Covid-19 Antibody Seroprevalence in Santa Clara County, California | Hacker News Reader

Covid-19 Antibody Seroprevalence in Santa Clara County, California

medrxiv.org

122 pointsbuboard200 comments

Caveat from the paper:

"This study had several limitations. First, our sampling strategy selected for members of Santa Clara County with access to Facebook and a car to attend drive-through testing sites. This resulted in an over-representation of white women between the ages of 19 and 64, and an under-representation of Hispanic and Asian populations, relative to our community. Those imbalances were partly addressed by weighting our sample population by zip code, race, and sex to match the county. We did not account for age imbalance in our sample, and could not ascertain representativeness of SARS-CoV-2 antibodies in homeless populations. Other biases, such as bias favoring individuals in good health capable of attending our testing sites, or bias favoring those with prior COVID-like illnesses seeking antibody confirmation are also possible. The overall effect of such biases is hard to ascertain."

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 positive. This should cause them to want to underweight the data from the underrepresented groups, but instead they overweighted it.

You can almost say that there were two populations that they were testing: White women who lived near a testing site who thought "what the hell, I'm bored with this quarantine so might as well get my finger pricked", and people who thought "I wonder if that thing I had a few weeks ago was COVID." The number you want is the prevalence from the bored white women. Bored white women might be somewhat skewed from other demographics, but probably not by that much. People who thought they might have had it is a massively skewed group. Effectively their correction removes the bored white women near a testing site, and only counts the people who have reason to think they had COVID.

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...

I agree that you have to make an age adjustment for IFR, and your numbers seem reasonable. However, in the U.S. you also have a lot more diabetes, obesity, and high blood pressure, which are major risk factors. I wouldn't be surprised if that required a bigger adjustment than the age difference.

Agreed that US as a whole can have a high IFR.

Santa Clara county though is one of the healthiest places in the US. Life expectancy of 84 exceeds Italy in fact.

Given that testing is still limited and you have to really make a case for yourself to get it, the group of people who have been tested (not for the antibodies, for infection) is at least a lower bound on the people who think they had it. And statewide we have been at around 12% positives, so it's reasonable to say at least 88% of people who think they had it didn't.

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

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 of their adjustment, and this other one would have produced results more consistent with my theories.”

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 goes up to the limit of 100% participation, where you have no selection bias.

What we see in this study is that groups with a lower participation rate (Men, nonwhites, people farther from a testing site) have a higher positivity rate. That is exactly the selection bias we would expect. They then proceed to "correct" the data by undercounting the group (white women that live near a testing site) that has the least selection bias.

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.

But that is the point. Selection bias towards those that are actually infected is minimized when a large portion of the population have the time and desire to get tested.

No his argument is that there are two sampling biases.

The demographics of a tested population do not match the demographics of the county.

If someone believes they've been infected they are more likely to seek testing. (self-selection bias)

The study adjusted for the first but not the second bias. He believes the second bias is more important, so when they adjusted for the first they moved the adjusted average farther away from the true average.

I think he's right, self-selection can be a far more impactful bias than things like sex and race.

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.

I agree that cases are probably giving maybe a 10-20x undercount. That means that NYC has maybe 3 million infections. That means NYC is maybe half way to herd immunity, assuming reinfection doesn't happen and that immunity is lasting.

The only problem is that to get the other half way there, another 20,000 people have to die, just in New York City. That doesn't even count the already existing infections that haven't died yet.

Maybe this time we can do a better job of concentrating the infections in the less-likely-to-die group, and cut down on the deaths a little that way.

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 Covid-19 on admission, and all 4 women tested positive for SARS-CoV-2 (Figure 1). Of the 211 women without symptoms, all were afebrile on admission. Nasopharyngeal swabs were obtained from 210 of the 211 women (99.5%) who did not have symptoms of Covid-19; of these women, 29 (13.7%) were positive for SARS-CoV-2. Thus, 29 of the 33 patients who were positive for SARS-CoV-2 at admission (87.9%) had no symptoms of Covid-19 at presentation.

So 15% infection rate among pregnant women. I would assume pregnant women would be more careful than average about avoiding getting infected. So overall infection rate in New York City must be even higher.

> So 15% infection rate among pregnant women. I would assume pregnant women would be more careful than average about avoiding getting infected.

Pregnant women would seem to be much less able to avoid certain activities that involve transmission (like visiting medical facilities and associated close contact with people who also work in hospitals with inadequate and improvised PPE), so that’s probably not a great assumption.

NYC has 30x the per capita death rate of Santa Clara county.

Santa Clara county only being 15% as infected is not consistent.

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.

Just a hypothesis, but if being more frequently exposed makes it more likely that the disease will be more severe, then this is plausible. There is some evidence that e.g. doctors/nurses were more likely to die compared to their age cohort, possibly because of higher viral load.

I don't think virologists thing frequency of exposure is very important to lethality. But there is evidence that dosage and stress play a roll which will both be higher in medical personnel.

I've definitely seen the opposite stated, that doctors/nurses are not actually more likely to die. I forget if it was Italy or NYC, but the fatality rate for doctors & nurses had been something like 0.5%. They are much more likely to be infected than anyone else, and there has (rightly) been lots written about the doctors and nurses who have died, which gives people this perception.

Take 30 times 0.15 you get 4.5. I wouldn't draw any conclusions due to the low quality of current data.

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

"We recruited participants by placing targeted advertisements on Facebook aimed at residents of Santa Clara County..."

I wonder if these ads mentioned the purpose of the study, or if that information was only given out after initial contact (I don't know enough about IRB requirements here).

A concern is that this could cause a selection bias for people who suspected they had the virus. People may respond to such an ad out of curiosity ("I think I had the virus, so it would be good to know for sure") or obligation ("I'm pretty sure I had the virus, so I should help out with this study").

It wouldn't have to be a large selection bias, either. Of the 3,330 people they tested, they found only 50 who tested positive.

I would like to see a bit of a better method of sample selection before drawing any conclusions.

Why would this matter as long as they got a good geographical sampling?

"We used Facebook to quickly reach a large number of county residents and because it allows for granular targeting by zip code and sociodemographic characteristics. We used a combination of two targeting strategies: ads aimed at a representative population of the county by zip code, and specially targeted ads to balance our sample for under-represented zip codes. In addition, we capped registrations from overrepresented areas."

Update: Okay, "Other biases, such as bias favoring individuals in good health capable of attending our testing sites, or bias favoring those with prior COVID-like illnesses seeking antibody confirmation are also possible."

If part of the value for participants was to know whether they had antibodies, then those that had gone through a flu (or covid) would be more likely to respond. This means that the overall sample likely includes more people who had COVID than a true random sampling would.

It looks like there's an argument to be made it could overcount or undercount depending on importance of each potential bias.

