[JetSetWilly]

Ah, but if you do a sample of 1000 then even low numbers of positives are still statistically significant. The poisson distribution on just 6 might be somewhat wide but you can give a reasonable confidence interval (ie 0.3-0.9%) - and either end of that interval still suggests a lot of undetected cases at a time when there was hardly any officially.

What you can't do - statistically - is say "oh there were just 6, that means it all could just be a coincidence and the prevlence is actually very low".

Especially when - if you dig into the numbers they make sense. First, they were spread over a 2 week period, with all the positives being in the second cohort - which is what is expected if there's an exponential explosion of cases going on. Secondly, 4 of the 6 positives are in the Edinburgh area which is the most internationally connected and affluent area of Scotland.

So I really don't think it can be breezily dismissed just because you don't like the data. But as you say, we'll have a lot more serological data before long so lets see.

[f38zf5vdt]

edit: removed for potentially being misleading.

[JetSetWilly]

Cross-reactivity is an issue in the ELISA test (the actual antibody test).

But they validated their results with a second step, the "pseudotype neutralisation assay". This test is highly specific for SARS-CoV-2 - as I understand it, they essentially introduce the virus to the hypothetically immune blood samples and then test that it is indeed neutralised by antibodies.

So, I think this Scottish serology test is pretty robust and accounts for these issues with cross-reactivity and specificity of the testing. In addition - they used a control sample from December 2019 none of which tested positive.

[tgb]

0 out of 100 controls and 6 out of 500 gives a very wide confidence interval. You could easily get those numbers with 1% false positives and no true positives in either group. Is your prior for false positives low enough to rule that out?

[JetSetWilly]

They did two independent assays so I would think the false positives rate should be far below 1% - certainly the authors seem to think so.

Secondly if they were false positives, I would expect them to be randomly distributed. Not concentrated in the affluent capital city, and not solely in the second cohort with none in the early march cohort (as you would predict if exponential growth is occurring).

[tgb]

Why even test 100 controls if your conclusion relies on a prior that false-positives are much less than 1%? I think that shows the author's are worried about false-positives and have failed to rule them out sufficiently.

Your second two points are fairly compelling.

[f38zf5vdt]

Yeah, I read more into their assay and I'm inclined to agree with you.