A thread on how to calculate a more accurate CFR (case fatality ratio) for 2019-nCoV.
tl;dr: the true CFR is closer to 9-56%, absolutely not 2-3% as reported by the media. Read on... #coronavirus
1/5
tl;dr: the true CFR is closer to 9-56%, absolutely not 2-3% as reported by the media. Read on... #coronavirus
1/5
2. Journalists ALWAYS make this error:
They compare current deaths to current confirmed cases. But it takes N days for the virus to kill, so deaths will lag confirmed cases by N days, and thus the calculated CFR is incorrect.
2/5
They compare current deaths to current confirmed cases. But it takes N days for the virus to kill, so deaths will lag confirmed cases by N days, and thus the calculated CFR is incorrect.
2/5
There are 2 methods to calculate a more accurate CFR.
Method #1: compare current deaths to cases confirmed N days ago. (N is the typical time-to-death.) Assuming N is 3-6 days, the CFR is between 8.88% and 25.47% see
3/5
Method #1: compare current deaths to cases confirmed N days ago. (N is the typical time-to-death.) Assuming N is 3-6 days, the CFR is between 8.88% and 25.47% see
3/5
Method #2: if N is hard to estimate, at least we know a patient eventually either recovers or dies. So "current deaths" divided by "current deaths + current recovered cases" is a good estimate. As of today China reported 132 deaths and 103 recoveries. So CFR = 132/235 = 56%!
4/5
4/5
Summary: a more accurate CFR for 2019-nCoV is 9-56% (near lower end according to method 1, near upper end according to method 2.)
For comparison, MERS had a CFR of 30-40%. However unlike MERS (R₀<1), 2019-nCoV is much, much more infectious (R₀=2.9.) Very scary.
5/5
For comparison, MERS had a CFR of 30-40%. However unlike MERS (R₀<1), 2019-nCoV is much, much more infectious (R₀=2.9.) Very scary.
5/5
By the way, the error journalists make in calculating the case fatality ratio is very well-known in epidemiology. See for example ncbi.nlm.nih.gov/pmc/articles/P… which calls it the "Naïve Estimator"
I am completely baffled that no epidemiologist has come forward to correct this.
6/n
I am completely baffled that no epidemiologist has come forward to correct this.
6/n
By the way, I assume the typical time-to-death is N = 3-6 days based on this chart:
When the infection is confirmed the patient is hospitalized (day 7) and the death typically takes place shortly after entering the ICU (day 11)
When the infection is confirmed the patient is hospitalized (day 7) and the death typically takes place shortly after entering the ICU (day 11)
Another paper from the American Journal of Epidemiology:
pdfs.semanticscholar.org/ebf2/48c9fc0a1…
How journalists compute CFR is called "simple estimate 1" and shown to underestimate.
Method #2 is called "simple estimate 2" and works: it is "reasonable at most points in the epidemic," see fig 3a
pdfs.semanticscholar.org/ebf2/48c9fc0a1…
How journalists compute CFR is called "simple estimate 1" and shown to underestimate.
Method #2 is called "simple estimate 2" and works: it is "reasonable at most points in the epidemic," see fig 3a
FINALLY! An article was published by a trusted source (medical doctor @maiamajumder) explaining what I tried to clumsily explain: that a CFR of 2% is nonsense, currently it's 56% and it should decrease as the outbreak progresses.
issues.org/clarity-please…
issues.org/clarity-please…
@maiamajumder Here is a really good chart showing that the naive case fatality rate is bound to start low then over time converge to the eventual CFR:
@maiamajumder Another example: naive CFR for SARS, which started at ~2% and eventually converged to 9.3%:
@maiamajumder Still not convinced that a naive CFR calculation is useless?
Here is a pure math simulation where the probability of dying 7-8 days after infection is CONSTANT at 30%. And yet the naive CFR starts at ~2% and rises to ~30% only at the end of the epidemic
Here is a pure math simulation where the probability of dying 7-8 days after infection is CONSTANT at 30%. And yet the naive CFR starts at ~2% and rises to ~30% only at the end of the epidemic