"This study had several limitations. First, our sampling strategy selected for members of Santa Clara County with access to Facebook and a car to attend drive-through testing sites. This resulted in an overrepresentation of white women between the ages of 19 and 64, and an under-representation of Hispanic and Asian populations, relative to our community. Those imbalances were partly addressed by weighting our sample population by zip code, race, and sex to match the county. We did not account for age imbalance in our sample, and could not ascertain representativeness of SARS-CoV-2 antibodies in homeless populations. Other biases, such as bias favoring individuals in good health capable of attending our testing sites, or bias favoring those with prior COVID-like illnesses seeking antibody confirmation are also possible. The overall effect of such biases is hard to ascertain."

Yes, the ads accurately described the study. Several friends of mine participated (I live in Santa Clara County).

Of course there is selection bias, which is why they make adjustments to make the sample better represent the population (see the abstract for details on this, or read the paper for all the details, there's well-established science behind these adjustments and it's up to the reviewers to be sure that they did it correctly). The point is that "they only found 50" is quite a large number compared to the number of confirmed cases.

What adjustments were made to account for selection bias of the sort described? It's not covered in the paper, and "well-established science" certainly doesn't answer the question.

These are very common statistical methods when dealing with surveys and populations. It isn't this paper's job to describe the background on something like that, just like it isn't this papers job to explain blood testing.

They describe their limitations and adjustments exactly like most other studies of this sort do. For example:

"Those imbalances were partly addressed by weighting our sample population by zip code, race, and sex to match the county. "

They said that they collected prior clinical symptoms in the survey, but have not mentioned that they adjusted for those.

Given that, I think it's fair to say that they have not adjusted for the selection bias of people with prior clinical symptoms being more likely to click the ad.

On the contrary - detecting selection bias of this sort can be very difficult. Do you measure it using a questionnaire? You might ask your subjects "Do you think you previously had COVID-19?" But what do you do with that response data? Exclude everyone that says yes? Then you're skewing one way. There's obviously no population data to match that response with like there is for general demographics such as race and age.

Many studies are only as good as their sampling methods. A good sampling method will alleviate concerns of selection bias.

I think you'd call individuals(using a method with far less self selection bias) and get them to fill out the same questionnaire and then do the same types of adjustments you'd do with race or gender.

But I also agree they 100% did not do any of these adjustments if they did it would be listed right here

> We report the prevalence of antibodies to SARS-CoV-2 in a sample of 3,330 people, adjusting for zip code, sex, and race/ethnicity. We also adjust for test performance characteristics using 3 different estimates: (i) the test manufacturer's data, (ii) a sample of 37 positive and 30 negative controls tested at Stanford, and (iii) a combination of both.

Adjusting for selection bias is really hard and potentially first-order.

Correct. For some things you could use baseline data, but there is zero baseline data to use here.

That reweighting doesn't address the most troubling selection bias: people choosing to participate because they want to get tested (e.g., because they think they might have had the disease).

In the data from Italy, ~14% of the >1M tests done for the virus (not antibodies) were positive. That prob includes ppl tested multiple times, or after recovery.

I wonder why this is so vastly different from the study above.

https://en.wikipedia.org/wiki/2020_coronavirus_pandemic_in_I...

There is also selection bias in whom Facebook will show the ads to due to Facebook's CTR models, other bidders' bids, and who is browsing Facebook more to see more impressions in general. They should have worked with an ad platform to run the ad as a PSA so they could use the ad platform's own randomization instead of relying on having the auctions they win result in anywhere close to a usable sample.

It’s hacky and I’m happy to be told how wrong it is. But based on these recent antibody studies in Germany, Finland, and now here in CA, I’ve been assuming an actual fatality rate of about 0.4, and thus an actual infected rate of 250x our known deaths, which is a much firmer number.

This is only a small comfort since it means we may have had about 8.75 million infected and presumably now immune in the USA. Or about 2.6%. There’s still a long way to go in that case.

It's known that the vast majority infected do not experience symptoms, or have mild symptoms. I don't think you can draw the conclusion that they had prior immunity.

Not concluding prior immunity but concluding that they now have immunity after having been infected. That is in progress towards herd immunity this number of people actually infected being revealed by these antibody studies is what is interesting. And based on my hacky heuristic described above that is currently around 2.6%. So we are still a long way off herd immunity. But at the same time closer than we thought we were.

> So we are still a long way off herd immunity. But at the same time closer than we thought we were.

Yes and no.

A disease is endemic when R0 x S = 1 [0]. R0 for SARS-CoV-2 is estimated to be 2.5 to 3.5 and some models predicting a much higher 5.7 [1].

The herd immunity threshold is given by (1 - S)%, which implies when R0 = 2.5, 60% of the population would need to be immune; similarly 71.42% and 82.45% for R0s, 3.5 and 5.7, respectively.

[0] https://en.wikipedia.org/wiki/Endemic_(epidemiology)

[1] https://news.ycombinator.com/item?id=22817942

What percentage of immunity can we expect to see a slowdown in the spread of the disease? While we may need 60% or more to achieve full herd immunity surely somewhere lower than that we can expect to see a noticeable reduction in the spread?

> ...noticeable reduction in the spread?

I believe the SIR model [0] answers those questions. 3blue1brown's video on the topic is super digestible and informative [1].

In short, the discovery of a prophylactic if not a therapeutic treatment are our best immediate bets, absent which, quarantines and lockdowns will have to persist to curb the spread of a disease as infectious as covid-19 whilst we patiently wait for an effective vaccine against it [2].

[0] https://en.wikipedia.org/wiki/Compartmental_models_in_epidem...

[1] https://www.youtube.com/watch?v=gxAaO2rsdIs

[2] https://en.wikipedia.org/wiki/Vaccine_efficacy

“mild” symptoms has been used to mean flu-like, i.e., not something I would call mild and certainly not the same as asympotomatic, but not requiring hospitalization.

I think it is stranger than that. Mild has been anything from a fever for a night, to flu like for a week or so.

Really want to get a test on myself and family. We had something that near hospitalized me, but the kids barely registered being sick.

Most of what I've seen suggests that at time of positive test results, 40% to 70% of people have not experienced symptoms. Have you seen numbers very different from that?

[Edit: 'No symptoms experienced at time of positive test results' is intended to mean the same as 'no symptoms experienced yet at time of positive test results'.]

That can't be true for the official confirmed numbers.

Like, it was for a while impossible to get a test if you are not experiencing multiple symptoms AND can make a case that you might have been exposed via interacting with someone who has traveled. Even now I don't think they're giving the test if you don't have a fever. It's implausible that 50% of the people tested by the official pathway had no symptoms because the official pathway is not available when you have no symptoms.

But if you mean that you've seen a 50% "asymptomatic transmission" rate, in other words studies like this that purport to test a bunch of people at random and see how many of them have COVID-19, note that this definition of "asymptomatic" may include many with symptoms not severe enough to be hospitalized -- in other words they might have been feverish and coughing but not moreso than a typical cold. I myself have had a really nasty cough but it is not "dry" but "productive" and it has not come with a fever -- I would really love to be tested but right now that does not seem to be possible! Maybe I am in this above grouping of "asymptomatic" folks.

