Read on Twitter
(1/8) Germany& #39;s #COVID19 death rate is rising. Where will it stop? Here& #39;s a suggestion. Case fatality rate (CFR) is calculated dividing number of deaths by number of confirmed cases. That& #39;s not the chance of an infected person to die, since many infections probably go unrecorded.
(2/8) In Germany, with treatment and testing consistent from early on and healthcare system functional, I& #39;d expect a relatively stable CFR. Why is it rising? Maybe CFR is rising because we ignore the time difference between diagnosis and death.
(3/8) Today& #39;s deaths were counted as cases some time ago. Wouldn& #39;t a more accurate CFR compare today& #39;s deaths with cases X days ago, so numbers correspond to the same group of people? Here I divided total deaths of every day by total cases 5 days before it:
(4/8) CFR on 23.3 is now significantly higher: 1% and not 0.4%. This is because number of deaths that day (123), which was previously divided by number of cases on the same day (29,056), was now divided by the number of cases 5 days before (12,327) – a much smaller number.
(5/8) The line also flattened a bit. Now let& #39;s check a 2 weeks offset. CFR on 23.3 tops 10%- this is surely exaggerated: we divided the deaths by cases of a date so early, that total cases were too low and resulted in an unrealistic rate. These dates don& #39;t seem to correspond.
(6/8) What& #39;s the right offset then? Given a person who died today- when on average were they tested and counted as a confirmed case? If we assume that there is a consistent German CFR, could we find the right offset by looking for the one that creates a somewhat stable CFR?
(7/8) Here are a few candidates. It seems like somewhere between 10 and 11-days offset, but closer to 10, CFR looks more or less stable throughout the period:
(8/8) Moreover, the variation of rates at a 10-days offset is the narrowest of all: 0.65% between lowest and highest, 3.25% and 3.9% respectively. Will 3.6%, the mean CFR at this offset, be Germany& #39;s final rate?
