4 ms·

> Look at: recorded deaths, excess deaths, hospital admissions, ICU admissions, serological studies. Please don't look at serological studies. Many of the sero

> Look at: recorded deaths, excess deaths, hospital admissions, ICU admissions, serological studies. Please don't look at serological studies. Many of the serological studies have frighteningly high false-postive rates that can lead to faulty assumptions that 20% or 50% of people have been infected already. This pretty much isn't true anywhere.

alkonaut 7y ago
So long as you keep it in mind, it’s just statistics. There have been studies with both good and bad tests. The test itself is probably easier to compensate for in a confidence interval, than selection bias which is usually unknown.
- joshuamorton 7y ago
  The problem is that for most serological studies, a correctly calculated 90% confidence interval includes 0%. That's not helpful.
  - alkonaut 7y ago
    Sero studies that take place where the expected result is <5% or where tests use sensitivity/specificty under 95% are indeed not helpful. But in those places perhaps it's too soon to even be doing sero studies. Look at hospital admissions there to track the evolution of the epidemic. Once you have at least 10 or 20% immunity sero studies can be used with some confidence, and there are now a couple of tests that claim 98% or better sens/spec, so using the earlier poor ones should be avoided now.
    - joshuamorton 7y ago
      There are basically no (larger than a single building) places that can legitimately claim 5% or higher infection rates though, which is the point.
      - alkonaut 7y ago
        Several cities are comfortably over 10% now even with pessimistic estimates of immunity/IFR. NYC and Stockholm are two examples. Even New York State as a whole is surely over 5% now.
        joshuamorton 7y ago
        No they can't. You can browse my comment history for why. Tldr we'd need to see much higher positive rates among the viral tests to support that theory, and we don't.
        alkonaut 7y ago
        I don’t understand. Not from your comment history either. What’s the problem disqualifying all studies? The Stockholm one is the one I’m most familiar with. It used a test with specificity over 98%. Predictive power depends on the prevalence, but as the floor of immunity you can use any PCR study for active infection. In the case of Stockholm there was 2.5% active infection over a month before the sero study that showed 10%. Obviously if a poor test with 85% or 90% specificity is used and you have low prevalence then you get a poor result. But if you can peg the floor of the prevalence at several percent and use an accurate test then what’s making the results useless? But even disregarding sero studies, you can draw some conclusions of immunity from death numbers. E.g if assuming a 0.5% IFR with 1600 dead in 2100000 you would have 15% immunity (assuming everyone infected is immune of course). IFRs so high that they yield under 5% immune (basically IFR over 1.5%) aren’t consistent with observation.
        joshuamorton 7y ago
        > IFRs so high that they yield under 5% immune (basically IFR over 1.5%) aren’t consistent with observation. But they are. The IFR in places like China, South Korea, and NZ that have things mostly under control at this point, the CFR is consistently above 2% (and as high as 5%), and those places are and were doing significant viral testing, so wouldn't have missed 50% or 150% of the infected. So a CFR of 2% or 1.5% isn't unreasonable (unless you're relying on sero studies as "observation"). Unless you're assuming that there's a huge swath of people (like the significant majority) of Covid-19 cases that are entirely asymptomatic (and we don't have any reason to believe this, we have upper bounds on pre-symptomatic cases at like 50% from randomized viral presence studies, and the asymptomatic numbers should be notably lower), the IFR and CFR should be reasonably close, in places that have things under control and that makes 1.5% IFR more likely than .5%. > In the case of Stockholm there was 2.5% active infection over a month before the sero study that showed 10%. Source for this? Sweden has 26,322 confirmed cases today. You're saying that 24000 of those were confirmed in Stockholm well over a month ago?

alkonaut 7y ago

So long as you keep it in mind, it’s just statistics. There have been studies with both good and bad tests. The test itself is probably easier to compensate for in a confidence interval, than selection bias which is usually unknown.

joshuamorton 7y ago
The problem is that for most serological studies, a correctly calculated 90% confidence interval includes 0%. That's not helpful.
- alkonaut 7y ago
  Sero studies that take place where the expected result is <5% or where tests use sensitivity/specificty under 95% are indeed not helpful. But in those places perhaps it's too soon to even be doing sero studies. Look at hospital admissions there to track the evolution of the epidemic. Once you have at least 10 or 20% immunity sero studies can be used with some confidence, and there are now a couple of tests that claim 98% or better sens/spec, so using the earlier poor ones should be avoided now.
  - joshuamorton 7y ago
    There are basically no (larger than a single building) places that can legitimately claim 5% or higher infection rates though, which is the point.
    - alkonaut 7y ago
      Several cities are comfortably over 10% now even with pessimistic estimates of immunity/IFR. NYC and Stockholm are two examples. Even New York State as a whole is surely over 5% now.
      - joshuamorton 7y ago
        No they can't. You can browse my comment history for why. Tldr we'd need to see much higher positive rates among the viral tests to support that theory, and we don't.
        alkonaut 7y ago
        I don’t understand. Not from your comment history either. What’s the problem disqualifying all studies? The Stockholm one is the one I’m most familiar with. It used a test with specificity over 98%. Predictive power depends on the prevalence, but as the floor of immunity you can use any PCR study for active infection. In the case of Stockholm there was 2.5% active infection over a month before the sero study that showed 10%. Obviously if a poor test with 85% or 90% specificity is used and you have low prevalence then you get a poor result. But if you can peg the floor of the prevalence at several percent and use an accurate test then what’s making the results useless? But even disregarding sero studies, you can draw some conclusions of immunity from death numbers. E.g if assuming a 0.5% IFR with 1600 dead in 2100000 you would have 15% immunity (assuming everyone infected is immune of course). IFRs so high that they yield under 5% immune (basically IFR over 1.5%) aren’t consistent with observation.
        joshuamorton 7y ago
        > IFRs so high that they yield under 5% immune (basically IFR over 1.5%) aren’t consistent with observation. But they are. The IFR in places like China, South Korea, and NZ that have things mostly under control at this point, the CFR is consistently above 2% (and as high as 5%), and those places are and were doing significant viral testing, so wouldn't have missed 50% or 150% of the infected. So a CFR of 2% or 1.5% isn't unreasonable (unless you're relying on sero studies as "observation"). Unless you're assuming that there's a huge swath of people (like the significant majority) of Covid-19 cases that are entirely asymptomatic (and we don't have any reason to believe this, we have upper bounds on pre-symptomatic cases at like 50% from randomized viral presence studies, and the asymptomatic numbers should be notably lower), the IFR and CFR should be reasonably close, in places that have things under control and that makes 1.5% IFR more likely than .5%. > In the case of Stockholm there was 2.5% active infection over a month before the sero study that showed 10%. Source for this? Sweden has 26,322 confirmed cases today. You're saying that 24000 of those were confirmed in Stockholm well over a month ago?