Social Heterogeneity Drives Complex Patterns of the COVID-19 Pandemic: Insights From a Novel Stochastic Heterogeneous Epidemic Model (SHEM)

Abstract

In addition to vaccine and impactful treatments, mitigation strategies represent an effective way to combat the COVID-19 virus and an invaluable resource in this task is numerical modeling that can reveal key factors in COVID-19 pandemic development. On the other hand, it has become evident that regional infection curves of COVID-19 exhibit complex patterns which often differ from curves predicted by forecasting models. The wide variations in attack rate observed among different social strata suggest that this may be due to social heterogeneity not accounted for by regional models. We investigated this hypothesis by developing and using a new Stochastic Heterogeneous Epidemic Model that focuses on subpopulations that are vulnerable in the sense of having an increased likelihood of spreading infection among themselves. We found that the isolation or embedding of vulnerable sub-clusters in a major population hub generated complex stochastic infection patterns which included multiple peaks and growth periods, an extended plateau, a prolonged tail, or a delayed second wave of infection. Embedded vulnerable groups became hotspots that drove infection despite efforts of the main population to socially distance, while isolated groups suffered delayed but intense infection. Amplification of infection by these hotspots facilitated transmission from one urban area to another, causing the epidemic to hopscotch in a stochastic manner to places it would not otherwise reach; whereas vaccination only in hotspot populations stopped geographic spread of infection. Our results suggest that social heterogeneity is a key factor in the formation of complex infection propagation patterns. Thus, the mitigation and vaccination of vulnerable groups is essential to control the COVID-19 pandemic worldwide. The design of our new model allows it to be applied in future studies of real-world scenarios on any scale, limited only by computing memory and the ability to determine the underlying topology and parameters.