Donate to Remove ads

Got a credit card? use our Credit Card & Finance Calculators

Thanks to johnstevens77,Bhoddhisatva,scotia,Anonymous,Cornytiv34, for Donating to support the site

Coronavirus - Modelling Aspects Only

The home for all non-political Coronavirus (Covid-19) discussions on The Lemon Fool
Forum rules
This is the home for all non-political Coronavirus (Covid-19) discussions on The Lemon Fool
Itsallaguess
Lemon Half
Posts: 9129
Joined: November 4th, 2016, 1:16 pm
Has thanked: 4140 times
Been thanked: 10023 times

Re: Coronavirus - Modelling Aspects Only

#309532

Postby Itsallaguess » May 18th, 2020, 5:56 am

zico wrote:
Itsallaguess wrote:
Just on that particular point regarding the idea of granular R-values though - doesn't it make a lot of sense to factor those three major areas separately in terms of then being able to compose and prescribe 'Covid-19 management-techniques' differently for each area?


I see your point, but it's important to remember R-values are just another way of describing the growth rate in simple terms (so people don't get confused by negative growth rates) so having 3 different ones defeats the point really. As Jenny Harries said, the scale and spread of an epidemic are more important. If we have 3,000 cases per day and an expected increase of 10% per week, in the short-term, that's a lot better than 300,000 cases per day increasing by 1% per week.

If we didn't have R-values, we'd be talking about number of cases and growth rates.


But that was sort of my point.

Take a hypothetical situation where there are zero new cases in the general community, and it was shown that all new cases were being seen in hospitals and care-homes...

Given the ability to perform much more stringent virus-management protocols within the hospital and care-home communities, then in that hypothetical situation, wouldn't it be more and more difficult to continue to justify highly-stringent lock-down protocols in the general community?

That was the point I was trying to make - that a 'single' R-value discussion that has recently shown it to be a slightly rising number, at the same time as being told that we can walk-down the more stringent community lock-down protocols seem, on the face of it, to be a confusing message to those of us that are interested in the wider picture.

But I was asking if there was the possibility that the wider data behind that 'single R-value' might actually show a 'rapidly improving' situation in the general community, with some legacy issues in the other two areas, which 'might' go some way to explain that confusing 'rising R-number & lowering lock-down protocols' situation?

Cheers,

Itsallaguess

zico
Lemon Quarter
Posts: 2139
Joined: November 4th, 2016, 12:12 pm
Has thanked: 1074 times
Been thanked: 1086 times

Re: Coronavirus - Modelling Aspects Only

#309687

Postby zico » May 18th, 2020, 4:03 pm

Itsallaguess wrote:
zico wrote:
Itsallaguess wrote:
Just on that particular point regarding the idea of granular R-values though - doesn't it make a lot of sense to factor those three major areas separately in terms of then being able to compose and prescribe 'Covid-19 management-techniques' differently for each area?


I see your point, but it's important to remember R-values are just another way of describing the growth rate in simple terms (so people don't get confused by negative growth rates) so having 3 different ones defeats the point really. As Jenny Harries said, the scale and spread of an epidemic are more important. If we have 3,000 cases per day and an expected increase of 10% per week, in the short-term, that's a lot better than 300,000 cases per day increasing by 1% per week.

If we didn't have R-values, we'd be talking about number of cases and growth rates.


But that was sort of my point.

Take a hypothetical situation where there are zero new cases in the general community, and it was shown that all new cases were being seen in hospitals and care-homes...

Given the ability to perform much more stringent virus-management protocols within the hospital and care-home communities, then in that hypothetical situation, wouldn't it be more and more difficult to continue to justify highly-stringent lock-down protocols in the general community?

That was the point I was trying to make - that a 'single' R-value discussion that has recently shown it to be a slightly rising number, at the same time as being told that we can walk-down the more stringent community lock-down protocols seem, on the face of it, to be a confusing message to those of us that are interested in the wider picture.

But I was asking if there was the possibility that the wider data behind that 'single R-value' might actually show a 'rapidly improving' situation in the general community, with some legacy issues in the other two areas, which 'might' go some way to explain that confusing 'rising R-number & lowering lock-down protocols' situation?

