Coronavirus - Modelling Aspects Only

[johnhemming]

I went for 30 days when I was originally looking at R^2 to determine "exponential vs linear" fits (as its a bit over twice the latency period I expect for effects to show)

For the original infection it was exponential for a while then moved to linear and then peaked.

For this second wave the question is the balance between a seasonal increase and a true second wave. I don't have a good idea as to how the seasonal component should react. An amount of exponential would be reasonable (as it arises from an increase in infectiousness).

It is useful looking at the position by NHS region as there remain clearly different patterns. The logic of something related to the seasons would see the more southerly areas picking up after the north. Hence if the north is currently peaking that may give a guide to the midlands and potentially more southernly regions.

[scotia]

finding the window for the data that is big enough to smooth the sampling artefacts and capture the timeliness of the system
- while still being short enough to capture any change response is always tricky

Agreed -its very difficult and possibly subjective.
So I don't put too much weight on the the report today that Covid-19 deaths increased by 60% over the previous week - which corresponds (by my mathematics) to a doubling time of 10.3 days

[servodude]

finding the window for the data that is big enough to smooth the sampling artefacts and capture the timeliness of the system
- while still being short enough to capture any change response is always tricky

Agreed -its very difficult and possibly subjective.
So I don't put too much weight on the the report today that Covid-19 deaths increased by 60% over the previous week - which corresponds (by my mathematics) to a doubling time of 10.3 days

Yup
- with a doubling time of 10days if you only looked at two weeks data you could draw a straight line pretty much though the points (with m=2)

That is why when it does slow down people can be forgiven for thinking it "becomes linear" when it's still the same geometric/exponential system just with different coefficients

-sd

[johnhemming]

That is why when it does slow down people can be forgiven for thinking it "becomes linear" when it's still the same geometric/exponential system just with different coefficients

Looking at earlier this year (Feb-March-April) it was initially exponential then went something like linear and then peaked.

I am not myself certain what will happen this time as I don't think it is the same as a disease encountering a population with anything like 50% susceptibility. (which is what I think earlier this year was). My underlying model is one of a disease becoming more infectious in a population with quite a bit of infection.

I do think we will have a materially better idea in a month.

sd: if you have the time could you do some charts using a logarithmic y axis (hospital admissions and deaths).

[Itsallaguess]

That is why when it does slow down people can be forgiven for thinking it "becomes linear" when it's still the same geometric/exponential system just with different coefficients

Looking at earlier this year (Feb-March-April) it was initially exponential then went something like linear and then peaked.

I am not myself certain what will happen this time as I don't think it is the same as a disease encountering a population with anything like 50% susceptibility. (which is what I think earlier this year was). My underlying model is one of a disease becoming more infectious in a population with quite a bit of infection.

I do think we will have a materially better idea in a month.

The current projections being given point to a lower peak of deaths, but one that is sustained for longer than the earlier wave, leading to higher overall numbers when compared to that initial wave -

Downing Street is privately working on the assumption that the second wave of coronavirus will be more deadly than the first, with the death toll remaining high throughout the winter.

An internal analysis of the projected course of the second wave is understood to show deaths peaking at a lower level than in the spring but remaining at that level for weeks or even months.

It is understood that the projection – provided by the Scientific Advisory Group for Emergencies (Sage) – has led to intense lobbying from Sir Patrick Vallance and other Government advisers for Boris Johnson to take more drastic action.

https://www.telegraph.co.uk/news/2020/10/27/second-covid-wave-forecast-deadly-first/

Cheers,

Itsallaguess

[johnhemming]

The difficulty with the SAGE position is that it is unclear what effect any restrictions would have. The virus is already spread widely over the population (endemic) hence any further restrictions are unlikely to have that much of an effect. The area with the worst problem is the north and that is continuing notwithstanding relatively severe restrictions.

We are currently at a rate of roughly 1,000 hospital admissions a day which one would assume would average out at around 200 deaths.

Professor Wendy Barclay, a Sage member and scientist from Imperial College London, on Tuesday said none of the current restrictions appeared to be having a significant impact on the spread of the virus.

"The total lockdown that we had back in late March was enough to turn the tide and get the virus back under control," she told Times Radio. "So far, none of the other restrictions that we've seen, and none of the other actions, seem to have done that."

Whereas I accept closing schools would reduce infections (but also harm children's long term futures) the evidence from earlier this year is that it was mainly infection that held back further infections not the lockdown.

[swill453]

I am not myself certain what will happen this time as I don't think it is the same as a disease encountering a population with anything like 50% susceptibility. (which is what I think earlier this year was).

Of course that's contentious, and you know there's a body of scientific opinion (maybe even a consensus) that would say we still have well over 50% susceptibility.

Scott.

[johnhemming]

I am not myself certain what will happen this time as I don't think it is the same as a disease encountering a population with anything like 50% susceptibility. (which is what I think earlier this year was).

Of course that's contentious, and you know there's a body of scientific opinion (maybe even a consensus) that would say we still have well over 50% susceptibility.

