Oxford Uni CV epidemiology study - good news?

New Topic
This topic has been archived, and won't accept reply postings.
 TobyA 25 Mar 2020

I've heard some discussion of this study - the FT has a level headed article on it https://www.ft.com/content/5ff6469a-6dd8-11ea-89df-41bea055720b and this seems to be the actual article https://www.dropbox.com/s/oxmu2rwsnhi9j9c/Draft-COVID-19-Model%20(13).pdf?d...

I don't understand epidemiology beyond on the simplest level and don't understand the maths that their model is based on, but from reading the coverage on it, it seems that if they are correct it is rather optimistic. Their work suggests many of us are potentially already infected and have recovered so should be immune now. 

But that almost feels like wishful thinking - it would be nice if my grotty feeling of flu-eyness a few weeks back and the tickley cough I've had since was "it". I could get out and go and help people in some way. But then you read of people who have tested positive and feel really poorly with it - fit young people. And of course we see the horror of the situation in Italy, increasingly in Spain too, and the increasing number of deaths here as well. Then things feel really dark.

Yet on the other hand, these are proper scientists at a proper university, putting their work out for everyone to check and, as the FT says, cooperating with teams at other unis to check their work. So perhaps a ray of light? 

 Liamhutch89 25 Mar 2020
In reply to TobyA:

I'm also very interested to read some critique / peer review of this study. In my non-medical brain I don't understand how this marries up to seeing an exponential growth in deaths right now, but as you say, these are proper scientists at a proper university. 

OP TobyA 25 Mar 2020
In reply to Liamhutch89:

Exactly my feeling Liam. I guess prepare for the worst (stay at home), hope for the best.

 skog 25 Mar 2020
In reply to TobyA:

> Yet on the other hand, these are proper scientists at a proper university, putting their work out for everyone to check

Yep - and proper scientists will do just that, and their work will contribute to the body of knowledge on the subject, and move science forward that way.

Even exploration of unlikely possibilities can be good science; a lot of good science has been done that pointed towards incorrect conclusions.

I'd love to believe this one, but it's really hard to square with the rate of spread and the situations in Spain and Italy.

 RomTheBear 25 Mar 2020
In reply to Liamhutch89:

> I'm also very interested to read some critique / peer review of this study. In my non-medical brain I don't understand how this marries up to seeing an exponential growth in deaths right now, but as you say, these are proper scientists at a proper university. 


I’ve read the paper. It has no conclusion really, all it says it that, under their simplistic model, it is possible that more than 50% of the population is already infected, but that is just one of the many scenario that produce a decent model fit.

It’s also very limited, there is no cross checking with more accurate models, incubation period isn’t even take into account.

It’s completely inconclusive, it just highlights a possibility.

Post edited at 09:14
1
 DaveHK 25 Mar 2020
In reply to TobyA:

We have 2 very different models from 2 respected institutions. Does this not just suggest that we don't have reliable enough data or robust enough models to model well?

Post edited at 09:19
 RomTheBear 25 Mar 2020
In reply to DaveHK:

Exactly. Relying on model for decision making at this point is reckless.

 wintertree 25 Mar 2020
In reply to TobyA:

> Yet on the other hand, these are proper scientists at a proper university, putting their work out for everyone to check and, as the FT says, cooperating with teams at other unis to check their work. So perhaps a ray of light? 

More signs that we desperately need a test to identify recovered individuals and solid data on recovered>immune status.

 Dave Garnett 25 Mar 2020
In reply to RomTheBear:

All models are wrong, but some are useful.  The ones that best fit the observed data trends.

This is interesting but it's only based on the first 15 days following the first reported death (for reasons they explain).  I'm no epidemiologist but that doesn't cover the period in which the curves will diverge very much, so the potential for errors is large.

This question of how many people have actually had COVID-19, with mild or even no symptoms, and have now recovered is hugely important.  The authors make the point that a broad survey of seropositivity is crucial and still not being done.  

I thought the general view was that it was very unlikely that the virus has been widespread in the population (especially outside China) much before February, partly because the original infection had been pretty accurately tracked.  If so, I don't see how the basic assumption of the paper can be right.

Post edited at 09:36
 RomTheBear 25 Mar 2020
In reply to Dave Garnett:

> All models are wrong, but some are useful.  The ones that best fit the observed data trends.

