Some more information about the study here: <a href="https://pressroom.usc.edu/what-a-usc-la-county-antibody-study-can-teach-us-about-covid-19/" rel="nofollow">https://pressroom.usc.edu/what-a-usc-la-county-antibody-stud...</a><p>Population is supposed to be representative of LA county. Tests are from Premier Biotech with some preliminary data on false negative/ false positive rates shown. I suspect assumptions related to the test itself, rather than population, may be a bigger factor for any inaccuracies. There's also the question of whether antibody response implies immunity for SARS-COV2<p>Another interesting aspect of the result was 6% of men showed antibodies but only 2% of women tested.<p>Also:
--2.4% of people between the ages of 18 and 34 had antibodies to the coronavirus
--5.6% of people between 35 and 54 had antibodies
--4.3% of between 55 and older had antibodies<p>Edit: The key assumption that really, really needs to be triply checked here is the specificity of the test. They're assuming 99.5%, similar to the Stanford study I believe. If it's closer to 98% then prevalence estimate confidence intervals should include 0%, I believe. Also should be noted that the same PI on the Stanford study, Jay Bhattacharya, is also involved in this. Same criticisms likely apply here - more caution needed!<p>Second edit: On further reflection, I think these studies could do serious damage based on how the results are being announced by press-release - in this case, afaik there's no pre-print. In our current environment this isn't <i>just</i> science. Policy is being enacted and people may die because of poorly framing results. Everyone involved needs to take a step back and think through this carefully. Also - did anyone figure out why a hedge fund guy was an author on the Stanford pre-print? And why that same guy then published a Wall Street Journal op-ed about the study without identifying himself as a co-author? Shady stuff.
Here is a basic question: how are the sensitivity and specificity (false negative and false positive rates, IIRC) of these tests determined?<p>It seems like a learning problem with noisy labels. Does anyone know what they do in practice? Especially chaotic situations like this where there aren't reference tests? Are there reference positive and negative samples?<p>One concern also referenced for the Stanford study was that, as mentioned in a USC popular writeup on this study linked elsewhere on this post,<p>> Premier Biotech, the manufacturer of the test that USC and L.A. County are using, tested blood from COVID-19-positive patients with a 90 to 95% accuracy rate.<p>They must be factoring this into the 2-5% infection rate number somewhere?
This LA study has one of the same authors as the Stanford study (Neeraj Sood). It uses the same test and so it has the same issues with the false positive rate. An honest person would address those but I didn't see that in the press release.<p>This post describes the issues: <a href="https://medium.com/@balajis/peer-review-of-covid-19-antibody-seroprevalence-in-santa-clara-county-california-1f6382258c25" rel="nofollow">https://medium.com/@balajis/peer-review-of-covid-19-antibody...</a><p>See also this thread: <a href="https://news.ycombinator.com/item?id=22924118" rel="nofollow">https://news.ycombinator.com/item?id=22924118</a>
Both the LA and Stanford studies are junk. Easily explained by false positives inherent to the test. Stuff like this is why pre-print services can have disastrous consequences.<p><a href="https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaws-in-stanford-study-of-coronavirus-prevalence/" rel="nofollow">https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaw...</a>
The Stanford study in Santa Clara that used Premier Biotech appears to be very misleading. Given the confidence intervals on the specificity all 50 positive cases out of the 3330 tested could just be false positives.<p>See Gelman's article on the topic.<p><a href="https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaws-in-stanford-study-of-coronavirus-prevalence/" rel="nofollow">https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaw...</a><p>I am similarly skeptical of the findings here.
