Developing Intervention Campaigns to Combat Anti-vaccine Opinion Propagation and their Impact on Disease Spread: A Computational Approach

Abstract

Information regarding vaccines from sources such as health services, media, and social networks can significantly shape vaccination decisions. In particular, the dissemination of negative information can contribute to vaccine hesitancy, thereby exacerbating infectious disease outbreaks. This study investigates strategies to mitigate anti-vaccine social contagion through effective counter-campaigns that disseminate positive vaccine information and encourage vaccine uptake, aiming to reduce the size of epidemics. In a coupled agent-based model that consists of opinion and disease diffusion processes, we develop different intervention strategies to mitigate the spread of vaccine misinformation subject to budget constraints. We first explore and compare different heuristics to design positive campaigns based on the network structure and local presence of negative vaccine attitudes. We examine two campaigning regimes: a static regime with a fixed set of targets, and a dynamic regime in which targets can be updated over time. We demonstrate that strategic targeting and engagement with the dynamics of anti-vaccine influence diffusion in the network can effectively mitigate the spread of anti-vaccine sentiment, thereby reducing the epidemic size. However, the effectiveness of the campaigns differs across different targeting strategies and is impacted by a range of factors. We find that the primary advantage of static campaigns lies in their capacity to act as an obstacle, preventing the large-scale clustering of emerging anti-vaccine communities, thereby resulting in smaller and unconnected anti-vaccine groups. On the other hand, dynamic campaigns reach a broader segment of the population and adapt to the evolution of anti-vaccine diffusion, not only protecting susceptible agents from negative influence but also fostering positive influence propagation within clusters of negative opinions. We then frame this as a budget-constrained optimization problem, developing optimization-based strategies. Specifically, we explore two optimization objectives: minimizing the number of anti-vaccine opinion adopters and maximizing the number of pro-vaccine opinion adopters. We demonstrate that minimizing anti-vaccine opinion adoption is a more efficient strategy, as it promotes vaccine uptake within negative regions and thus blocks the diffusion of anti-vaccine influence. In contrast, the maximization approach operates near positive areas in the network, neither helping contain negative influence nor spreading positive influence throughout the network, resulting in poor control over the spread of anti-vaccine opinions and the corresponding epidemic size. Furthermore, we find that control performance deteriorates when interventions are delayed. This is due to the early formation of anti-vaccine communities that continue to grow, making them increasingly difficult to contain. We also introduce cluster-based optimization interventions aimed at curbing the expansion of existing communities. These campaigns identify and target critical regions in the network, effectively limiting the growth of the communities and thus reducing epidemic size. Next, we systematically study the influence of network topology on the effectiveness of control strategies. We find that the efficiency of campaigns depends on both the network structure and the timing of the intervention. For early intervention, we demonstrate that minimizing anti-vaccine opinion adoption is an efficient strategy, particularly when incorporating detailed knowledge of individual’s neighbourhoods, which helps improve the campaign by targeting agents who can positively influence their hesitant peers. For late intervention, shielding critical regions is an effective strategy in small-worlds and lattices. However, it is less effective in scale-free and random networks due to the distinct patterns of cluster formation in these networks which tend to produce large clusters from the early stages of their emergence. In addition, although controlling negative diffusion becomes more challenging the longer the intervention is delayed, we find that it can be managed more efficiently and with fewer resources in small-world and regular lattice networks than in scale-free and random networks. Last, we study interventions in multi-layer networks considering the different venues of social interactions. We find that, virtual-based interventions generally outperform physical-based interventions across most campaigns. An exception is the cluster-based campaigns, where controlling physical anti-vaccine communities performs better than controlling virtual anti-vaccine communities. In addition, missing social contact data from the virtual layer can degrade control performance, particularly when the layer has a scale-free structure. Furthermore, campaign behavior in multi-layer networks is consistent with that in single-layer networks, suggesting that the choice of intervention strategy should align with the structure of the dominant information layer. In conclusion, we make several practical recommendations on targeted interventions to counter the diffusion of vaccine misinformation and its impact on public health. We provide insights into effective implementation for policymakers and public health administrators, while also highlighting the limitations and operational challenges that may affect the effectiveness of these interventions.