#Epitwitter Fun with Numbers:
It’s possible to put bounds on the case-fatality rate (CFR) for this outbreak with the available data.
A quick #tweetorial on the partial identification of absolute risk bounds!
1/n
It’s possible to put bounds on the case-fatality rate (CFR) for this outbreak with the available data.
A quick #tweetorial on the partial identification of absolute risk bounds!
1/n
According to the quoted tweet we have this data:
31 cases with known outcome
6 of whom died
227 with an unknown outcome
First, the CFR among known data: 6/31 = 19%
2/n
31 cases with known outcome
6 of whom died
227 with an unknown outcome
First, the CFR among known data: 6/31 = 19%
2/n
Next, we include unknown outcomes, taking them to their logical extremes (all live or all die).
The lower risk bounds is calculated under the “all live” extreme:
6/(31+227) = 2.3%
Given the available data, this is the lowest possible CFR (i.e., if no new cases arise).
3/n
The lower risk bounds is calculated under the “all live” extreme:
6/(31+227) = 2.3%
Given the available data, this is the lowest possible CFR (i.e., if no new cases arise).
3/n
The upper bound is calculated under the “all die” extreme:
(6+227)/(31+227) = 91%
Given the available data, this is the highest possible CFR (again, if no new cases arise).
So the case-fatality rate must be between 2.3% and 91% though the current observed CFR is 19%
4/n
(6+227)/(31+227) = 91%
Given the available data, this is the highest possible CFR (again, if no new cases arise).
So the case-fatality rate must be between 2.3% and 91% though the current observed CFR is 19%
4/n
These “worst case bounds” are very wide but provide some information which can aide decision making by clarifying the range of the CFR.
They could easily be updated as new data come in or narrowed with simple assumptions (eg, risk in unknow is 1/2 that of current known).
5/n
They could easily be updated as new data come in or narrowed with simple assumptions (eg, risk in unknow is 1/2 that of current known).
5/n
For an applied example in a different context you can check out our (free) paper which uses absolute risk bounds to account for competing risks!
Fin!
journals.lww.com/greenjournal/f…
Fin!
journals.lww.com/greenjournal/f…
Side note, in this case the lower bound is problems the most useful part 👇🏻
