I’m trying to make some educated guess about the situation, but it looks like the data are very lacking. Could someone validate my logic please?
1. Some people claim that SARS-CoV-19 could have been around for ages, “everybody’s bad flu last autumn was this thing” and basically nothing is happening except panic. People are dying not because of the virus, but because everybody is going to hospitals making them overcrowded. People dying of other reasons just happen to also have almost-harmless COVID-19. When I’ve heard this couple weeks ago I chuckled, then gave it a second thought and decided that it could be possible, though unlikely. Now, given we have some antibody tests results (San Miguel County, David Friedberg in SF − 1, 2), we can conclude that “the virus has been around long before Dec 2019″ theory is false. Otherwise much more than 1% of people would be positive for antibodies as it is very infectious. Sounds right?
A note. The tests mentioned above are very unreliable (San Miguel county didn’t explain the protocol, in the second case the test kits were provided to volunteers who would test their friends or whoever they find appropriate—so non-representative sample by any means). They show that it’s not 50% herd immunity, but don’t really provide data whether it’s 0.01% or 1%.
2. To understand what will happen next, we need to know CFR (Case Fatality Rate) or better yet IFR (Infection Fatality Rate). CFR is deaths / cases, IFR is the same, but tries to account for asymptomatic cases. And that’s the problem: we don’t know percentage of asymptomatic cases. Maybe 50 people of every 1′000 infected die, and that’s 5%, pretty scary. But maybe another 99′000 are also infected, but don’t even feel it, then we get 0.05%, less than flu.
I personally watch closely Iceland statistics: https://www.covid.is/data They have a government body testing those in high-risk groups and with symptoms, and a private biotech firm testing everybody. They have conducted about 29′000 tests, 8% of the country population. This data does not contain details about whether those tested had symptoms or not, whether positives developed symptoms later or not etc, but it seems to be the closest to “test big random sample of seemingly healthy population” research I want. 28′992 samples, 1′586 infected, 6 deaths. I feel an urge to count IFR, but it would still be highly inaccurate (there is a skew because partly they test high risk groups, because older age population probably strictly follows isolation and is not exposed, and other factors). Data from Italy, or Diamond Princess cruise ship, or anywhere is very different.
I see two different options here:
1) Low IFR less than 0.5%. Maybe less than flu (~0.1%). It is still dangerous because nobody is immune, and if everybody gets infected at the same time, hospitals are overcrowded and even young patients with typically good prognosis will have complications and die without adequate care. Here we can employ FlattenTheCurve ideas (because majority of cases are mild, don’t require hospitalisation and we can “burn through” the population fast), nobody should really worry going for groceries and the quarantine may finish soon.
2) High IFR. It’s very dangerous for older generation, it’s not nice for younger generation—one will probably not die, but have a reasonable chance of visiting ICU, which is not pleasant. Business as usual, stay at home, wash your hands, wait for a vaccine next year or two.
Which case looks more probable to you? Have you seen any high-quality data suggesting one of the cases? I stress high-quality here, there are lots of reports, but mostly data look unreliable or plainly false.
If you go through my LW comment history you’ll find that I’m in the camp of “The IFR is definitely >0.3%, and very plausibly >0.8%” and that I seem to care somewhat strongly about conveying this to others. :) Maybe you’ll find some of the discussions (or links therein) useful. (Unfortunately I can’t recommend any single resource that looks super convincing all on its own.)
Edit: By “very plausibly” I mean 25% likely rather than 50% likely. By “definitely” I mean 97% likely.
We’re finally getting some results on this. An antibody test in an Italian town an hour outside Milan has been done. 2000 people out of a population of 6169 have been tested.
Results: 13-14% of the population tested positive for the antibodies (~832 people).
The town had 27 confirmed cases, with 4 confirmed deaths. 6 deaths of all causes were recorded in March.
So this is arguably good news. It implies an IFR of about .5%. If the population here is older/unhealthier than average, as seems to be true for Northern Italy as a whole, then we could see that number dipping down further. Covid being less than an order of magnitude worse than the flu is starting to look likely but there is still a long way to go to reach herd immunity and I personally think that they should at least try for actual suppression unless we get really good numbers on the death rate and are more confident that survivors won’t have lasting side effects.
Someone in that twitter thread points out that with subtracting false positives, it implies that 10% would be the better guess, as opposed to 13-14%. Does that make sense? Then 4 Covid-confirmed deaths per 620 people would be 0.66%.
And what about sampling bias? I read that the tests were voluntary. Unless someone was extremely meticulous about trying to somehow get a representative sample, I don’t think it’s reasonable to treat this as random. It’s really quite obvious that people who had flu-like symptoms for a couple of days will be more curious to go among people and have a needle stuck into them. .
I share your rough estimates of IFR in your other comment here although I was concerned about how high IFR might be with overwhelmed hospitals.
Sampling bias at its worst here would mean that IFR is 3 times more than those calculations (i.e. 1.5-2%). If this is the worst case in Lombardy where the hospitals are overwhelmed then it is something of a relief to me that higher rates are unlikely.
Very interesting, thanks. I think it’s 13% of tests, not 13% of entire population of 6′169, so not 832.
I don’t speak Italian and struggle to find any details. Actually they didn’t mention that all 2′000 samples were processed. On 5th April the mayor of the town posted a photo of newspaper mentioning “29% of 38 persons” positive. So I would not be surprised if they have taken blood from 2′000 people, processed 100 of them so far and this results got it to newspaper.
Very promising (we get more test data!), but I wouldn’t draw any conclusions on this yet.
