SARS-CoV-2 testing

Covid-19 discussion, bring your own statistics
Post Reply
User avatar
Bird on a Fire
Stummy Beige
Posts: 2956
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: SARS-CoV-2 testing

Post by Bird on a Fire » Thu May 14, 2020 3:24 pm

shpalman wrote:
Thu May 14, 2020 3:10 pm
lpm wrote:
Thu May 14, 2020 1:19 pm
Mancester Uni: over 25% of UK has had it...

https://www.manchester.ac.uk/discover/n ... 9-already/
i.e. 26% https://www.medrxiv.org/content/10.1101 ... 20072264v1
The analysis is based on reported number of confirmed cases up to 14th April, without taking into account changes in the number of tests conducted nor criteria for who gets tested. That approach seems likely to exaggerate the rate of transmission.

Judging by their Figure 1, their results suggest that the proportion of infections that get confirmed by testing declined steadily from mid-March.

Based on a skimread, they are taking the reported numbers at face value and don't acknowledge any effect of testing. Seems like a clearcut case of garbage-in-garbage-out to me.

Not many epidemiology departments on the author list, either.
Born at 356.32 ppm CO2

User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Thu May 14, 2020 4:17 pm

They also seem to uncritically cite that Santa Clara study.

It's probably the case that back at the beginning of April there were far more cases out there compared to the positive test results, just because there was so little testing (on the 14th of April, 302599 people had been tested and there had been 93873 cases).

The daily case rate has hardly come down since then, though; if you tried to follow their numbers through to the present day you'd probably end up >100%, but I'm not going to try to do this right now.

I reckon the total number of true cases in the UK might be of the order of a few million? Let's say 3 million. 4.5% of the population, then. That would get the CFR down to about 1%.
molto tricky

User avatar
Bird on a Fire
Stummy Beige
Posts: 2956
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: SARS-CoV-2 testing

Post by Bird on a Fire » Thu May 14, 2020 5:23 pm

shpalman wrote:
Thu May 14, 2020 4:17 pm
It's probably the case that back at the beginning of April there were far more cases out there compared to the positive test results, just because there was so little testing (on the 14th of April, 302599 people had been tested and there had been 93873 cases).
Indeed, but they don't allow for undetected cases. They assume that all the reported cases in five days' time resulted from the reported cases five days previously, which is obviously going to massively inflate the transmission rate. Then they extrapolate that totally crazy number for a month to get the 25% figure, but it's obviously bollocks.

Preprints are just PDFs. Maybe I'll submit one using a random number generator to model covid cases - couldn't be much worse.
Born at 356.32 ppm CO2

User avatar
Gfamily
Dorkwood
Posts: 1512
Joined: Mon Nov 11, 2019 1:00 pm

Re: SARS-CoV-2 testing

Post by Gfamily » Thu May 14, 2020 6:58 pm

As I had a sore throat and a touch of a headache yesterday, I reported them on the Covid symptom reporting app. Today it asked me if I would apply for a test.

I didn't fancy having to drive to a centre (working full time as it is), so I waited until the next batch of Home Test kits became available at 5pm and managed to get one.

I don't think I have Covid, but if it'll help them to develop machine learning for symptoms...
My avatar was a scientific result that was later found to be 'mistaken' - I rarely claim to be 100% correct
ETA 5/8/20: I've been advised that the result was correct, it was the initial interpretation that needed to be withdrawn
Meta? I'd say so!

User avatar
Woodchopper
Dorkwood
Posts: 1593
Joined: Sat Oct 12, 2019 9:05 am

Re: SARS-CoV-2 testing

Post by Woodchopper » Thu May 14, 2020 7:20 pm


User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Thu May 14, 2020 8:13 pm

Woodchopper wrote:
Thu May 14, 2020 7:20 pm
Data from Belgium
https://www.gva.be/cnt/dmf20200514_0495 ... -overleven

IFR of 0.97
0.0097

or, 0.97%
molto tricky

User avatar
lpm
Dorkwood
Posts: 1438
Joined: Mon Nov 11, 2019 1:05 pm
Location: IMPEACH AND EXTERMINATE

Re: SARS-CoV-2 testing

Post by lpm » Thu May 14, 2020 9:56 pm

Bird on a Fire wrote:
Thu May 14, 2020 5:23 pm
shpalman wrote:
Thu May 14, 2020 4:17 pm
It's probably the case that back at the beginning of April there were far more cases out there compared to the positive test results, just because there was so little testing (on the 14th of April, 302599 people had been tested and there had been 93873 cases).
Indeed, but they don't allow for undetected cases. They assume that all the reported cases in five days' time resulted from the reported cases five days previously, which is obviously going to massively inflate the transmission rate. Then they extrapolate that totally crazy number for a month to get the 25% figure, but it's obviously bollocks.

