CovidSIMVL -- Transmission Trees, Superspreaders and Contact Tracing in Agent Based Models of Covid-19

The agent-based model CovidSIMVL (github.com/ecsendmail/MultiverseContagion) is employed in this paper to delineate different network structures of transmission chains in simulated COVID-19 epidemics, where initial parameters are set to approximate spread from a single transmission source, and R0ranges between 1.5 and 2.5. The resulting Transmission Trees are characterized by breadth, depth and generations needed to reach a target of 50% infected from a starting population of 100, or self-extinction prior to reaching that target. Metrics reflecting efficiency of an epidemic relate closely to topology of the trees. It can be shown that the notion of superspreading individuals may be a statistical artefact of Transmission Tree growth, while superspreader events can be readily simulated with appropriate parameter settings. The potential use of contact tracing data to identify chain length and shared paths is explored as a measure of epidemic progression. This characterization of epidemics in terms of topological characteristics of Transmission Trees may complement equation-based models that work from rates of infection. By constructing measures of efficiency of spread based on Transmission Tree topology and distribution, rather than rates of infection over time, the agent-based approach may provide a method to characterize and project risks associated with collections of transmission events, most notably at relatively early epidemic stages, when rates are low and equation-based approaches are challenged in their capacity to describe or predict. MOTIVATION - MODELS KEYED TO CONTEMPLATED DECISIONSOutcomes are altered by changing the processes that determine them. If we wish to alter contagion-based spread of infection as reflected in curves that characterize changes in transmission rates over time, we must intervene at the level of the processes that are directly involved in preventing viral spread. If we are going to employ models to evaluate different candidate arrays of localized preventive policies, those models must be posed at the same level of granularity as the entities (people enacting processes) to which preventive measures will be applied. As well, the models must be able to represent the transmission-relevant dynamics of the systems to which policies could be applied. Further, the parameters that govern dynamics within the models must embody the actions that are prescribed/proscribed by the preventive measures that are contemplated. If all of those conditions are met, then at a formal or structural level, the models are conformant with the provisions of the Law of Requisite Variety1 or the restated version of that law - the good regulator theorem.2 On a more logistical or practical level, the models must yield summary measures that are responsive to changes in key parameters, highlight the dynamics, quantify outcomes associated with the dynamics, and communicate that information in a form that can be understood correctly by parties who are adjudicating on policy options. If the models meet formal/structural requirements regarding requisite variety, and the parameters have a plausible interpretation in relationship to real-world situations, and the metrics do not overly-distort the data contents that they summarize, then the models provide information that is directly relevant to decision-making processes. Models that meet these requirements will minimize the gap that separates models from decisions, a gap that will otherwise be filled by considerations other than the data used to create the models (for equation-based models) or the data generated by the simulations. In this work, we present an agent-based model that targets information requirements of decision-makers who are setting policy at a local level, or translate population level directives to local entities and operations. We employ an agent-based modeling approach, which enables us to generate simulations that respond directly to the requirements of the good regulator theorem. Transmission events take place within a spatio-temporal frame of reference in this model, and rates are not conditioned by a reproduction rate (R0) that is specified a priori. Events are a function of movement and proximity. To summarize dynamics and associated outcomes of simulated epidemics, we employ metrics reflecting topological structure of transmission chains, and distributions of those structures. These measures point directly to dynamic features of simulated outbreaks, they operationalize the "efficiency" construct, and they are responsive to changes in parameters that govern dynamics of the simulations.