No. Not even remotely. I could give you easily dozens different models that fit the observed data perfectly and all predict something completely different.

In fact that kind of what this studies does, they show that a wide range of possible parameters fit the data perfectly.

As Von Neumann famously said: « With four parameters I can fit an elephant, and with five I can make him wiggle his trunk »

I’m not criticising the paper, it’s a good idea to try to explore the range of parameters that fit the little data that we have, and the approach us sound, although limited.

But the treatment and misrepresentation of it in the press is dreadful.

Post edited at 09:36
1
 Dave Garnett 25 Mar 2020
In reply to RomTheBear:

Sorry, I hit send prematurely!

 RomTheBear 25 Mar 2020
In reply to Dave Garnett:

No probs. Makes more sense now ! Thanks.

I can see that Petra Keplac, one of the prominent expert in infectious disease modelling in the U.K., also bashed the FT headline on Twitter:

”Really irresponsible, tabloid-like, headline from @FT - the Oxford study doesn't claim that half of the population might have been infected by coronavirus. “

 summo 25 Mar 2020
In reply to TobyA:

I would put it in the box called "dangerous unproven speculation that bears no correlation with existing evidence"... I knew there were plenty Muppets in career education. This just proves my own thoughts, it's reckless and counter productive. 

5
 mondite 25 Mar 2020
In reply to TobyA:

Its all going to be guesswork until a decent level of testing is done across the population and not just those who have ended up in hospital.  There does seem to be highly variable responses but without the testing no one knows the split.

As an aside on the symptoms side. Good to see the tree pollen starting to pick up. Thats going to be having a fair few people feeling pretty shitty and wondering if it ticks the low end infected.

 RomTheBear 25 Mar 2020
In reply to mondite:

> Its all going to be guesswork until a decent level of testing is done across the population and not just those who have ended up in hospital.  There does seem to be highly variable responses but without the testing no one knows the split.

> As an aside on the symptoms side. Good to see the tree pollen starting to pick up. Thats going to be having a fair few people feeling pretty shitty and wondering if it ticks the low end infected.

Interestingly I’ve learned recently that there is a lot of evidence that allergies and viral infections have synergistic effects. Having allergies apparently makes you more prone to viral infections, and possibly, vice versa.

Always amazes me how very little we understand about the human body.

OP TobyA 25 Mar 2020
In reply to summo:

Aren't you a farmer or something? Do you understand the maths or the medicine behind this?

It would be helpful if people who weigh in on this issue so strongly, you, Rom, Wintertee for example, would be willing to explain just a tiny bit why your opinions are worth listening to. I know Dave G's background so know his opinion is worth listening to even if this wasn't his field of research, but are you a scientist of some type?

 Dave Garnett 25 Mar 2020
In reply to RomTheBear:

> Interestingly I’ve learned recently that there is a lot of evidence that allergies and viral infections have synergistic effects. Having allergies apparently makes you more prone to viral infections, and possibly, vice versa.

It's true that an antiviral immune reaction requires TH1 type T cell response, and asthma is classically characterised by a predominantly TH2 response but it's complicated.  I wouldn't want to draw any conclusions about susceptibility to COVID-19 complications.

 summo 25 Mar 2020
In reply to TobyA:

> Aren't you a farmer or something? Do you understand the maths or the medicine behind this?

Are farmers all uneducated illiterates? There's probably more science in farming and forestry now than ever before. I've only a couple of degrees, in the sciences; physics, environmental and earth science related, but I struggle on by. Not precisely immunology, but they do dip in and out of biology and chemistry a fair bit, no science field runs in isolation. 

> It would be helpful if people who weigh in on this issue so strongly, you, Rom, Wintertee for example, would be willing to explain just a tiny bit why your opinions are worth listening to. I know Dave G's background so know his opinion is worth listening to even if this wasn't his field of research, but are you a scientist of some type?

My A level maths(pure and stats) alone, makes me sceptical when folk throw out reports based on a multitude of estimates. The more unprecise figures you put in, the errors compound, they don't cancel each other out. 

Just look at Deaths, infection rates,  etc across the world there is fairly clear correlation, the graphs (no shortage in the media ) trend in the same direction, albeit with regional variation. Just like a UK graph will have variation across the country. 

There is nothing suggesting the UK is an exception to global trends. It's the same virus, largely similar society structures. I fail to see how they draw such a varied conclusion. 