Cheers,

Itsallaguess


Yes, you're perfectly right and it's a good point. I've probably just got the wrong end of the stick in our discussion. If I was a government scientist working on this, I'd want as much data as possible broken down by all sorts of factors, so I could better understand what the key drivers and relationships were, to help develop strategy and focus resources on the most important areas.

However, having sold the public a nice simple idea of "R-value is all-important and we must keep the R-value below 1" government spokespeople are now referring to a multiplicity of R-values, and saying they need to be considered along with the number of infections. This seems to be setting the groundwork for ignoring potential increases in R-values in the coming weeks - but maybe I'm just being too cynical!

zico
Lemon Quarter
Posts: 2139
Joined: November 4th, 2016, 12:12 pm
Has thanked: 1074 times
Been thanked: 1086 times

Re: Coronavirus - Modelling Aspects Only

#309704

Postby zico » May 18th, 2020, 4:25 pm

Interesting findings here from Kings College London/Boston University on a self-reporting app that's regularly used by 2.5 million people in USA and UK. (I also started using it a couple of weeks ago).
(This is the data that's behind today's updated UK advice about loss of smell being an important symptom).

The aim of the app is to predict Covid-19 infections without the need for a test, based on self-reported symptoms, and claims to be 80% accurate. Presumably they've validated their prediction model, but nothing about this in the links below.
The big benefit of this approach is that you don't need vast amounts of physical testing to get a decent estimate of the numbers of people affected in various areas.


Obviously the 2 weaknesses of this approach are :-
a) not a random sample of the population, and people feeling unwell may be more likely to update their symptoms daily;
b) self-reported so relying on people being accurate about their symptoms.

https://covid.joinzoe.com/post/detectin ... out-a-test

Highlights of the above link are :-

From 2.5m users, 800,000 (32%) said they felt unwell in some way.
Of these 800,000 people, 17.4% were predicted to have Covid-19.
That's 5.6% of the population. Applying 5.6% to 57 million UK adults gives 3.1 million people infected in the UK - with symptoms.
Obviously people infected but without symptoms will be reporting they feel fine. Assuming 50% asymptomatic gives 6.2 million people infected in the UK.

This link is to UK-specific estimates of various areas of the UK, using an interactive map.

https://covid.joinzoe.com/data#levels-over-time

Nimrod103
Lemon Half
Posts: 6467
Joined: November 4th, 2016, 6:10 pm
Has thanked: 939 times
Been thanked: 2255 times

Re: Coronavirus - Modelling Aspects Only

#309842

Postby Nimrod103 » May 19th, 2020, 8:20 am

It will be very interesting if and when a reliable figure for the number of people who have had CV is ever established for the UK. The problem I have with the Kings College London/Boston University self-reporting app, is that in the UK until the testing rates were ramped up, typically 10,000 people/day were being tested on the basis of suspicious symptoms, but only c.3000 were positive cases. I infer from that that a lot of people thought they were victims, but actually had caught some other bug.

zico
Lemon Quarter
Posts: 2139
Joined: November 4th, 2016, 12:12 pm
Has thanked: 1074 times
Been thanked: 1086 times

Re: Coronavirus - Modelling Aspects Only

#309886

Postby zico » May 19th, 2020, 10:24 am

Nimrod103 wrote:It will be very interesting if and when a reliable figure for the number of people who have had CV is ever established for the UK. The problem I have with the Kings College London/Boston University self-reporting app, is that in the UK until the testing rates were ramped up, typically 10,000 people/day were being tested on the basis of suspicious symptoms, but only c.3000 were positive cases. I infer from that that a lot of people thought they were victims, but actually had caught some other bug.


