3 ms·

How did you calculate your 0.1%? NYC is 8,340,000 people[1]. They report 14,753 confirmed deaths, plus 5,178 probable[2]. That's 0.24%. Excess deaths (above exp

How did you calculate your 0.1%? NYC is 8,340,000 people[1]. They report 14,753 confirmed deaths, plus 5,178 probable[2]. That's 0.24%. Excess deaths (above expected baseline mortality) are yet higher. I don't believe a binary symptomatic/asymptomatic distinction is terribly useful. I agree that most cases that are tracked eventually report symptoms, just symptoms that are mild enough they wouldn't normally have sought medical care (in normal times, because they wouldn't bother; or now, because they're afraid to go to hospital). That seems entirely consistent with the serology studies. You originally said: > There are basically no (larger than a single building) places that can legitimately claim 5% or higher infection rates I believe that's very, very false--NYC's official numbers are around 4x that already, and those imply an IFR consistent with (and even higher than) estimates like Verity's. To claim even 2% IFR seems borderline-irresponsibly high to me, let alone the >4% that implied. If you post links to the studies you mentioned, I can take a look. I believe the Massachusetts one you mention is Chelsea, which indeed found a very high rate; but subjects were approached in a public place (thus oversampling people with higher-risk behavior), in a particularly hard-hit neighborhood. So I'd guess its results are correct for the people sampled, but not representative of the overall city. Have you studied how the specificity of antibody tests is validated, like on blood banked before the emergence of the virus? Why don't you think that validation is adequate? I agree that the correctly-calculated confidence interval for something like the Santa Clara study is near-uselessly wide, but I don't see any reason to doubt serology from harder-hit areas. 1. https://www1.nyc.gov/site/planning/planning-level/nyc-population/nyc-population.page https://www1.nyc.gov/site/planning/planning-level/nyc-popula... 2. https://www1.nyc.gov/site/doh/covid/covid-19-data.page https://www1.nyc.gov/site/doh/covid/covid-19-data.page

joshuamorton 6y ago
> Have you studied how the specificity of antibody tests is validated, like on blood banked before the emergence of the virus? Why don't you think that validation is adequate? I agree that the correctly-calculated confidence interval for something like the Santa Clara study is near-uselessly wide, but I don't see any reason to doubt serology from harder-hit areas. Yes, my understanding is two-fold. One, that independent verification (like https://covidtestingproject.org/ https://covidtestingproject.org/) found less specificity than the manufacturer reported results. And it found that some tests had false positive rates of well over 10%, which is just completely useless. The Santa Clara study is useless with pretty much any test, but even some of the NYC numbers have error bars down to zero, depending on the serology test used. > in a particularly hard-hit neighborhood. So I'd guess its results are correct for the people sampled, but not representative of the overall city. I can and do also totally believe conclusions like this are possible: I'm sure that there are population subsets that have achieved near total immunity. I don't particularly doubt the conclusions of the Boston homeless study, for example, although I expect if we check back in on them, many will have reported symptoms and some will be hospitalized. But again, drawing conclusions about "we should reopen" from that kind of thing is irresponsible. > I believe that's very, very false--NYC's official numbers are around 4x that already, and those imply an IFR consistent with (and even higher than) estimates like Verity's. To claim even 2% IFR seems borderline-irresponsibly high to me, let alone the >4% that implied. It looks like you're correct, the data I was looking at was wrong/outdated/I can't read or something. Given that, I would agree that NYC basically has to have 5% infected, and 10% is reasonable, although 20 and 25% I'd still have doubts about (and even more doubts about them given that the data is from ~3 weeks ago). So yes, I'd agree that an IFR of 4% plus is irresponsible to suggest, and I didn't intend to do so.
- tripletao 6y ago
  I agree that testing project is providing valuable information. A few of the tests were indeed quite bad; but I take that as a reason to look carefully at the particular test used, not to dismiss the whole field. 6/9 of their lateral flow assays were >98% specific for IgG. Of course, the bottom of the confidence interval is lower; but the top is higher, and I don't see what would make the low side more likely than the high side. This is unlike e.g. a typical drug study, where your prior should be strongly biased to the drug not working, since most drugs don't. Perhaps your prior is biased here; but I don't see why, since the serology is roughly consistent with the Diamond Princess and other prior results (like per Verity). NYC appears to have used an in-house IgG assay for which unfortunately no paper exists; but the Wadsworth Center seems reputable, with no obvious incentive to overstate the prevalence. To any extent they're under political pressure, that would probably go the other way (to show the success of the lockdown, and justify its continued effect). So perhaps they made a simple mistake, but their IFR ends up consistent with serology performed by other means in Europe, and with the few populations universally tested by PCR--actually on the high end, especially if we include probables or use full excess death. The harm of the coronavirus certainly shouldn't be minimized. The harm of the continuing lockdown shouldn't be either--a generation of schoolchildren may lose months of public education, one of very few social equalizers in American society. Given the harm on both sides, I believe we should be making policy based on our best estimates of the IFR, and not either extreme of a confidence interval. That seems unquestionably <2% to me, and probably <1% based on European serology (though the NYC may be slightly higher).

