5 ms·

That's a lot of interpretation from 6 cases. There's a lot of serological testing being done right now, and we'll have hard data soon, but from what I see from

That's a lot of interpretation from 6 cases. There's a lot of serological testing being done right now, and we'll have hard data soon, but from what I see from expert epidemiologists (disclosure: I'm not one myself), the idea that we have a vast number of completely asymptomatic cases is wishful thinking.

JetSetWilly 7y ago
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 7y ago
  edit: removed for potentially being misleading.
  - JetSetWilly 7y ago
    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.
    - f38zf5vdt 7y ago
      Yeah, I read more into their assay and I'm inclined to agree with you.
    - tgb 7y ago
      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 7y ago
        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 7y ago
        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.
- jdjrirkfkfk 7y ago
  Doesnt poisson distribution require no correlation? Is it correct to use it for an epidemic with chains of transmission?
raphlinus 7y ago
Here's an expert thread (for example): https://twitter.com/richardneher/status/1250533354560159745 https://twitter.com/richardneher/status/1250533354560159745

JetSetWilly 7y ago

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 7y ago
edit: removed for potentially being misleading.
- JetSetWilly 7y ago
  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.
  - f38zf5vdt 7y ago
    Yeah, I read more into their assay and I'm inclined to agree with you.
  - tgb 7y ago
    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 7y ago
      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 7y ago
        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.
jdjrirkfkfk 7y ago
Doesnt poisson distribution require no correlation? Is it correct to use it for an epidemic with chains of transmission?

raphlinus 7y ago

Here's an expert thread (for example): https://twitter.com/richardneher/status/1250533354560159745 https://twitter.com/richardneher/status/1250533354560159745