Coronavirus - Modelling Aspects Only

[servodude]

Cases is a useless measure because it normally depends upon how many of which people are tested.

In your NSH opinion.

Personally I am happy with the figure, because I doubt the basis for selecting subjects and the nature of the testing is varying dramatically from day to day. IOW I feel the variation in methodology is limited enough to observe real trends.

GS

"Cases" is just a recorded measure for some sampling scheme, nothing more nor less; you need to interpret anything from it with context (sampling strategy/count/selection criteria, reliability of test/collection)
Granted the context can be difficult to ascertain, or corrupted because it's been put through the filter of politics, but you only get the data that you have; being noisy doesn't make it useless, just makes it harder to make out the real picture

- sd

[GoSeigen]

Cases is a useless measure because it normally depends upon how many of which people are tested.

In your NSH opinion.

Personally I am happy with the figure, because I doubt the basis for selecting subjects and the nature of the testing is varying dramatically from day to day. IOW I feel the variation in methodology is limited enough to observe real trends.

GS

"Cases" is just a recorded measure for some sampling scheme, nothing more nor less; you need to interpret anything from it with context (sampling strategy/count/selection criteria, reliability of test/collection)
Granted the context can be difficult to ascertain, or corrupted because it's been put through the filter of politics, but you only get the data that you have; being noisy doesn't make it useless, just makes it harder to make out the real picture

- sd

Agreed, as I said, " I doubt the basis for selecting subjects and the nature of the testing is varying dramatically from day to day. " and so for watching trends I'm satisfied its good enough.

GS

[servodude]

Agreed, as I said, " I doubt the basis for selecting subjects and the nature of the testing is varying dramatically from day to day. " and so for watching trends I'm satisfied its good enough.

GS

Totally with you on that GS!

I've found a lot of certainty and assertation creeping in to this modeling thread; I think that should be left to the politicians (and hopefully done elsewhere)

-sd

[Backache]

It is my understanding from a number of sources that Covid-19 infection can be resisted without the creation of antibodies. I think this paper is compatible with that conclusion. If you disagree I am happy to spend some time going through it, but probably not today because I have a lot to do.

The paper may be compatible but it is not supportive as the only cases that they followed up fully developed antibodies.
However this thread appears to be about modelling so I won't labour the point about immunity.

[zico]

Some worrying stuff here about how coronavirus immunity is likely to be only temporary. One of the links talks about how the amount of exposure to Covid-19 is likely to be more significant than whether or not people have previously had the disease.
It's based on study on other coronaviruses, and important to note that I'm not aware that anyone has been proved to have been re-infected so far, so it may be unduly pessimistic, but there are a number of links and sources within this link.

https://threadreaderapp.com/thread/1273 ... 27008.html

[bruncher]

Just a brief reminder about keeping this thread on topic to discussions about modelling and data, and avoiding getting into potentially political arguments about opposing views - it's easy to get sidetracked, and I'm probably as guilty as anyone.

Anyway, right back on topic, here's a link to the latest ONS mortality statistics on Covid-19 by ethnic groups.

The summary report is clear & easy to read, but here's my summary and take on the main findings.

Black and Asian people have 4 times the White death rate, reducing to 1.9 when adjusted for age, demographics and underlying health.
ONS study doesn't correct for occupation type - seems intuitively likely that this factor could explain much of the difference, as London NHS and care homes and other public-facing jobs likely to have a higher proportion of BAME staff than most other occupations - but that's just my theory, unsupported by proper data.

ONS shows Chinese people have a very similar death rate to White people. This is important because it means the stats from China (if they can be trusted) will apply equally to the White ethnic group.


https://www.ons.gov.uk/peoplepopulation ... 0april2020

The figures for difference between 'white' and 'BAME' given in the press is up to X4 difference, but there's often a conflation of "more likely to contract Covid-19" and "more likely to die from Covid-19 when infected".

