Coronavirus - Modelling Aspects Only

[Mike4]

In the New York epidemic, I remember the Cuomo saying the longer someone was on a ventilator, the worse their chances of survival.
I agree it's very strange that a death after 29+ days with Covid on the death certificate is NOT counted as a Covid death.

Somehow I imagined virtually everyone who dies from COVID-19 in hospital graduated to a ventilator first, and spent some considerable time on it before succumbing. Is that not the case?

What proportion of people dying of/with COVID in a hospital die without ever going 'on the vent', does anybody know?

[servodude]

In the New York epidemic, I remember the Cuomo saying the longer someone was on a ventilator, the worse their chances of survival.
I agree it's very strange that a death after 29+ days with Covid on the death certificate is NOT counted as a Covid death.

Somehow I imagined virtually everyone who dies from COVID-19 in hospital graduated to a ventilator first, and spent some considerable time on it before succumbing. Is that not the case?

What proportion of people dying of/with COVID in a hospital die without ever going 'on the vent', does anybody know?

Around half.
- elderly people who have underlying conditions will have had "the chat" and decided on a DNR (do not resuscitate) order
- that doesn't mean they don't get to hospital but they won't be put in to an induced coma for invasive ventilation

They would be candidates for NIV though but that's not considered "on the vent"

-sd

[Mike4]

Around half.
- elderly people who have underlying conditions will have had "the chat" and decided on a DNR (do not resuscitate) order
- that doesn't mean they don't get to hospital but they won't be put in to an induced coma for invasive ventilation

They would be candidates for NIV though but that's not considered "on the vent"

-sd

So the brutal and callous assertion we hear sometimes that those who succumb to COVID don't matter "because they were about to die anyway', is arguably part way to the truth in around half the cases. And not in the other half.

Those who get intubated and ventilated are broadly those who are younger and more robust? (Or were before getting COVID)?)

[servodude]

Around half.
- elderly people who have underlying conditions will have had "the chat" and decided on a DNR (do not resuscitate) order
- that doesn't mean they don't get to hospital but they won't be put in to an induced coma for invasive ventilation

They would be candidates for NIV though but that's not considered "on the vent"

-sd

So the brutal and callous assertion we hear sometimes that those who succumb to COVID don't matter "because they were about to die anyway', is arguably part way to the truth in around half the cases. And not in the other half.

Those who get intubated and ventilated are broadly those who are younger and more robust? (Or were before getting COVID)?)

Well that's one way of looking at it. but I'd say they were "more likely to die" than "about to"
It's a bit like the way co-morbidities are mentioned as if they are unusual

Intubation is very hard on the patient (and is expensive)
- and with COVID in the early days it only had a success rate of about 20%. Those were mostly the young and robust ones

-sd

[servodude]

It's been a week since the graph below (and it looks like I must have posted what was going to be an update on the wrong board)
- so seeing as this is where modelling is I'll re-do one now

I think it takes about 2 weeks for any intervention to show in any figures
- so it will be interesting to see how wrong this is then

sliding the averaging window forward to the last week and fitting - it looks like it might be slowing a bit
- but less than the change from the original plot
- I'll probably remove the original linear extrapolation because it just isn't

incidentally - google sheets - "free" software that I can tinker with from any browser
- how great is that!

-sd

[scotia]

When comparing a model against Covid-19 admissions data for England in an earlier post
(https://www.lemonfool.co.uk/posting.php?mode=reply&f=83&t=22737#pr347221)

I noticed data features that would be interesting to follow with later data, and I also thought it would be useful to examine Covid-19 associated deaths for possible correlations. The most recent data (downloaded 18/10/20) is displayed below. The green points are daily admissions for England, modified by creating a moving average of 7 days to reduce statistical noise and to eliminate day-of-the-week bias. The blue points are the daily deaths – also plotted as the 7-day moving average. Both are normalised to 1 at their start points (i.e. summed over week ending 7/9/20). The un-normalised numbers for the first and last points are displayed on the graph. The deaths data extended to the week ending 17/10/20, while the admissions data ended at 15/7/20.



I previously noticed a levelling off around day 20 (week 20/9/20 to 26/9/20), followed by a continuing upward trend – and in the new data, this trend has continued upwards, with no further signs of levelling. Looking at the deaths data, there is also a brief levelling off around day 20, and another around day 29. I wonder if there are possible changes in policy that have produced these breaks in gradient – or is it just random?
However my main concern is the current rapid increase, particularly in deaths, with no sign of any controlling feature at the moment.

