Most look at infection rates among Wuhan evacuees. Since almost all evacuees were tested, it's a good proxy to estimate the city-wide infection rate. See the 4-5 papers linked in my blog post's last section: https://blog.zorinaq.com/case-fatality-ratio-ncov/#updates--...
Interesting, this is always my argument against these fatality ratio estimates, can you explain how the IFR can take undetected mild cases into account? Does it just assume at random how many of those cases exist?