The issue is increasingly taken up in politics, hence the importance of understanding what is going on, as some groups understandably claim they need special protection.

What I have not seen in any modelling is any recognition of the potential differences in the way different ethnic/cultural groups have responded to the Government's instructions for social isolation during lockdown, which I feel could be significant just as I agree that occupation is going to be a major factor.

[Itsallaguess]

Coronavirus was already in Italy by December, waste water study finds -

Italian scientists say sewage water from two cities contained coronavirus traces in December, long before the country's first confirmed cases.

The National Institute of Health (ISS) said water from Milan and Turin showed genetic virus traces on 18 December.

It adds to evidence from other countries that the virus may have been circulating much earlier than thought. In their study, ISS scientists examined 40 sewage samples collected from wastewater treatment plants in northern Italy between last October and February.

Samples from October and November came back negative, showing that the virus had not yet arrived, ISS water quality expert Giuseppina La Rosa said. Waste water from Bologna began showing traces of the virus in January.

https://www.bbc.co.uk/news/world-europe-53106444

Cheers,

Itsallaguess

[modellingman]

Just a brief reminder about keeping this thread on topic to discussions about modelling and data, and avoiding getting into potentially political arguments about opposing views - it's easy to get sidetracked, and I'm probably as guilty as anyone.

Anyway, right back on topic, here's a link to the latest ONS mortality statistics on Covid-19 by ethnic groups.

The summary report is clear & easy to read, but here's my summary and take on the main findings.

Black and Asian people have 4 times the White death rate, reducing to 1.9 when adjusted for age, demographics and underlying health.
ONS study doesn't correct for occupation type - seems intuitively likely that this factor could explain much of the difference, as London NHS and care homes and other public-facing jobs likely to have a higher proportion of BAME staff than most other occupations - but that's just my theory, unsupported by proper data.

ONS shows Chinese people have a very similar death rate to White people. This is important because it means the stats from China (if they can be trusted) will apply equally to the White ethnic group.


https://www.ons.gov.uk/peoplepopulation ... 0april2020

The figures for difference between 'white' and 'BAME' given in the press is up to X4 difference, but there's often a conflation of "more likely to contract Covid-19" and "more likely to die from Covid-19 when infected".

The issue is increasingly taken up in politics, hence the importance of understanding what is going on, as some groups understandably claim they need special protection.

What I have not seen in any modelling is any recognition of the potential differences in the way different ethnic/cultural groups have responded to the Government's instructions for social isolation during lockdown, which I feel could be significant just as I agree that occupation is going to be a major factor.

Whilst I agree with zico's comments about the study being easy to read, it is section 4 of the technical appendix ( https://www.ons.gov.uk/peoplepopulation ... ethodology ) which contains the detailed results. These are presented in the form of two charts separately for males and females and show the "odds ratio" for the likelihood of death for specific ethnic groups compared to the White ethnic group. Death in these circumstances is defined as death from/with Covid-19 in the 6 week study period to 10 April 2020. Courtesy OGL, the charts are shown below.



The study attempts to correct for a variety of factors that are known to have a general influence on mortality from Covid-19. If an ethnic group is over-represented at the unfavourable end of the scale for a particular factor (or under-represented at the favourable end) the modelling approach adjusts the mortality measure to account for this. After adjusting for all 6 factors, Blacks had a mortality rate greater than Whites by a factor of 1.93 for males and 1.89 for females. For Bangladeshi/Pakistani people the factors reduce to 1.81 (males) and 1.61 (females). For males the factors decline across the ethnic groups in order Other(1.34), Indian(1.32), Chinese(1.18) and Mixed(1.03). For females it is Indian (1.43), Other(1.08), MIxed(1.03) and Chinese(0.75).