Of course it's just a theory, until you test a fair proportion of any population we'll not know with more certainty. 

Post edited at 10:43
4
 summo 25 Mar 2020
In reply to TobyA:

Ps. If they are right and I'm wrong. The death rate should fall dramatically over the next week, as if so many already have covid, then each day that passes, proportionally fewer new people are left to catch it. 

Why is the UK death rate climbing? Yes I know there is a lead in time. But it should still be slowing. 

Post edited at 10:48
 Offwidth 25 Mar 2020
In reply to summo:

Well, I think given your educational qualifications and your ability to read the linked paper and comment fairly on it, what you said about the paper is plain insulting (something you unfairly chose to do and that being nothing at all to do with your job). What is disproportionate here is what the newspapers made of it (their headlines were not claimed by the authors) and the wish fullfilment in what some people made of the headlines.

The best confirmed peer reviewed science is still matching what WHO said about China and slightly tweeked since then. Mortality rates below 1% if the health system can cope and maybe as much as 3% if it can't (with risks lower with age but still some reported cases of healthy young, with no pre-conditions, falling seriously ill and needing a ventilator). The stated 'mild illness' means like flu that wouldn't require hospital help (ie a high fever and feeling really rotten but not struggling to breathe)

4.. 3...2 ...on the data, even the government seem to have finally accepted we are 2 weeks behind Italy (with no guarentee we will stay that way for weeks of course, especially given many infected in Lombardy ran to other parts of their country just before the lock-down... and other different factors in Italy... some positive, as their hospitals were less stressed and better equipt before).

The deaths are still climbing approximately exponentially, but with daily variability, as we are still on the early part of the exponential infection growth, and cases now were almost certainly based on infections around a week or more ago, prior to serious curtailment in social mobilty. Levelling might take longer than 2 weeks, as when it starts to level in hotspots that might be masked by where it takes off elsewhere (as indicated in Italy). All exponential growth patterns will look the same at some point irrespective of the mortality rate until a significant fraction of the population gain immunity....the way to find out mortality rates is to test nearly all for covid 19 or, more likely, a simpler antibody test for having had it. Irrespective, it's nothing like seasonal flu as it overwhelms health systems even with some strict population control.

Like some I know, and many others have reported, I've had a long lasting infection since December with initial symptoms like a bad cold and months after that with a nagging cough, but the overall symptoms don't match Covid 19 at all... not that this stops the conspiracy theorists claiming its been here since Xmas.

Post edited at 13:45
1
 wintertree 25 Mar 2020
In reply to Offwidth:

> weeks, as when it starts to level in hotspots that might be masked by where it takes off elsewhere (as indicated in Italy).

Also hopefully masked by a bump in detection as we ramp up testing rates in the UK.  

 summo 25 Mar 2020
In reply to Offwidth:

We'll agree to disagree on the paper. I still think they wrote the conclusion first. 

6
OP TobyA 25 Mar 2020
In reply to summo:

> Are farmers all uneducated illiterates?

I'm quite certain you really know that that wasn't what I was implying - but certain people in these discussions have very strong views one way or the other. You at times, along with plenty of others will give strong opinions on, for instance, this study.  I just want to know why people are so certain. I think most of us without any applicable knowledge - that's definitely me - just aren't sure what to believe, or maybe more accurately how others are interpreting what information there is so differently. 

A mate has just become an associate professor at one of the Russell Group universities, he has had a Royal Society Fellowship for some years funding his research. A smart guy who clearly knows his stuff in his field, plant science. I asked him when this was all kicking off a week or so ago if he thought there would be any chance experienced scientists in other fields might start getting seconded into efforts to fight CV - I think the NHS in London had been appealing for people with lab skills to help ramp up the testing, something like that around then. He reckoned his skills and knowledge weren't applicable in any real helpful way. 

So of course you get self appointed internet experts just like on anything, but not everyone is necessarily that. I'm not accusing you or Rom or Wintertree of being Russian trolls, or whatever the opposite to a Russian troll is, etc. I'm just interested in why you are so certain. 

 wintertree 25 Mar 2020
In reply to TobyA:

> I'm not accusing you or Rom or Wintertree of being Russian trolls, or whatever the opposite to a Russian troll is, etc. I'm just interested in why you are so certain. 

