Estimating unobserved SARS-CoV-2 infections in the United States
pnas.orgIt seems cases and pandemic related deaths are likely dramatically under reported and detected:
https://austingwalters.com/u-s-covid19-less-tests-more-death...
Real deaths seem to be closer to 250-300k at this point (whereas officially it’s 170k-180k).
Also testing is down for those curious... about 25% down from a month ago.
edit: added “pandemic related deaths” - not all deaths are necessarily covid, but could be from lack of healthcare availability, etc.
I know for a fact deaths are dramatically under reported (at least in FL) because I have relatives that died of COVID and the nursing homes put congestive heart failure and respiratory failure as cause of death and not COVID. After receiving positive COVID test results...
It's much less clear cut than that. In fact death with COVID ≠ death from COVID every single time, even for elderly people.
Well sure, and that is a perfectly valid comment at the micro level. However, we can relatively clearly see that the COVID-linked death reporting is under-reported from mortality baselines.
Therefore it's highly likely that for a given death it is more likely to be incorrectly categorise as non-COVID when COVID was responsible, than to be incorrectly designated COVID.
For contrast the UK overcounted its COVID deaths
https://www.washingtonpost.com/world/britain-says-it-overcou...
It overcounted some of the test based deaths, but in terms of excess mortality there were probably still more covid deaths than have been reflected in test-positive death numbers, so it’s still an undercount.
Here is some data here on excess deaths: https://www.economist.com/graphic-detail/2020/07/15/tracking...
The parent is giving an example where it is clear cut.
"Your honor, the man didn't die because I pushed him down the stairs, he died because he hit his head against a wall at the bottom of those stairs.".
In the US if he tested positive for COVID after death, he would have been recorded as a COVID death. Not even joking.
From all the varying responses and processes related to COVID-19, there is no single 'the US' policy in any such matter.
Clearly not, given the grandparent.
https://www.cdc.gov/nchs/nvss/vsrr/covid19/excess_deaths.htm... has a similar graph as your link, but you can mouseover to get weekly numbers.
You would expect those deaths to show a signal in the all-cause, excess mortality. When I last checked, somehow, US still had lower excess mortality than the UK per capita. Although numbers are very low in the UK, whereas they are still pretty high in the US.
> 108,689 (95% posterior predictive interval [95% PPI]: 1,023 to 14,182,310)
So between 1000 and 14 million. Got it.
Not all of those numbers are equally likely though.
Interesting. An estimate that infections were multiple orders of magnitude higher than detected in mid-March is higher than I'd heard before, although I suppose it's probably implied by the "basically everyone in New York caught it" theory.
As nice as it would be, seroprevalence studies have ruled out the "basically everyone in New York caught it" theory.
It’s statistically impossible for literally everyone in NY to catch COVID.
What recent studies have shown is that while the naive herd immunity threshold is perhaps around 60% (working backwards from R0 estimates), after you factor in clustering of populations, baseline hygienic improvements, and moderate social distancing, that NY did reach an effective level of herd immunity.
I hope so. We'll find out either way.
A substantial percentage of infections are believed to not show up in seroprevalence studies. One of the major areas of research right now is determining how large this population is.
This study proposes 10:1 instead of 6:1 or so found in New York from seroprevalence. So still not anywhere near a majority getting it even if you adjusted by the difference.
> A substantial percentage of infections are believed to not show up in seroprevalence studies
What is the reason behind this?
The thresholds for antibody tests were established using known true negative samples (e.g., blood banked before the pandemic) and known true positive samples (e.g., patients who tested positive by PCR). But patients with worse symptoms are over-represented among people who tested positive by PCR (since they're more likely to seek a test), and patients with worse symptoms will generally have higher levels of antibodies in the blood. So if anything, the sensitivity of the test is probably an overestimate, which would make the number infected an underestimate.
I've seen a few papers testing asymptomatic patients (identified by contact tracing or other mass testing), with mixed results. NYC uses an in-house test for which I don't believe any paper exists, so I don't think we can say anything there. The IFR from NYC's serology is higher than most other estimates, which could imply under-ascertainment but could also be real (e.g., because they forced nursing homes to accept positive patients, because they were doing early intubation that we now know is harmful, etc.).
T-cell immunity is known to be a relevant factor, though. It'd be unreasonable to assume it fully plugs the gap, but I think there's gotta be some explanation for why only areas hit hard in April were able to keep cases low over the last few months.
Because the population has been more cautious since then?
There's nothing like seeing your friends and family struggling to breath to keep you being careful.
