A Review and Agenda for Integrated Disease Models Including Social and Behavioural Factors

"Integrated modelling that better incorporates social and behavioural dynamics will advance predictive accuracy and thereby more effectively guide policy and response measures..."

In light of large-scale epidemics of new pathogens and global health threats such as Ebola and COVID-19, the field of disease modelling has a considerable and growing influence on questions related to public health policy. Such outbreaks have highlighted the importance (and complexities) of developing models that integrate social and behavioural dynamics, such as community attitudes to disease response. These outbreaks have also sparked growth in capacity, coordination, and prioritisation of social science research and of risk communication and community engagement (RCCE) practice. This paper provides a review of modelling methodologies and describes the challenges and opportunities for integrating them with social science research and RCCE practice. It also sets out an agenda for advancing transdisciplinary collaboration for integrated disease modelling and for more robust policy and practice for reducing disease transmission.

Existing modelling approaches that attempt to integrate human behaviour outlined in the paper include:

• Economic epidemiology - Behavioural influence in observed disease dynamics, and vice versa, is captured through the notion of prevalence-elastic behaviour, which is quantifies how population-level infection rates and personal infection status influence the adoption of behavioural recommendations, such as vaccination. Economic epidemiologists model how rational individuals would behave given some level of disease prevalence. However: "Affective (and not necessarily conscious) factors, including a range of emotions such as fear and anger, coupled with pervasive errors in appraising risks..., all compounded by network conformity effects, can combine to produce behaviours that are far from canonically rational, with far-from-optimal results."
• Behavioural change as network dynamics - Modelling disease spread in contact networks provides a representation of the heterogeneity and complexity of human behaviour in the form of the network, or graph, in which an epidemic can occur. Interventions in this context can be any piece of information (for example, a vaccination campaign) that can reduce the risks of transmission. For top-down approaches, the objective of the model is to inform targeting strategies, which can be optimised, for example, to find highly connected individuals. Bottom-up approaches tend to be more descriptive, often modelled as extensions of classic disease models, assuming random contact within populations and letting disease and information spread and interact. However: "[D]ata informing the complexities of human physical and communication networks (and their overlap) are scarce and difficult to collect. Importantly, such efforts also tend to assume a static structure with fixed behaviour, knowledge and contact patterns."
• Coupled contagion models with an extension to Agent_Zero - Fear is modeled as a contagion that influences behavioural decisions, which in turn impact disease transmission. The cognitive underpinnings of fear, as well as deliberative biases, heuristics, and network effects are represented in Epstein's Agent_Zero, "who" is endowed with affective, deliberative, and social modules. The framework could be used to represent collective behaviours such as fear-driven refusal of safe vaccines. However: "Data-based parameterization will...depend on experimental findings - absent field studies that can capture relevant information during outbreaks."
• Individual-level, or agent-based models (ABMs) - Disease transmission is modelled across agents that are representative of the unique, underlying sociodemographic, clinical, and other characteristics that make up a population being affected by an outbreak. "However, in these frameworks, individual behaviour is sometimes considered in isolation of contextual factors, for example, community-level dynamics or policy-level factors that can influence decision-making."

As noted here, "[e]fforts have accelerated in recent years to garner a consensus on the essential role of social science research and RCCE practice in understanding and containing disease transmission." In recent years, social science research has generated data on perceptions and behaviours related to outbreaks of Ebola, Zika, polio, and COVID-19. RCCE approaches used in these situations are designed to support communities to understand new disease threats, interpret and adapt top-down directives, identify community priorities and actions, amplify and replicate community-derived protective actions, and provide feedback on response measures.

The paper looks specifically at the role of community engagement (CE) in supporting communities to identify and address their most pressing issues. Meanwhile, risk communication (RC) is a cross-cutting tool that has evolved to take into consideration subjective and objective risks determined by social, cultural, economic, and psychological factors. RC approaches use a range of messaging platforms and methodological approaches, including communication for behavioural change, mass media, social media, health education, and health promotion. The experience of the COVID-19 epidemic has highlighted the pernicious nature of misinformation as a risk in itself and has spurred the emerging field of infodemiology.