For me, in this case because I was quite certain because I believe I understand enough of what I have read and seen to make a judgement.  Enough that about 6 weeks ago I opened the zombie day preparation checklist for Wintertree Towers and followed it.  The only other time it started to be followed was when the government and EU were playing chicken over a no deal Brexit.

I have been disappointed so far to see things go the way I expected.  I note that part of what I read and saw was strong public dissent from many experts and leaders in key, relevant fields and part of it was private discussions with people much closer to the directly relevant fields than me.  It was not my interpretation of the reports from China alone, although that was enough to convince me...

I have had various derogatory comments thrown at me for taking a stand on here over the years.  The key to me is that I try and explain my reasons as well as I can and that I leave the reader to make their own mind up.   There are plenty of discussions on here where I have but don’t share my opinions as I don’t believe I have enough justification to stand by them.

In terms of staff from plant sciences - I think the most useful will be lab technicians who can run qPCR and who can be trained to do so at the covid testing centres.   There is no shortage of scientific talent in the areas relevant to antivirals and vaccines so nothing productive to be gained in trying to convert people from other specialities (condensing 10-40 years of work and knowledge into 2 weeks) but those people are often well placed to understand and interpret what we see.

The most worrying thing to me is that Rom, Summo and I all appear to be in close agreement.  (Edit: agreement includes the general situation.  I’m not taking a view on publishing this particular paper; I see pros and cons from both scientific and media/public angles)

Edit: The OP of this thread just called back to it to note how their views have changed.  I was one of the first replies and I think I gave a level headed take on the situation that was never “certain” about what would happen here - https://www.ukhillwalking.com/forums/off_belay/coronavirus_panic-716660?v=1#x9...

Post edited at 15:45
 summo 25 Mar 2020
In reply to TobyA:

I'm not certain. But neither should they be publishing data which isn't verified, has many estimations, that the national media latch onto. It creates false optimism and complacency, on day 2 of a fairly modest restriction. The virus will likely kill several hundred, potentially 1000s in London, suggesting half the population have it already, implies quarantining isn't worth the hassle. They were very irresponsible to release it publically. 

4
 Robert Durran 25 Mar 2020
In reply to summo:

> Ps. If they are right and I'm wrong. The death rate should fall dramatically over the next week, as if so many already have covid, then each day that passes, proportionally fewer new people are left to catch it. 

> Why is the UK death rate climbing? Yes I know there is a lead in time. But it should still be slowing. 

If half the population have had it, then the infection rate will only be falling if R (the average number of people infected by each ill person if there were no immunity) is less than 2. I think most estimates without good social distancing were between 2 and 3. If it has now dropped lower, there should be a drop in infection rates after the incubation period.

 summo 25 Mar 2020
In reply to Robert Durran:

I did say there would be a lead in period and it would of course depend on what date the UK reached 50% infection. 

It's a nice topic to bounce around the uni halls to prove or disprove, but not something appropriate for national release where many will interpret it as fact. 

3
OP TobyA 25 Mar 2020
In reply to wintertree:

So are you a scientist in the field or a related field?

 Timmd 25 Mar 2020
In reply to TobyA:

> I'm quite certain you really know that that wasn't what I was implying - but certain people in these discussions have very strong views one way or the other. You at times, along with plenty of others will give strong opinions on, for instance, this study.  I just want to know why people are so certain. I think most of us without any applicable knowledge - that's definitely me - just aren't sure what to believe, or maybe more accurately how others are interpreting what information there is so differently. 

> A mate has just become an associate professor at one of the Russell Group universities, he has had a Royal Society Fellowship for some years funding his research. A smart guy who clearly knows his stuff in his field, plant science. I asked him when this was all kicking off a week or so ago if he thought there would be any chance experienced scientists in other fields might start getting seconded into efforts to fight CV - I think the NHS in London had been appealing for people with lab skills to help ramp up the testing, something like that around then. He reckoned his skills and knowledge weren't applicable in any real helpful way. 

> So of course you get self appointed internet experts just like on anything, but not everyone is necessarily that. I'm not accusing you or Rom or Wintertree of being Russian trolls, or whatever the opposite to a Russian troll is, etc. I'm just interested in why you are so certain. 

You've put it as well as I hoped to, and better than I could do, I'm interested as well. 

