Thread on the importance of critical thinking:
A new preprint of 1099 cases of novel coronavirus (medrxiv.org/content/10.110…) points out the case fatality rate they observed is lower than 2 other studies of respectively 41 and 99 cases. It doesn't explain why.
Let's find out why.
A new preprint of 1099 cases of novel coronavirus (medrxiv.org/content/10.110…) points out the case fatality rate they observed is lower than 2 other studies of respectively 41 and 99 cases. It doesn't explain why.
Let's find out why.
They noted that only 15 of the 1099 cases (or 1.4%) died, see image.
For comparison the two other studies (thelancet.com/journals/lance… & thelancet.com/journals/lance…) reported a case fatality rate (CFR) of respectively 15% and 11%.
For comparison the two other studies (thelancet.com/journals/lance… & thelancet.com/journals/lance…) reported a case fatality rate (CFR) of respectively 15% and 11%.
But wait! Look at the number of patients that, as of the time the study was prepared, still remained in the hospital:
1029 out of 1099 (93.6%)
A vast majority of the cases are still in the hospital! They could either eventually recover or die. Their outcome isn't known yet.
1029 out of 1099 (93.6%)
A vast majority of the cases are still in the hospital! They could either eventually recover or die. Their outcome isn't known yet.
Let's compare with the other study that reported a CFR of 15%:
7 out of 41 (17%) remained in the hospital
So only a minority of cases are unresolved.
7 out of 41 (17%) remained in the hospital
So only a minority of cases are unresolved.
And let's compare with the other study that reported a CFR of 11%:
57 out of 99 (58%) remained in the hospital
So about half of cases are unresolved.
57 out of 99 (58%) remained in the hospital
So about half of cases are unresolved.
To sum up all 3 studies:
1099-case study:
1.4% died, 93.6% still in hospital
99-case study:
11% died, 58% still in hospital
41-case study:
15% died, 17% still in hospital
Essentially, the more UNRESOLVED cases you have, the LOWER the case fatality rate will be.
1099-case study:
1.4% died, 93.6% still in hospital
99-case study:
11% died, 58% still in hospital
41-case study:
15% died, 17% still in hospital
Essentially, the more UNRESOLVED cases you have, the LOWER the case fatality rate will be.
Until the outcome (death or recovery) of most patients is known, the CFR calculated as "deaths/cases" is grossly inaccurate, and will underestimate the true CFR. Epidemiologists call it a "naïve CFR".
In other words, the 1.4% naïve CFR they report would only be accurate if EVERY SINGLE ONE of the 1029 patients still hospitalized does eventually recover.
In fact, we know some WILL die.
In fact, we know some WILL die.
This is why epidemiologists recommend instead to use the "resolved CFR" formula:
deaths/(deaths+recoveries)
As only cases for whom the outcome is known (death or recovery) are taken into account.
deaths/(deaths+recoveries)
As only cases for whom the outcome is known (death or recovery) are taken into account.
I wrote a blog post comparing naïve vs resolved CFR: blog.zorinaq.com/case-fatality-…
The resolved CFR is virtually always a much more accurate estimate of the true CFR.
The resolved CFR is virtually always a much more accurate estimate of the true CFR.
A more insightful comparison would be to look at the resolved CFR across all 3 studies:
1099-case: 15/(15+55)=21%
99-case: 11/(11+31)=26%
41-case: 6/(6+28)=18%
And across all mainland China as of Feb 9:
908/(908+3281)=22%
All in a very tight range, 18-26%!
1099-case: 15/(15+55)=21%
99-case: 11/(11+31)=26%
41-case: 6/(6+28)=18%
And across all mainland China as of Feb 9:
908/(908+3281)=22%
All in a very tight range, 18-26%!
This suggest that so far all studies support each other, and country-wide data support them too, that the true CFR, the true case fatality ratio is about 20%.
The end.
The end.
