CFR is case fatality ration and is literally defined as "people who we know are sick to people who died". It does not mean overall lethality and it never meant that. CFR is not prediction nor model, it is input into the mode.
Asymptomatic non tested population is literally not supposed to be counted into CFR.
'asymptomatic non tested population' is an interesting way to refer to people with covid 19 antibodies found via immunoassays. This sizeable population shows that our cfr estimates have enough error to be meaningless.
CFR does not have "error rate" this way. It is just defined differently then you thought it is defined. You thought it is mortality, it never was. It is the only number available at the beginning of epidemic (no matter which one).
"asymptomatic non tested population" is the most exact description of people who were not tested, had no symptoms and therefore don't count as "cases". Yes, we measure them by measuring antibodies if precise tests are available.
No, there are tests that have high false-positive rates. The studies in both the US and Europe used similarly flawed tests and didn't correctly account for the false-positive rates of those tests. [0]
There are two ways to think through how these results are flawed: the first is that the false positive rates of the tests is vastly higher than the prior (which we'll assume is approximately the confirmed case rate per capita) is. There were three studies, NYC, Santa Clara, and one in Europe that I know of. All three tests had false positive rates over 2%, and one test had a false positive rate of like 10%. So when a study shows that a large percent of the population have the virus, and that percentage is suspiciously similar to the false-positive rate of the test, you pause.
The second is a much more grokkable one. Let's assume that 2% of people in Santa Clara were infected (per the study), then 40,000 people are infected, and of them around 2000 people were symptomatic enough to get tested and be hospitalized/tested (per available stats). Or 5% of infections are symptomatic enough to get noticed, ish. Then we'd see that NYC would have ~3.5 million people infected now, or something like 40%.
So we'd expect to see 40% (or more actually) of people tested in NY and NYC test positive for covid. But we don't[1]. We see that as NY adds more tests, the positive rate of the more reliable tests is going down, because they can test more people and because NY appears to be over the hump. At peak, 20% of people were testing positive daily, now it's down to closer to 10%.
And note that most of the tests are going to people who have some symptoms of something, which is why it's that high. 40% of NY residents were infected, we'd expect to see more than 40% of the tests positive, but we don't.
If you do the same calculation but with death rate, you find that there are more people in NYC with Covid than live in NYC. In conclusion: those studies are bad because the tests they use are unreliable.
Are you equally skeptical that we have a test that reliably detects Covid?
And I suppose all of the 2+ dozen studies done are completely inaccurate as well? List below includes the revised study from Ioannidis. Your criticisms are from a preprint.
May 4: 370 workers test positive in Missouri. None had symptoms.
May 1: Dutch Royal Institute of Health 13884 healthcare workers test positive in the Netherlands. 9 have died. 0.06% CFR among healthcare workers 3% hospitalization rate 6/9 underlying conditions 9/9 above 45
Right, and my entire point is that most of the antibody tests appear to be inaccurate with false positive rates that are egregiously high. Quoting a bunch of antibody tests with egregiously high false positive rates doesn't address that at all. You'd need to show that the antibody tests are accurate, and independent tests have shown that they aren't.[0]
The conclusion there is that the serological tests have false-posdtive rates of 15-20% in the worst cases, and 2-3% in the best cases, on blood samples from before covid-19 existed.
So using those tests, on a population where not a single person has covid-19, you'd see results showing that 2-20% of the population was previously infected.
These independent numbers often differ from the official reported numbers from the test makers. And the seroprevalence studies you cute either use those numbers, or worse just ignore the possibility of false positives entirely.
Again, if 20% of New Yorkers had covid, (much more than) 20% of people tested using reliable tests would show that.
> Are you equally skeptical that we have a test that reliably detects Covid?
No, because we have evidence that the seroprevalence tests are god awful. We don't have that for viral presence tests, unless you believe the god awful seroprevalence tests.
To humor you, I'll run through some of the links you posted:
May 4: 370 workers test positive in Missouri. - Reliable (PCR, not serological) tests, people may be presymptomatic, since this just happened.
May 1: revised Santa Clara study - From the paper There is one important caveat to this formula: it only holds as long as (one minus) the specificity of the test is higher than the sample prevalence. If it is lower, all the observed positives in the sample could be due to false-positive test results, and we cannot exclude zero prevalence as a possibility.