The OP article suggests that as many as 98-99% of cases might not involve hospitalization and therefore might be cases like mine (assuming I do indeed have COVID-19, maybe I don't). Now it is likely that there is some sort of selection bias in how they recruited candidates, but still it tends to push that number above your high end of 70%, suggesting maybe it's closer to 80 or 90%. On the other hand while the paper says that it did its due diligence subtracting out the test's false-positive rate, a rather small error here can have a big impact on that number because there are so few true positives right now.

In some ways the paper's claim is a bit of good news, it means that this disease has much lower mortality than we were originally told. (There may be lower bounds on mortality -- I have heard of one town in Italy where 1% of its population is gone due to COVID-19 in which case that would seem to be a good lower bound.) In other ways it is bad news, it means that this disease, still with nontrivial mortality, is going to be much less affected by quarantine countermeasures and the social distancing stuff is really the only reason it is spreading as slowly as it is, so that we need to endure the pain of isolation for much longer -- potentially until testing becomes widespread and cheap.

That may not be too much longer, as mentioned in a previous HN comment/story [1] that if you can get testing to work with the old Sanger sequencers used in the Human Genome Project you can maybe ramp up to 200,000 samples/day at first with possibilities to go up to 1M/day once you solve additional logistical problems. Before that the production of kits and sourcing of reagents may be your limit.

[1] https://news.ycombinator.com/item?id=22808208

Ok, we can't conclude anything about the rate of symptoms from testing that was only conducted on people showing severe symptoms. I don't think we can conclude much of anything from the OP either, because its rate of positives seems compatible with the rate of false positives from the test. They could have 0% infected, they could have 5% infected.

An example where testing was conducted without requiring symptoms first: https://taskandpurpose.com/news/uss-theodore-roosevelt-sailo...

Yes, although note that it's not clear which kind of test is being used by the Navy there.

If it's the nasal swab-PCR test that most civilians are getting, then the claim of "asymptomatic" must be accompanied by a "yet". We can't really conclude anything for at least two weeks after the positive result -- after seeing whether those people ever develop symptoms.

If it's blood-antibody testing, then we can give the statement a little more weight, because a positive result would mean the infection is in a later stage or even past. I have no idea whether the Navy is using these tests.

Also, they weren't able to factor out a bias towards people seeking confirmation tests due to perceived COVID symptoms.

This is in the bay area, where people were very quick to self-quarantine. Most people have been working from home since early March. They were asking people to break their self-quarantine to get tested without having any symptoms?

Have not experienced symptoms yet. I haven't seen any information about testing people and then following through until a negative test, to see whether they stay asymptomatic. It's more likely that we're testing them in the window between being infected and showing symptoms.

Confirmed population infection rate in Santa Clara is like 0.1%, and has been on a stay at home order for a month.

Highly unlikely these are all just people who have produced antibodies as part of a yet-to-be overrun immune response.

But by immune, do you mean asymptomatic and contagious?

Only talking about where we are in terms of progress to herd immunity. Edited it to say “presumably [now] immune” which hopefully is clearer. See my other comment below for elaboration.

I've been using 0.5% for mental math, rounding down from the lower bound from this study in The Lancet:

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

I would be surprised if, after this current wave of infections, the percentage of people with antibodies in the US is higher than the low single digits.

Herd immunity without a vaccine is a pipe dream. Our best bet is to massively ramp up testing and contact tracing and really start pushing the number of infections down to a point where parts of society can start functioning again.

So a 0.5% fatality rate sounds like "only" five times the rate of the flu. EXCEPT that I would bet that if you thoroughly traced all the asymptomatic but positive for the flu, it's fatality rate would decline also and it still causes significant deaths.

Total Covid deaths are already at a "mild flu season" level already after a month and continuing some level of exponential growth after social distancing.

It's sad and awful but it seems like with the current administration approach, the response of ramped up contract tracing has pretty much flown away and with these figures, we're looking 500K deaths, 250K over the next two month and another 250K over a longer period. Hide your parents...

Except that a huge portion of those deaths are in places that may have already reached herd immunity, if the ratio of comfirmed / actual cases is as small as this study suggests.

NYC, for example, has 123k confirmed cases. Multiply by 50, and you have 6.1M, which is close to 70% of the total population.

Where are you getting the assumption that only 1 in 50 infected are confirmed? That seems like an especially low confirmed case to infection ratio.

Directly from the article. It’s the lower-bound estimate for confirmed to infected found by the study.

You should read it before commenting.

I’m not a virologist or epidemiologist, so I might be off-base here, but: the flu’s really well understood isn’t it? Medicine’s been studying it basically forever. I’d be really surprised if our fatality rate for it wasn’t pretty accurate and didn’t already have asymptomatic cases factored in.

Herd immunity isn't a pipe dream it is just one hell of a sacrifice. And we'd better check the immunity actually lasts before we try that. Frankly I've never been a great fan of the 'flatten the curve' explanation, sure the graphs seem nice but I reckon the peak in those graphs is drawn too low by several orders of magnitude.

In the Netherlands a fatality rate of 0.5% would mean it'd take about a year for herd immunity to kick in assuming we managed sustain the peak that occurred about 2 weeks ago (and it's somewhat dubious whether the healthcare system actually can sustain that peak).

Let me quantify that sacrifice in a US-centric way with some napkin math to ballpark:

328.2 million (Approximate population of the US [1]) * 80% for herd immunity (based on an R0 5.7 [2]) * 1% = 2.6 million people die.

If the fatality rate is 0.5% instead, then 1.3 million people die If the fatality rate is 0.1%, we're still looking at ~260k-ish people.

That's pretty grim.

This situation is still developing so hopefully those numbers get adjusted downwards as later information becomes available.

For reference, ~1.2 million people died to heart attacks and cancer in 2017. [3]

[1] https://www.census.gov/popclock/ [2] https://wwwnc.cdc.gov/eid/article/26/7/20-0282_article [3] https://www.cdc.gov/nchs/fastats/leading-causes-of-death.htm

One aspect that people seem to be ignoring with the whole 'healthcare capacity' argument is the mental health of the healthcare workers. The capacity everyone talks about is IMO the acute capacity, not really the chronic sustained capacity.

That's because just like in a war, it isn't realistic to expect the frontline soldiers to stay in combat for longer than weeks at a time before they start to exhibit PTSD and their effectiveness begins to drop.

Unless we have large reserves of doctors and nurses to rotate in and out I think we should be very careful not to take them for granted.

It's not a pipe dream. It's just that it needs to happen in a different way. We need to massively increase number of isolation rooms, icu beds etc. (just like China did building that hospital in 10 days), massively increase number of clinicians (take every one in their last year of nursing school and med school and get them in the hospital, not making decisions but carrying out basic tasks; use a draft if you have to, just like Italy did), and massively increase testing capacity (~4x from today's capacity in US), and massively increase ventilators, add ubiquitous temperature checking (to enter any building other than your house, you need to confirm temp) and then just open up the economy and let it rip. With enough testing and enough surge healthcare capacity, we can do this.

> Herd immunity without a vaccine is a pipe dream. Our best bet is to massively ramp up testing and contact tracing and really start pushing the number of infections down to a point where parts of society can start functioning again.

