[prepend]

> You can't compute fatality rates simply by deaths/recoveries, because recoveries lag deaths by several weeks (once you're dead you're dead, but if you survive it takes a few weeks to recover and start testing negative).

There are limitations with this method, but I think it’s the best possible at the time. What method do you think is better?

[9nGQluzmnq3M]

Count deaths & recoveries a few weeks in for cohorts that showed first symptoms on the same days. But data collection for this is hard and it takes a while too.

[prepend]

This is mentioned in the article and is more accurate, but isn’t possible throughout an outbreak for the reason you state.

[joosters]

You could count the number of deaths by date 'n', compared to the number of recoveries by date 'n+14', say, picking a date related to the average recovery time?

[prepend]

The recovery date isn’t fixed so there’s not a single “+14” that is as useful as picking the dead / known outcome.

I think this is because it’s hard to know date of infection, but it’s definite to know whether the case resulted in death or has been determined “recovered.”