I find this immensely frustrating

There is actually not really any evidence at all that serology surveys are 'flawed' or 'dramatically underestimate' how many people have had COVID-19

Sigh

This headline is based on an editorial in the BMJ. So, firstly, not a study per se, this is a few academics' opinions about something (granted, they sound very clever)

https://www.bmj.com/content/370/bmj.m3364

Thing is, the editorial is, well, pretty speculative. They are certainly not making any hard claims here, just raising the potential possibility that serology tests are missing mild/asymptomatic cases of COVID-19

But even there, I actually disagree with the authors of this editorial quite a bit

For example, they cite a study conducted in Luxembourg as evidence that IgA serology tests would pick up more people than the more commonly-used IgG

This sounded odd to me, because I've read that Luxembourg paper - it's in my review - and I couldn't remember it making this argument

Because it doesn't, really

In fact, the Luxembourg paper DID find that more people had IgA responses than IgG responses, but this was expected because the IgA test WAS NOT AS GOOD

The specificity was VERY low for just IgA, meaning there were many false positives

In fact, if you correct for test characteristics, you find that the IgA test predicts a seroprevalence of ~1%

Not that different after all!

Another study the authors cite as supportive of the idea that seroprevalence studies are missing loads of patients is this Austrian study that found a seroprevalence of 42% in an outbreak location

The authors of the editorial say that this is a red flag, because other serosurveys of Austria have found lower seroprevalence but used only IgG testing, so it's the IgA tests that are pushing up this number (and therefore many IgG-only studies are wrong)

Except, this doesn't make any sense whatsoever

The seroprevalence in Ischgl was 42% because it was the epicenter of the entire COVID-19 outbreak in Austria and indeed Europe!

Moreover, if you actually look at the IgA vs IgG positives in the study, you'll see that there are only a tiny fraction (~4%) who are IgA positive but not IgG positive

Completely contradicting the editorial

So, the evidence that this editorial cites in favor of its argument appears to completely contradict the main point, which is...not great

And yet, reported as fact

As I said, sigh

More broadly, we know that the antibody tests for COVID-19 are pretty good, and probably don't miss a huge proportion of people who have actually had the disease https://www.bmj.com/content/370/bmj.m2516

I wrote this up in a rapid response to the article which has now been published https://www.bmj.com/content/370/bmj.m3364/rapid-responses