, defined as follows: social reflect the fact that the maximum ASP is obtained for varying x ranging from 0 to 1 (for fixed p k ) and varying P from 0 to 1(for fixed x l ), respectively. Figure 7 presents the stepwise procedure of finding SED and quantifying the dual dilemma in the proposed method.

Step 1. Construct the ASP phase diagram based on the evolutionary game approach (figure 7(i)).

To this end, implement both the vaccination and treatment games and obtain the appropriate ASP at the NE, along with a certain vaccine effectiveness (e) and treatment efficacy (ε). The ASP associated with the fraction of vaccinators and probability of treatment (x l , p k ) at the NE can also be observed for a precise (e i , ε j ).

Step 2. For this fixed (e i , ε j ), evaluate ASP To address whether a dual dilemma exists in the vaccination and treatment games, we plot the SED diagrams in the vaccination effectiveness (e) versus treatment efficacy (ε) planes at different costs. Panels A and B of figure 8 are plotted under the strategy update rules IB-RA and SB-RA, respectively, and plots (a-*) and (b-*) present the SEDs in the vaccination game SED v and the treatment game SED T , respectively. Moreover, the combined vaccination and treatment costs (C r , C T ) were varied as (0.1, 0.1), (0.1, 0.9), (0.9, 0.1) (0.1, 0.1), and (0.9, 0.9) in plots (*-i), (*-ii), (*-iii) and (*-iv), respectively. As demonstrated in the above SED formulation, the dilemma situation (non-white areas in the plots) appeared in all cases, but interestingly depended on e and ε. Nodilemma regions, in which either the vaccination or treatment game became trivial, were also observed. Now, comparing panels (a-i) and (a-ii) for C r = 0.1 with panels (a-iii) and (a-iv) for C r = 0.9, one finds that the non-dilemma region (whiteout region) expanded with increasing vaccination cost. At the smaller vaccination cost (C r = 0.1), the region of larger SED appeared at a relatively low vaccination effectiveness (around e = 0.4; dotted red box in panel (a-i)). The huge gap between NE and the social optimal results from the lower NE due to a lower efficiency which makes fewer people commit vaccination when compared with a situation allowing a relatively higher efficiency which makes much more people commit. By contrast, a relatively high vaccination efficiency entices people to vaccinate, thus increasing the NE. Meanwhile, increasing the vaccination cost (C r = 0.9) shifted the region of reasonably high SED to the maximally high vaccination effectiveness (dotted blue box in panel (a-iv)). This occurred because despite the high vaccination effectiveness, low vaccination efficiency hampers the commitment to vaccination. Reliable vaccination provides a high commitment incentive, but high cost encourages free-riding on the herd immunity of the devoted others. In summary, we have numerically demonstrated that the SED precisely and easily detects the social dilemma in our dual-dilemma coevolutionary model.

This paper developed an SITR/V epidemic model that combines the effects of proactive vaccination and retroactive treatment on the control and prevention of infectious viral diseases. The model building and its investigation were presented in this work. The most important contribution is that our new model gives a brand-new framework in which both pre-emptive vaccination and treatment as an ex post provision having different evolutionary time scales, which dovetails the ideological dynamics with the dynamics of the human decision-making process. This concept has been never studied. Also, our model gives a clear procedure to quantify the social dilemmas, respectively, entailed by 'vaccination game' and 'treatment game'. The most novel aspect of our model is the simultaneous implementation of two social dilemma games: the antiviral treatment game and the vaccination game, which none of the previous work has tried to implement. The vaccination game implements on the global time scale under two strategy adaptation rules: IB-RA and SB-RA, assuming an infinite and well-mixed population. Meanwhile, the treatment game describes the behaviour of antiviral administration with resistant-strain emergence. The treatment game is updated on the local time scale by presuming the DC rule, and precisely integrating a feedback loop between the sensitive and resistant strains. The outcome of antiviral and vaccination use depends on the effectiveness of the vaccine, the efficiency of treatment and their corresponding costs. Increasing the effectiveness of the vaccine and lowering its administration cost reduced the final epidemic size (increased vaccination coverage). Lowering the treatment cost and enhancing the treatment efficacy exerted a similar effect. Thus, by applying retroactive antiviral use with pre-emptive vaccination, we can deeply understand and investigate individual decisions regarding vaccination and implement proper strategies that lessen the diffusion of infection or recommend appropriate and careful administration of both antivirals and vaccination. We also introduced the social optimum point P c that distinguishes the conditions under which treatment resistance emerges under antibiotic overuse and its associated factors. Besides evolving the voluntary vaccination game, our model introduces a new game aspect with two provisions: vaccination as a proactive measure and treatment as a retroactive measure. Presuming seasonal influenza-like diseases, the (pre-emptive) vaccination works over repeated seasons on global time scales, whereas the ex post treatment works seasonally on local time scales and depends on the antiviral cost, prescription behaviour and resistant-strain emergence. In the present model, we successfully established a 'double-layer' game structure of pre-emptive vaccination and ex post treatment. Unlike the vaccination game model, which only considers whether the vaccine is accepted or declined, and whether an ex post provision is taken in a single season, the treatment game includes an aspect that depends on the antiviral resistance evolution and prescribing behaviour. To explicitly prove the dual-dilemma situation in the 'double-layer' game, we proposed the SED indicator, which quantifies whether the dynamics develop a social dilemma structure. This indicator is measured by the gap between the NE and the social optimal state. The dilemma strength [64] [65] [66] , which explains the dilemma structure in simple two-player and two-strategy (2 × 2) games, is too simplistic for realistic dilemma games with substantially complex and time-dependent structures, such as vaccination games and traffic flow analysis. However, SED can quantify whether a game intrinsically has a social dilemma or not. We applied the SED to the present social dual-dilemma game, in which both vaccination and treatment dilemmas are inevitable.