Preprints are just PDFs. Maybe I'll submit one using a random number generator to model covid cases - couldn't be much worse.
The Mirror front page is leading with this Manchester thing.
I'll miss him after he's died in the pandemic

User avatar
Bird on a Fire
Stummy Beige
Posts: 2956
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: SARS-CoV-2 testing

Post by Bird on a Fire » Thu May 14, 2020 10:39 pm

Oh dear.
Born at 356.32 ppm CO2

plebian
Stargoon
Posts: 121
Joined: Wed Sep 25, 2019 4:45 pm

Re: SARS-CoV-2 testing

Post by plebian » Thu May 14, 2020 11:14 pm

Cue headlines of "almost half way to herd immunity".
Pithy, well, a right helmet at least.

User avatar
Brightonian
Fuzzable
Posts: 285
Joined: Mon Nov 11, 2019 3:16 pm

Re: SARS-CoV-2 testing

Post by Brightonian » Thu May 14, 2020 11:46 pm


monkey
Sindis Poop
Posts: 58
Joined: Wed Nov 13, 2019 5:10 pm

Re: SARS-CoV-2 testing

Post by monkey » Fri May 15, 2020 3:15 pm

Bird on a Fire wrote:
Thu May 14, 2020 3:24 pm
shpalman wrote:
Thu May 14, 2020 3:10 pm
lpm wrote:
Thu May 14, 2020 1:19 pm
Mancester Uni: over 25% of UK has had it...

https://www.manchester.ac.uk/discover/n ... 9-already/
i.e. 26% https://www.medrxiv.org/content/10.1101 ... 20072264v1
The analysis is based on reported number of confirmed cases up to 14th April, without taking into account changes in the number of tests conducted nor criteria for who gets tested. That approach seems likely to exaggerate the rate of transmission.

Judging by their Figure 1, their results suggest that the proportion of infections that get confirmed by testing declined steadily from mid-March.

Based on a skimread, they are taking the reported numbers at face value and don't acknowledge any effect of testing. Seems like a clearcut case of garbage-in-garbage-out to me.

Not many epidemiology departments on the author list, either.
There's another problem (I think, I'm no expert). Their 25% infected value comes from an extrapolation of the linear fit shown in Fig 4. However they assume that the daily infection rate goes to zero when 100% of the poulation is infected. But shouldn't that be when herd immunity is achieved? I've read elsewhere that proportion is 50-60%. So the 25% should be reduced by this amount, so 12.5 - 15.6%.

For a back of the fag packet calculation, it looks like the number of tests increased of the time period used by about 33% (estimated by eyeballing a graph), if you also account for that, you'd reduce the value by a further third, giving ~8-10% infected.

User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Fri May 15, 2020 4:24 pm

monkey wrote:
Fri May 15, 2020 3:15 pm
There's another problem (I think, I'm no expert). Their 25% infected value comes from an extrapolation of the linear fit shown in Fig 4. However they assume that the daily infection rate goes to zero when 100% of the poulation is infected. But shouldn't that be when herd immunity is achieved? I've read elsewhere that proportion is 50-60%. So the 25% should be reduced by this amount, so 12.5 - 15.6%.
As I understand it, if you didn't introduce any particular social measures (so R hasn't been changed), the infection peaks at about 1-1/R0, i.e. the point at which out of the R0 people which each infected person could infect (and would infect in the case of a completely susceptible population) there's exactly 1 who is still susceptible.
molto tricky

User avatar
sTeamTraen
Snowbonk
Posts: 589
Joined: Mon Nov 11, 2019 4:24 pm

Re: SARS-CoV-2 testing

Post by sTeamTraen » Fri May 15, 2020 5:04 pm

lpm wrote:
Thu May 14, 2020 9:56 pm
Bird on a Fire wrote:
Thu May 14, 2020 5:23 pm
shpalman wrote:
Thu May 14, 2020 4:17 pm
It's probably the case that back at the beginning of April there were far more cases out there compared to the positive test results, just because there was so little testing (on the 14th of April, 302599 people had been tested and there had been 93873 cases).
Indeed, but they don't allow for undetected cases. They assume that all the reported cases in five days' time resulted from the reported cases five days previously, which is obviously going to massively inflate the transmission rate. Then they extrapolate that totally crazy number for a month to get the 25% figure, but it's obviously bollocks.

