(1/13) Very excited to regroup with (extended) @IDMOD_ORG ers: @BMAlthouse @WengerEdward @joel_c_miller @svscarpino @LHDnets @all_are with inputs from @famulare_mike. This is a draft on super spreading events, stochasticity and heterogeneity of COVID-19. https://covid.idmod.org/data/Stochasticity_heterogeneity_transmission_dynamics_SARS-CoV-2.pdf

(2/13) It's likely that the role of super-spreading events in transmission dynamics is similar as SARS-CoV-1 and MERS-CoV. We should not overlook the heterogeneity in numbers of secondary infections, frequently modeled as a negative binomial distribution.

(3/13) Our mental picture should not be that most people transmit to 2-3 others (for example, the concept of Figure 1 here is wrong: https://jamanetwork.com/journals/jama/fullarticle/2765665), but instead a small number of infections dominate the transmission while most others fail to have secondary infections.

(4/13) The distribution is over-dispersed with a high probability of extinction on the lower end, and a long tail on the higher end.

(5/13) At the early stage we will see more randomness and stochasticity, and it seems more explosive with huge number of cases reported in the first few generations. But once it takes off, it still becomes a stable exponential as per classic deterministic models.

(6/13) This distribution has been widely used over years, and is already in most COVID-19 models. We wanted to highlight it because it explains many puzzles we had earlier, such as the early stochastic and seemingly explosive dynamics at hotspots and at the population level.

(7/13) Again, super-spreading not only exist, but if transmission follows a negative binomial, these SSEs and hotspots drive and fuel the continuous onward transmission. It has important implications about control measures of COVID-19.

(8/13) We classified hotspots into biological (i.e. high viral load), behavioral (i.e. many contacts), high-risk facilities (i.e. health care facilities, long term care, meatpacking plants), and opportunistic (i.e. cruise ships, party, crowded bus, choir, other mass gatherings).

(9/13) If we could do targeted intervention and significantly reduce transmission in these types of hotspots to "cut the long tail", then overall transmissibility is greatly reduced, then we don't need to rely on a single population-level intervention to bring Reff below 1.

(10/13) All measures to lower overall R0 like masks or behavioral change, contact tracing and/or isolation should be used as well. Even if they can't bring Reff below 1, they increase the extinction probability and increase the probability of success for targeted interventions.

(11/13) If one could cut the long-tail of secondary infections, lower population-wide R0 as much as possible through masks or behavioral change, and do aggressive contact tracing and testing/isolations, these stacked measures will be an effective strategy.

(12/13) This heterogeneity means contact tracing should have a clear goal of identifying the hotspot. Once there is a case, there could have been a large number of infections somewhere 1-2 generations ago, and we might be behind the frontier.

(13/13) Glad to see recent discussions on SSEs, such as: https://www.sciencemag.org/news/2020/05/why-do-some-covid-19-patients-infect-many-others-whereas-most-don-t-spread-virus-all, demonstrating the urgent need to understand mechanisms of transmission, while posing an opportunity that outbreak can be effectively contained with targeted interventions to eliminate SSEs.

(14/13) Special thanks to another former IDMer @philipwelkhoff. This draft would not exist without the many insightful discussions with him.