I remain to be convinced about arguments related to occupation type, simply because the White group numerically dominates the statistics. For example, this source ( https://www.ethnicity-facts-figures.ser ... rce/latest ) suggests that whilst Whites are under-represented in the NHS the numbers are still very large at 79.2% of the NHS workforce compared to 85.6% of the working age population. The largest ethnic group from this source is Asian and comprises 10% of the NHS workforce compared to 7.2% of the working age population. Blacks in the same terms are 6.1% of the NHS and 3.4% of the WA population. There are some caveats of course, the working age population is based on the 2011 census (as is much of the ONS study data) whilst the NHS figures are from March 2019. There are, as zico notes,other occupational groups (such as care home workers) that might be important. The ONS technical appendix notes that the Bangladeshi/Pakistani ethnic group is by far the biggest ethnic group in the transport and drivers and operatives standard occupational classification sub-major group, which encompasses bus, coach and taxi drivers and those driving other types of industrial and agricultural vehicles, though takes this observation no further in the analysis (probably because of data limitations).

In terms of Bruncher's remarks, elements of the press and some politicians have jumped on the "4 times as likely" headline. Bruncher himself raised this on Polite Discussions a while ago and received these modelling related responses from ursaminotaur and myself
https://lemonfool.co.uk/viewtopic.php?f ... 60#p318231
https://lemonfool.co.uk/viewtopic.php?f ... 60#p318252
https://lemonfool.co.uk/viewtopic.php?f ... 80#p318722

The suggestion of a differential response across ethnic groups is an interesting one I have not seen elsewhere. The lack of modelling is, I suspect, due to a paucity of data on which to base appropriate models. Other hypotheses are based around Vitamin D, and again from Polite Discussions there is relevant comment here:
https://lemonfool.co.uk/viewtopic.php?f ... 00#p319202
and in the following few posts.

[GoSeigen]

The suggestion of a differential response across ethnic groups is an interesting one I have not seen elsewhere. The lack of modelling is, I suspect, due to a paucity of data on which to base appropriate models. Other hypotheses are based around Vitamin D, and again from Polite Discussions there is relevant comment here:
https://lemonfool.co.uk/viewtopic.php?f ... 00#p319202
and in the following few posts.

If there's any significant variation the overwhelming explanation for that fact will be the very act of inventing artificial "ethnic groups" and then people treating those groups unequally in law and in daily action. This should be blindingly obvious and has been discussed ad nauseam of course.

Person A takes white mice, feeds them well and lets them run around in a natural environment; takes black rats and puts them on a starvation diet and keeps them in a dark metal box. They are then mixed together and person B comes along, makes a careful statistical study and discovers differentials between the white and black mice. What a genius! He concludes that white fur must confer special advantages on mice, and readers of his study speculate as to what scientific explanation there may be.

GS

[johnhemming]

The suggestion of a differential response across ethnic groups is an interesting one I have not seen elsewhere.

I am pretty sure I posted links to this on TLF a few months ago.

[ReformedCharacter]

If there's any significant variation the overwhelming explanation for that fact will be the very act of inventing artificial "ethnic groups" and then people treating those groups unequally in law and in daily action. This should be blindingly obvious and has been discussed ad nauseam of course.

Person A takes white mice, feeds them well and lets them run around in a natural environment; takes black rats and puts them on a starvation diet and keeps them in a dark metal box. They are then mixed together and person B comes along, makes a careful statistical study and discovers differentials between the white and black mice. What a genius! He concludes that white fur must confer special advantages on mice, and readers of his study speculate as to what scientific explanation there may be.

GS

The researchers also found that people with blood type A had a 45 percent increased risk of contracting the coronavirus and developing respiratory failure compared to people with other blood types. On the other hand, people with blood type O had a 35 percent lower risk of developing severe COVID-19 illness.

https://www.news-medical.net/news/20200 ... irmed.aspx

Of course the distribution of blood groups differs between populations in different countries (or - ahem - 'ethnic groups'):

https://en.wikipedia.org/wiki/Blood_typ ... by_country

RC

[dealtn]

If there's any significant variation the overwhelming explanation for that fact will be the very act of inventing artificial "ethnic groups" and then people treating those groups unequally in law and in daily action. This should be blindingly obvious and has been discussed ad nauseam of course.