I’m trying to make some educated guess about the situation, but it looks like the data are very lacking. Could someone validate my logic please?
1. Some people claim that SARS-CoV-19 could have been around for ages, “everybody’s bad flu last autumn was this thing” and basically nothing is happening except panic. People are dying not because of the virus, but because everybody is going to hospitals making them overcrowded. People dying of other reasons just happen to also have almost-harmless COVID-19. When I’ve heard this couple weeks ago I chuckled, then gave it a second thought and decided that it could be possible, though unlikely. Now, given we have some antibody tests results (San Miguel County, David Friedberg in SF − 1, 2), we can conclude that “the virus has been around long before Dec 2019″ theory is false. Otherwise much more than 1% of people would be positive for antibodies as it is very infectious. Sounds right?
A note. The tests mentioned above are very unreliable (San Miguel county didn’t explain the protocol, in the second case the test kits were provided to volunteers who would test their friends or whoever they find appropriate—so non-representative sample by any means). They show that it’s not 50% herd immunity, but don’t really provide data whether it’s 0.01% or 1%.
2. To understand what will happen next, we need to know CFR (Case Fatality Rate) or better yet IFR (Infection Fatality Rate). CFR is deaths / cases, IFR is the same, but tries to account for asymptomatic cases. And that’s the problem: we don’t know percentage of asymptomatic cases. Maybe 50 people of every 1′000 infected die, and that’s 5%, pretty scary. But maybe another 99′000 are also infected, but don’t even feel it, then we get 0.05%, less than flu.
I personally watch closely Iceland statistics: https://www.covid.is/data They have a government body testing those in high-risk groups and with symptoms, and a private biotech firm testing everybody. They have conducted about 29′000 tests, 8% of the country population. This data does not contain details about whether those tested had symptoms or not, whether positives developed symptoms later or not etc, but it seems to be the closest to “test big random sample of seemingly healthy population” research I want. 28′992 samples, 1′586 infected, 6 deaths. I feel an urge to count IFR, but it would still be highly inaccurate (there is a skew because partly they test high risk groups, because older age population probably strictly follows isolation and is not exposed, and other factors). Data from Italy, or Diamond Princess cruise ship, or anywhere is very different.
I see two different options here:
1) Low IFR less than 0.5%. Maybe less than flu (~0.1%). It is still dangerous because nobody is immune, and if everybody gets infected at the same time, hospitals are overcrowded and even young patients with typically good prognosis will have complications and die without adequate care. Here we can employ FlattenTheCurve ideas (because majority of cases are mild, don’t require hospitalisation and we can “burn through” the population fast), nobody should really worry going for groceries and the quarantine may finish soon.
2) High IFR. It’s very dangerous for older generation, it’s not nice for younger generation—one will probably not die, but have a reasonable chance of visiting ICU, which is not pleasant. Business as usual, stay at home, wash your hands, wait for a vaccine next year or two.
Which case looks more probable to you? Have you seen any high-quality data suggesting one of the cases? I stress high-quality here, there are lots of reports, but mostly data look unreliable or plainly false.
Thank you.
If you go through my LW comment history you’ll find that I’m in the camp of “The IFR is definitely >0.3%, and very plausibly >0.8%” and that I seem to care somewhat strongly about conveying this to others. :) Maybe you’ll find some of the discussions (or links therein) useful. (Unfortunately I can’t recommend any single resource that looks super convincing all on its own.)
Edit: By “very plausibly” I mean 25% likely rather than 50% likely. By “definitely” I mean 97% likely.
We’re finally getting some results on this. An antibody test in an Italian town an hour outside Milan has been done. 2000 people out of a population of 6169 have been tested.
Results: 13-14% of the population tested positive for the antibodies (~832 people).
The town had 27 confirmed cases, with 4 confirmed deaths. 6 deaths of all causes were recorded in March.
So this is arguably good news. It implies an IFR of about .5%. If the population here is older/unhealthier than average, as seems to be true for Northern Italy as a whole, then we could see that number dipping down further. Covid being less than an order of magnitude worse than the flu is starting to look likely but there is still a long way to go to reach herd immunity and I personally think that they should at least try for actual suppression unless we get really good numbers on the death rate and are more confident that survivors won’t have lasting side effects.
Someone in that twitter thread points out that with subtracting false positives, it implies that 10% would be the better guess, as opposed to 13-14%. Does that make sense? Then 4 Covid-confirmed deaths per 620 people would be 0.66%.
And what about sampling bias? I read that the tests were voluntary. Unless someone was extremely meticulous about trying to somehow get a representative sample, I don’t think it’s reasonable to treat this as random. It’s really quite obvious that people who had flu-like symptoms for a couple of days will be more curious to go among people and have a needle stuck into them. .
I share your rough estimates of IFR in your other comment here although I was concerned about how high IFR might be with overwhelmed hospitals.
Sampling bias at its worst here would mean that IFR is 3 times more than those calculations (i.e. 1.5-2%). If this is the worst case in Lombardy where the hospitals are overwhelmed then it is something of a relief to me that higher rates are unlikely.
Both good points. Hopefully we get more tests of the sort reported soon.
Very interesting, thanks. I think it’s 13% of tests, not 13% of entire population of 6′169, so not 832.
I don’t speak Italian and struggle to find any details. Actually they didn’t mention that all 2′000 samples were processed. On 5th April the mayor of the town posted a photo of newspaper mentioning “29% of 38 persons” positive. So I would not be surprised if they have taken blood from 2′000 people, processed 100 of them so far and this results got it to newspaper.
Very promising (we get more test data!), but I wouldn’t draw any conclusions on this yet.