Simulating Vaccine Decisions in an Agent-Based Model of Disease Spread

Abstract

Agent-based models (ABMs) are used to simulate the spread of disease. Compared to traditional mathematical disease models, ABMs can capture movement, heterogeneous characteristics, and behaviors of individuals, all of which play a role in disease dynamics. However, ABMs often oversimplify or ignore the health behavior component. Models typically use spatially aggregated data to determine a probability of adopting a behavior and apply it uniformly to the entire agent population. This ignores the underlying mechanisms that drive health decisions. Therefore, this study seeks to develop a more data-driven approach to modeling health behaviors in ABMs of disease spread. First, we generate a realistic population of agents for Virginia with characteristics including age, gender, race, income, education, and social influence. Using individual level survey data that asks questions about vaccine decisions in 2021, we train a logistic regression to predict vaccine uptake based on the characteristics above. Next, in the agent-based simulation of disease spread, the agents apply the trained model to themselves using their own characteristics as an input to determine their vaccine decision. We compare our method for simulating behavior with a uniform probability of vaccine uptake. This study advances disease simulations by allowing for data driven health behaviors.