I know it is contentious. However, I would think SAGE's position (September 90% susceptibility) has been disproven by the doubling time not being short enough (for either hospital admissions or dealths, positive cases are not a reliable statistic).

SAGE's position also does not fit with the known earlier starting times for the virus in the UK (proven beyond reasonable doubt January, balance of probabilities December).

[swill453]

I know it is contentious. However, I would think SAGE's position (September 90% susceptibility) has been disproven by the doubling time not being short enough (for either hospital admissions or dealths, positive cases are not a reliable statistic).

SAGE's position also does not fit with the known earlier starting times for the virus in the UK (proven beyond reasonable doubt January, balance of probabilities December).

Do you still give credence to the Yeardon paper, that said its propositions would be shown wrong if the pandemic re-ignited in London?

Scott.

[johnhemming]

I know it is contentious. However, I would think SAGE's position (September 90% susceptibility) has been disproven by the doubling time not being short enough (for either hospital admissions or dealths, positive cases are not a reliable statistic).

SAGE's position also does not fit with the known earlier starting times for the virus in the UK (proven beyond reasonable doubt January, balance of probabilities December).

Do you still give credence to the Yeardon paper, that said its propositions would be shown wrong if the pandemic re-ignited in London?

Hospital admissions in London are running at 14% of the peak for London.

[servodude]

I am not myself certain what will happen this time as I don't think it is the same as a disease encountering a population with anything like 50% susceptibility. (which is what I think earlier this year was).

Of course that's contentious, and you know there's a body of scientific opinion (maybe even a consensus) that would say we still have well over 50% susceptibility.

I know it is contentious. However, I would think SAGE's position (September 90% susceptibility) has been disproven by the doubling time not being short enough (for either hospital admissions or dealths, positive cases are not a reliable statistic).

SAGE's position also does not fit with the known earlier starting times for the virus in the UK (proven beyond reasonable doubt January, balance of probabilities December).

That's like saying the road isn't long enough because the speed your car is travelling is too slow
SAGE are overstating their position by relying exclusively on extrapolation of serology testing
- but that has nothing to do with the growth rate in the metrics

-sd

[servodude]

I know it is contentious. However, I would think SAGE's position (September 90% susceptibility) has been disproven by the doubling time not being short enough (for either hospital admissions or dealths, positive cases are not a reliable statistic).

SAGE's position also does not fit with the known earlier starting times for the virus in the UK (proven beyond reasonable doubt January, balance of probabilities December).

Do you still give credence to the Yeardon paper, that said its propositions would be shown wrong if the pandemic re-ignited in London?

Hospital admissions in London are running at 14% of the peak for London.

And doubling roughly every 12 days at present: which is how stuff ignites

-sd

[johnhemming]

And doubling roughly every 12 days at present: which is how stuff ignites

We are currently missing one days scheduled data. I think you and I may agree that any lumpiness in the data probably averages out over a period of more than a week, but less than two weeks. Hence we need to look for when over that period admissions are not going up either substantially or in fact at all. That is likely to give us the peak.

We can also I think agree that there will be a peak. We, however, don't know when that is going to be.

[servodude]

And doubling roughly every 12 days at present: which is how stuff ignites

We are currently missing one days scheduled data. I think you and I may agree that any lumpiness in the data probably averages out over a period of more than a week, but less than two weeks. Hence we need to look for when over that period admissions are not going up either substantially or in fact at all. That is likely to give us the peak.

We can also I think agree that there will be a peak. We, however, don't know when that is going to be.

Totally agree!
And all of that is irrelevant to the fact that something that has a positive exponential growth rate can quite acceptably be considered to be igniting

-sd

[johnhemming]

Probably best we agree on what we agree about and we see over time as to what the truth is in terms of "igniting".

[servodude]

sd: if you have the time could you do some charts using a logarithmic y axis (hospital admissions and deaths).

deaths is proving a bit tricky through the API at the moment
- not sure if I've chosen the "wrong" deaths metric
- or if the code's barfing on the returned data (I should look to sanitise it for null, nan etc - ditch wales and work out how to handle the always revised last few days in NI also!)

but admissions.. currently on a log_y plot (again doubling times in the titles going back in 7 day steps)

Nations


Regions


-sd

[johnhemming]

Thanks for doing the work.

ditch wales

Hospital admissions for Wales are produced on a different basis and are best not included as they will, therefore, undermine any conclusions.

[tjh290633]

It might help if you looked at the first derivative of your figures. i.e. the slope of your graphs.

TJH

[servodude]

It might help if you looked at the first derivative of your figures. i.e. the slope of your graphs.

TJH

The doubling time is shown for the 3 fits in the graph legends (as ln(2)/b)
- which is a pretty good human readable metric for the "slope" of geometric model
TBH it's the change in this that's most interesting to myself

-sd

[servodude]

A week on since viewtopic.php?p=349695#p349695
- (well a week of my time but not necessarily a week's worth of data!)

Nations

- no real change in the English national picture
- Scotland now growing at a comparable rate (down from "quite a bit faster")

Regions

- only the North West and South West show any reduction in growth rate
- only the North West and North East/Yorks show a doubling period > two weeks

- sd