A policy tool for COVID-19 workplace mitigation strategies
1 Executive summary
Assessing the risk of Covid-19 infections in a workplace setting involves evaluating the propagation of the disease among employees. Propagation depends on a large number of variables, including employee interactions in the larger community, adherence to social distancing policies, local environmental factors, demographics, and workplace policies. Given the enormous complexity and the difficulty in calibrating a complicated model from mostly non-existent or highly limited data, we focus planning efforts around a branching model that requires three user inputs: 1. The frequency of human-to-human interactions involving two or more people; 2. The 95% upper bound on interaction sizes; and 3. A mitigation factor. Using a web interface to make choices with these three variables, the decision maker builds a propagation scenario starting from one infected individual. Through establishing policy, the decision maker can affect all three variables. Our propagation model is rooted in the academic literature and clinical data.
Perhaps the greatest risk facing a decision maker is an asymptomatic individual who arrives at work and begins propagating the infection throughout the workplace to others who also remain asymptomatic for many days. In such an event, the infection may only be discovered after a sizable fraction of the workforce has become infected. We focus on this scenario as a mechanism to gauge the risk of policy choices.
The decision maker can restrict the frequency and size of interactions by controlling office dynamics. For example, the decision maker might restrict the number of meetings and ensure that meeting sizes remain small. Furthermore, the decision maker can enforce mitigation procedures, including the wearing of masks, frequent hand-washing, and physical distancing. Once the decision maker sets a policy, which in turn sets a value for each of the three variables, the model assumes that unbeknownst to the decision maker, one asymptomatic infected individual appears in the workplace. That individual will start a chain of infections across the workforce. Given the three input variables, the model yields: 1. The average number of individuals an infected person will infect during their infectious period; 2. The average infected population growth as a function of time across the workforce; 3. The probability that the infection will stop on its own through good luck; and 4. Simulations which show possible trajectories of the infected population. From these outputs, the decision maker can choose a workplace policy that leads to the three input variables that generate an acceptable outcome. In addition, the decision maker can select workplace sizes (e.g., floor population sizes) by observing how fast an undetected outbreak will spread across the workforce.
The model does not provide a multitude of selection variables for complicated simulation runs—it is not an agent based model—rather, the model provides the dynamics of the propagation in the workforce starting from a single infected individual. Reality can be much worse, but this model bounds the outcomes. If, with a certain set of parameter choices, the risk from starting with one infected person is too large, starting with more infected people will only make the situation worse. Decision makers can use this model, along with their judgment and other inputs, to determine an acceptable range of policy choices.