Covid-19 Antibody Seroprevalence in Santa Clara County, California • Hacker News Board

Re: Covid-19 Antibody Seroprevalence in Santa Clara County, California

#171

Post by Gibbon1 » Fri, Apr 17, 2020, 11:24 PM UTC

ysleepy wrote: post #62

microdrum wrote: post #47

Earlier quoted context omitted.

Yep. You've got nothing. CFR is low, and this has been knowable for a long time. You're scared. It's OK. It's illogical and anti-scientific, but it's OK. But you won't be able to hurt ordinary people any longer. Here's Germany's imputed CFR: https://spectator.us/covid-antibody-test-german-town-shows-1... Here's Denmark: https://www.dr.dk/nyheder/indland/doedelighed-skal-formentli... Here's Iceland: https://reason.com…

You are oversimplifying. While these numbers are indicators, they are non-representative samples. Different communities spread the disease to different demographics. A CFR of 0.4% is still huge, especially if it is very non-uniform and would mean 1.2mio deaths in the US if it saturates. And if that isnt enough for you, we still don't have reliable information about long term effects of an infection. It could end up b…

CFR rate of 0.4% probably also means 2-4% lingering morbidity rate. Friend of mine almost died of sepsis. And it wrecked her health.

Re: Covid-19 Antibody Seroprevalence in Santa Clara County, California

#172

Post by usaar333 » Fri, Apr 17, 2020, 11:28 PM UTC

cameldrv wrote: post #153

guenthert wrote: post #136

Earlier quoted context omitted.

With the average adult getting two bouts of common cold each year and mild COVID symptoms resembling a common cold, I'd think a large fraction of the population might think they had it. I wouldn't dear to guess in which way this shifts the estimate. I think it's understood, that with such a small sample size, results are to be taken cautiously. You have to start somewhere though.

There was another serological survey of blood donors in Castiglione d'Adda in Italy. That one came up 70% positive. That town had 1.4% of its population die in March. That suggests an IFR of 2%, not counting people who were sick and might die from COVID in April.

We do need to be careful talking about "IFR" as with a disease with such high age-dependence demographic skews heavily alter population IFR.

Castiglione d'Adda has more people over 70 than Santa Clara County has over 65; population over 65 is about 60% higher relatively. Population under 18 conversely is about 22% lower.

Hard to do the math exactly, but if you simply switch 8.1% of your population from being children to being 80+, you raise IFR by 0.6% per the Imperial College China estimates. Combined with the hospital triaging Italy was doing, I don't think sub-1% IFRs in Santa Clara county are improbable. (though I do think this survey's claims are improbably low)

sources: https://www.citypopulation.de/php/italy-localities-lombardia...

https://www.census.gov/quickfacts/santaclaracountycalifornia

https://www.thelancet.com/journals/laninf/article/PIIS1473-3...

Re: Covid-19 Antibody Seroprevalence in Santa Clara County, California

#173

Post by mlyle » Fri, Apr 17, 2020, 11:45 PM UTC

irq11 wrote: post #143

cameldrv wrote: post #103

Earlier quoted context omitted.

Yes, and look at the way they corrected for this. They rebalanced for demographics, and that doubled their estimate. This is exactly the opposite of what they should have done. For example, zip codes farther from the testing sites were less likely to get tested, and more likely to be positive. What you see is that certain groups were less likely to go get tested, UNLESS they had a good reason to think that they were…

That is just nonsense. You don’t adjust for demographic bias in a sample by conditioning on a second parameter. There’s no universe in which underweighting the underweighted group makes any sense at all. Moreover, there isn any support for the idea that “bored white women” are more or less likely to get tested for symptomatic illness than anyone else. Your entire argument boils down to “I don’t like the implications…

I do think it's likely the study does overreport the infection rate. People who are really curious about their past COVID-19 status are not representative of the population as a whole. And stratified sampling can help to try and correct this, but... we all know that you can't really "save" or "fully condition" a convenience sample.

On the other hand, it's hard to reconcile their finding with anything less than 90% or more of infections being missing from the case count. And 10% of the infection fatality rate is really exciting news. It also means that it's exceptionally likely that jurisdictions like New York are a huge slice of the way (15% of population infected?) to herd immunity.

Re: Covid-19 Antibody Seroprevalence in Santa Clara County, California

#174

Post by mlyle » Fri, Apr 17, 2020, 11:50 PM UTC

JamesBarney wrote: post #164

Fjolsvith wrote: post #120

Earlier quoted context omitted.

What if there are other, undiscovered factors that could explain the difference? Such as NYC populace having had exposure to some other virus that predisposed them to a greater vulnerability. Or perhaps, air pollution exposure predisposing them to greater vulnerability.

We have two possible hypotheses. There is an unknown quality about New York that causes the virus to be 4 times more severe. A self-selected group from the internet demonstrated self-selection sampling bias. I know where I would put my money.

I think there's significant error bars all around. Note that the New York data is older than the SCC data, so there's delay related effects, too.

We can quibble a whole lot with exact effect magnitudes and data analysis techniques, but...

In any case, we have a whole lot of data (Iceland RTPCR, Gangelt RTPCR, Switzerland RTPCR, Netherlands blood donation serology, Santa Clara County serology) that indicates infection rates likely exceed case rates by 10x or more in areas with relatively good testing availability.