Preprints are just PDFs. Maybe I'll submit one using a random number generator to model covid cases - couldn't be much worse.
The Mirror front page is leading with this Manchester thing.
The Manchester paper Is utter, utter shite. Grab yer popcorn and settle in.
Sitting in a sleazy snackbar sucking sickly sausage rolls

monkey
Sindis Poop
Posts: 58
Joined: Wed Nov 13, 2019 5:10 pm

Re: SARS-CoV-2 testing

Post by monkey » Fri May 15, 2020 5:22 pm

shpalman wrote:
Fri May 15, 2020 4:24 pm
monkey wrote:
Fri May 15, 2020 3:15 pm
There's another problem (I think, I'm no expert). Their 25% infected value comes from an extrapolation of the linear fit shown in Fig 4. However they assume that the daily infection rate goes to zero when 100% of the poulation is infected. But shouldn't that be when herd immunity is achieved? I've read elsewhere that proportion is 50-60%. So the 25% should be reduced by this amount, so 12.5 - 15.6%.
As I understand it, if you didn't introduce any particular social measures (so R hasn't been changed), the infection peaks at about 1-1/R0, i.e. the point at which out of the R0 people which each infected person could infect (and would infect in the case of a completely susceptible population) there's exactly 1 who is still susceptible.
That means I'm wrong about herd immunity doesn't it? That tells you how many infections you have at the peak, not the total. But am I right thinking that not 100% of the (susceptible) population would get infected? That doesn't seem right to me.

User avatar
lpm
Dorkwood
Posts: 1438
Joined: Mon Nov 11, 2019 1:05 pm
Location: IMPEACH AND EXTERMINATE

Re: SARS-CoV-2 testing

Post by lpm » Fri May 15, 2020 6:52 pm

When about 67% have immunity, if R=3, it fizzles naturally.

But if 20% have caught it and are immune, the "neutral" point isn't R=1, it's R=1.25. Which us useful - lockdowns slow it faster and/or the lockdown can be a bit easier.

But the current 5% isn't much help. Still, London and New York have a nice bit of herd immunity.
I'll miss him after he's died in the pandemic

Millennie Al
Clardic Fug
Posts: 207
Joined: Mon Mar 16, 2020 4:02 am

Re: SARS-CoV-2 testing

Post by Millennie Al » Sat May 16, 2020 2:10 am

lpm wrote:
Thu May 14, 2020 1:19 pm
Mancester Uni: over 25% of UK has had it...

https://www.manchester.ac.uk/discover/n ... 9-already/
It is highly implausible that lots of people (substatially more than half of those infected) have been asymptomatic cases, regardless of any modelling or guesswork. This is based on contact tracing experiences. Let Sx represent symptomatic cases and Ax asymptomatic cases and X->Y mean that X infects Y. If we have S0->S1 then contact tracing easily explains how S1 got infected. If we have S0->A0->S1 or S0->A0->A1->S1, it's still easy - get plausible contacts for S0 and S1 and see the link that might have passed on the infection. Beyond that, things get difficult and it starts to look like S1 got infected for no reason.

If there are equal numbers of S and A, then the probability of S->S is 0.5, S->A->S is 0.25, S->A->A->S is 0.125, which means all other possibilites are 0.125. As the proportion of A rises, this value also rises, so we can estimate the ratio of A to S by the proportion of symptomatic cases which seem to have no cause, and we can get a general feel for this by reading the news to see if in some parts of the world where contact tracing is done there are lots of unexplained cases or few. Lots of unexplained cases would be very alarming to anyone trying to do contact tracing, so would be very newsworthy. Since we read of contact tracing being very successful, we can conclude that there cannot be many A for each S. This means it's hardly worth bothering to read any paper which claims there is a huge number of asymptomatic cases.