Post edited at 16:09
 wintertree 25 Mar 2020
In reply to TobyA:

> So are you a scientist in the field or a related field?

I’ve a background including physics and bio physics and lately I am doing work with toxicity (including bacterial but not viral) and drug discovery.  I’m not an epidemiologist or a virologist but I can generally follow their publications and communications well and I know when - and who - to ask when I’m not clear.

I have also seen the same points as me made clearly by people with different backgrounds.

I don’t normally mention a background as I don’t like to call to perceived authority, partly as I know enough scientists whose opinion I would probably listen to and then do the opposite...

Post edited at 16:12
 Richard J 25 Mar 2020
In reply to TobyA:

The FT's science reporting is usually pretty good but I think this article, and the headline even more, was epically misjudged.  From my reading of the paper, all they're actually saying is that in the early stages of the epidemic, when you're seeing exponential growth, there are many combinations of parameters (like the fraction of covert infections, transmission rate, etc) that will give you the observed results.  

Undoubtedly there will be an unknown fraction of undocumented infections (see eg this paper using much more sophisticated modelling to estimate this number from the Chinese experience - https://science.sciencemag.org/content/early/2020/03/24/science.abb3221.ful...), but the indeterminacy of the fitting to the model used by the Oxford group means that this is an unknown that needs to be determined by some other means.

The models will be much more sensitive to the parameters individually when the numbers starts to level off - and that's when we'll need the modelling most, to work out what the exit strategy out of the current lock-down needs to be.

Post edited at 16:33
Deadeye 25 Mar 2020
In reply to TobyA:

Thanks for the link to the paper; hadn't seen the full text.

The Table 1 is pertinent (page 5) - 11 variables, the majority of which are not well known yet as we don't have population-based data.  The majority of the data we do have is from hospitals and concerns infections and deaths, not antibodies (recovered). 

My read (and I'm not am epidemiologist) is that by tweaking the variables you can get more than one set to produce a curve that fits what we've seen.

The data outside hospitals is confounded by some people just self-isolating (and so no record) and others calling 111 and getting scored as positive when in fact they had 'flu.

Overlay that we've had a variety of interventions which have been adopted with a wide range of rigour over a variable time.

I'd compare just two hypotheses:

1. Oxford is right and we're seeing the very small tip of a largely benign iceberg of current/past infection (death 0.01% say) that's been around longer

2. ICL is right and we're seeing most of the iceberg with quite high mortality (1-4%) and infection rate

On balance I'd favour #2 as more likely because:

- The symptoms appear to be quite distinctive; distinctive enough to be noticed as unusual in a high quality hospital environment.  So multiple cases much earlier would have prompted review; especially so in the current mild 'flu season and where some early patient deaths should have benefitted from 'flu vaccination.  Severe illness or death in younger patients would also red flag - and we didn't see that.

- The early tracing work established both the duration from exposure to sysmptoms (reasonably clearly) and that there is quite high transmisability.   If there was widespread immunity, I'd expect those traces to have run into far more dead ends.

- Countries are getting distinct waves depending on how their society works, what interventions are made and, perhaps, their genetics.  If the virus had covered significant proportions of the populations across the world, I'd expect to have seen less distinct local epidemics.

- I can't model this, but I would expect that if we have 50% immunity, then the impact of staged lockdown measures would have been more effective and more rapidly effective.

So, of course I hope we're almost done with it; but I fear not - and whilst it's in any way unclear the only sensible option is distancing.

Edit. You asked for credentials.  I was a virologist, although I  know more about HIV and herpes than C19 and it was all quite a time ago!

Post edited at 17:01
 SouthernSteve 25 Mar 2020

In reply:

There is an interesting review of the food and mouth outbreak modelling with post-outbreak analysis which shows even in a good model there are considerable uncertainties. There is currently considerable debate about the current model and that the model code is unavailable for review has led to scientific disquiet. The Oxford paper adds to this debate.

https://royalsocietypublishing.org/doi/10.1098/rspb.2008.0006

 wintertree 25 Mar 2020
In reply to SouthernSteve:

> There is currently considerable debate about the current model and that the model code is unavailable for review has led to scientific disquiet

I believe it’s going on GitHub and getting some attention...

https://twitter.com/neil_ferguson/status/1241835454707699713?s=20

Being the naive person I am, I’d imagined that the model was a “paper” description of the maths and a clean implementation made with high level maths libraries like python+numpy+scipy or matlab.  Spoiler: no.