The problem is, a vaccine is also a pipe dream. So we're going to have to all get infected over some sort of timeline that doesn't cause societal collapse. Also known as flattening the curve.

This is what I've been assuming - 200 to 250x. But 250xdeaths gives you the infected number from two weeks ago.

To get current infected you then have to multiply it by communication^(days to death), something like 1.1^14, with the first term varying depending on population density and lockdown measures, and the second on how on the ball that locality is for detecting cases and keeping people alive.

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 sign up for a Google account.[1]) Now that California hospital admissions and deaths for COVID-19 have flattened, growth should be linear for a while. But the rate of increase is still not really known.

Still, we're probably looking at a year and a million deaths in the US, within a factor of 2 either way.

[1] https://verily.com/stories/the-project-baseline-covid-19-pro...

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 classic COVID-19 symptoms or those who haven’t.

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.

This is what we know from other human corona viruses. SARS-CoV-2 could be different, but without evidence I would not want to bet on long immunity.

I made this comment the last time a serological testing study came out: they have 30 negative controls and claim 1.5% positives among their tests. So the number of controls is insufficient to rule out this being entirely false positives. You have to then rely on the test manufacturer's claims of false positives rates, which comes from 371 negative controls. They say:

> our estimates of specificity are 99.5% (95 CI 98.1-99.9%) and 100% (95 CI 90.5-100%)

Look at the those confidence intervals, even the narrower one (from the manufacturer's data)! The bottom is a four-fold increase in false positives, compared to the point-estimate, and is greater than the total positives they had in their finding.

This (I think?). I am not a doctor or biologist. Anyone who knows more than me, which is not a high bar, please step in and correct me.

Based on quick amateur reading of the paper, the rate of crude positives (1.5%) might be entirely real cases, entirely false positives, or anywhere in between. The real rate of infection could be anywhere from 0% to 5%.

The authors report around 3,500 real tests and 30 known-negative tests. We need 30,000 real data points (or 300 known-negative tests) to conclude anything. (Right?)

I am not criticizing the experts who conducted the study. Thank you for doing the study. I'm only trying to understand what the study reveals. It may reveal that the group needs 10x more funding right away.

To repeat you in plainer English: the serum test isn't infallible and reports "infected", incorrectly, for uninfected patients. It does this at some rate that is very near, or perhaps larger than, the fraction of "true" infected patients they report in their results.

Teasing out data from all that noise requires that the false positive error rate be measured very accurately. And they didn't do that, so really this doesn't tell us much.

Edit: Alternatively, borrowing jargon from a more common field around here: the measured infection rate of ~3% is very close to the noise floor of the experiment. It might be that, or it might be near zero, and we can't tell the difference. This study is very good evidence that the infection fraction is not much larger, however. We can easily rule out high infection rates like the 30% numbers that seems to get thrown around.

I don't see anything in the methods about potential crossreactivity with NL63, OC43, or HKU1 coronaviruses. Presumably this was done by the company, but crossreactivity is an extremely important control when it comes to this type of test. It is common, and if it's not vetted thoroughly, you may be measuring something else completely.

Add this to growing pile of evidence that this coronavirus is more widespread than most testing to date would suggest. How much more widespread still appears very unclear. There is a major ramping up of antibody testing happening right now, and multiple new tests being introduced, so we should have a better idea very soon.

But the takeaway should be that this is a good thing. Every unknown case we uncover decreases the overall rate of hospitalization and death, and potentially decreases the effective transmission rate (assuming most people who had the virus develop immunity and that the immunity lasts, which is also still unclear).

I don't follow your argument. If the epidemic is more widespread than initially thought, why do you assume that the effective transmission rate would potentially decrease? If anything, the transmission rate was higher than assumed and, due to the higher number of infections and how they spread exponentially,it would only ramp up.

And additionally, there are reports of reinfections among covid19 patients who were found to be cured.

There's good news and bad news: the good news is the more cases that are found from people who didn't know they had it, the less deadly the disease is. The bad news: if it's all over the place it will be really hard to stamp it out.

I assume it's endemic at this point and we just need to learn how to live with it. Society can brace for a storm, but at some point you need to adapt and learn to live with the weather.

"If the epidemic is more widespread than initially thought, why do you assume that the effective transmission rate would potentially decrease?"

Herd immunity. If testing is only detecting 1/50th of cases in NYC, for example, it means that there have been about 6M cases, or about 69% of NYC's population. That's basically the threshold where we'd expect to see the infection counts level off naturally.

there is no herd immunity if reinfection can occur, which appears to be the case

There is no credible scientific evidence that reinfection can occur. There have been, at most, anecdotal reports from unreliable sources.

And there are good reasons to believe reinfection shouldn't be possible, to be clear - it's not just being dismissed out of hand.

It's the difference between the effective reproductive rate and the basic reproductive rate. Yes, more cases means a higher basic reproductive rate, which is an innate characteristic of the virus. But more immunity in the world means a lower effective reproductive rate, which is the same rate we're trying to manipulate with physical distancing and masks, etc.

"And additionally, there are reports of reinfections among covid19 patients who were found to be cured."

Like I said in my comment, it is still unclear how immunity works.

> there are reports of reinfections

AFAICT those reports pertain to testing positive again, not to actual reinfection. The leading hypothesis from scientists in the studies I've seen is that the tests are picking up RNA from dead virus, which will neither cause symptoms nor be transmissible to others. I certainly hope that hypothesis turns out to be the correct one, because if not we're in for even more trouble.

The problem with every one of these studies I've seen is that they come up with around 3% of the population having antibodies, while a typical false-positive rate for these tests is also around 3%. In this case they claim they adjust for the specificity (basically false-positive rate), but I didn't see that they reported what the specificity was.

If the mortality really is very low, then it should be easy to prove. Just go to NYC, do some random testing, and show us an infection rate that far exceeds the false-positive rate.

This is certainly not a study. And certainly shouldn't be taken as good data of any kind, but it is interesting: https://www.livescience.com/coronavirus-in-pregnant-woman-hi...

Sure as hell isn't random (all pregnant women... so all women for one. All at the same hospital is another).

"Between March 22 and April 4, those hospitals screened 215 pregnant women for SARS-CoV-2 (the virus that causes COVID-19), and 33 women, or 15%, tested positive. Of these who tested positive, 29 women — or nearly 14% — showed no symptoms."

Anyone know if there's been a follow-up to see if the women eventually developed symptoms?

I don't see anything about sampling bias here. I would assume that people who have been ill, experienced mild symptoms, or who were exposed to confirmed cases would be more likely to respond to the facebook ads recruiting testing volunteers.

I'm sure we are significantly undercounting cases but I highly doubt we are off by a magnitude of 50x.

I know. 50-85x is hard to believe. How could we be that far off?

It’s not hard to believe when you consider testing criteria.

The # of cases reported that I believe they are comparing against is the number of confirmed positive tests, not the number of assumed cases in the county based on some other model that is trying to project # infected. I don't think anyone thought that we had 100% coverage where all cases were tested. So we're not "wrong" if such a study as this sees a gap between reported confirmed cases and expected infected, it just means we have a better understanding of the testing gap, which is self evident in existence but the magnitude of which we don't know.