They appear to be using an antibody test from Premier Biotech. However, I can't gain clarity on whether we know if these tests are even accurate or reliable.<p>"But some COVID-19 antibody tests, including those being used by public health departments in Denver and Los Angeles and provided to urgent care centers in Maryland and North Carolina, were supplied by Chinese manufacturers that are not approved by China's Center for Medical Device Evaluation, a unit of the National Medical Product Administration, or NMPA, the country's equivalent of the U.S. Food and Drug Administration, NBC News has found.<p>Two U.S. companies — Premier Biotech of Minneapolis and Aytu Bioscience of Colorado — have been distributing the tests from unapproved Chinese manufacturers, according to health officials, FDA filings and a spokesman for one of the Chinese manufacturers. Many of the unapproved tests appear to have been shipped to the U.S. after the FDA relaxed its guidelines for tests in mid-March and before the Chinese government banned their export just over two weeks later.<p>If COVID-19 antibody tests are unreliable, they can produce false results, either negative or positive, health officials said. The use of such tests has been widely discussed as a way to ensure that employees are healthy enough to go back to work and to find COVID-19 survivors who may be able to provide blood plasma to severely ill patients.<p>Officials at the Association of Public Health Laboratories have expressed concern about the reliability of the numerous antibody tests being sold or used across the country with little scrutiny. "<p>see: <a href="https://www.nbcnews.com/health/health-news/unapproved-chinese-coronavirus-antibody-tests-being-used-least-2-states-n1185131" rel="nofollow">https://www.nbcnews.com/health/health-news/unapproved-chines...</a>
Does anyone have a link to the actual study? LA Times link doesn't include it. Other than the selection criteria for participants, it doesn't sound from the pressroom.usc.edu link that it's any better than the Santa Clara study, and perhaps worse, given the smaller number of participants. The pressroom link mentions an "accuracy" number, but not specificity and sensitivity.
NYC has 20x the deaths / 100k people, so under the assumption that number of deaths are on comparable trajectories (due to both locations being under lockdown, maybe not as unrealistic as a few weeks ago), NYC's prevalence could be 56%-112%. This could be one explanation for why numbers are going down in NYC, or the 112% being larger than 100% could suggest that either the situations aren't comparable or that the LA study's numbers are a bit high.
There's a logical jump from the 2nd to 3rd paragraph that bugs me, but upon reflection, I don't know if my objection is valid:<p>> The initial results from the first large-scale study tracking the spread of the coronavirus in the county found that 2.8% to 5.6% of adults have antibodies to the virus in their blood, an indication of past exposure.<p>> That translates to roughly 221,000 to 442,000 adults who have recovered from an infection, according to the researchers conducting the study, even though the county had reported fewer than 8,000 cases at that time.<p>So, in the 2nd paragraph, there, a match for SARS-COV2 antibodies means an exposure. In the 3rd, it means an infection. I have always understood that exposure != infection. Yes, the former is a necessary precondition to the latter, but an exposure doesn't automatically and unconditionally become an infection.<p>But then it occurred to me that the body probably isn't going to be generating antibodies unless a pathogen got past the body's initial lines of defense and required a more active response from the immune system. In which case then, perhaps it would be accurate to say antibodies == infection. Are we speaking then of asymptomatic carriers or those who never experienced more than mild symptoms and might not have realized they were infected?<p>Am I splitting hairs here? Is this simply shorthand for "exposure that subsequently became an infection?", much in the same way I might freely use "LDAP Server", "Name Server", "Domain Controller" "Auth Server", to all refer to the same system in an on-prem Windows environment? Or, am I rightly objecting to unclear language that would lead to incorrect conclusions?
<a href="https://covid19tracker.health.ny.gov/views/NYS-COVID19-Tracker/NYSDOHCOVID-19Tracker-Fatalities?%3Aembed=yes&%3Atoolbar=no&%3Atabs=n" rel="nofollow">https://covid19tracker.health.ny.gov/views/NYS-COVID19-Track...</a><p>89% of total fatalities in New York have at least 1 comorbidity. I wonder to what extent anyone with unknown cause of death or death due to cancer, but also had COVID-19 gets lumped into the death due to COVID-19 category.