The general vibe I get with this and other models is of domain specialists (virologists/epidemiologists) having gone and made some code.  I’m a great fan of a two step process of the domain specialists producing maths, and then someone else implementing the maths in the language closest to maths.  I’ve unpicked a few one-step solutions over the years.  The worst have invariably been done in MS Excel and its common to find all sorts of bugs in there.

This isn’t because I don’t trust the domain specialists ability to write code - that would frankly be insulting given how easy programming is compared to many scientific specialities.  It’s more that if one person does it all themselves, any broken assumptions (easily done), logical flaws or maths errors pass unnoticed.  It’s really hard not to make mistakes - there are infinitely more ways to get something complex wrong than right (the thermodynamics of bugs).  Having a break between people and a clear interface has a way of bringing many of these hidden problems into clarity.   Ideally when something becomes of national importance like this, you get two separate people to each make clean reference implantations of the maths into code and you cross check them, and you have a separate domain specialist validate the epidemiology model and its translation to maths.  

My other vibe I get looking at papers is that there are differential equation models for all of these things and that with a couple of measurements of the time evolution of the models, you could do phase space plots of the parameters vs the measurements and identify tipping points and high sensitivities in the models which would tell you where to look to tighten up real world measurement; the recently discussed Oxford paper is a prime example of not doing that.

The main value of the modelling I’ve seen to date is to hi-light how many unknowns there are, and the value of being over-cautious and potentially over-reacting in the early stages, lest we get it wrong and things grow faster than we can sensibly recover from.  Another poster who I suspect is a crack statistics expert (Richard J) noted elsewhere today that the models become most important for informing strategy on how to get out of lockdown.  It’s critical because there is so much time lag between cause (change of policy) and effect (detection of infection or death) that if we get it wrong it may be too late to fix it by the time we find out.

Post edited at 22:48
cb294 26 Mar 2020
In reply to RomTheBear:

That. You make a plausible mechanistic model based on parameters known to influence epidemy dynamics, then feed it the parameters you know and fit the others to make the model fit the overall data. In a dynamic situation like now, you repeat that as often as possible. The purpose of modelling is not (yet) to make predictions, but to constrain parameters and possibly detect whether early measures have any effect at all.

CB

 Richard J 26 Mar 2020
In reply to wintertree:

>...Another poster who I suspect is a crack statistics expert (Richard J) ...

I'm a grubby experimental physicist, thus not a crack expert at anything!

I think lots of eyebrows were raised at "thousands of lines of undocumented C".

On the "differential equation vibe" of the models, these are continuum models that in their simplest form look very much like chemical rate equations, appropriate for what chemical engineers would call "continuously stirred tank reactor" conditions.  Cities and nations aren't continuously stirred, though, so there are issues about how the spatial and stochastic characteristics of the spread are handled.  I know there's a bit of an effort starting now to coordinate contributions from modellers from other fields, for example in network dynamics, traffic modelling, some using agent based techniques and handling the very big data sets about personal mobility and interactions mobile phones give you.

 MG 26 Mar 2020
 DaveHK 26 Mar 2020
In reply to TobyA:

I've been working on some teaching resources for high school pupils on this stuff and these threads have been very useful.

OP TobyA 26 Mar 2020
In reply to Richard J:

Actually, to everyone and no one in particular - the Guardian science editor wrote about the problems of models yesterday https://www.theguardian.com/science/2020/mar/25/coronavirus-exposes-the-pro...

Quite helpful for people like me who don't understand the maths being used!

Deadeye 26 Mar 2020
In reply to Deadeye:

I'd add to my argument the short graph-movie half way down this page:

https://www.bbc.co.uk/news/world-us-canada-52033863

Anna McKay 27 Mar 2020
In reply to TobyA: Interesting thread you started.  There is an article from Stamford academics on the same topic in yesterday's Wall Street Journal Opinion https://archive.fo/sAElA It seems possible that we have got the modelling wrong.  We will only know when antibody tests are widely available.

I think that the prudent course of action is to behave as if the worst predictions are accurate, whilst subjecting them to intelligent scrutiny and analysing data as it becomes available, and reacting to new findings fast.