Given all the other factors, an order of magnitude or more gap between tests and sick people doesn't seem completely out of the question. I would be curious if there are other models using a different methodology which could help us get a handle on the conditional probability chain leading to tests being done or not on an individual, to see if there is a similar set of conclusions.

It's impossible to believe. Currently, Santa Clara's crude CFR is 3.7%.

NYC has similar demographics (and as bad nursing home hits?) -- 0.14% of the population has died from covid. That puts an upper bound of 26x (and that's if the entire population was infected)

You have no reason to believe that the Santa Clara CFR is correct, and dividing the CFR of one city by the IFR of another is meaningless.

They mention it, but state it's hard to ascertain. I agree with that assessment. Given the magnitude of the delta between confirmed cases and projected cases based on this study tho, it seems intuitively obvious there's a pretty big gap. (And not surprising, given the lack of testing, the asymptomatic nature of the disease, etc.)

If the vast majority of people contracting the virus never have symptoms, or have symptoms mild enough that they don't seek a doctor, or they DO seek a doctor but can't get a test because of testing criteria restrictions, I think the 50-85x number starts to make a lot of sense.

With only 30 people in the negative control group, the confidence interval on the false positive rate would easily swamp out the results of the study for 1.5% positive sample ratio, no?

Seems kind of useless unless your control group is a lot larger, and the false positive rate can be shown to be <<1.5%.

"Participants were recruited using Facebook ads targeting a representative sample of the county by demographic and geographic characteristics."

How is it representative when it's only Facebook users in this sample?

In the south bay? Pretty good, unless you want to argue for a demographic skew because of Facebook's poor uptake among iOS users.

I mean, no. It's not perfect. It will miss demographics like the elderly with lower social media use, but that group tends to be well-sampled already due to their risk profile. It will probably miss some immigrants too, which seems like a bigger problem.

But really, it's a pandemic. It doesn't care about your socioeconomic status. One of its defining qualities is the extent to which it does not cluster in particular communities like more typical epidemics.

They also were unable to remove any bias towards people self-selecting to be part of the test because they felt they had symptoms of COVID. The bay area has been in lock down since early March, its really hard to imagine someone participating if they didn't feel they had some symptoms and wanted to 'know for sure'.

Unless you think FB users for some reason would be infected at significantly different rates, that shouldn't be much of a problem.

I'm much more concerned about selection bias toward prior sick people; IIRC, Stanford offered to report positive results to the patient.

There would potentially be an age bias, though probably less in FB than a lot of other services you could cherrypick from. It'd also arguably bias older, since FB has been aging up in audience from what I can tell.

Without reading the study, though, possible they actually controlled that in the demographic profile for the ad.

Edit:

Going to the other thread confirmed this. From the paper,

Unless nursing homes are represented well in Facebook, I'd wager the bias is against what we believe to be the biggest risk pool.

The link was widely spread around via Nextdoor, email lists, other social networks. You could say it went viral.

There's also some data coming out of Wuhan around antibody testing. The methodology is not as well formed as the study in the OP but there is still value in the data.

> Wuhan’s Zhongnan Hospital found that 2.4% of its employees and 2% to 3% of recent patients and other visitors, including people tested before returning to work, had developed antibodies, according to senior doctors there.

Additional antibody testing is also underway in Wuhan.

Source: https://www.wsj.com/articles/wuhan-starts-testing-to-determi...

It is far more credible that Wuhan hit 3% infection rate (that's about double the PCR testing rate of the evacuees) than Santa Clara county, with an order of magnitude lower death count, having a 2% rate.

Except for people bored out of their minds, wanting to get away from their kids or spouse for an hour, etc. Also, there is an altruistic motivation - it feels like you are doing SOMETHING to help by being in this study.

I think the numbers will coma out to show this ripped through the USA in February; we slammed the barn doors closed well after the horse was gone.

Of course that's not the narrative that fits agendas so just like with "why cant we test people?" back then, there will be many official reasons not to look for antibody data and to doubt what data gets gathered.

Based on the difference between CA and NY, I'd be willing to bet that it got here earlier, maybe in December.

You would have seen more reports in hospitals of interstitial pneumonia, as well as a higher YoY death rate during January.

The CDC is reporting an abnormally high pneumonia death rate in January, and it can’t be explained by confirmed flu cases:

https://www.cdc.gov/flu/weekly/#S6

Look carefully at the black and red graph after the phrase “mortality surveillance data”.

The graph and text clearly state it refers to pneumonia and influenza, if you refer to the legend where it says Percent P&I. The P&I for the first 8 weeks of the year is clearly within normal ranges. It spikes from baseline in the first week of March continuing on to the present day.

If you look at the first set of graphs, you'll see a spike around the same time for influenza tests. I don't know how fast pneumonia occurs in people weakened by influenza, but I would expect some delay.

I think the spike in both influenza and in death due to influenza and pneumonia is probably due to testing. If you presented with influenza like symptoms in March, you were definitely given a flu test to rule out influenza.

it sounds like there was. Look at the 2019/20 flu season from the cdc. it was about 3rd place in last 10 years despite the country basically shutting down.

Anecdotally, I remember in December / January reading about this being the worst flu season in years (time). it would have been interesting to know if the flu was effecting the normal cohort or ignoring children.

I myself got the 2nd worst sick I have ever been at the end of November visiting San Francisco. (really bad dry cough, fever). It was strange in the sense that I had mild symptoms for a bout two weeks and then 'took my breath away'. went to the hospital, no pneumonia, but low oxygen levels. my friends kid got the same thing, mild cough for a few days and fine. didn't really recover until mid January. Was it covid19? probably not, was probably the flu.

That doesn't discount the possibility this was running around the globe even in December. I don't stick my head in the sand and wait for people to tell me so to consider that's at least a very possible scenario.

https://www.cdc.gov/flu/about/burden/preliminary-in-season-e...

https://time.com/5758953/flu-season-2019-2020/

California has 2X the population of New York. NY has more than 17X the number of deaths and while the daily deaths rate there is slowing it still far outpaces CA's daily rate. I don't know how you can look at those numbers and conclude the disease was spreading unchecked in California for multiple months.

Maybe a milder form of it came directly from Wuhan to CA, spreading like a bad flu. And then the version in NY came from Wuhan to Europe to NY, mutating along the way, and making it more deadly.

I'm not sure why we should believe the strain that originated in Wuhan was milder, but lets go with this theory for a minute. How does the nursing home in Washington fit into that picture? It was one of the earliest known hotspots in the US and it was extremely deadly. Was it the mild strain that you think might have already been spreading through California or the severe strain from Europe that spread to Washington first before New York? If it was the former, why didn't we see similar situations in California among vulnerable populations?

Not that I necessarily agree with the theory, an influenza outbreak in a nursing home could be extremely deadly. This is why they are nearly 100% vaccinated. And they are (or are supposed to be) quick to quarantine and lockdown if it seems like flu has been detected. Which is to say that even a "mild" strain would certainly pose a larger risk to a nursing home.