With the NYC analysis here, everyone is missing one unknowable piece of information.<p>We know that the initial major outbreak in NYC was with a strain from Europe, while the initial major outbreak in California was with an Asia-origin strain.<p>These might be differently dangerous. If the NYC/Europe strain is just 20%-30% more fatal than the Asian one, the maths all start being believable (you start saying that 30-45% of NYC has been infected -- which, unlike 60-100+%, is quite believable.<p>It seems likely that some vulnerable neighborhoods in NY have effectively everyone infected, but that is probably not universally true.<p>It also seems quite believable that we are nearing the point where the majority of folks in NYC have been exposed. If nothing else, the subway is still seeing quite active use, and must be a breeding ground for infection -- and yet, hospitalization has started declining, meaning that new infections are slowing down -- meaning that at least among folks that still go out and about, there must just be fewer people to infect.<p>With subway use still being as high as it is, it seems like it must be true that, amongst the population using it, R could only fall below 1 if herd immunity (in the rider population) was beginning to form.<p>This also jives with the numbers we are seeing, where reported cases are down despite both increased testing and looser counting.
Interesting that this article says only 2~3% with antibodies globally. I'm left wondering how reliable the testing (of all kinds) is.<p><a href="https://www.theguardian.com/society/2020/apr/20/studies-suggest-very-few-have-had-covid-19-without-symptoms" rel="nofollow">https://www.theguardian.com/society/2020/apr/20/studies-sugg...</a>
This recent Swedish study was done in (what seems) to be a way that's much less prone to bias than soliciting people via Facebook posts.<p>It was also based on tests collected from blood donor samples, so none of the specific bias of the Stanford study (which to me also seems like a huge flaw even if other sources of evidence are pointing to similar IFR numbers and undercounting)<p>In any case, the Swedish findings cited for Stockholm in the analysis linked to in this COVID19 subreddit post indicate a similarly low overall IFR of far below 1%.<p>It is worth noting that blood donors would probable tend more towards youth and general good health, which could skew its demographic profile. Also, depending on donors' motives for donating (maybe they thought it would include a COVID test) could skew results in different and unpredictable ways.<p><a href="https://www.reddit.com/r/COVID19/comments/g4znbg/at_least_11_of_tested_blood_donors_in_stockholm/" rel="nofollow">https://www.reddit.com/r/COVID19/comments/g4znbg/at_least_11...</a>
Understanding this study is very difficult because they haven't released all of the details about statistical procedures, recruitment, raw data, etc, but I would be worried that this study will have similar issues to the recently published Santa Clara study (both use the same lab test and share several authors).<p>I highly recommend that people read <a href="https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaws-in-stanford-study-of-coronavirus-prevalence/" rel="nofollow">https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaw...</a> for information about for information on how the Santa Clara study messed up. TLDR: Incorrect statistics, incorrect analysis, and incorrect sampling probably led to a vast overestimate of both prevalence and certainty in that high prevalence estimate.
Tl;dr: It looks like ~4.2% of Los Angeles residents may have already been affected.<p>Current stats from [<a href="https://covid-19.direct/county/CA/Los%20Angeles" rel="nofollow">https://covid-19.direct/county/CA/Los%20Angeles</a>]:
Confirmed cases: 12,349
Deaths: 601
Population: 4M
Implied fatality rate: 4.87%<p>If this study is correct (~4.2% prevalence):
Cases: 167,580
Deaths: 601
Implied fatality rate: 0.36%<p>There is massive difference between the appropriate response to an illness with a 5% fatality rate and an illness with a 0.36% fatality rate.
863 people were in this study. A market data firm selected those people.<p>LA proper has about 16,000,000 people.<p>This study should be taken with a grain of salt either way for now.
Not sure what significance there is. Sure, some intermediate statistics change (e.g. what is the true percentage of bad reactions). But nothing changes the true death rate, which is what, the single largest mortality stat in America right now? Arguing over this fraction and that, isn't going to make this less deadly.
3.4% mortality rate reported by WHO seems to be not as accurate then <a href="https://www.who.int/dg/speeches/detail/who-director-general-s-opening-remarks-at-the-media-briefing-on-covid-19---3-march-2020" rel="nofollow">https://www.who.int/dg/speeches/detail/who-director-general-...</a><p>---
Globally, about 3.4% of reported COVID-19 cases have died
---<p>That was used as a basis quite a few decisions, how to treat patients, when to close / open various locations, etc. It does say "reported" in the report, but I think often that part of ignored and many officials assumed it to be as a rate based on infected cases.<p>The article does mention this issue, and I wonder what the new / updated mortality rate is then based on this data.