COVID-19

[shpalman]

The biggest issue is that the model is the wrong one, since it's based on active infections spreading the virus but if someone with symptoms results positive to the test (less than 10% of those considered worth testing) then they are either told to self-isolate or put in hospital (about 50-50) so that's the point at which they stop spreading it. The assumption is also made that asymptomatic carriers don't spread it, which just feels like a bit of an excuse to not test everybody (which is of course unfeasible anyway). The model doesn't even feature an incubation period.

The number of active infections spreading the virus is therefore completely unknown, as is the number of people who quietly got better from a mild infection of it. To fit the data even a little bit I need a very long recovery time which means R0 is far too large. This might just be because recoveries aren't declared until doctors are extremely sure.

[Bird on a Fire]

The biggest issue is that the model is the wrong one, since it's based on active infections spreading the virus but if someone with symptoms results positive to the test (less than 10% of those considered worth testing) then they are either told to self-isolate or put in hospital (about 50-50) so that's the point at which they stop spreading it. The assumption is also made that asymptomatic carriers don't spread it, which just feels like a bit of an excuse to not test everybody (which is of course unfeasible anyway). The model doesn't even feature an incubation period.

The number of active infections spreading the virus is therefore completely unknown, as is the number of people who quietly got better from a mild infection of it. To fit the data even a little bit I need a very long recovery time which means R0 is far too large. This might just be because recoveries aren't declared until doctors are extremely sure.

Sounds like this would benefit from a hidden Markov model, which would separate your observation process from the "state process" of interest, i.e. number of cases.

The number of infections today is dependent on the number of infections yesterday, with some multiplying process. But we don't know any of the numbers involved. What we have instead are imperfect observations, which we know a bit about (e.g. number of people tested, sensitivity of the test) which would inform Bayesian priors and you can then use maximum likelihood to find the parameter estimates for the underlying disease process that would most likely generate that sequence of observations, along with associated measures of confidence.

Given the amount of stochasticity involved I'd be surprised if you need a particularly mechanistic model to get something adequately predictive.

[bmforre]

NYTimes right now:

Coronavirus Live Updates: Fatality Rate of Covid-19 Is Higher Than the Flu
The World Health Organization announced on Tuesday that the global death rate of the disease caused by the new coronavirus was 3.4 percent.

RIGHT NOW Deaths outside China exceeded those inside the country for the first time since the start of the outbreak.

[jimbob]

An Iranian colleague's wife is still in touch with other medics who stayed in Iran after qualifying as doctors. A
They told her that there were many odd pneumonia cases in Qom in November.

[shpalman]

The biggest issue is that the model is the wrong one, since it's based on active infections spreading the virus but if someone with symptoms results positive to the test (less than 10% of those considered worth testing) then they are either told to self-isolate or put in hospital (about 50-50) so that's the point at which they stop spreading it. The assumption is also made that asymptomatic carriers don't spread it, which just feels like a bit of an excuse to not test everybody (which is of course unfeasible anyway). The model doesn't even feature an incubation period.

The number of active infections spreading the virus is therefore completely unknown, as is the number of people who quietly got better from a mild infection of it. To fit the data even a little bit I need a very long recovery time which means R0 is far too large. This might just be because recoveries aren't declared until doctors are extremely sure.

Sounds like this would benefit from a hidden Markov model, which would separate your observation process from the "state process" of interest, i.e. number of cases.

The number of infections today is dependent on the number of infections yesterday, with some multiplying process. But we don't know any of the numbers involved. What we have instead are imperfect observations, which we know a bit about (e.g. number of people tested, sensitivity of the test) which would inform Bayesian priors and you can then use maximum likelihood to find the parameter estimates for the underlying disease process that would most likely generate that sequence of observations, along with associated measures of confidence.

Given the amount of stochasticity involved I'd be surprised if you need a particularly mechanistic model to get something adequately predictive.

Ok I tried to move from the SIR model to the SEIR model (a person passes from Susceptible to Exposed depending on the number of Infected people, and there is a time constant for passing from Exposed to Infected) but I probably need to extend that to a model in which some of the Exposed can pass directly to Undetected without being Infected and so aren't Susceptible anymore but don't count in the stats of people who are Recovered... oh and a Quarantine rate in which Infected people stop contributing to new infections.

[tenchboy]

Fire Flood Pestilence Famine Two Popes and now og hod just pass me the revolver and a glass of port I'll meet you on the other side.
Turn it on turn it on again...

[lpm]

An Iranian colleague's wife is still in touch with other medics who stayed in Iran after qualifying as doctors. A
They told her that there were many odd pneumonia cases in Qom in November.

None of these "ancient Covid" claims make sense.

