Read on Twitter
Today in NEJM, a new #covid19 study: https://www.nejm.org/doi/full/10.1056/NEJMoa2026116
Let's see what they find about:
- percentage who develop antibodies
- antibody levels (not) changing over time
- Infection Fatality Ratio (IFR)
Spoiler: age-adjusted IFR is 0.57%, supporting CDC estimate of 0.65%
1/n
Let's see what they find about:
- percentage who develop antibodies
- antibody levels (not) changing over time
- Infection Fatality Ratio (IFR)
Spoiler: age-adjusted IFR is 0.57%, supporting CDC estimate of 0.65%
1/n
Their main finding is that antibodies do NOT decline:
«Our results indicate that antiviral antibodies against SARS-CoV-2 did not decline within 4 months after diagnosis»
This result does not support vague claims circulating on Twitter that antibodies decline over time.
2/n
«Our results indicate that antiviral antibodies against SARS-CoV-2 did not decline within 4 months after diagnosis»
This result does not support vague claims circulating on Twitter that antibodies decline over time.
2/n
Next, they find that most infections do develop antibodies: «1107 of the 1215 who were tested (91.1%) were seropositive»
Compare with RKI who found only 66% did ( https://twitter.com/zorinaq/status/1298308988015009793). But RKI's study had only ~300 infected individuals, so the NEJM study is more reliable
3/n
Compare with RKI who found only 66% did ( https://twitter.com/zorinaq/status/1298308988015009793). But RKI's study had only ~300 infected individuals, so the NEJM study is more reliable
3/n
Next, they find an IFR of 0.3%. It's low because it's an artifact of the young having been more infected than average:
0.91% of ages 0-70 were infected
0.39% of >70 were infected
See age distribution in Supplementary Appendix 1 table S7: https://www.nejm.org/doi/suppl/10.1056/NEJMoa2026116/suppl_file/nejmoa2026116_appendix_1.pdf
4/n
0.91% of ages 0-70 were infected
0.39% of >70 were infected
See age distribution in Supplementary Appendix 1 table S7: https://www.nejm.org/doi/suppl/10.1056/NEJMoa2026116/suppl_file/nejmoa2026116_appendix_1.pdf
4/n
IOW 0.3% is the *prevalent* IFR, skewed by uneven prevalence among ages.
We calculate an *age-adjusted* IFR of 0.57% in Iceland, using table S7 data.
And 0.57% does support the US CDC estimate of 0.65% (from table 1 in https://www.cdc.gov/coronavirus/2019-ncov/hcp/planning-scenarios.html)
For the math, see next tweet
5/n
We calculate an *age-adjusted* IFR of 0.57% in Iceland, using table S7 data.
And 0.57% does support the US CDC estimate of 0.65% (from table 1 in https://www.cdc.gov/coronavirus/2019-ncov/hcp/planning-scenarios.html)
For the math, see next tweet
5/n
Per table S7:
0-70: 3 deaths, 3011.9 infected, 0.91% prevalence
>70: 7 deaths, 165.4 infected, 0.39% prevalence
Prevalent IFR:
(3 + 7) / (3011.9 + 165.4) = 0.3%
There's a .91/.39=2.33 difference in prevalence
Age-adjusted IFR:
(3 + 7*2.33) / (3011.9 + 165.4*2.33) = 0.57%
6/n
0-70: 3 deaths, 3011.9 infected, 0.91% prevalence
>70: 7 deaths, 165.4 infected, 0.39% prevalence
Prevalent IFR:
(3 + 7) / (3011.9 + 165.4) = 0.3%
There's a .91/.39=2.33 difference in prevalence
Age-adjusted IFR:
(3 + 7*2.33) / (3011.9 + 165.4*2.33) = 0.57%
6/n
And here's the kicker:
An age-adjusted IFR of 0.57% in Iceland was nearly perfectly predicted, 3 months ago, by the Spanish IFR serosurvey.
3 months ago I wrote a tool to apply the Spanish age-stratified IFR to the population Pyramid of any country: https://twitter.com/zorinaq/status/1270761731682258945
7/n
An age-adjusted IFR of 0.57% in Iceland was nearly perfectly predicted, 3 months ago, by the Spanish IFR serosurvey.
3 months ago I wrote a tool to apply the Spanish age-stratified IFR to the population Pyramid of any country: https://twitter.com/zorinaq/status/1270761731682258945
7/n
I wanted to see what my script predicts for Iceland.
So I go to Statistics Iceland to get the population pyramid for Iceland ( https://www.statice.is/statistics/population/inhabitants/ → Population by sex and age 1841-2020 → https://px.hagstofa.is/pxen/pxweb/en/Ibuar/Ibuar__mannfjoldi__1_yfirlit__yfirlit_mannfjolda/MAN00101.px)
I add the data to my script: https://github.com/mbevand/covid19-age-stratified-ifr/commit/b366f3f7ba20729ee12542224e7a126d03afa034
8/n
So I go to Statistics Iceland to get the population pyramid for Iceland ( https://www.statice.is/statistics/population/inhabitants/ → Population by sex and age 1841-2020 → https://px.hagstofa.is/pxen/pxweb/en/Ibuar/Ibuar__mannfjoldi__1_yfirlit__yfirlit_mannfjolda/MAN00101.px)
I add the data to my script: https://github.com/mbevand/covid19-age-stratified-ifr/commit/b366f3f7ba20729ee12542224e7a126d03afa034
8/n
I edit the script to set:
pyramid_target = pyramid_iceland
And run it:
$ ./calc_ifr.py
[...]
IFR on target country assuming disease prevalence equal among ages: 0.584%
Holy cow! 0.584% is nearly identical to the 0.57% they found in the NEJM study!
9/n
pyramid_target = pyramid_iceland
And run it:
$ ./calc_ifr.py
[...]
IFR on target country assuming disease prevalence equal among ages: 0.584%
Holy cow! 0.584% is nearly identical to the 0.57% they found in the NEJM study!
9/n
Worth noting: my calc_ifr.py script previously correctly predicted the IFR in the US.
It calculated 0.658% on June 10 ( https://twitter.com/zorinaq/status/1270761731682258945), matching nearly perfectly the US CDC estimate of 0.65% published in early July ( https://twitter.com/zorinaq/status/1282163890147614720)
10/n
It calculated 0.658% on June 10 ( https://twitter.com/zorinaq/status/1270761731682258945), matching nearly perfectly the US CDC estimate of 0.65% published in early July ( https://twitter.com/zorinaq/status/1282163890147614720)
10/n
Anyway, a thing to keep in mind about this NEJM study is that the total number of deaths in the whole country of Iceland is N=10. Very small. Hence the wide confidence interval around the IFR.
11/n
11/n
