Modeling Policies to Mitigate an Epidemic and the Role of Non-Compliance

Abstract

The COVID-19 pandemic has significantly affected the way of life for almost every person around the world, and it has created a lot of uncertainty about how a population can optimally navigate an epidemic without a high human or economic cost. To evaluate the effectiveness of different policies on mitigating an epidemic, an agent-based model was created to model the interactions between agents that form a local economy, and a SEIRS epidemiological model was added to simulate the spread of a virus through a population. In addition, a set of virus mitigation strategies was defined, and a Policy Gradient neural network was trained to apply these strategies in a way that minimized the effect that the epidemic had on both the population and the economy. The learned policies found that it is possible to stop the spread of the virus through the use of contact tracing, quarantining, and locking down specific locations. This optimal policy created strong economic growth by stopping the virus quickly and allowing the economy to return to normal as soon as possible. Policies that did nothing to slow the spread of the virus resulted in lower economic growth because the spread of the virus made the economy unproductive for a longer period. Additionally, if a small proportion of the population was allowed to not comply with the policy, the learned policies were only able to achieve a slight reduction in the number of deaths without causing total economic collapse. Even a small amount of non-compliance in a population can make any response policy ineffective, thus addressing non-compliance is a key step to improving the outcomes of an epidemic.