Unless, there's a bad assumption, namely that S->A is just like A->S. If instead A->S is impossible, then we could have each S infect three more S and sixty A. That would mean we would never see unexplained infections, but there are twenty times as many asymptomatic cases as symptomatic cases. The problem with this hypothesis is that it should show up during contact tracing whereby when the contacts of an infected person are found and tested, in cases where there has been excessive delay in locating the individuals, we would see many people who test positive, are asymptomatic, and have not infected anyone else. That too would be quite remarkable, so I'd expect it would be newsworthy, so the lack of stories about it suggests it doesn't occur, undermining the plausibility of this mechanism.

Another odd case that would work is that S->A is rare, but A->A is common and A->S is impossible. That would make it hard to notice the S->A infections, while allowing the numbers of A to grow wildly. Of course this must mean that something very odd is happening in the S->A step - the disease is mutating into an asymptomatic variant. This seems extreely unlikely.

So overall, I consider the hypothesis that there are lots of asymptomatic cases to be sufficiently implausible that it requires very strong evidence to support it - that means actual testing and not just modelling.
Covid-19 - Don't catch it: don't spread it.

User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Sat May 16, 2020 11:20 am

I've put a simple SIR (susceptible->infected->recovered) model in a spreadsheet: https://docs.google.com/spreadsheets/d/ ... sp=sharing

I think you should be able to copy it and play with the contagion factor and recovery time, which together give R0, and changes how many people are infected at the peak and how many people get infected in total.

It's won't really fit the COVID data since in that case you would need a more complicated
susceptible->exposed->infectious->quarantined(maybe)->recovered/died
model.
molto tricky

User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Sat May 16, 2020 2:59 pm

Lombardy's numbers yesterday said that 325701 people had been tested, and there have been 84199 total positives (only 27746 are still positive). So maybe 25% of the people tested in Lombardy so far have ever been positive. I would have trouble with assuming that could be scaled up to mean 25% of the entire population of Lombardy (which would be 2,500,000 people).
molto tricky

User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Sat May 16, 2020 3:32 pm

shpalman wrote:
Sat May 16, 2020 2:59 pm
Lombardy's numbers yesterday said that 325701 people had been tested, and there have been 84199 total positives (only 27746 are still positive). So maybe 25% of the people tested in Lombardy so far have ever been positive. I would have trouble with assuming that could be scaled up to mean 25% of the entire population of Lombardy (which would be 2,500,000 people).
Oh and the equivalent number for the whole of the UK is about 14%, based on about 3% of the population having been tested.
molto tricky

AMS
Fuzzable
Posts: 233
Joined: Mon Nov 11, 2019 11:14 pm

Re: SARS-CoV-2 testing

Post by AMS » Sat May 16, 2020 4:23 pm

shpalman wrote:
Sat May 16, 2020 2:59 pm
Lombardy's numbers yesterday said that 325701 people had been tested, and there have been 84199 total positives (only 27746 are still positive). So maybe 25% of the people tested in Lombardy so far have ever been positive. I would have trouble with assuming that could be scaled up to mean 25% of the entire population of Lombardy (which would be 2,500,000 people).
We'd need to know how much of the testing is of symptomatic patients, or of medical staff with higher chance of exposure, and how much is random testing out in the community.

User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Sun May 17, 2020 5:36 am

https://www.vox.com/2020/5/16/21259492/ ... s-immunity
The new Spanish survey (which is still underway) has a much higher-quality research design [compared to the flawed studies from California], and unfortunately, it has a much less reassuring result at this point: Only about 5 percent of people in the study have tested positive for Covid-19 antibodies, suggesting that just 5% of the Spanish population has had the coronavirus.

What they found was that about 10-14% of the population in and around Madrid has antibodies, along with about 7% of the population of Barcelona and smaller numbers outside of Spain’s two major cities. Across the country, it averages out to roughly 5%.
molto tricky

User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Sun May 17, 2020 7:39 am

shpalman wrote:
Sun May 17, 2020 5:36 am
https://www.vox.com/2020/5/16/21259492/ ... s-immunity
The new Spanish survey (which is still underway) has a much higher-quality research design [compared to the flawed studies from California], and unfortunately, it has a much less reassuring result at this point: Only about 5% of people in the study have tested positive for Covid-19 antibodies, suggesting that just 5 percent of the Spanish population has had the coronavirus.

What they found was that about 10-14% of the population in and around Madrid has antibodies, along with about 7% of the population of Barcelona and smaller numbers outside of Spain’s two major cities. Across the country, it averages out to roughly 5%.
Apparently there have been 230,698 confirmed cases and 27,563 deaths as of 16th May, while 5% of the population would be about 2 million. There may also been 5,700–6,000 more deaths than officially confirmed, according to the excess mortality.