It's easy to backtrack to 1st Dec 2019 and a handful of cases. By 31 Dec there would have been in the region of 500 cases in Wuhan, including about 50 in the hospitals - and it was only at the end of Dec that any doctor noticed the anomoly, with that whistle-blower doctor. Of those 50, two-thirds were directly linked to the Wuhan market. It's simply impossible for that profile to fit in with any other origin.

[Boustrophedon]

So the government are only going to release infection figures once a week and redacted of geographical information.

https://twitter.com/DHSCgovuk/status/12 ... 4407802880

So there's 6 days for those in the know (IE ministers.) to do a nice little bit of insider trading on companies likely to be affected.

[jimbob]

So the government are only going to release infection figures once a week and redacted of geographical information.

https://twitter.com/DHSCgovuk/status/12 ... 4407802880

So there's 6 days for those in the know (IE ministers.) to do a nice little bit of insider trading on companies likely to be affected.

Way to avoid panic

[bob sterman]

So the government are only going to release infection figures once a week and redacted of geographical information.

https://twitter.com/DHSCgovuk/status/12 ... 4407802880

So there's 6 days for those in the know (IE ministers.) to do a nice little bit of insider trading on companies likely to be affected.

The tweet says...

"As of today, due to the number of new cases, we will no longer be tweeting information on the location of each new case. Instead, this information will be released centrally in a consolidated format online, once a week. We are working on this now and plan to share on Friday."

Doesn't this mean they will still tweet the numbers (perhaps each day) but just not tweet the locations.

[headshot]

NYTimes right now:

Coronavirus Live Updates: Fatality Rate of Covid-19 Is Higher Than the Flu
The World Health Organization announced on Tuesday that the global death rate of the disease caused by the new coronavirus was 3.4 percent.

RIGHT NOW Deaths outside China exceeded those inside the country for the first time since the start of the outbreak.

How is it possible to confidently quote a death rate when they have absolutely no idea how many people have been infected - given they've already said that many people are asymptomatic or only having mild symptoms that wouldn't have been reported or tested as Covid-19?

[shpalman]

If I want to model this slowdown I could reduce the susceptible population to some number which isn't several orders of magnitude greater than the number of cases, i.e. 6000 instead of 60 million. But to get the same exponential increase in the first week or so I need to increase the contact parameter so R_0 is now 11.3.

Reducing the susceptible population fudges the idea of quarantine and travel restrictions, so that the virus can't actually reach the entire population of Italy.

It predicts that we'll have 2625 active infections by the end of today.

Figure_1.png (37.5 KiB) Viewed 83235 times

... and we actually have 2706.

It predicts 3028 for tomorrow. I haven't changed any parameters.

[jimbob]

An Iranian colleague's wife is still in touch with other medics who stayed in Iran after qualifying as doctors. A
They told her that there were many odd pneumonia cases in Qom in November.

None of these "ancient Covid" claims make sense.

It's easy to backtrack to 1st Dec 2019 and a handful of cases. By 31 Dec there would have been in the region of 500 cases in Wuhan, including about 50 in the hospitals - and it was only at the end of Dec that any doctor noticed the anomoly, with that whistle-blower doctor. Of those 50, two-thirds were directly linked to the Wuhan market. It's simply impossible for that profile to fit in with any other origin.

Oh this wasn't doubting the Whuhan origin. There were a lot of Chinese in Qoms though. But as you say, it's not that likely that it got to Iran quite so early.

This is interesting on the origin:

https://nextstrain.org/narratives/ncov/ ... 03-04?n=11

Starting point mid November to mid December.

[lpm]

Thank you that's a very interesting read, including a couple of links within it.

Says approx 500 undetected cases in Seattle area, similar to the Wuhan position of 500 on 1 Jan.

[lpm]

Those guys from that link are pretty dismissive of this latest "new 2nd strain has emerged" theory.

[sTeamTraen]

Fire Flood Pestilence Famine Two Popes and now og hod just pass me the revolver and a glass of port I'll meet you on the other side.
Turn it on turn it on again...

As David Baddiel tweeted, this might not be a great time for three 70-year-olds to spend time with massive crowds of people.

[dyqik]

Fire Flood Pestilence Famine Two Popes and now og hod just pass me the revolver and a glass of port I'll meet you on the other side.
Turn it on turn it on again...

As David Baddiel tweeted, this might not be a great time for three 70-year-olds to spend time with massive crowds of people.

You say that, but...

[Pucksoppet]

Fire Flood Pestilence Famine Two Popes and now og hod just pass me the revolver and a glass of port I'll meet you on the other side.
Turn it on turn it on again...

Perhaps you need a supernatural anæsthetist to come?

I've just realised that Defector was not Hackett's first album after leaving the group. My memory was confused, as I remember people walking around with 'Hackett//Defector" lapel badges, which I presumed was a comment on what they thought about him leaving. Hi ho.