May 1: Dutch Royal Institute - Unreliable sero test
May 1: 15.5% seroprevalence in German Community - Unreliable sero test
April 30: partial results give 61% of population infected - Unreliable sero test (likely compounded by bad stats, but I can't read french)
April 27: 36% infected in homeless shelters - Reliable test, but unclear if the people were asymptomatic or presymptomatic. Needs followup.
April 27: Governor of NY claims their study indicate 24.7% - Based on sero test
April 26: 96% symptom free out of 3300 positive - No info on which to base an opinion
April 26: 17,000 healthcare workers tested positive - Unclear, can't read french
April 26: Seroprevalence in Kobe Japan 300x to 800x - Seroprevalence says it all
April 26: 25% of seroprevalence in Iran - "Seroprevalence"
April 24: between 123k and 221k cases in Miami-Dade - Blood (sero) test
April 24: Denmark 1.7% seroprevalence - Sero test
April 23: governor of ny tweets that 13.9% of the state could have virus - Based on sero testing
April 22: Sweden health official claims 20% of Stockholm could be infected - Based on sero testing
April 22: LA County could have between 223k and 442k cases - Based on sero testing
April 22: 20% of Stockholm infected - Based on sero
April 20: Up to 10% of the people in Wuhan developed antibodies - Antibodies (sero)
April 19: 49% of the population of Orties developed antibodies - Sero
April 17: 32% of the residents tested for antibodies - Sero
April 16: 3% of blood donors have antibodies - Sero
April 15: 50% homeless test positive in a Boston Shelter - You cited this already
April 15: Coronavirus probably dozens of times - Sero
So, removing duplicates and serological tests, we're left with 3 things:
May 4: 370 workers test positive in Missouri. - Reliable (PCR, not serological) tests, people may be presymptomatic, since this just happened.
April 27: 36% infected in homeless shelters - Reliable test, but unclear if the people were asymptomatic or presymptomatic. Needs followup.
This third study is by far the most reliable, and it's conclusions were that 50% experienced no symptoms, and 1.4% of a working class population (the 2.1% included non-residents, which is less controlled) was infected, which would trend higher than average (Ninety percent of the people who were PCR positive had no capability of working from home during shelter in place.)
The other two studies just show that people in close proximity can all be infected quickly, and so be presymptomatic at the same time.
Can you provide evidence that every antibody test used in all of the studies above are as inaccurate as you claim? Your 1-word opinion on each of them is useless.
> We don't have that for viral presence tests, unless you believe the god awful seroprevalence tests.
> Can you provide evidence that every antibody test used in all of the studies above are as inaccurate as you claim? Your 1-word opinion on each of them is useless.
To my knowledge, there are basically no reliable antibody tests, which is the problem. Some are more reliable, but most are terrible, and there are very few places that publish exactly which test they use have, from my reading, used unreliable tests.
The point, which you seem to keep ignoring is that seroprevalence tests aren't reliable. So citing a bunch of them doesn't convince me. You're just citing something which I've already explained isn't reliable.
So let's assume this is true. My understanding is that PCR tests have gotten a bit better since February when that paper was published, but it sets a decent lower bound.
Let's assume the test has a 30% false-negative rate, and a .05% false-positive rate. If 1% of people are infected, and you test everyone, you'll find that (.01 * .70) + (.99 * .0005) = .75% of the population will test positive.
What about 10% of the population? (.1 * .70) + (.9 * .0005) = 7% of the population is infected according to the test. Given that PCR tests are reporting ~1% of the population is infected, we can expect that no more than 2% of the population is infected. And that assumes an incredibly good false-positive rate.
If instead the PCR test has a more reasonable false-positive rate of half a percent, and we assume an underlying 1.5% infection rate, then with a 30% false negative rate, then P(FP|N) > P(FN|P), or in other words, your test would overestimate the true infection rate (to be 1.54%, precisely).
That is to say, when the underlying infection rate is relatively low, false-positives are much more impactful than false negatives, because there are many more chances to be a false positive. This is the base rate fallacy in action.
So the upshot? Even assuming a better than real world[0] false-positive rate of the PCR tests, and a likely worse than real world false-negative rate, the PCR tests show that the true infection rate is still far below what serological tests show.
Asymptomatic non tested population is literally not supposed to be counted into CFR.