[servodude]

I wonder if there are possible changes in policy that have produced these breaks in gradient – or is it just random?

you would expect a temporary change in policy to show up as a period of slowed growth
- working backwards from the knees of the changes might give some indications of when it happened and thus what it might have been
- i suspect though that with the data being at a national level and interventions being more localised it might be hard to de-convolve

Can I ask what you're using for the plots?

I've decided that it's no longer a job for a spreadsheet so I've dipped my feet back in to python today

Here's recent admissions and deaths with exponential extrapolations based on the 30 data points ending 14 and 7 days from the most recent



- sd

[scotia]

I wonder if there are possible changes in policy that have produced these breaks in gradient – or is it just random?

you would expect a temporary change in policy to show up as a period of slowed growth
- working backwards from the knees of the changes might give some indications of when it happened and thus what it might have been
- i suspect though that with the data being at a national level and interventions being more localised it might be hard to de-convolve

Thinking it over, the numbers on the deaths curve are smallish - so I probably should have put some error bars on them ( a square root as for a Poisson distribution?) , and it would probably have showed the little blips to be statistical noise. Maybe I'll spend a bit more time on it today - its too wet for gardening. And maybe I'll get round to test-fitting an exponential. I thought about splitting it down to NHS regions - but the death numbers get rather small, although maybe the admissions numbers may be adequate.

Can I ask what you're using for the plots?
I've decided that it's no longer a job for a spreadsheet so I've dipped my feet back in to python today

I'm an ancient programmer - having played with lots of languages on lots of computer systems for over 50 years. For the current work (on a PC) I extract the data from the government Covid-19 data site using Curl, and decompress it with 7-zip into a CSV file. I then read this, process it, and display it using VBNet in Visual Studio 2019 (the free version). I take screenshots of the graphs, and tidy them up by clipping. All a bit clunky - but it works. I'll also confess to an intermediate state of playing around with the CSV file in Excel, getting ideas on the areas to be investigated.

[scotia]

Yes - its a rainy day, so I had another look at the Covid-19 deaths data.



The vertical lines on the data points are the plus and minus standard deviations (square root of the raw number) - assuming the data points follow a Poisson distribution. I'm not sure it provides much more information.
I also plotted an exponential exp(Lambda*t) where Lambda=0.069 and t is in days (the Blue line). I selected Lambda by getting a starting value from the average of a few data point computations, then adjusting it to produce a reasonable fit.
If the data continues to follow this exponential curve, then in a further 40 days there will be around 10,000 deaths per week - i.e. as high as the April peak. Hence my previously expressed main concern.

[Backache]

This is the immediately interesting bit for me

died within (equal to or less than) 28 days of the first positive specimen date

- so it doesn't get reset on subsequent tests

The median time on a ventilator should you require one is ~3weeks (varies across the world a bit based on resources and policy)
I find it strange to think that you could wait out that period being kept alive and then not be counted... but there's strange everywhere

-sd

The ICNARC data for the UK showed the median time on the ventilator was 14 days in the first wave.

[servodude]

This is the immediately interesting bit for me

died within (equal to or less than) 28 days of the first positive specimen date

- so it doesn't get reset on subsequent tests

The median time on a ventilator should you require one is ~3weeks (varies across the world a bit based on resources and policy)
I find it strange to think that you could wait out that period being kept alive and then not be counted... but there's strange everywhere

-sd

The ICNARC data for the UK showed the median time on the ventilator was 14 days in the first wave.

Thanks
- do you know what the recovery rate was in the UK for mechanically ventilated patients?
- was it around the 20% seen elsewhere?

- sd

[Backache]

Thanks
- do you know what the recovery rate was in the UK for mechanically ventilated patients?
- was it around the 20% seen elsewhere?

- sd

Can't find the full study just now but I read it recently, there was quite a variation over the course of first wave (Feb till August). The overall survival following mechanical ventilation was around 50% but it improved considerably over the course of the wave.
I don't think it was ever as low as 20% though the very early reports may have had a figure around that number because they were reporting completed episodes and people took longer to recover than to die.
The very early report from the second wave may be a little optimistic for the opposite reason as people appear to be recovering quicker than they are dying so the initial survival figures of 80% reported recently may be a tad optimistic.
This is from a quick perusal of the data rather than definitive opinion.