[El Pollo Diablo]

So the government are only going to release infection figures once a week and redacted of geographical information.

https://twitter.com/DHSCgovuk/status/12 ... 4407802880

So there's 6 days for those in the know (IE ministers.) to do a nice little bit of insider trading on companies likely to be affected.

The tweet says...

"As of today, due to the number of new cases, we will no longer be tweeting information on the location of each new case. Instead, this information will be released centrally in a consolidated format online, once a week. We are working on this now and plan to share on Friday."

Doesn't this mean they will still tweet the numbers (perhaps each day) but just not tweet the locations.

And we all know if it's not on twitter it isn't being released at all

[bob sterman]

Those guys from that link are pretty dismissive of this latest "new 2nd strain has emerged" theory.

I had a quick look at the paper - the part where they present an argument about selection pressures leading to a change in the prevalence of different strains is not very clearly written. So it's hard to follow their line of reasoning.

[Boustrophedon]

So the government are only going to release infection figures once a week and redacted of geographical information.

https://twitter.com/DHSCgovuk/status/12 ... 4407802880

So there's 6 days for those in the know (IE ministers.) to do a nice little bit of insider trading on companies likely to be affected.

The tweet says...

"As of today, due to the number of new cases, we will no longer be tweeting information on the location of each new case. Instead, this information will be released centrally in a consolidated format online, once a week. We are working on this now and plan to share on Friday."

Doesn't this mean they will still tweet the numbers (perhaps each day) but just not tweet the locations.

Anyway they have done a U-turn and decided to carry on as before.

[lpm]

Those guys from that link are pretty dismissive of this latest "new 2nd strain has emerged" theory.

I had a quick look at the paper - the part where they present an argument about selection pressures leading to a change in the prevalence of different strains is not very clearly written. So it's hard to follow their line of reasoning.

Paper here. S and L types.
https://academic.oup.com/nsr/advance-ar ... 36/5775463

Although the L type (∼70%) is more prevalent than the S type (∼30%), the S type was found to be the ancestral version. Whereas the L type was more prevalent in the early stages of the outbreak in Wuhan, the frequency of the L type decreased after early January 2020. Human intervention may have placed more severe selective pressure on the L type, which might be more aggressive and spread more quickly. On the other hand, the S type, which is evolutionarily older and less aggressive, might have increased in relative frequency due to relatively weaker selective pressure.

[cvb]

So the government are only going to release infection figures once a week and redacted of geographical information.

https://twitter.com/DHSCgovuk/status/12 ... 4407802880

So there's 6 days for those in the know (IE ministers.) to do a nice little bit of insider trading on companies likely to be affected.

The tweet says...

"As of today, due to the number of new cases, we will no longer be tweeting information on the location of each new case. Instead, this information will be released centrally in a consolidated format online, once a week. We are working on this now and plan to share on Friday."

Doesn't this mean they will still tweet the numbers (perhaps each day) but just not tweet the locations.

Anyway they have done a U-turn and decided to carry on as before.

Link to update

[bob sterman]

Those guys from that link are pretty dismissive of this latest "new 2nd strain has emerged" theory.

I had a quick look at the paper - the part where they present an argument about selection pressures leading to a change in the prevalence of different strains is not very clearly written. So it's hard to follow their line of reasoning.

Paper here. S and L types.
https://academic.oup.com/nsr/advance-ar ... 36/5775463

Although the L type (∼70%) is more prevalent than the S type (∼30%), the S type was found to be the ancestral version. Whereas the L type was more prevalent in the early stages of the outbreak in Wuhan, the frequency of the L type decreased after early January 2020. Human intervention may have placed more severe selective pressure on the L type, which might be more aggressive and spread more quickly. On the other hand, the S type, which is evolutionarily older and less aggressive, might have increased in relative frequency due to relatively weaker selective pressure.

Yes - that's the bit that I don't think is very clear. E.g. using the term "aggressive" rather than clarifying whether they are referring to harm to the patient or transmissability.

[Bird on a Fire]

I had a quick look at the paper - the part where they present an argument about selection pressures leading to a change in the prevalence of different strains is not very clearly written. So it's hard to follow their line of reasoning.

Paper here. S and L types.
https://academic.oup.com/nsr/advance-ar ... 36/5775463

Although the L type (∼70%) is more prevalent than the S type (∼30%), the S type was found to be the ancestral version. Whereas the L type was more prevalent in the early stages of the outbreak in Wuhan, the frequency of the L type decreased after early January 2020. Human intervention may have placed more severe selective pressure on the L type, which might be more aggressive and spread more quickly. On the other hand, the S type, which is evolutionarily older and less aggressive, might have increased in relative frequency due to relatively weaker selective pressure.

Yes - that's the bit that I don't think is very clear. E.g. using the term "aggressive" rather than clarifying whether they are referring to harm to the patient or transmissability.

I'm pretty sure they mean transmissability.