Data accessibility. This article has no additional data. The supplementary material provides the basic code of our study.

Fundamental of evolutionary game theory and its applications

Risk assessment for infectious disease and its impact on voluntary vaccination behavior in social network

Imperfect vaccine and hysteresis

Dynamical behavior of a stochastic SVIR epidemic model with vaccination

Effect of information spreading to suppress the disease contagion on the epidemic vaccination game

Evolutionary vaccination game approach in metapopulation migration model with information spreading on different graphs

Impact of imperfect vaccination and defense against contagion on vaccination behavior in complex networks

Social learning of prescribing behavior can promote population optimum of antibiotic use

Tamiflu report comes under fire

A contribution to the mathematical theory of epidemics

Dynamics of stochastically perturbed SIS epidemic model with vaccination

Dynamic behaviors of a modified SIR model in epidemic diseases using nonlinear incidence and recovery rates

Global stability for the SEIR model in epidemiology

A delayed SEIQR epidemic model with pulse vaccination and the quarantine measure

Analysis of SIR epidemic model with information spreading of awareness

Analysis of epidemic outbreaks in two-layer networks with different structures for information spreading and disease diffusion

A brief history of the antibiotic era: lessons learned and challenges for the future

Emergence of drug resistance: implications for antiviral control of pandemic influenza

Emergence of drug-resistant influenza virus: population dynamical considerations

A delay differential model for pandemic influenza with antiviral treatment

A population-dynamic model for evaluating the potential spread of drug-resistant influenza virus infections during community-based use of antivirals

Prophylaxis or treatment? Optimal use of an antiviral stockpile during an influenza pandemic

Impact of emerging antiviral drug resistance on influenza containment and spread: influence of subclinical infection and strategic use of a stockpile containing one or two drugs

Potential impact of antiviral use during influenza pandemic

Antiviral resistance and the control of pandemic influenza

Transmission dynamics of an influenza model with vaccination and antiviral treatment

Transmission dynamics of an influenza model with age of infection and antiviral treatment

Modeling the effects of vaccination and treatment on pandemic influenza

Stability and bifurcation of an SIVS epidemic model with treatment and age of vaccination

Optimal strategy of vaccination and treatment in an SIR epidemic model

Antiviral treatment for pandemic influenza: assessing potential repercussions using a seasonally forced SIR model

Swarm intelligence inspired cooperation promotion and symmetry breaking in interdependent networked game

/journal/rspa Proc. R. Soc

Information cascades in complex networks

Incorporating latent constraints to enhance inference of network structure

Statistical physics of vaccination

Vaccination and the theory of games

Can influenza epidemics be prevented by voluntary vaccination?

Three-strategy and four-strategy model of vaccination game introducing an intermediate protecting measure

To vaccinate or not to vaccinate: a comprehensive study of vaccination-subsidizing policies with multi-agent simulations and mean-field modeling

Influence of breaking the symmetry between disease transmission and information propagation networks on stepwise decisions concerning vaccination

Investigation of epidemic spreading process on multiplex networks by incorporating fatal properties

Suppression of epidemic spreading process on multiplex networks via active immunization

Discerning influential spreaders in complex networks by accounting the spreading heterogeneity of the nodes

Coevolution of vaccination opinions and awareness affecting the spread of epidemics

Virus propagation and patch distribution in multiplex networks: modeling, analysis, and optimal allocation

Community detection in temporal networks via a spreading process

Coupled disease-behavior dynamics on complex networks: a review

Effects of stubborn decision-makers on vaccination and disease propagation in social network

Effect of noise-perturbing intermediate defense measures in voluntary vaccination games

Realistic decision-making processes in a vaccination game

Effect of intermediate defense measures in voluntary vaccination games

Which is more effective for suppressing an infectious disease: imperfect vaccination or defense against contagion?

Dynamical behaviors for vaccination can suppress infectious disease-a game theoretical approach

Oseltamivir-resistant influenza A viruses circulating in Japan

Dilemma game structure observed in traffic flow at a 2-to-1 lane junction

Dilemma game structure hidden in traffic flow at a bottleneck due to a 2 into 1 lane junction

/journal/rspa Proc. R. Soc

Dangerous drivers foster social dilemma structures hidden behind a traffic flow with lane changes

Social dilemma structures hidden behind a traffic flow with lane changes

Social dilemma structure hidden behind traffic flow with route selection

Complex traffic flow that allows lane-changing and hampering intrinsically contains social-dilemma structures

Improvement of traffic flux with introduction of a new lane-change protocol supported by intelligent traffic system

Localized prosocial preferences, public goods, and common-pool resources

Relationship between dilemma occurrence and the existence of a weakly dominant strategy in a two-player symmetric game

Universal scaling for the dilemma strength in evolutionary games

Scaling the phase-planes of social dilemma strengths shows gameclass changes in the five rules governing the evolution of cooperation

Acknowledgements. We would like to express our gratitude to the funders.