People have been saying 10x for months: https://www.washingtonpost.com/health/2020/06/25/coronavirus...
I would not be surprised if it were much higher.
It's not that hard to work backward from the observed fatality rate and the death curve to the infection curve two weeks ago. That's basically all they're doing in this paper.
It does not suggest that everyone in new york caught the virus. Just that in the early days of the pandemic, when schools were open and no one knew about the virus, the model for growth was exponential with about a 25% increase in cases per day, the best estimate for the number of infections was 2^(14/3)*(1/1.5%)x deaths. Each death corresponded to about 2000 cases.
Not 2000 cases, 200 cases (1 / 0.53%). If it were 2000 cases then 40 million new yorkers got COVID (which is larger than the population of NYC).
Back of the envelope based on this
* https://www.cdc.gov/nchs/nvss/vsrr/COVID19/
* https://science.sciencemag.org/content/368/6498/eabd4246
* https://worldpopulationreview.com/us-cities/new-york-city-ny...
-> about half of NYC has been infected.
you have completely missed the point. You cannot compare the number of current deaths with the number of current infections without taking into consideration the rate at which the virus is spreading.
At the beginning, the rate at which the number of deaths was growing was 26% per day, or doubling approximately every 3 days. This means that in the two weeks that it takes for the average person that is going to die of covid to die of covid, the number of people infected has grown by a factor of 2^4 to 2^5. So by the time that 30 people have died, It is reasonable to suspect that that the number of infections had grown by an order of magnitude since those people were infected, and those people are 1.5% of the people who had been infected two weeks ago. (This back of the envelope calculation is very sensitive to changes in the time to death distribution for people who have contracted covid, particularly to number of people that die fast.)
Furthermore, your infection fatality ratio is entirely wrong. My 1.5% was very optimistic. South Korea has the most exhaustively tested population on earth, and their case fatality rate is 2%, and it's worse among cases that have reached an endpoint. The virus could have mutated and attenuated since then, but other evidence suggests that the New York strain was more lethal than the SK strain, not less.
The Sciencemag paper that you have linked relies on a "seroprevalence of 3%", despite the parenthetical statement right next to their assumption that the confidence interval on that seroprevalence is between 0 and 3 percent. So not only have they chosen the maximum value for seroprevalence in that interval as their assumption, but the interval actually includes zero. Antibody testing cannot say with 95% confidence that any of its positive results were not false positives. That's a pretty bad test.
Ok acknowledged, but what's the point again? Aren't we trying to figure out if all of NYC has been infected by the virus or not? 20k deaths divided by fatality rate (give or take demographic breakdown) gets you the answer today as new cases/deaths are minimal (virus spread is minimal).
You don't like that fatality rate, so be it, but which one you believe in is all that really matters for this exercise given the virus spread <-> delay till death is not a huge factor at the moment in NYC.
I don't know where you get your numbers from, but they are off by a huge factor. The worldwide IFR is around 0.3% (stable for months), the CFR is multiplied by factor 10.
Korea's CFR is 2.3%, the IFR 0.2%. Some special countries which could not protect the elderly have much higher IFR's of 1 - 1.5%. Remember, the IFR is 10% for over 80 years and neglectable for everybody under 50.
So far there's no evidence on different strains, but lot of evidence of different policies leading to the differences. Esp. for Belgium, Spain, Italy, US and Brazil to name the worst.
https://www.medrxiv.org/content/10.1101/2020.07.23.20160895v...
I haven't exhaustively evaluated it, but they claim to be actually simulating the transmission, not just tracing the death curve.
I had wondered how the current new cases / day in NY were so low as compared to other areas. This seems likely to be at least part of it.
What's the theory on why other places that were hit hard early are still quite high...like California? Lower density?
California wasn't hit very hard early. In terms of total reported cases, they're only just now starting to catch up with New York (21 per thousand for the state vs 17 per thousand for California) - and as we catch up we've had a pretty sharp improvement from the peak in July.
The theory is that California flattened the curve in the original sense, delaying the peak in order to ensure it's moderately lower and protect the capacity of the healthcare system.
Louisiana, then?
I don't know enough about Louisiana to offer more than the calculation that they've indeed seen more reported cases per capita than New York.
Seems that at least for cases on a per-capita basis, LA got it worse than CA during both humps:
http://91-divoc.com/pages/covid-visualization/?chart=states-...
I’m sure overall prevalence in CA has a ways to go to catch up to NY (NY “cases” is a massive undercount compared to CA’s moderate undercount) but they’ll get there soon enough.