Person A takes white mice, feeds them well and lets them run around in a natural environment; takes black rats and puts them on a starvation diet and keeps them in a dark metal box. They are then mixed together and person B comes along, makes a careful statistical study and discovers differentials between the white and black mice. What a genius! He concludes that white fur must confer special advantages on mice, and readers of his study speculate as to what scientific explanation there may be.

GS

Not sure I would go as far as "overwhelming explanation".

Imagine amongst those "black rats" there was a pre-ponderence to something else, lets call it, say "sickle cell anaemia", that didn't affect white mice to the same extent. A second experiment was undertaken that swapped the rats and mice, and it remained the case that sickle cell anaemia was still disproportionately affecting the rats, as it did in a third experiment where rats and mice were treated equally.

Sometimes genetics do have an influence. The cause of that genetic difference might well be due to "conditions" such as "starvation", "dark metal box" type environments (or exposure to malaria) multi-generations back, but can still be an influence a long time after such "conditions" ceased, and indeed were never exposed to the current "subjects".

[GoSeigen]

The suggestion of a differential response across ethnic groups is an interesting one I have not seen elsewhere.

I am pretty sure I posted links to this on TLF a few months ago.

Where?

[GoSeigen]

Of course the distribution of blood groups differs between populations in different countries (or - ahem - 'ethnic groups'):

Only blood group? I bet other things differ. Like skin colour perhaps? Curly hair? Big rubber lips? A propensity to blow oneself up? Criminality? But of course none of these differ within countries (or - ahem - 'ethnic groups'), no no no...



GS

[johnhemming]

Where?

viewtopic.php?f=76&t=21665&p=285920&hilit=ACE2#p285920

[ReformedCharacter]

Of course the distribution of blood groups differs between populations in different countries (or - ahem - 'ethnic groups'):

Only blood group? I bet other things differ. Like skin colour perhaps? Curly hair? Big rubber lips? A propensity to blow oneself up? Criminality? But of course none of these differ within countries (or - ahem - 'ethnic groups'), no no no...



GS

You had the opportunity to respond to my comment that there are genetic differences among people in different countries which affect the rate of infection and outcome of Coronavirus on a board with the subject 'Coronavirus - modelling aspects only please' but you seem to have sullied the debate with offensive stereotypes rather than trying to defend your proposition that:

If there's any significant variation the overwhelming explanation for that fact will be the very act of inventing artificial "ethnic groups" and then people treating those groups unequally in law and in daily action.

I can understand that you may have an objection to the term 'ethnic groups' but you seem to be letting that prevent you from accepting that there are perfectly scientific and rational explanations for the different outcomes from Coronavirus between people with different genetics.

RC

[johnhemming]

Sometimes genetics do have an influence.

In the case of Covid-19 it is the prevalence of ACE2 cells.

[Mike4]

Sometimes genetics do have an influence.

In the case of Covid-19 it is the prevalence of ACE2 cells.

And the number of ACE2 cells from what I've read seems to be largely dependent on the size of the patient. The 'larger' the patient, the more blood vessels wall area they will have and therefore the more ACE2 cells they will have. Hence the reason for obese patients having a higher representation in ICUs than in the general population.

I think this is still speculation rather than peer reviewed research, so could be coincidence rather than cause and effect.

[zico]

Having looked at the latest ONS figures, it seens there could be a modelling & scientific case for ending lockdown/herd immunity.

Note : based on modelling, not politics! As ever, use Polite Discussions board for political comment please.


I've made assumptions about particular starting figures, but grateful if people could discuss the general approach, rather than individual starting figures. I've set out the approach so you can easily insert your preferred figures/assumptions/wild guesses and get a different set of results based on your assumptions.