joshuamorton 6y ago

> Have you studied how the specificity of antibody tests is validated, like on blood banked before the emergence of the virus? Why don't you think that validation is adequate? I agree that the correctly-calculated confidence interval for something like the Santa Clara study is near-uselessly wide, but I don't see any reason to doubt serology from harder-hit areas. Yes, my understanding is two-fold. One, that independent verification (like https://covidtestingproject.org/ https://covidtestingproject.org/) found less specificity than the manufacturer reported results. And it found that some tests had false positive rates of well over 10%, which is just completely useless. The Santa Clara study is useless with pretty much any test, but even some of the NYC numbers have error bars down to zero, depending on the serology test used. > in a particularly hard-hit neighborhood. So I'd guess its results are correct for the people sampled, but not representative of the overall city. I can and do also totally believe conclusions like this are possible: I'm sure that there are population subsets that have achieved near total immunity. I don't particularly doubt the conclusions of the Boston homeless study, for example, although I expect if we check back in on them, many will have reported symptoms and some will be hospitalized. But again, drawing conclusions about "we should reopen" from that kind of thing is irresponsible. > I believe that's very, very false--NYC's official numbers are around 4x that already, and those imply an IFR consistent with (and even higher than) estimates like Verity's. To claim even 2% IFR seems borderline-irresponsibly high to me, let alone the >4% that implied. It looks like you're correct, the data I was looking at was wrong/outdated/I can't read or something. Given that, I would agree that NYC basically has to have 5% infected, and 10% is reasonable, although 20 and 25% I'd still have doubts about (and even more doubts about them given that the data is from ~3 weeks ago). So yes, I'd agree that an IFR of 4% plus is irresponsible to suggest, and I didn't intend to do so.

tripletao 6y ago
I agree that testing project is providing valuable information. A few of the tests were indeed quite bad; but I take that as a reason to look carefully at the particular test used, not to dismiss the whole field. 6/9 of their lateral flow assays were >98% specific for IgG. Of course, the bottom of the confidence interval is lower; but the top is higher, and I don't see what would make the low side more likely than the high side. This is unlike e.g. a typical drug study, where your prior should be strongly biased to the drug not working, since most drugs don't. Perhaps your prior is biased here; but I don't see why, since the serology is roughly consistent with the Diamond Princess and other prior results (like per Verity). NYC appears to have used an in-house IgG assay for which unfortunately no paper exists; but the Wadsworth Center seems reputable, with no obvious incentive to overstate the prevalence. To any extent they're under political pressure, that would probably go the other way (to show the success of the lockdown, and justify its continued effect). So perhaps they made a simple mistake, but their IFR ends up consistent with serology performed by other means in Europe, and with the few populations universally tested by PCR--actually on the high end, especially if we include probables or use full excess death. The harm of the coronavirus certainly shouldn't be minimized. The harm of the continuing lockdown shouldn't be either--a generation of schoolchildren may lose months of public education, one of very few social equalizers in American society. Given the harm on both sides, I believe we should be making policy based on our best estimates of the IFR, and not either extreme of a confidence interval. That seems unquestionably <2% to me, and probably <1% based on European serology (though the NYC may be slightly higher).