The takeaway to me is that unless you are willing to exert extremely strict border controls, localized quarantines and hard lockdowns of hotspots, and pervasive test & trace, indefinitely, then you wont keep Rt below 1 without the benefit of some herd immunity.
Hence lockdowns should either be as minimal as possible, encouraging low-risk populations to be out and about while high-risk populations shelter.... Or, you need an extreme and extraordinary response until widespread effective vaccination.
Small island nations may find they can choose the second path (although it’s not guaranteed, Hawaii has recently failed at it) but the vast majority of the world should choose the first.
And we should stop politicizing it, because largely everyone is in the same boat and will hit all the same endpoints.
That might be part of the problem; if it turns out that everyone is in the same boat and will hit the same endpoints, then it's not going to be possible to politicize it, and there is an impending presidential election in the United States where the challenger and a lot of the press are relying on the idea that the US could've done massively better if not for the current incumbent to try and unseat him.
I don't know a single person in NY who wasn't coughing for a week at some point between january and march.
In the last few days of January and two weeks in early February in Manhattan, I (26M) got quite sick. The first few days were pretty bad and I had 102F fever. I had lots of difficulty breathing and a dry cough for around 2 weeks after. A friend/co-worker who sits next to me at work had essentially identical symptoms 12 hours before me lasting for similar amount of time. I think his breathing problems were less bad than mine.
The worst hit me on the weekend and I returned to work on Monday with the cough.
Around 1 week prior I attended a company holiday party at a large museum with (probably?) one thousand attendees from all over the world. They mostly came from the EU, but at least some from APAC as well.
I have no evidence that this was COVID-19, but in retrospect the symptoms matched reasonably well and I haven't been sick since.
Get an IgG antibody test if you want to know, should still be visible there
It isn’t a fully reliable test though. There are both false positives and false negatives.
For tests in the real world, there are always false positives and false negatives. No test is 100% perfect. The question is always how high those rates are, and thus how well you can rely on the results.
The antibody test has a failure rate of 10%, the PCR test a failure rate of <0.1%.
In civilized countries the PCR test is free for critical cases (like travelling abroad or in contact with a positive), or 60 EUR if not.
Sure, but if you had symptoms AND test positive, it is quite likely (above 95%) the result is correct.
The problem with this is that people have many reasons to cough. You could say that same statement for any year and it would be accurate.
Is it possible that your NY social network at the time was part of a fairly insular clique?
That might explain why your sample is (potentially) not representative of the overall city or state.
I have friends all over NYC and Long Island and Jersey City, plus colleagues, their friends and family, etc etc. Collectively we know thousands of people.
The CDC officially became aware of coronavirus on Jan 1, and the first documented case in the USA was Jan 20. Now, we could have all had the flu, but really the flu doesn't have such a horrible deep chest cough like we all had in my opinion.
Given that some people say coronavirus was circulating as early as November 2019, I think we all had something more than the seasonal flu. By march 2020, the only people I knes who were sick were in nursing homes.
Seems unlikely. This is showing a 10:1 estimate, where New Yorkks, based on zero positivity, was 6:1 or so. Using 10:1 it would be around 25% infected in New York State.
Delhi, at least, appears to have around a 5% detection rate over the past month.
There was a population-wide serosurvey conducted from Aug 1st to 7th, which resulted in a 29.1% prevalence estimate, and an earlier one from June 27th to July 10th, which resulted in a 22.8% estimate. Assuming a 2 week period for IgG to be detected after infection, these surveys correspond to prevalence as of 20th July and 20th June roughly. With a population of 20M, that's around 1.25M new infections in that period. The confirmed cases by PCR testing increased by ~60000, yielding (roughly) a detection rate of 5%.
I've been looking at the infection rates in the US falling over the past few weeks, and I'm wondering if it's because we achieved herd immunity for the current r0, people changed their behavior when they saw cases rising, or fewer people are getting tested because with the lag in processing time, a positive result isn't actionable.
The way that the areas that are now most resilient are the same areas with high population density which were hit very hard early on (NY, NJ, for example) tells me that we have somehow hit herd immunity. There is some evidence coming out that other coronaviruses (which cause the common cold) could have causes immunity in some unknown percentage of the population, meaning the percentage of SARS-CoV-2 infection required to reach herd immunity is much lower than thought.