I realize that, but it skips over the last question in my previous post. Why didn't we see a similar situation in any California nursing homes if this mild strain was spreading unchecked for months in a population with zero immunity?

I think the more likely explanation is that NY simply is a lot more dense and less reliant on cars vs the less dense and more reliant on cars California.

Especially if the infectious dose also affects severity, as is the case with other respiratory illnesses: https://www.ncbi.nlm.nih.gov/pubmed/25416753

Basically, if you're in a suburban area and pick up a few viral particles from one coffee-shop counter, it's possible that you're not in nearly as much danger as someone touching handrails and inhaling subway air all day long.

At 3-4%? Herd immunity needs 70% in US. COVID-19 can not have spread freely if only 3% people are infected.

https://www.journalofinfection.com/article/S0163-4453(20)301...

I keep hearing this sentiment, but I cannot square it with the fact that a non-trivial percent of infected people need to be on ventilators for roughly 20 days. If, back in December, 10k people had what we now know as Covid, there would have been dozens of people on ventilators for weeks at a time in January. The average flu patient needing intubation is only on a ventilator for about 5 days. I just don't see a situation where we have dozens of people testing negative for the flu yet requiring a higher level of ICU care and no alarm bells are raised.

My optimistic side wants the truth to be that there are many strains, and the worst one spread in the last few months. This would mean that there are other strains that provide adequate anti-bodies and yet are mild enough to go unnoticed by professionals.

"a non-trivial perfect of known infected people"

All of the flu surveillance numbers suggested a historically bad season this year, and there was a big peak in flu-like mortality in February. It's entirely possible that we were mis-characterizing the data:

https://gis.cdc.gov/grasp/fluview/mortality.html

The data from that time period seems to coincide with positive influenza test results:

https://gis.cdc.gov/grasp/fluview/fluportaldashboard.html

Stanford did a study looking for Coronavirus cases in Jan and Feb in the Bay area.

http://med.stanford.edu/news/all-news/2020/04/testing-pooled...

> The researchers found that the burden of COVID-19 in the Bay Area prior to mid-February was low. Only two of nearly 3,000 people with respiratory-disease symptoms who were tested in early 2020 at Stanford Health Care or affiliated clinics for common respiratory viruses were infected with SARS-CoV-2, the virus that causes COVID-19.

Sure, but the bay area has very few detected infections, in general.

It seems pretty obvious that we need to do the same experiment on samples from New York, Washington, Michigan, etc.

CA and WA had the earliest cases of COVID,I believe in Jan or early Feb. If this ripped in Feb, why did we not see similar hospitalizations and death rates in CA and WA compared to NY,Italy, Spain, etc ?

my friend probably had the virus (2 inconclusive tests) and tried to get into this study. some group asked her for $125 to get tested. i'm not sure if she saw any of the facebook ads. so, grain of salt about representativeness of the sample. there's for sure a response bias towards people who think they had it.

Just curious, did your friend have pretty severe symptoms? As in, was it worse perceivably than say a case of bad flu?

The Stanford study was free. They didn't charge $125.

There could be a meaningful distinction between exposure and infection. I would define the first as anyone who was exposed to the virus (or even just parts of the virus) and generated antibodies. The second would be an actual infection and presumably a sufficient antibody response to repel the virus and confer immunity to another infection. Really you can generate antibodies without even getting exposed to live virus, just like you can develop allergy to pollen or other foreign objects that enter your body. I would say only the second case is meaningful for public health. For that you need to look at how people do convalescent plasma. You need to test for specific antibodies that actually confer immunity.

There are now multiple sources of evidence pointing to a very low true infection fatality rate, as low as 0.1% in some regions. Iceland has been testing a semi-random sample using PCR since early in the epidemic, and there have been several small antibody surveys in “hot spots” showing infection rates over 15%. You can also use CDC surveillance of flu-like illnesses to estimate the total number of infected nationwide; this morning I posted an analysis of this data estimating only 1 in 40 cases were detected in March, consistent with the Stanford results: https://fivetran.com/blog/covid-19-count

It may have a low overall fatality rate, yet it seems to have a high rate of hospitalization in clinical cases.

And it’s more contagious than the flu, so even if the fatality rate is identical, more people are at risk.

> it seems to have a high rate of hospitalization in clinical cases.

The raw hospitalization rate is inflated because both treatment and testing are mostly limited to very severe cases in places where there's a lot of confirmed cases (e.g. NYC.)

Still, comparing it to influenza... we dont see this many severe cases typically.

Its pretty obvious this is worse than the flu. I don't think anyone is really disputing that at this point.

CDC flu surveillance data suffers from even more selection bias than this study. It's basically useless.

Iceland PCR data is reliable but it is suggestive of an IFR of more like 0.3%. Except Iceland also managed to successfully isolate their elderly so if you correct for that, we're back at 0.7% or so for the general population.

How do you square this with 0.14% of all of NYC and 0.25% of all of Bergamo county being dead from confirmed Coronavirus? Surely this puts a strong lower limit on IFR.

Deaths in both places can only increase and deaths will inevitably be retroactively assigned a Coronavirus cause.

A large percentage of the population in NYC and Italy have been probably already been infected -- for NYC specifically, I don't think its too implausible to imagine 20-30%. That would give an IFR of ~0.6% which is in line with a lot of other estimates. An IFR of <0.3% seems unlikely though.

That gives a current IFR of 0.6%. Most cases haven't resolved yet and South Korea is showing that the death rate inches up over time as deaths take much longer to resolve than recoveries.

Saw this comment [1] on Reddit that outlines the limitations of the study well:

This is the most poorly-designed serosurvey we've seen yet, frankly. It advertised on Facebook asking for people who wanted antibody testing. This has an enormous potential effect on the sample - I'm so much more likely to take the time to get tested if I think it will benefit me, and It's most likely to benefit me if I'm more likely to have had COVID. An opt-in design with a low response rate has huge potential to bias results.

Sample bias (in the other direction) is the reason that the NIH has not yet released serosurvey results from Washington:

We’re cautious because blood donors are not a representative sample. They are asymptomatic, afebrile people [without a fever]. We have a “healthy donor effect.” The donor-based incidence data could lag behind population incidence by a month or 2 because of this bias.

Presumably, they rightly fear that, with such a high level of uncertainty, bias could lead to bad policy and would negatively impact public health. I'm certain that these data are informing policy decisions at the national level, but they haven't released them out of an abundance of caution. Those conducting this study would have done well to adopt that same caution.

If you read closely on the validation of the test, the study did barely any independent validation to determine specificity/sensitivity - only 30! pre-covid samples tested independently of the manufacturer. Given the performance of other commercial tests and the dependence of specificity on cross-reactivity + antibody prevalence in the population, this strikes me as extremely irresponsible.