[servodude]

Thanks
- do you know what the recovery rate was in the UK for mechanically ventilated patients?
- was it around the 20% seen elsewhere?

- sd

Can't find the full study just now but I read it recently, there was quite a variation over the course of first wave (Feb till August). The overall survival following mechanical ventilation was around 50% but it improved considerably over the course of the wave.
I don't think it was ever as low as 20% though the very early reports may have had a figure around that number because they were reporting completed episodes and people took longer to recover than to die.
The very early report from the second wave may be a little optimistic for the opposite reason as people appear to be recovering quicker than they are dying so the initial survival figures of 80% reported recently may be a tad optimistic.
This is from a quick perusal of the data rather than definitive opinion.

Would it be this data from April? https://www.icnarc.org/DataServices/Attachments/Download/c9b491af-ea80-ea11-9124-00505601089b
- gives 67% (dying) for those receiving ARS (mechanical ventilation)
- wouldn't have wanted to have required support for my kindeys though!

50% dying on ARS would not have raised too many eyebrows - and I know many were raised
- this paper quotes 72% for older patients (the main cohort) https://www.thelancet.com/journals/lanres/article/PIIS2213-2600(20)30316-7/fulltext
- this is the 80% from NY that reached the mainstream media (https://www.businessinsider.com.au/coro ... ?r=US&IR=T)

fank thuck we've learned a bit more since then

stay well
-sd

EDIT: for url escaping

[Backache]

Would it be this data from April? https://www.icnarc.org/DataServices/Attachments/Download/c9b491af-ea80-ea11-9124-00505601089b
- gives 67% (dying) for those receiving ARS (mechanical ventilation)
- wouldn't have wanted to have required support for my kindeys though!

50% dying on ARS would not have raised too many eyebrows - and I know many were raised
- this paper quotes 72% for older patients (the main cohort) https://www.thelancet.com/journals/lanres/article/PIIS2213-2600(20)30316-7/fulltext
- this is the 80% from NY that reached the mainstream media (https://www.businessinsider.com.au/coro ... ?r=US&IR=T)

fank thuck we've learned a bit more since then

stay well
-sd

EDIT: for url escaping

There are weekly reports of the ICNARC data, last weeks or the week befores forget which had a link to a paper that summarised the changes through the first wave.
The overall figures for the first wave are on the website in the latest report found here https://www.icnarc.org/Our-Audit/Audits/Cmp/Reports
I'm not sure how to get the older ones.
I was working in ITU through the first wave, it was certainly a bit scary.

[scotia]

I have returned to look at the English admissions data - and particularly at the apparent break in the gradient.



As before, the daily data points are the sums of the previous weeks number, normalised to a starting point of unity. The vertical bars on the points are plus and minus the standard deviation ( square root of number for a poisson distribution) I have plotted exponential curves for the data on both sides of the break in gradient. The data point at day 20 corresponds to the week ending 27/9/20. I wondered if this was a feature of all of the NHS regions, so I looked at those with the larger numbers of admissions - North East, North West, Midlands and London - and all showed a similar pattern. In London, the gradient not only levelled off - it turned down wards:-



So what has caused this feature? Is it possible there was a change in the rule which defined how a Covid-19 admission was counted? Or is it an actual real fall in the admission rate? The data point at day 20 corresponds to the week ending 27/9/20. So its mid-point is around 23/9/20. Guessing that the time from infection to admission as being around 10 days, this suggests some happening in days around the 13th of September that had a controlling effect of the infection rate. Any ideas?

[servodude]

There are weekly reports of the ICNARC data, last weeks or the week befores forget which had a link to a paper that summarised the changes through the first wave.
The overall figures for the first wave are on the website in the latest report found here https://www.icnarc.org/Our-Audit/Audits/Cmp/Reports
I'm not sure how to get the older ones.
I was working in ITU through the first wave, it was certainly a bit scary.

Thanks for the link.
I hope things might be less scary in the ITU this time around; I've spent most of this year seconded to a BiPAP project as a result of the pandemic and I've seen the impact it's had at times on the clinicians we work with.