That's from https://en.wikipedia.org/wiki/COVID-19_ ... c_in_Spain

The https://en.wikipedia.org/wiki/COVID-19_ ... _of_Madrid page hasn't been updated in a while so somebody else can check what 10-14% of Madrid would mean.
molto tricky

User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Sun May 17, 2020 7:46 am

AMS wrote:
Sat May 16, 2020 4:23 pm
shpalman wrote:
Sat May 16, 2020 2:59 pm
Lombardy's numbers yesterday said that 325701 people had been tested, and there have been 84199 total positives (only 27746 are still positive). So maybe 25% of the people tested in Lombardy so far have ever been positive. I would have trouble with assuming that could be scaled up to mean 25% of the entire population of Lombardy (which would be 2,500,000 people).
We'd need to know how much of the testing is of symptomatic patients, or of medical staff with higher chance of exposure, and how much is random testing out in the community.
Well, I don't know that, but I don't think there has been any random testing out in the community apart from places like which would only count for a few thousand people (I seem to remember the swabs indicating about 3% of the population had it) or Ortisei (in which a few hundred people were tested for antibodies and about half had them).

I think since testing capacity has been ramping up steadily, it's more likely that medical staff are being tested, and I anecdotally know of one asymptomatic case who was told to isolate at home.
molto tricky

User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Sun May 17, 2020 9:02 am

Bird on a Fire wrote:
Thu May 14, 2020 3:24 pm
shpalman wrote:
Thu May 14, 2020 3:10 pm
lpm wrote:
Thu May 14, 2020 1:19 pm
Mancester Uni: over 25% of UK has had it...

https://www.manchester.ac.uk/discover/n ... 9-already/
i.e. 26% https://www.medrxiv.org/content/10.1101 ... 20072264v1
The analysis is based on reported number of confirmed cases up to 14th April, without taking into account changes in the number of tests conducted nor criteria for who gets tested. That approach seems likely to exaggerate the rate of transmission.

Judging by their Figure 1, their results suggest that the proportion of infections that get confirmed by testing declined steadily from mid-March.

Based on a skimread, they are taking the reported numbers at face value and don't acknowledge any effect of testing. Seems like a clearcut case of garbage-in-garbage-out to me.

Not many epidemiology departments on the author list, either.
I went and had a look at the actual data for England for the same period as their Fig. 1 and I plotted it on the same scale and with the same colours, only I used a seven-day-either-side average to smooth out the weekly fluctuations rather than using a polynomial trendline (there is no physical basis for the trend to be polynomial, aside from the polynomials being a completely set so that if you went to infinite order to could describe any function with a polynomial).
England-Fig1.png
England-Fig1.png (40.26 KiB) Viewed 437 times
Two things you will notice; firstly that at the time they did their analysis, the death data had not been updated to include covid deaths outside hospital (which is why my deaths data goes off the scale). Secondly, the case data does not fall off at the end of the graph. There was possibly the artefact in their case of late reporting, i.e. the past few days' case data hasn't come in yet (or maybe it has for the official UK total but hadn't been attributed properly to region in the table) so the numbers are too low. There's a dip on the 12th of April because that was a Sunday, and weekend reporting is always lower, but it comes back up.

If we extend it to all the data I have as of now it looks like this:
England-Fig1-20200517.png
England-Fig1-20200517.png (62.52 KiB) Viewed 437 times
Notice how it seems like cases per day has dropped off during the last few days. This doesn't happen if you look at the data for the whole of the UK:
UK-Fig1-20200517.png
UK-Fig1-20200517.png (83.67 KiB) Viewed 437 times
In fact the cases per day has a completely different trend in April and May. I can't account for that difference; Scotland and Wales were both at about 300 cases per day at the peak dropping to 100-200 now, NI is about 100. So I don't know how there can currently be more than 3000 cases per day in the UK if there are only about 1000 cases per day in England.
molto tricky

User avatar
shpalman
Dorkwood
Posts: 1558
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: SARS-CoV-2 testing

Post by shpalman » Sun May 17, 2020 4:17 pm

Ah it's because the UK total "Includes tests carried out by commercial partners which are not included in the 4 National totals" according to https://coronavirus.data.gov.uk/

So using the England national total for any kind of analysis is worthless since there might be nearly 130,000 tests which don't show up in it?
molto tricky

Post Reply