Reporting rate

This seems like one of the hardest things to measure: how many people actually have coronavirus, given your estimate? It's wildly dependent upon testing strategy and delays in the system.

• This paper uses simulations from Wubei to derive a maximum likelihood reporting rate of 0.14
  • This is the most up-to-date estimate we've seen, so we're currently using this
• This paper used sampling from evacuated Japanese individuals to place the rate at ~ 0.1
  • We used this in the app before 19th March 2020

Disease dynamics data

We use the SIR model, a standard epidemiological model, to estimate the disease dynamics.

Compartmental models in epidemiology

The specific flavour we use is the Kermack-McKendrick model:

https://mathworld.wolfram.com/Kermack-McKendrickModel.html

The crucial factor governing disease spread is R0 (the basic reproduction rate), which is the average number of people somebody with the disease infects. This is a function of the number of susceptible people, the infection rate β and the recovery rate γ.

You can decompose this into how many people an infected person contacts a day, and the transmission probability from a given contact:

β = Probability of transmission x Number of contacts

We allow you to control Number of Contacts per day in the webapp. We have estimated Probability of Transmission from the following numbers:

• Number of infectious days from https://github.com/midas-network/COVID-19/tree/master/parameter_estimates/2019_novel_coronavirus, and https://www.medrxiv.org/content/10.1101/2020.03.05.20030502v1
• Number of contacts per day and R0 from Table 1 and 2 of https://www.mdpi.com/2077-0383/9/2/462/htm

We can then formulate R0 as:

R0 = Probability of transmission x Number of Contacts per day x Number of infectious days

Subbing in our numbers:

                              **2.5 = Probability of Transmission x 14 x 10**

we solve for Probability of transmission, which comes out as ~0.018, or 1.8%.