This paper elides the fact that other rigorous serosurveys are neither consistent with this level of underascertainment nor the IFR this paper proposes. Many of you are familiar with the Gangelt study, which I have criticized. Nevertheless, it is an order of magnitude more trustworthy than this paper (both insofar as it sampled a larger slice of the population and had a much much higher response rate). It also inferred a much higher fatality rate of 0.37%. IFR will, of course, vary from population to population, and so will ascertainment rate. Nevertheless, the range proposed here strains credibility, considering the study's flaws. 0.13% of NYC's population has already died, and the paths of other countries suggest a slow decline in daily deaths, not a quick one. Considering that herd immunity predicts transmission to stop at 50-70% prevalence, this is baldly inconsistent with this study's findings.

For all of the above reasons, I hope people making personal and public health decisions wait for rigorous results from the NIH and other organizations and understand that skepticism of this result is warranted. I also hope that the media reports responsibly on this study and its limitations and speaks with other experts before doing so.

[1] https://www.reddit.com/r/COVID19/comments/g32wjh/covid19_ant...

And last, but definitely not least: you can't draw the conclusion that presence of antibodies equates immunity. It will correlate but there is no guarantee.

Is there a better method?

Yes, but not one that is ethical. You do a challenge experiment to those with antibodies and see if they get infected.

The next best alternative is monitoring a large number of people with antibodies (and a control group) for new infections. This is slow and expensive, but if you have enough people then it will work.

It’s better to have data and know its limitations, than not to have data at all.

This may be a true statement when applied to public health professionals, but is quite likely dangerous to public welfare when in the hands of disinformation campaigns and propaganda.

“These prevalence estimates represent a range between 48,000 and 81,000 people infected in Santa Clara County by early April, 50-85-fold more than the number of confirmed cases. Conclusions The population prevalence of SARS-CoV-2 antibodies in Santa Clara County implies that the infection is much more widespread than indicated by the number of confirmed cases.”

This thread critiques this article in a way I find convincing: https://twitter.com/DiseaseEcology/status/125122527387113472...

Great to see actual testing on a representative population sample, though I’m not sure what the consequences ought to be:

> These prevalence estimates represent a range between 48,000 and 81,000 people infected in Santa Clara County by early April, 50-85-fold more than the number of confirmed cases.

So with 69 deaths so far and 9 deaths last week in Santa Clara County, projecting that to an extra 4 weeks to 105 deaths for those infected in early April, that means a 0.16% death rate?

Lots of reasons I don't think you can extrapolate meaningful death rates from this yet. One big one is that time from infection to death covers a very wide range: in Wuhan time from hospitalization to death had both a mean and a standard deviation of about 2 weeks, and time from infection to hospitalization isn't well known yet, but is likely at least a week. Add to that the fact that some cases are diagnosed post-mortem and you might not know about all of the fatalities for early April cases until well into May.

In our small town, I know of 3 people who came into direct contact with the first confirmed case, exhibited symptoms, but were refused testing. I can definitely believe their guess that the number of cases is an order of magnitude greater than the number confirmed.

That is certainly alarming that it is so much more widespread than expected - it really highlights the failure of our controls against the virus and the importance of early testing. It I'd good news that this implies the death rate is less severe.

However, a caveat is that antibody tests might have high rates of false positives, which would be a particularly large problem when very few people are actually positive. So read this with a grain of salt. The paper isn't peer reviewed yet etc.

On the contrary, this is not alarming at all.

It would point out that actual Covid-19 mortality could be close to that of seasonal flu, as it tells us the number of people infected is 50X-85X than the number of confirmed cases.

Assuming the data is reliable, of course.

the authors infer an IFR of 0.12 - 0.2% which is smaller than other antibody studies (like the one in germany which inferred 0.37% IIRC). While it s indeed small, it reflects the current situation, not what will happen 2 weeks down the road, plus the sensitivity of their antibody testing is not precisely known yet.

Most signs so far suggest that the virus is basically uncontainable, short of everyone on the planet taking aggressive measures to stop it escaping China back in January or December.

That was back when China was claiming it didn’t transmit between humans, and ordering labs to destroy their samples.

https://www.nationalreview.com/the-morning-jolt/chinas-devas...

Right, that's why people say the Chinese government's lies (and the WHO's acceptance of them) are at fault. The rest of the world certainly should have reacted sooner, ramping up PPE production and researching treatments and trying to slow it down. But the idea of complete containment, forming widespread green zones where the virus is not merely suppressed but eradicated, probably became impossible as early as February. (Some island or basically-island countries appear to be making it work, but it's not clear what their long-term plan is - are they just going to enact sakoku for the next 18 months?)

Well and I say this not even to indite China. Maybe they really and truly believed at the time there wasn’t human transmission. Maybe a virus originating in France or the US would have played out mostly the same way. (I doubt it, but that’s irrelevant)

My point was just that with the information we had in December, there was no way politically to enact the extraordinary (and extraordinarily damaging) measures that would have been required to prevent an epidemic.

Not only was it politically untenable (Trump was only acquitted on February 5th!) but it was simply scientifically unsupported and unjustifiable at that time.

IMO even with hindsight being 20/20 I can’t say given the information we had on January 1 that we could or should have done any differently than the travel ban on January 31st, which itself was extraordinary. Everything post January 31st however is an absolute boondoggle of the highest degree.

As far as PPE and testing capacity, I believe those are both systemic issues which needed to be dealt with years ago. Good luck with the US trying to procure a billion masks from China right at the onset of their own health crisis — imagine how that would play out with 1,000s of deaths in Wuhan and 0 known cases in the US. We would have been accused of hoarding and causing the spread in China.

The testing issue was mainly an easily foreseeable and preventable regulatory problem at the FDA which has existed for some time. It took too long to remove the regulatory hurdles, but that there was not already a pandemic regulatory response framework ready to activate is totally inexcusable since we already went through this in 2009.

Your dates are off.

SARS-2 was circulating outside China in SE Asia before Dec. 15.

BlueDot warned subscribers Dec. 30.

WHO notification was Dec. 31.

Disneyworld Shanghai was closed Jan. 24.

You can't hear anything when you don't listen.

> Your dates are off.

I think not.

The Lancet reported on January 24th the Time of Onset of the earliest known patients in Wuhan in a nice graphic [0]. The first case onset in Wuhan is reported as December 1st.

"Chinese epidemiologists with the Chinese Center for Disease Control and Prevention published an article on 20th January 2020 stating that the first cluster of (~40) patients with ‘pneumonia of an unknown cause’ had been identified on 21st December 2019". [1]

Much later, just in the last few weeks, new reports have come out trying to trace earlier possible cases of COVID. Here is one such case dating back to November 17th, in Hubei [2]. This was not known in December or January.

There is a pre-print available as of April 8th claiming to have found an earlier case through genome tracing dating as early as mid-September in Guangdong, a southern coastal province in China. [3]

> SARS-2 was circulating outside China in SE Asia before Dec. 15.

Citation needed.

My other dates are straight from the National Review, and I believe they are accurate.

The first known case in the US was discovered on January 21st.