For the Kings College study, simply reporting "I feel unwell" doesn't automatically generate a Covid-19 "diagnosis". The app then asks further questions about your health, and makes a prediction of whether or not you've got Covid-19. For example, last week I had a very bad allergic reaction, so answered "I feel unwell" then no to the specific questions along the lines of "high temperature?", "loss of sense of smell?", "fatigue?" etc.

zico
Lemon Quarter
Posts: 2139
Joined: November 4th, 2016, 12:12 pm
Has thanked: 1074 times
Been thanked: 1086 times

Re: Coronavirus - Modelling Aspects Only

#309912

Postby zico » May 19th, 2020, 12:18 pm

England & Wales updated ONS figures for excess deaths upto including week 19 (w/e 7th May).

Good news at last.
Excess deaths reduced sharply to 3,082 for week 19 only. Although even this lower number is equivalent to the death toll from the Hillsborough disaster happening 4 times per single day, just to give an idea of the scale of this.

Excess deaths (England & Wales only) are 49,575 more than the 5-year average since the start of the epidemic.

For the first time, the number of non-Covid deaths was about 900 below the 5-year average, which may support the theory that some Covid-19 deaths are simply resulting in slightly earlier deaths amongst people who would have died soon in anycase.

Once again, we see the great age-related disparity.
People aged 19-64 make up 81% of the adult population, but account for only 9% of deaths.
For people aged 19-59 it's 76% of population, but only 5.6% of deaths.
For people aged 19-44, 55% of population, but only 1% of deaths.


https://www.ons.gov.uk/peoplepopulation ... ng8may2020

Does anyone know how to convert this graph below so it can be seen on my post?

<iframe height="893px" width="100%" src="https://www.ons.gov.uk/visualisations/dvc825/fig1/index.html"></iframe>

GoSeigen
Lemon Quarter
Posts: 4349
Joined: November 8th, 2016, 11:14 pm
Has thanked: 1590 times
Been thanked: 1579 times

Re: Coronavirus - Modelling Aspects Only

#309926

Postby GoSeigen » May 19th, 2020, 12:36 pm

zico wrote:Does anyone know how to convert this graph below so it can be seen on my post?

Short of saving and uploading a screenshot you can't: it is not an image, it's actually rendered by the browser along with the rest of the page.

GS

Alaric
Lemon Half
Posts: 6031
Joined: November 5th, 2016, 9:05 am
Has thanked: 20 times
Been thanked: 1398 times

Re: Coronavirus - Modelling Aspects Only

#309936

Postby Alaric » May 19th, 2020, 12:50 pm

zico wrote:For the first time, the number of non-Covid deaths was about 900 below the 5-year average, which may support the theory that some Covid-19 deaths are simply resulting in slightly earlier deaths amongst people who would have died soon in anycase.


The alternative theory put forward in the linked document is that they are classifying by date of notification and the Bank Holiday causes a delay.

zico
Lemon Quarter
Posts: 2139
Joined: November 4th, 2016, 12:12 pm
Has thanked: 1074 times
Been thanked: 1086 times

Re: Coronavirus - Modelling Aspects Only

#309942

Postby zico » May 19th, 2020, 12:56 pm

Thanks - annoyingly the most informative graphs on ONS site can't be copied & pasted.
Of the ones that can be pasted, I think the one below is the most informative. Shows there were several days at the peak when Covid-19 deaths were above 1,000 deaths per day.

Image

dealtn
Lemon Half
Posts: 6072
Joined: November 21st, 2016, 4:26 pm
Has thanked: 441 times
Been thanked: 2324 times

Re: Coronavirus - Modelling Aspects Only

#309946

Postby dealtn » May 19th, 2020, 12:59 pm

zico wrote:England & Wales updated ONS figures for excess deaths upto including week 19 (w/e 7th May).



Careful. I think its w/e 8th May, which also happens to be a bank holiday.

There is a well observed "weekend" effect in the numbers, and effectively there is a 3 day weekend in this set of numbers, so noiser than usual.

zico
Lemon Quarter
Posts: 2139
Joined: November 4th, 2016, 12:12 pm
Has thanked: 1074 times
Been thanked: 1086 times

Re: Coronavirus - Modelling Aspects Only

#309951

Postby zico » May 19th, 2020, 1:04 pm

dealtn wrote:
zico wrote:England & Wales updated ONS figures for excess deaths upto including week 19 (w/e 7th May).