The herd immunity threshold depends on behavior. It may be at 30% infected when the population is wearing masks and social distancing, but could be 70% if restrictions are removed. I agree that parts of New York City have probably reached herd immunity for current behavior patterns, so restrictions can probably be slightly relaxed, but 70% of the population is still vulnerable to infection, so you can't throw caution to the wind.
It also depends on interconnectedness of social networks. If the 30% that had it is highly concentrated in one sub group, then you might see low rates now, but high rates when groups start intermingling. Herd immunity at a national level won't save nursing homes, because almost by definition, there won't be herd immunity in that local community.
This also means large, economy-driving cities, that thrive because of interconnected social networks will be slower to reopen, while rural communities might have never effectively shut down. NYC will be interesting to watch in this regard. If interconnectedness is the economic driver we think it is, we'll see slower growth until people return to cities.
In the US, there's been some push for more national action around preventing/managing the spread of covid. While I agree that more national leadership is needed, and standard guidelines around reporting and degrees of "open" would be helpful, the US is so diverse a nation-wide lockdown never made sense, and there are enough complicated factors that the decision really has to be made at local levels.
Herd immunity at a national level while protecting the vulnerable will protect nursing homes to the extent it makes introduction of the virus into nursing homes less likely. You're correct that it does nothing to help the nursing homes once the virus gets introduced there, though.
Share the evidence or delete the comment as misinformation.
Multiple papers (including one in Nature[1]) have reported T-cell immunity to SARS-CoV-2 in something between a third and half the samples collected before the pandemic. That doesn't mean the virus won't replicate in people with that pre-existing immunity, and there's evidence from homeless shelters and such of almost everyone testing positive (both by PCR and later for IgG in their blood); but people with that T cell immunity very likely get less sick, and might be less infectious (though near-certainly still infectious, and that's more speculative).
And it seems from the hostility of your response ("delete the comment as misinformation") that you find it incredible that herd immunity could exist with less than 1 - 1/R0 of the population infected? But even ignoring any pre-existing immunity, that calculation assumes a homogeneous and well-mixed population. That's clearly not the case, since some people (a nurse in a crowded ER, a police officer, a store clerk, a nightlife enthusiast, etc.) have far more contacts than others (a remote worker who gets stuff delivered). People with more contacts will get infected first, with disproportionate harm, and then become immune first, with disproportionate benefit. Many papers have modeled[2] this; though no one's found a great way to measure that heterogeneity yet, so for now, it's hard to say much beyond that the effect exists, and is potentially big.
And finally, herd immunity isn't a binary threshold, especially in a heterogeneous population. As others have noted, even places without enough immunity for R < 1 will still have slower spread than in a naive population, or may get to R < 1 from the immunity plus slightly more cautious behavior. Conversely, places that do have R < 1 overall may still have pockets of spread in sub-populations with unusually high R0. In any case, it's no conspiracy theory to believe that NYC developed significant amounts of immunity along the way to its ~24k (about 0.3% of the population!) deaths.
1. https://www.nature.com/articles/s41586-020-2550-z
2. https://www.medrxiv.org/content/10.1101/2020.02.10.20021725v...
The evidence is not hard to find. [1] Took me about 60 seconds.
I would suggest in future supplying this to the conversation yourself as it makes for a more productive discussion.
[1] https://www.nih.gov/news-events/nih-research-matters/immune-...
https://www.nih.gov/news-events/nih-research-matters/immune-...
Can't comment on the other stuff.
In my area, the lag in testing results is only 24-36 hours (that’s for free city-provided testing without a requirement to suspect you have it or have been exposed [so utterly routine processing])
You're looking at garbage data. The data has never been accurate or consistent.
Citation needed. I agree that it's not perfect, and it can actually be hard to count, but the aggregate numbers for the US tell a very consistent story, and it doesn't scream "garbage data."
I seem to be hitting trying to read this. Can someone help summarize the results of this analysis?
The core result is
> Simulating from 1 January, we obtained 108,689 (95% PPI: 1,023 to 14,182,310) local infections cumulatively in the United States by 12 March (Fig. 1A).
What that means is that they don't really know - that confidence interval is absurd - but they have reason to think there could have been 100k Americans with the coronavirus on March 12.
Results with such confidence intervals should not be allowed to be published.
As if we don't have enough fake news/hype going around.
I don't understand the comments on this thread that seem to want to suppress preliminary data about COVID.
Sure, the confidence interval is ridiculously large but the paper is open about its methods and conclusions. It's definitely in the "more work is needed" category, but I don't see how this information should not be published.
Who’s the gatekeeper we trust to decide what can and can’t be published?
100k was estimated to be infected by early March.