Re: Covid-19 Antibody Seroprevalence in Santa Clara County, California

#175

Post by hcknwscommenter » Fri, Apr 17, 2020, 11:50 PM UTC

usaar333 wrote: post #63

sxp wrote: post #14

There are similar high infection rates from other semi-randomized samples. https://www.nejm.org/doi/full/10.1056/NEJMc2009316 > Between March 22 and April 4, 2020, a total of 215 pregnant women delivered infants at the New York–Presbyterian Allen Hospital and Columbia University Irving Medical Center . All the women were screened on admission for symptoms of Covid-19. Four women (1.9%) had fever or other symptoms of…

NYC has 30x the per capita death rate of Santa Clara county. Santa Clara county only being 15% as infected is not consistent.

Isn't it consistent though? There is a confidence interval associated with all of these numbers and the intervals are quite wide.

Re: Covid-19 Antibody Seroprevalence in Santa Clara County, California

#176

Post by SubiculumCode » Fri, Apr 17, 2020, 11:50 PM UTC

guenthert wrote: post #136

cameldrv wrote: post #103

Earlier quoted context omitted.

Yes, and look at the way they corrected for this. They rebalanced for demographics, and that doubled their estimate. This is exactly the opposite of what they should have done. For example, zip codes farther from the testing sites were less likely to get tested, and more likely to be positive. What you see is that certain groups were less likely to go get tested, UNLESS they had a good reason to think that they were…

With the average adult getting two bouts of common cold each year and mild COVID symptoms resembling a common cold, I'd think a large fraction of the population might think they had it. I wouldn't dear to guess in which way this shifts the estimate. I think it's understood, that with such a small sample size, results are to be taken cautiously. You have to start somewhere though.

3,330 tests isn't generally considered a small sample

Re: Covid-19 Antibody Seroprevalence in Santa Clara County, California

#177

Post by SubiculumCode » Sat, Apr 18, 2020, 12:00 AM UTC

cameldrv wrote: post #152

irq11 wrote: post #143

Earlier quoted context omitted.

That is just nonsense. You don’t adjust for demographic bias in a sample by conditioning on a second parameter. There’s no universe in which underweighting the underweighted group makes any sense at all. Moreover, there isn any support for the idea that “bored white women” are more or less likely to get tested for symptomatic illness than anyone else. Your entire argument boils down to “I don’t like the implications…

Not in this case. It's basically Simpson's paradox. Honestly I think the study is basically useless, because instead of a random sample, it's a sample of people who, when they saw an ad on Facebook for a free Coronavirus test, decided to sign up, and then actually show up for the test. The point is that the higher your (voluntary) participation rate is within a group, the less chance you have of selection bias. This…

3,300 people signed up for this study and completed this within weeks. Have you recruited for a study before? That kind of participation is hard to achieve...practically unheard of, unless you have widespread interest and time: Something which a whole lot of people have right now given layoffs, fear, and the bit of monetary compensation for participation, I bet sampling was much wider and less biased than you think.

Re: Covid-19 Antibody Seroprevalence in Santa Clara County, California

#178

Post by Animats » Sat, Apr 18, 2020, 12:01 AM UTC

danieltillett wrote: post #154

Animats wrote: post #144

Earlier quoted context omitted.

There’s still a long way to go in that case. Yes. Stanford's recent study says 2-4% of the population of Santa Clara County has been infected. Those are probably reasonable bounds. Herd immunity is around 80% for this. So about 20 to 40 times as many people need to die before it's over. That test needs to be repeated weekly to get the growth rate. (Preferably not with the scheme from Verity, which requires that you s…

The problem with herd immunity to this virus is how short the immunity to corona virus is - around a year. If we wait for herd immunity to build from low level natural spread then we will never get there. There is a very major risk with extrapolating the Santa Clara data in the motivation of those being tested has not been controlled. Who is more likely to want to know they have been infected - those who have had cla…

How short the immunity to corona virus is - around a year.

Source? Right now, that seems unknown - too soon to tell. A few months of measuring antibody levels and we'll know more.

Re: Covid-19 Antibody Seroprevalence in Santa Clara County, California

#179

Post by mlyle » Sat, Apr 18, 2020, 12:03 AM UTC

Medicalidiot wrote: post #129

jacquesm wrote: post #101

Earlier quoted context omitted.

They use the correct tests, but they use them on a group that isn't even close to a random sample of the population, which is what you'd need to do to get to their conclusion.

I'm not going to comment on that because I have a basic understanding of public health and epidemiology on the diagnostic side of things. Why do you think that this isn't a "good" random sample?

It's not a random sample; it's a convenience sample of people who responded to Facebook ads.

Re: Covid-19 Antibody Seroprevalence in Santa Clara County, California

#180

Post by cameldrv » Sat, Apr 18, 2020, 12:24 AM UTC

SubiculumCode wrote: post #177

cameldrv wrote: post #152

Earlier quoted context omitted.

Not in this case. It's basically Simpson's paradox. Honestly I think the study is basically useless, because instead of a random sample, it's a sample of people who, when they saw an ad on Facebook for a free Coronavirus test, decided to sign up, and then actually show up for the test. The point is that the higher your (voluntary) participation rate is within a group, the less chance you have of selection bias. This…

3,300 people signed up for this study and completed this within weeks. Have you recruited for a study before? That kind of participation is hard to achieve...practically unheard of, unless you have widespread interest and time: Something which a whole lot of people have right now given layoffs, fear, and the bit of monetary compensation for participation, I bet sampling was much wider and less biased than you think.

Yeah but most studies aren't testing for a virus that has everyone locked in their houses causing a global pandemic.