Spatial Clustering in Vaccination Hesitancy: The Role of Social Influence and Social Selection

"Understanding...difference[s] in cluster configuration and distribution is critical to prediction of outbreak potential..., and suggests the benefit of incorporating the social context of a community in public health programs."

Vaccine hesitancy behaviour tends to cluster spatially, creating pockets of unprotected sub-populations that can be hotspots for outbreaks of disease. This paper hypothesises that landscape-level spatial clustering in hesitancy may be produced by (i) social selection: geographically proximate communities may independently adopt similar behaviours because they share characteristics that promote the behaviour; or (ii) social influence: vaccination behaviour may diffuse between socially proximate areas through learning of social norms and practices. The researchers validate their models of both processes with empirical data on vaccine hesitancy, social determinants, and social connectivity in the United States (US). They then propose and evaluate two intervention strategies to slow the rise in hesitancy and diminish the increased risk of outbreak emergence due to geographic clustering.

The paper is based on consideration of a society as being made up of spatially and socially interconnected communities. The researchers represent this society as a landscape-level network in which communities (i.e., US counties) are represented by network nodes, and spatial proximity or social interactions between communities are represented by network edges. They define spatial proximity based on shared land borders between US counties. The researchers integrate three empirical data sources: (i) an infection case study using relevant empirical vaccine hesitancy data based on school exemptions for the measles vaccine; (ii) from the US Census, each US county's socioeconomic attributes relevant to the social selection model; and (iii) social connectivity between communities relevant to the model of social influence, based on social media activity data.

The analysis finds that both social selection and social influence are independently capable of generating hesitant behaviour clustering, but they are configured differently. In network structures ranging from spatial to aspatial, social selection tends to be a more robust process, where significant changes in the network structure are needed in order to impact the spatial clustering. On the other hand, spatial clustering in hesitancy driven by social influence depends on the network structure. Thus, small changes in the distribution of edges have an impact on the spatial clustering.

Together, these processes result in unique spatial configurations of hesitant clusters. Theoretical findings suggest that when a society trusts hesitancy propaganda, social selection plays an important role, and many smaller clusters of hesitancy appear. On the other hand, when a society tends to be more skeptical about propaganda, social influence overcomes, and a few larger clusters appear, despite the same overall frequency of vulnerable communities. Conversely, social influence can take advantage of a society already affected by social selection to spread hesitancy easily, generating observed patterns of spatial cluster distribution.

Informed by these findings, the researchers propose policies that are designed to not only reduce the prevalence of hesitancy but also reduce spatial clustering in hesitancy, with the goal of reducing pockets of vulnerable communities rather than simply eliminating isolated counties of high hesitancy. Namely:

• Strategy to reduce clustering caused by social selection: This approach focuses on prioritising communities that are vulnerable due to their own socio-economic traits but also are surrounded by a socio-cultural environment with a high tendency towards hesitancy. Once the communities are identified, traditional public health measures of reducing vaccine hesitancy, such as healthcare provider training and community health outreach programmes, can be implemented. As reported here, all targets are found to be effective in diminishing clustering relative to no intervention. Targeting social selection by targeting traits and conformity results in a larger reduction in spatial clustering for lower values of social selection. Within a context of high segregation, the strategy recommended here is to target high-hesitancy counties, which implies having available hesitant data.
• Strategy to reduce clustering caused by social influence: This approach involves exploiting the impact of network structure on spatial clustering, with the goal of reducing clustering by altering the structure of the social connectivity between communities (i.e., rerouting social connections to make them less spatial). That is, it proposes to manipulate the spatial connectivity that underlies the social influence process for communities with observed high hesitancy. While such a strategy is likely to be impractical in traditional social networks, the social connectivity between communities due to social media usage may be amenable to manipulation by expanding upon the geo-targeted and connection-targeted digital marketing techniques that are already common. The researches find that as the social network is made less spatial, there is a larger reduction in spatial clustering, and that this impact increases with larger values of social influence.

The researchers note that, while they chose the case study of measles to ground their analysis and validation, spatial clustering in vaccination patterns has been found for a range of vaccines (e.g., pertussis, Hepatitis B, polio, human papillomavirus, and COVID-19). The mechanism of social selection has been demonstrated through associations of vaccination with socio-economic factors and media environment in a variety of vaccination settings, and evidence exists for social influence being relevant to a number of vaccines. Thus, they expect their methodology and findings to be generalisable and adaptable to different vaccination systems. For a fast-evolving and complex vaccination setting like COVID-19 vaccination, for example, one could include medical distrust or political mistrust as a trait in the social selection model, and one might model the spread of misinformation-fueled vaccine hesitancy with the social influence model parameterised based on a society's exposure to misinformation.

The researchers conclude that understanding how social behaviour impacts spatial clustering can be a step toward designing mitigation strategies to reduce clustering of vulnerable populations. Generative models of social behaviour such as the one proposed here can also inform dynamical behaviour-disease models, which have been limited to assuming vaccine hesitancy in a non-spatial context and only through the lens of social influence. Because the threat that vaccine hesitancy poses to local elimination of vaccine-preventable childhood diseases is growing, the researchers advocate for continued progress on mathematical modeling of this phenomenon from both a social and spatial perspective.