Careful. I think its w/e 8th May, which also happens to be a bank holiday.

There is a well observed "weekend" effect in the numbers, and effectively there is a 3 day weekend in this set of numbers, so noiser than usual.


You're right, I forgot to mention that. Even allowing for that, the steep downward trend is still very encouraging.

Alaric
Lemon Half
Posts: 6031
Joined: November 5th, 2016, 9:05 am
Has thanked: 20 times
Been thanked: 1398 times

Re: Coronavirus - Modelling Aspects Only

#309959

Postby Alaric » May 19th, 2020, 1:11 pm

zico wrote:Of the ones that can be pasted, I think the one below is the most informative. Shows there were several days at the peak when Covid-19 deaths were above 1,000 deaths per day.


That seems to show up a failure to consider Care Homes as places that the disease needed to kept away from. Shutting down or restricting many normal activities will presumably have reduced the numbers seeking treatment in hospitals and their mortality rate. Those in Care Homes are mostly isolated in normal circumstances from public transport, pubs and restaurants, sporting events, workplaces etc. Family visits were banned or severely restricted early on, so how did the virus take a hold?

modellingman
Lemon Slice
Posts: 614
Joined: November 4th, 2016, 3:46 pm
Has thanked: 594 times
Been thanked: 364 times

Re: Coronavirus - Modelling Aspects Only

#309964

Postby modellingman » May 19th, 2020, 1:21 pm

Itsallaguess wrote:
Take a hypothetical situation where there are zero new cases in the general community, and it was shown that all new cases were being seen in hospitals and care-homes...

Given the ability to perform much more stringent virus-management protocols within the hospital and care-home communities, then in that hypothetical situation, wouldn't it be more and more difficult to continue to justify highly-stringent lock-down protocols in the general community?

That was the point I was trying to make - that a 'single' R-value discussion that has recently shown it to be a slightly rising number, at the same time as being told that we can walk-down the more stringent community lock-down protocols seem, on the face of it, to be a confusing message to those of us that are interested in the wider picture.

But I was asking if there was the possibility that the wider data behind that 'single R-value' might actually show a 'rapidly improving' situation in the general community, with some legacy issues in the other two areas, which 'might' go some way to explain that confusing 'rising R-number & lowering lock-down protocols' situation?



I admire the optimism inherent in your thinking. However...

R is simply a population characteristic and, as you surmise, if that population is divided into sub-populations (such as care homes, hospitals and "the rest") that characteristic may well take different values in each sub-population. If one of those sub-populations is much larger than the others then the characteristic of the population as a whole is largely going to be determined by that of the dominant sub-population. It seems highly unlikely that R is rising generally because it is rising in the two smaller sub-populations (care homes and hospitals) but falling in "the (much larger) rest."

The conflict you point at between an R that is possibly going in the wrong direction and the moves to relax non-pharmaceutical interventions is symptomatic of the difficulties inherent in managing any complex system. Inevitably, those accountable for managing such systems (ie politicans in this context) must deal with competing objectives and goals with far from perfect data, information, knowledge and resources. How well or how badly they perform is, rightly, subject to lots of debate elsewhere.

My own take on R at sub-population level is that it only makes much sense if those sub-populations are reasonably well self-contained. Without this, there will be leakage effects across sub-population boundaries. So whilst R makes sense at a broad regional level, it makes makes much less sense to separate out hospitals and care homes because of their dense connections to "the rest" (primarily through the staff who work within them). It makes far more sense to assess disease transmission in these sub-populations in terms of internal and external transmission. Internal transmission effectively keeps a reservoir of infection sustained. Perhaps somebody (not me) already has a model based on these notions and is looking at the effect of different NPIs on managing these transmission rates.

zico
Lemon Quarter
Posts: 2139
Joined: November 4th, 2016, 12:12 pm
Has thanked: 1074 times
Been thanked: 1086 times

Re: Coronavirus - Modelling Aspects Only

#309967

Postby zico » May 19th, 2020, 1:24 pm

Alaric wrote:
zico wrote:Of the ones that can be pasted, I think the one below is the most informative. Shows there were several days at the peak when Covid-19 deaths were above 1,000 deaths per day.


