Modeling the Coupled Dynamics of COVID-19 Transmission and Protective Behaviors

Project Summary/Abstract A growing number of COVID-19 transmission models have been developed to help forecast the on-going epi- demic and compare outcomes of different non-pharmaceutical interventions (NPIs) in terms of cases, deaths, and medical supply needs. Most of these models do not include adaptive behavioral effects describing how risk perceptions and fatigue influence engagement with social distancing and transmission reduction. Decisions on mask-wearing, levels of social contact, and vaccination will define whether the epidemic is controlled or enters annual circulation. We propose the development of population-based (PBM) and agent-based (ABM) transmis- sion models to study the interplay between individual behavior and transmission dynamics, while considering the many uncertainties which still surround the virus, such as seasonal effects and the loss of immunity. Addition- ally, our models will be used to study how COVID-19 and seasonal influenza and respective behaviors interact, exacerbate outcomes, and potentially overwhelm the health care system. These models will build upon our prior research. Since Fall 2016 we have conducted regular longitudinal surveys investigating attitudes towards, risk perceptions of, and propensity to vaccinate for seasonal influenza. The ABM models constructed from these data account for adaption and memory of past experiences, peer effects, and population heterogeneity. Using machine learning methods, we have augmented a synthetic network representative of a small US city with this behavioral data. We have continued to conduct modified versions of these surveys to track how these beliefs translate to COVID-19. In parallel, we have developed a compartmental population-based model of COVID-19, which models transmission and the effects of NPI intensity and timing on both health and economic outcomes. We propose to extend our current compartmental PBM and build a new individual-level ABM, informed by longitudinal surveys. We will conduct a four-year longitudinal panel survey to construct an empirical behavioral model for decisions to socially distance, engage in transmission reduction measures (such as mask-wearing), and vaccinate. This information will be combined with our existing synthetic network data-set to enable us to build an individual level ABM of the spread of COVID-19 in a representative US city, integrated with our influenza ABM. This model will capture both how individual behaviors impact macro-level disease transmission and how influenza and COVID-19 could interact. Insights and data from our individual-level model will be used to inform and parameterize adaptive behavior within our compartment-level model, allowing for policy comparisons across a range of US states. In addition, we will consider which policies are robust to key behavioral and technological uncertainties, such as the extent of behavior change in response to perceived risk and the timing and effectiveness of vaccines. Finally, we will develop web-based interactive tools that allow for the exploration and comparison of different policies in a variety of potential futures.