Modeling layered non-pharmaceutical interventions against SARS-CoV-2 in the United States with Corvid

Dennis L Chao; Assaf P Oron; Devabhaktuni Srikrishna; Michael Famulare

We introduced and demonstrated the use of a new agent-based model, Corvid, to simulate a variety of NPIs against SARS-CoV-2. Simulations of single interventions help us understand their contribution to reducing transmission, but we do not consider them realistic, since interventions can affect the lives of non-targeted people. For example, if schools are closed, parents may need to stay home to care for their children, creating a de facto work-from-home policy. Multiple simultaneous NPIs could suppress transmission for long periods of time, but stochasticity ("bad luck") can allow SARS-CoV-2 to persist for months under these conditions even if R effective is slightly below 1. Therefore, when NPIs were relaxed, the epidemic could rebound. Although we did not include importation of cases in our simulations, the same sort of rebound could be expected if new cases arrived in the population after NPIs are relaxed. Combinations of interventions that bring R effective far below 1 might be required for elimination in a shorter time frame [9] . But even if elimination does not occur, combinations of NPIs can drastically reduce the prevalence of infections, essentially "resetting" the epidemic to an earlier and more manageable state. Of course, these interventions are costly to society, so adaptively modulating them to maintain incidence within a tolerable range is an appealing option if surveillance is sufficiently broad and timely [9, 10] .