1. There's a relatively new study out of Denmark that estimates seroprevalence from nearly 10,000 blood donations by people age 17-69 to be 1.7%

(CI: 0.9-2.3).

medrxiv.org/content/10.110…

(CI: 0.9-2.3).

medrxiv.org/content/10.110…

2. Based on these numbers, the authors estimate an infection fatality rate considerably lower than many US estimates—lower even than that from the Santa Clara and LA county studies: 0.082% (CI: 0.059-0.154%).

3. Methodologically this study seems to avoid many of the problems that plague the Californian studies, most notably because sampling bias is reduced by using blood donations instead of soliciting patients specifically for serotype testing.

4. As the authors clearly state, that this paper estimates the infection fatality rate for people ages 17-69 only. This is very important, because the fatality rate skyrockets in people 70 and up.

Thus the 0.082 IFR for this age group is an underestimate of the population IFR.

Thus the 0.082 IFR for this age group is an underestimate of the population IFR.

5. How big of an underestimate is it?

We can get a ballpark figure by looking at the relative age-specific CFR rates from the Wu and McGoogan JAMA study, weighting each by UN data about the demographic structure of Denmark.

We can get a ballpark figure by looking at the relative age-specific CFR rates from the Wu and McGoogan JAMA study, weighting each by UN data about the demographic structure of Denmark.

6. When you go through this process, you find that the population IFR should be almost exactly twice the IFR in the 20-69 category. (I don't have data to extend down to age 17).

This yields a population IFR estimate of something like 0.16% with approximate CI 0.12-0.31.

This yields a population IFR estimate of something like 0.16% with approximate CI 0.12-0.31.

7. This is very close to the data coming out of California, data that I have publicly criticized.

This is puzzling. With 0.15% of New York City's population dead from COVID19, that seems to be a very hard lower bound on IFR. Realistically, 0.7-1.0% seems more likely for NYC.

This is puzzling. With 0.15% of New York City's population dead from COVID19, that seems to be a very hard lower bound on IFR. Realistically, 0.7-1.0% seems more likely for NYC.

8. Other estimates of IFR also land in that range 0.5%-1%. While the authors of the Danish study discuss some limitations, I don't see obvious glaring flaws. Lag times to death may come into play.

What, if anything, am I missing?

What, if anything, am I missing?

9. @kearnsneuro quickly picked up one thing. The age distribution of blood donor population is probably not representative of the age distribution in the population at large. twitter.com/i/status/12557…

10. Why does this matter? If younger people are more likely to be infected *and* more likely to give blood, the procedure described in the paper would yield an underestimate of IFR.

Still I'd be surprised if this gets us all the way up to the NYC numbers.

Still I'd be surprised if this gets us all the way up to the NYC numbers.

11. @BendyGardiner correctly points out that I made things much too complicated with my silly IFR scaling to the population as a whole, and in the process got an odd result. I think I know why, but first his objection:

twitter.com/BendyGardiner/…

twitter.com/BendyGardiner/…

12. We still need the demographic pyramid, but there's no need to use Chinese CFR estimates to make this correction. I see slightly different numbers than Ben reports, but let's work with these.

13. Approximately 73% of the Danish population is aged 20-69 (again, I don't have the data for 17-69).

IFR 20-69 = deaths 20-69 / cases 20-69.

IFR pop = all deaths / all cases.

Assume cases are uniform across age.

The paper gives us total deaths as well as deaths 17-69.

IFR 20-69 = deaths 20-69 / cases 20-69.

IFR pop = all deaths / all cases.

Assume cases are uniform across age.

The paper gives us total deaths as well as deaths 17-69.

14. To get the population IFR, we take the IFR for 20-69 year olds and numerator by a factor of 370/53 and the denominator by 1/0.73.

This gives us a 5.4-fold increase for a population IFR of 0.44%, CI (0.32-0.83%), closer in line with what I would have expected.

This gives us a 5.4-fold increase for a population IFR of 0.44%, CI (0.32-0.83%), closer in line with what I would have expected.

15. Why is this higher than the estimate I got using the Chinese figures? I'm guessing that CFR in the 60-69 age group in China is substantially higher than CFR for the same group in Denmark.

16. Another possibility is that I screwed up the math; I'm open to corrections. It's 1:30 AM and I'm unwisely doing back-of-the-envelope math in real time and public view on a combined 10 hours of sleep over the past 3 nights.

Consider this your disclaimer for the entire thread.

Consider this your disclaimer for the entire thread.

17. Thanks to everyone who helped me work through this.

tl;dr conclusion: the population IFR is several times higher than the age 17-69 IFR, so the Danish study here is closer in line with the results out of New York than the results from the Santa Clara and LA county studies.

tl;dr conclusion: the population IFR is several times higher than the age 17-69 IFR, so the Danish study here is closer in line with the results out of New York than the results from the Santa Clara and LA county studies.

18. It's almost frustrating. The original thread, posts 1-8, was my effort to be transparent by highlighting a piece of solid work that did not support my prior beliefs. But then you folks showed up and proved that I was right all along.

Maybe there's still an error somewhere.😀

Maybe there's still an error somewhere.😀