That seems to show up a failure to consider Care Homes as places that the disease needed to kept away from. Shutting down or restricting many normal activities will presumably have reduced the numbers seeking treatment in hospitals and their mortality rate. Those in Care Homes are mostly isolated in normal circumstances from public transport, pubs and restaurants, sporting events, workplaces etc. Family visits were banned or severely restricted early on, so how did the virus take a hold?


Grateful if we could stay away from opinions and speculation on government's response on this thread (Polite Discussions is the best place for that), and focus on data and modelling please. So in your above comments, it's fair enough to point out changes in behaviour that could have changed the numbers, but I'd like to focus discussion on evidence/data to look at the issues.

To try and directly answer your question above, there seems (and I emphasise "seems") to be increasing evidence from various sources that the care home epidemic was caused by staff shortages leading to agency workers visiting several care homes, and also by the government policy of discharging people from hospital directly into care homes without a Covid-19 test. If anyone has any links to data/figures on these occurrences, I'd like to see it - because until then, it's just a reported viewpoint that may or may not be true.

For example, it seems that 5-7% of Covid-19 infections were actually picked up in hospitals by patients in hospital for another reason.

https://www.theguardian.com/world/2020/ ... s-covid-19

Similarly, I should have made the point that the figures below will be skewed, because older people are far more likely to be self-isolating and protecting themselves from the spread of the virus - which means the risk to younger people is even less than the figures show.

Once again, we see the great age-related disparity.
People aged 19-64 make up 81% of the adult population, but account for only 9% of deaths.
For people aged 19-59 it's 76% of population, but only 5.6% of deaths.
For people aged 19-44, 55% of population, but only 1% of deaths.

Itsallaguess
Lemon Half
Posts: 9129
Joined: November 4th, 2016, 1:16 pm
Has thanked: 4140 times
Been thanked: 10023 times

Re: Coronavirus - Modelling Aspects Only

#310008

Postby Itsallaguess » May 19th, 2020, 3:40 pm

modellingman wrote:
My own take on R at sub-population level is that it only makes much sense if those sub-populations are reasonably well self-contained.

Without this, there will be leakage effects across sub-population boundaries. So whilst R makes sense at a broad regional level, it makes makes much less sense to separate out hospitals and care homes because of their dense connections to "the rest" (primarily through the staff who work within them).

It makes far more sense to assess disease transmission in these sub-populations in terms of internal and external transmission. Internal transmission effectively keeps a reservoir of infection sustained. Perhaps somebody (not me) already has a model based on these notions and is looking at the effect of different NPIs on managing these transmission rates.


That makes sense - thanks.

Where I was perhaps wondering slightly differently to the above was where you mention 'dense connections to the rest', where presumably the 'general population' element is clearly having to be 'intimately involved' with the care-home and hospital environments, but where I thought we might be able to separate them out a little more firmly, given the assumed PPE elements creating 'natural boundary-zones' to compliment those separate 'elements'..

Obviously in the early days, and certainly at care-home level, this seems to have been a real issue, possibly exacerbated by the lack of PPE quantity/quality, but certainly by the inherent use of bank and agency care-home staff who might have a tendency to cross-pollinate care-homes in much more diverse ways, that hospitals perhaps don't suffer from in the same magnitude.

Thanks again for your detailed explanation.

Cheers,

Itsallaguess

9873210
Lemon Slice
Posts: 984
Joined: December 9th, 2016, 6:44 am
Has thanked: 226 times
Been thanked: 296 times

Re: Coronavirus - Modelling Aspects Only

#310018

Postby 9873210 » May 19th, 2020, 4:31 pm

modellingman wrote:R is simply a population characteristic and, as you surmise, if that population is divided into sub-populations (such as care homes, hospitals and "the rest") that characteristic may well take different values in each sub-population. If one of those sub-populations is much larger than the others then the characteristic of the population as a whole is largely going to be determined by that of the dominant sub-population. It seems highly unlikely that R is rising generally because it is rising in the two smaller sub-populations (care homes and hospitals) but falling in "the (much larger) rest."

