[timr]

> What, the linked study proves the opposite. A massive amount of death in Manaus in 2021, despite 'herd immunity' measured in October 2020.

That is not a study. It is an editorial. It "proves" nothing, and instead advances a few alternative theories that might explain the discrepancy. However, it is not a complex situation. The following quote from the editorial is consistent with my personal belief after reading the original paper [1] that made the herd immunity claim:

"The 76% estimate of past infection might have been biased upwards due to adjustments to the observed 52·5% (95% CI 47·6–57·5) seroprevalence in June, 2020, to account for antibody waning."

This is true, but it understates the extent of the authors "adjustments". If you look at the original paper (Figure 2; linked below), they adjusted the raw data upward by a factor of about 3x. It is an...aggressive...modification to the raw data.

The parsimonious explanation for what is being observed in Manaus is that the Science paper claiming herd immunity was wrong. The authors of this Lancet editorial go to great lengths to advance alternative explanations involving "variants", but these are all unfounded, convoluted alternatives. The simplest explanation is that the original paper was wrong.

[1] https://science.sciencemag.org/content/371/6526/288

[ttul]

The latest episode of TWiV massacres the paper. It’s a good listen.

https://www.microbe.tv/twiv/twiv-722/

[disgruntledphd2]

I definitely found an epidemiologist on Twitter giving out about that study. The Science study used a convenience sample, which makes me believe that it's a very biased estimate of sero-prevalance.

[timr]

Yeah, there are possible issues with the way the Science paper collected their sample, as well. The authors of the Lancet editorial allude to this in the next sentence:

"furthermore, comparisons of blood donors with census data showed no major difference in a range of demographic variables, and the mandatory exclusion of donors with symptoms of COVID-19 is expected to underestimate the true population exposure to the virus."

Here, the Lancet writers are trying to argue that the sample wasn't that biased, and therefore the Science paper was right, and therefore the current infections are due to "variants".

It's a very weird and convoluted argument. Comparison to demographic variables is irrelevant, and the second sentence doesn't address the actual concern with the Science paper (i.e. testing in Brazil was rare at the time of the study, so people with Covid sought out the study). Regardless, I think it's a red herring. It doesn't matter if the sample was biased or not -- you can look at the data as collected, and see that the upward adjustments were so aggressive that the conclusion of the Science paper is likely to be wrong.