I've used England & Wales only figures below, because that's what was in latest ONS weekly stats.

First assumptions is that we need 70% for herd immunity and Infection Fatality Rate is 0.8% (Based on Cambridge modelling given to SAGE).
England & Wales Herd immunity deaths = 250,000 deaths (45m popn x 70% x 0.8%)
(Note - the recent steroid treatment breakthrough was estimated to have potentially saved 5k lives out of 50k so would be a 10% reduction - I haven't included this to keep it simple).

(If you disagree with the above figures, you can easily swap your preferred figures).

Based on ONS week 24 figures (w/e 12th June) there were 160 deaths per day.
67% of deaths were people aged 80+
21% were people aged 70-79
7% were people aged 60-69
5% were people aged under 60.

If we assumed the 250k herd immunity deaths followed the same pattern, then
168k deaths for people aged 80+
52k deaths for people aged 70-79
18k deaths for people aged 60-69
14k deaths for people under 60

People aged 60+ account for 18% of the total England & Wales adult population.
Now, let's make a big (but perhaps reasonable) assumption that people aged 60+ are much more careful than average, so "only" 30% of them get infected instead of 70%.
That brings the total deaths down to 108,000.
There have already been 60,000 Covid-related deaths in Eng & Wales, so if only 30% of over 60s get infected, there'd be 48,000 more deaths.

A related point is that we're currently at a low level of Covid-19 deaths (week 24 figures were 160 deaths per day).
If, for the next 6 months we can keep to an average deaths of 160 per day, there'd be 29,000 more deaths.

Also, in 6 months time, there's a reasonable chance there could be a vaccine, which would be given initially to over 60s (the group most at risk). So over 60's would just need to be very careful over the next 6 months, and there's a fighting chance they could be immunised against Covid-19 which should give people motivation to be extra-careful.

I've assumed an average daily deaths figure of 160, but what will happen in practice is that the figure as of today's date will be lower than that, and gradually increase as lockdown measures are lifted, so government will have plenty of time to take measures to combat increases in infections.

If a vaccine isn't available for another 12 months (or possible never) than that creates a different set of challenges, because elderly people would need to limit their social contacts indefinitely.

The ONS week 24 death figures are still showing a large number of deaths amongst people over 70s despite the fact that they were probably initially infected one month ago (in early May) when public awareness should have been at its height.

Incidentally, there's been a lot of talk about the "cure being worse than the disease". The only bit of evidence I've found for that is the Times on Saturday referring to an ONS report given to SAGE on the health impacts of a L-shaped recession caused by coronavirus, and they estimated this to be 12,000 lives lost per year due to this kind of recession.

[dealtn]

I've assumed an average daily deaths figure of 160, but what will happen in practice is that the figure as of today's date will be lower than that, and gradually increase as lockdown measures are lifted, so government will have plenty of time to take measures to combat increases in infections.

All quite reasonable, but I will just focus on this bit.

I don't think we know what will happen to the R number (which in itself isn't perfect at low numbers) or the daily deaths as a result of lockdown changes.

Partly this will depend on how those different (age) groups (and other groups of differing outcomes if infected) respond. In fact if those least affected actually acted speedily and unwisely it might even be of benefit to the other groups, so long as they weren't affected.

As one of the scientific advisors very early on put it (can't remember who, and the exact quote), if everyone who was likely to die were isolated together in say the Hebrides, and everyone else mingled freely such that the virus spread as fast as possible, but only amongst that "unaffected" group, the virus would run its course quickly and no lives would be lost.

Of course that isn't possible, and the quick route to herd immunity this represents would in practice lead to lots of deaths in the "Hebrides" group in practice.

But if the vulnerable groups remained isolated/shielded/careful (which is exactly the point you make) lockdown easing doesn't necessarily translate into an increase in the current daily deaths of 160 (and probably lower given the known lags).