In light of this analysis, the paper elaborates on the key challenges and opportunities for advancing a collaborative agenda for integrated modelling. Such an approach would entail multidisciplinary collaboration among epidemiologists, clinicians, social scientists, mathematical modellers, RCCE practioners, and members of communities directly affected by disease. Through an iterative design process (see figure above), the team decides on research questions and modelling frameworks, making sure to let questions of the greatest relevance to affected communities and first responders drive decisions about modelling approaches. The idea is that explicitly incorporating behaviour would enable assessment of the epidemiological impact of interventions such as RCCE. All outcomes warrant accountability in terms of sharing results with stakeholders and affected communities.

In short, for modellers, a key goal of such a collaboration is to use complex information to develop integrated models that capture the factors that underlie observed disease trends (vs. models that use traditional sources of epidemiological data alone). For social scientists and RCCE practitioners, the hope is that more explicit, data-informed modelling of social and behavioural processes - and the extrinsic and intrinsic drivers influencing them - will generate quantitative evidence on the roles and importance of social learning and community action, as well as the impacts non-pharmaceutical interventions such as RCCE.

Recommendations to advance better integration of social science research and RCCE practice in epidemiological modelling include:

1. Build a community of practice for networking, partnership-building, and collaboration.
2. Collaboratively develop protocols for the integration of contextual, sociocultural, and behavioural factors into modelling, with a focus on the theories that underpin decision-making processes, as well as existing historical data on social norms, cultural practices, existing "baseline" political economy, and other factors.
3. Build or expand on open-source model and code repositories that can focus on integrated disease modelling.
4. Develop prepositioned sets of RCCE data frameworks, including thematic priorities, measures, and indicators.
5. Preposition data sharing agreements between researchers, governments, and other response actors to facilitate rapid data movement from field-based data collection centres to centres of modelling and research, particularly at the country level.
6. Establish and sustain transdisciplinary teams to routinely work together during an outbreak response.
7. Facilitate training and peer exchange to develop mutual understanding and shared skill sets among epidemiologists, modellers, social scientists, RCCE practitioners and decision-makers.
8. Review research undertaken during recent epidemics from the perspectives of epidemiology and social science.
9. Ensure that restitution to local communities is mainstreamed into modelling practices.
10. Jointly and clearly communicate with policymakers to co-create recommendations for immediate, mid-term, and long-term actions.
11. Advocate for increased investment in integrated disease modelling.

Full list of authors, with affiliations: Jamie Bedson, independent consultant; Laura A. Skrip, Bill & Melinda Gates Foundation, and University of Liberia; Danielle Pedi, Bill & Melinda Gates Foundation; Sharon Abramowitz, independent consultant; Simone Carter, United Nations Children's Fund (UNICEF); Mohamed F. Jalloh, Karolinska Institutet; Sebastian Funk, London School of Hygiene & Tropical Medicine (LSHTM): Nina Gobat, University of Oxford, and Global Outbreak Alert and Response Network (GOARN); Tamara Giles-Vernick, Institut Pasteur, and Sonar-Global Network; Gerardo Chowell, Georgia State University; João Rangel de Almeida, Wellcome Trust; Rania Elessawi, UNICEF; Samuel V. Scarpino, Northeastern University, and Santa Fe Institute, and University of Vermont; Ross A. Hammond, Santa Fe Institute, and Washington University in St Louis, and Brookings Institution; Sylvie Briand, World Health Organization (WHO); Joshua M. Epstein, Santa Fe Institute, and New York University; Laurent Hébert-Dufresne, University of Vermont; and Benjamin M. Althouse Bill & Melinda Gates Foundation, and University of Washington, and New Mexico State University. Jamie Bedson and Laura A. Skrip contributed equally.