-sd

[servodude]

Guessing that the time from infection to admission as being around 10 days, this suggests some happening in days around the 13th of September that had a controlling effect of the infection rate. Any ideas?

from: https://www.gov.uk/government/news/coro ... -september

From Monday 14 September, you must not meet with people from other households socially in groups of more than 6.

- that would just be too neat wouldn't it?
- and wouldn't explain what looks like a discontinuity in trend; it should have manifested in a growth rate change - unless everyone went all "Barnard Castle" after 5 days and thought sod this!

if the change in "guidance" were accompanied in a tightening of admission criteria it might

- sd

[scotia]

Guessing that the time from infection to admission as being around 10 days, this suggests some happening in days around the 13th of September that had a controlling effect of the infection rate. Any ideas?

from: https://www.gov.uk/government/news/coro ... -september

From Monday 14 September, you must not meet with people from other households socially in groups of more than 6.

- that would just be too neat wouldn't it?
- and wouldn't explain what looks like a discontinuity in trend; it should have manifested in a growth rate change - unless everyone went all "Barnard Castle" after 5 days and thought sod this!

if the change in "guidance" were accompanied in a tightening of admission criteria it might

- sd

I think you may be on the correct track. A Government Minister makes an announcement about the increasing problem of Covid-19, and the intention to increase restrictions. This could have been the source of the discontinuity. The initial tone of the announcement may have made persons take more care, however this may have only lasted a few (5?) days, but the group 6 restriction may have stuck, and has resulted in the 9.5 day doubling period increasing to a 13-day doubling period
And another piece of supporting evidence is the Scotland admissions data which I have just plotted - no sign of any discontinuity - it just keeps climbing with no sign of a reducing gradient through the area of interest. Clearly we don't pay the slightest attention to pronouncements from London, since its a devolved issue. However we don't seem to pay much attention to the Scottish first minister either - from the week ending 6/9/20 to the week ending 16/10/20 the weekly Covid-19 admissions have risen unchecked from 29 to 514.

[servodude]

I think you may be on the correct track. A Government Minister makes an announcement about the increasing problem of Covid-19, and the intention to increase restrictions. This could have been the source of the discontinuity. The initial tone of the announcement may have made persons take more care, however this may have only lasted a few (5?) days, but the group 6 restriction may have stuck, and has resulted in the 9.5 day doubling period increasing to a 13-day doubling period
And another piece of supporting evidence is the Scotland admissions data which I have just plotted - no sign of any discontinuity - it just keeps climbing with no sign of a reducing gradient through the area of interest. Clearly we don't pay the slightest attention to pronouncements from London, since its a devolved issue. However we don't seem to pay much attention to the Scottish first minister either - from the week ending 6/9/20 to the week ending 16/10/20 the weekly Covid-19 admissions have risen unchecked from 29 to 514.

Actually that sounds plausible; this could just be a clunky shift in gears

When I posted earlier I wasn't really thinking about the fact that the growth was slower after
- In my head i expected "the rule of 6" to have had a bigger effect
- but I suppose there was probably a reduced amount of interaction (vs "normal") beforehand

I am ashamed to say I haven't looked at the Scotland data - I get my status updates from there when my mum calls
- I should have a look

- sd

[servodude]

I think you may be on the correct track. A Government Minister makes an announcement about the increasing problem of Covid-19, and the intention to increase restrictions. This could have been the source of the discontinuity. The initial tone of the announcement may have made persons take more care, however this may have only lasted a few (5?) days, but the group 6 restriction may have stuck, and has resulted in the 9.5 day doubling period increasing to a 13-day doubling period
And another piece of supporting evidence is the Scotland admissions data which I have just plotted - no sign of any discontinuity - it just keeps climbing with no sign of a reducing gradient through the area of interest. Clearly we don't pay the slightest attention to pronouncements from London, since its a devolved issue. However we don't seem to pay much attention to the Scottish first minister either - from the week ending 6/9/20 to the week ending 16/10/20 the weekly Covid-19 admissions have risen unchecked from 29 to 514.

Actually that sounds plausible; this could just be a clunky shift in gears

When I posted earlier I wasn't really thinking about the fact that the growth was slower after
- In my head i expected "the rule of 6" to have had a bigger effect
- but I suppose there was probably a reduced amount of interaction (vs "normal") beforehand

I am ashamed to say I haven't looked at the Scotland data - I get my status updates from there when my mum calls
- I should have a look

- sd

looks like the doubling time might be pushing out


-sd