R is essentially new case/existing cases. In calculating R for the general population, sub-populations are weighted by number of cases in the sub-population, not the size of the sub-population.

If infections are entirely or largely restricted to a sub-population then R of the general population is the R of the sub-population, no matter how small the sub-population. This will continue until cases are no longer restricted to the sub-population, which will happen when the disease is controlled in the sub-population or it breaks out into the general population.

modellingman
Lemon Slice
Posts: 614
Joined: November 4th, 2016, 3:46 pm
Has thanked: 594 times
Been thanked: 364 times

Re: Coronavirus - Modelling Aspects Only

#310029

Postby modellingman » May 19th, 2020, 5:04 pm

9873210 wrote:
modellingman wrote:R is simply a population characteristic and, as you surmise, if that population is divided into sub-populations (such as care homes, hospitals and "the rest") that characteristic may well take different values in each sub-population. If one of those sub-populations is much larger than the others then the characteristic of the population as a whole is largely going to be determined by that of the dominant sub-population. It seems highly unlikely that R is rising generally because it is rising in the two smaller sub-populations (care homes and hospitals) but falling in "the (much larger) rest."

R is essentially new case/existing cases. In calculating R for the general population, sub-populations are weighted by number of cases in the sub-population, not the size of the sub-population.

If infections are entirely or largely restricted to a sub-population then R of the general population is the R of the sub-population, no matter how small the sub-population. This will continue until cases are no longer restricted to the sub-population, which will happen when the disease is controlled in the sub-population or it breaks out into the general population.


Fair point. A quick search at UK level suggests that there are about 500,000 care beds and 150,000 hospital beds, so the two sub-populations amount to about 1% of the population as a whole. Unless current rates of infection in these small sub-populations are approaching 100 times the rates in the general population, the general population will dominate the determination of R. As others have noted, R is about rates of change. Your point underlines that scale within populations is also an important measure.

servodude
Lemon Half
Posts: 8271
Joined: November 8th, 2016, 5:56 am
Has thanked: 4435 times
Been thanked: 3564 times

Re: Coronavirus - Modelling Aspects Only

#310112

Postby servodude » May 20th, 2020, 12:39 am

9873210 wrote:If infections are entirely or largely restricted to a sub-population then R of the general population is the R of the sub-population, no matter how small the sub-population


That is a key point that I think often gets lost

With a simple lumped R the size of the population only affects the maximum number that can be infected - not how it happens

It seems sadly appropriate to consider John Horton Conway's Game of Life for this
- the size of the population is simply the visible field
- what's going on is a function of the current state

or if you drop a bit of oil in a bucket of water
- it spreads out the same irrespective of the surface area
- with enough oil you cover the surface and that's when the surface area affects the behaviour

- sd

Nimrod103
Lemon Half
Posts: 6467
Joined: November 4th, 2016, 6:10 pm
Has thanked: 939 times
Been thanked: 2255 times

Re: Coronavirus - Modelling Aspects Only

#310239

Postby Nimrod103 » May 20th, 2020, 12:23 pm

The curious case of all the countries whose death statistics are bogus because of ignoring care home fatalities:

https://order-order.com/2020/05/20/eu-r ... /#comments

servodude
Lemon Half
Posts: 8271
Joined: November 8th, 2016, 5:56 am
Has thanked: 4435 times
Been thanked: 3564 times

Re: Coronavirus - Modelling Aspects Only

#310254

Postby servodude » May 20th, 2020, 1:39 pm

Nimrod103 wrote:The curious case of all the countries whose death statistics are bogus because of ignoring care home fatalities:

https://order-order.com/2020/05/20/eu-r ... /#comments


Do you reckon that they would double the results if included?
6dB margin is something I'm quite comfortable working with
- doesn't make then bogus
- just makes them what they are (and you can garner the higher order metrics )

-sd


Return to “Coronavirus Discussions”

Who is online

Users browsing this forum: No registered users and 9 guests