Overdispersion in COVID-19 increases the effectiveness of limiting nonrepetitive contacts for transmission control

Abstract

Increasing evidence indicates that superspreading plays a dominant role in COVID-19 transmission. Recent estimates suggest that the dispersion parameter k for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is on the order of 0.1, which corresponds to about 10% of cases being the source of 80% of infections. To investigate how overdispersion might affect the outcome of various mitigation strategies, we developed an agent-based model with a social network that allows transmission through contact in three sectors: "close" (a small, unchanging group of mutual contacts as might be found in a household), "regular" (a larger, unchanging group as might be found in a workplace or school), and "random" (drawn from the entire model population and not repeated regularly). We assigned individual infectivity from a gamma distribution with dispersion parameter k We found that when k was low (i.e., greater heterogeneity, more superspreading events), reducing random sector contacts had a far greater impact on the epidemic trajectory than did reducing regular contacts; when k was high (i.e., less heterogeneity, no superspreading events), that difference disappeared. These results suggest that overdispersion of COVID-19 transmission gives the virus an Achilles' heel: Reducing contacts between people who do not regularly meet would substantially reduce the pandemic, while reducing repeated contacts in defined social groups would be less effective.

Keywords: mitigation strategies; overdispersion; pandemic; social networks; superspreading.

Fig. 1.

(A) Schematic representation progression of disease in our agent-based model. Individual agents become infectious 2.5 d before symptom onset on average. Agents enter the recovered state after an average of 3 d of symptoms, giving an average total infectious period of 5.5 d. (B) Schematic representation of the connectivity between 150 agents. Individuals are represented as nodes, with shading indicating age (light = young, dark = older). Edges represent social connections, with bright yellow denoting close contacts, orange denoting regular contacts between adults, and red edges denoting regular contacts involving children. Random contacts are not pictured. The network diagram was generated by running our simulation on a smaller population of just 150 individuals, with the same rules for connectivity as in the full-scale simulations.

Fig. 2.

The impact of mitigation on modeled incidence. Simulation of epidemic trajectories with mitigation starting when 1% of the population has been infected. In each panel, we show three trajectories corresponding to the unmitigated epidemic, the case where we completely restrict all regular contacts, and the case where we restrict all random contacts. When superspreading is not present (i.e., k is infinite; A), the effect of eliminating regular and random contacts is similar; however, when superspreading is a factor in transmission (i.e., when k = 0.1; B), the effect of eliminating random contacts is dramatically enhanced. We did not consider mitigation by limiting close contacts as this would not be a credible mitigation strategy.

Fig. 3.

Sensitivity of model results to dispersion factor k. When an epidemic size of 1% is reached, a mitigation scheme consisting of restricting all random contacts is initiated. We explore the epidemic trajectories for different values of the overdispersion factor k. As k decreases (i.e., transmission heterogeneity increases), eliminating random contacts has a progressively greater effect.