Here is WHO writing on January 5, 2020 [4];

"On 31 December 2019, the WHO China Country Office was informed of cases of pneumonia of unknown etiology (unknown cause) detected in Wuhan City, Hubei Province of China. As of 3 January 2020, a total of 44 patients with pneumonia of unknown etiology have been reported to WHO by the national authorities in China. Of the 44 cases reported, 11 are severely ill, while the remaining 33 patients are in stable condition. According to media reports, the concerned market in Wuhan was closed on 1 January 2020 for environmental sanitation and disinfection."

I claimed that there was nothing that could have been done by the US in the December time frame, and I believe that is self-evident from the timeline. Looking at that statement from the WHO on January 5th, there is nothing in that which could possibly motivate a national quarantine response at that time. Not even in China, but certainly not in the US.

[0] - https://els-jbs-prod-cdn.jbs.elsevierhealth.com/cms/attachme...

Full Text:

(https://www.thelancet.com/journals/lancet/article/PIIS0140-6...)

[1] - https://bfpg.co.uk/2020/04/covid-19-timeline/

[2] - https://www.livescience.com/first-case-coronavirus-found.htm...

[3] - https://www.pnas.org/content/early/2020/04/07/2004999117

[4] - https://www.who.int/csr/don/05-january-2020-pneumonia-of-unk...

How do you square that with the fact that there is no outbreak in any overseas Chinese community? Not Taiwan or Hongkong or Macau, nor in the epicenter of Italy (https://www.reuters.com/article/us-health-coronavirus-italy-...), nor San Francisco Chinatown (https://www.nytimes.com/2020/04/17/us/san-francisco-coronavi...)? All of these communities have far more ties to mainland China.

It could have already passed through all of those communities basically unnoticed, and/or confused with other ILIs. There's no way to rule out that possibility, and when you consider how rare cardiovascular diseases (the most common comorbidity) are in those communities, you have to conclude that it's not vanishingly unlikely: https://minorityhealth.hhs.gov/omh/browse.aspx?lvl=4&lvlid=4...

Obviously mitigation efforts still play some part in limiting the spread, but I would not bet that all of those communities were able to suppress such a contagious disease before a big chunk of the population was exposed to it.

Maybe Chinese emigrants in Italy are young so they didn't notice the wave. But Taiwan/Hongkong/Macau? They are testing and tracing. When their recent travelers to Europe and America came back, a lot of them were infected and caused a spike in cases. https://www.nytimes.com/interactive/2020/04/09/world/asia/co...

Note coauthor name "John Ioannidis". That gives the paper a healthy boost in my view.

Careful, though, because just being a critic doesn't mean he isn't susceptible to the same sorts of biases and issues that cause a lot of other medical papers to fail. I'm hoping he's on the paper because they ran it by him for a thumbs-up before submitting it.

I hope he is not on the paper for that reason as this is basically honorary authorship. If he just reviewed the paper before submission then he should be in the acknowledgements.

I suspected the asymptomatic rate to be high, but not by a factor of 50-85! That's very great news no?

This is very strong evidence that the "realists" like Alex Berensen were correct. The shutdowns probably should end, while encouraging at-risk populations (mostly retired) to stay at home.

nope.

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/2020/04/03/what-we-should-have-learned-fr...

More studies coming next week from LA.

Why quote Denmark but ignore Sweden? Sweden are doing what you're calling for, and they have pretty high numbers of deaths.

It's not "pretty high", it's middle of the pack and nowhere near the apocalypse that was supposed to be the outcome of their chosen path.

Deaths per 1m population in Sweden is currently 139. For Denmark it's 58, and for Norway it's 30.

This is a total failure as a talking point - it's amazing how quick it's spreading. Go compare it to Italy and Spain, which have both got the most severe restrictions. In Spain children are under house arrest, yet their outcomes are much worse than Sweden.

As I said, Sweden is in the middle of the pack. Your response is not a rebuttal to that. Lockdown is meant to avoid deaths by ensuring everyone has a bed, which in Sweden they do. Beyond that death rates will diverge for other reasons - it appears Sweden is seeing large racial disparities in deaths for some reason, as is America. So as a country that took in a lot more immigrants than Norway that would by itself be a contributing factor. But deaths due to not getting hospital beds is something we can safely say isn't happening, which means the justification for the lockdown is invalidated.

I'm really uncertain why so many people are resisting this outcome. It's a good one! Lockdowns are bad! If they have no effect then that means they should end and that's great for absolutely everyone.

No they don't. Stop being scared. Start being logical.

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 being persistent like herpes or HIV, make people sterile or reduce lung capacity and or increase chance of cancer.

With these, possibly small, risks, it is ethically incredibly irresponsible to allow the disease spread through the whole population. You cannot gamble with a conceivable downside like that.

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.

"if it saturates"

That sequence of words shows you do not know the basics of epidemiology. Epidemics do not affect 100% of a population.

Few non-peer reviewed papers with biased samples, questionable tests and questionable methodology don't make strong evidence. CFR just can't be that low. There are towns in Italy where the virus killed 1% of total population

https://www.ecodibergamo.it/stories/valle-brembana/il-grido-...

Given that there was a disaster (in the sense of failed containment) in the hospital of Alzano Lombardo early on (which is in the area), I wonder if that meant that more vulnerable segments of the population were exposed to the virus.

It sure can be that low. Pretty simple reason, too:

1) Studies linked above are properly statistically randomized 2) Italy population dead = old smokers, Chinese from Wuhan working in factories

The study is a random population sample. A single tiny town Italy has a highly characterized population. If you disagree, you've never been to Italy.

Stop being scared. Start being logical.

Data from Wuhan was that in the early phase of the epidemic about 70% of infections were transmitted at home. Once they realized this, they went to rapid isolation of positive tests and quarantine of contacts in “COVID hotels” away from home.

So your suggestion is entirely unworkable unless there is a plan to remove all at-risk persons from contact with low-risk persons.

Please show your work and let us know what that plan looks like. Who will care for these people, how will they be housed, how will you prevent the care force from bringing infection to them ala nursing homes, etc.

Guess what? We have been doing it your way for 45 days. Everyone infected is now locked in a house with others. They too are now infected. Intrafamilial and nosocomial infection is much more powerful than community -- that has been proven here and in earlier epidemics.

And yet anti-science morons in SF clutch their scarf over their mouths when you jog past them.

The anti-science on HN is astounding.

Antibody tests are not as accurate as RT-PCR, but the sensitivity that was observed here was 91% and 99% specificity when compared to confirmed COVID cases. I just want to point out that these researchers are using the corrects tests to do this.

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.

What about Facebook ads doesn't mean that they're an adequate sample? Is there a confounding factor that you have in mind?

Many confounding factors. Socioeconomic factors that can't be fixed with stratified sampling (access to transportation, usage of Facebook, etc.. conversely the willingness of a $10 amazon gift card to motivate people to participate will vary). People who are concerned about possible past exposure to COVID-19 may have a different probability to respond.

Of course, the bigger issue is that we can't rule out that the antibody tests have a false positive rate that can explain the whole result. We need serology tests in places with a higher positive rate to know.

Basically, 50 positive results out of 3330 is a really small signal. Just a small false positive rate or sampling issue could explain all the positive results.

Reply on news.ycombinator.com