If infection and immunity have  built up on the scale predicted by both Oxford and Stamford, this tells us interesting things about our immune system.  I would indicate, I thunk, the merits of supporting our immune system rather than relying solely on slow to develop vaccines which may have short-lived efficacy (virus mutations...).  

 Toerag 27 Mar 2020
In reply to Deadeye:

I like how the USA was neck and neck with France and Germany whose controls then kicked in whilst the USA did naff-all and rose to the top of the chart.  They are stuffed.

 wintertree 27 Mar 2020
In reply to Anna McKay:

> I think that the prudent course of action is to behave as if the worst predictions are accurate, whilst subjecting them to intelligent scrutiny and analysing data as it becomes available, and reacting to new findings fast.

Exactly.

 wintertree 28 Mar 2020
In reply to Richard J:

Sorry, very slow reply.  

>  these are continuum models that in their simplest form look very much like chemical rate equations, appropriate for what chemical engineers would call "continuously stirred tank reactor" conditions.  Cities and nations aren't continuously stirred, though, so there are issues about how the spatial and stochastic characteristics of the spread are handled.

It's a bit of a nightmare I suspect - you can make a Markov model and DEQ solve it for the continuum (mean) behaviour or Monte Carlo it for the stochastic behaviour (variance about the mean in different runs).  I'm aware of supercomputer time in at leat one UK university being pooled for the modelling consortium; this suggests a lot of Monte Carlo modelling to me.

A coarse "population-level" model misses the hierarchical and clustered structure of the population and its transmission "bridges", and a fine grained model rapidly starts to accumulate so many free parameters that its predictive power is no better.  I do wonder if the datasets people like Google have snooped over the years for their advertising are a gold mine for this.

The benefit of these models to me seems to be in using them to identify the precedence in which parameters (social changes) improve the situation, although that always seems to come back to the transmission "bridges" aka R0 and reducing it - not a surprise as that's the key parameter in an exponential phase.

My intuition is that you want to balkanise the population into lots of islands with minimal bridges between them - the same household member going to the same shop every time, the same delivery people doing the same rounds to the same houses, the same food suppliers going to the same shops etc.   This creates islands with minimal chance of infection spreading between them and is a layer of structure on top of the current emergency measures.  You can also then try and improve the training of the key "bridge" people in terms of protocols to avoid passing on infection [1]. This is how you keep a hot object from heating a cool object - minimise bridging structures.  The thermal analogy is by no means identical to a pandemic, but it's not a bad starting point.   I'm hoping make some basic models and also that the ICL one lands on GitHub soon so I can dive in to making a reference version.  

Markov stuff is close to my heart if not my best expertise - I've had a side project I've been trying to finish for years with a ~1e8 state system. I have an (apparently) novel algorithmic approach to computationally efficiently finding the null space of the rate matrix to give the balance vector(s) of the system.  Much more enjoyable stuff than serious work but languishing for lack of time.

[1] I've been out one since lockdown for a perishables shop,  I split the world into "household" and "other", use disposable gloves and careful ordering in time and space of tasks to minimise contamination, with all my outer clothes coming off outside and going straight in to a boil wash and the shoes and car going into 72 hour quarantine.  A good time to develop careful OCD.

 pneame 28 Mar 2020
In reply to wintertree:

> Sorry, very slow reply.  

> My intuition is that you want to balkanise the population into lots of islands with minimal bridges between them - the same household member going to the same shop every time, the same delivery people doing the same rounds to the same houses, the same food suppliers going to the same shops etc.  

Interestingly, states in the US seem to be starting this route - FL requires anyone flying in from NY to quarantine, likely to do the same for louisiana - hopelessly inefficient and too late but a good effort.  And rhode island actively looking for new yorkers to quarantine! To a certain extent, the sheer size of the US makes for lots of islands, all semi-autonomous. When all this calms down in a few months the data modelers will have a field day. 

 Richard J 28 Mar 2020
In reply to wintertree:

> A coarse "population-level" model misses the hierarchical and clustered structure of the population and its transmission "bridges", and a fine grained model rapidly starts to accumulate so many free parameters that its predictive power is no better.  I do wonder if the datasets people like Google have snooped over the years for their advertising are a gold mine for this.

Yes indeed.  I'd be very surprised if this wasn't been studied extensively in China, where the snooping is pretty much state mandated.  And I expect our friends in Cheltenham know more about this than they're saying.  

> My intuition is that you want to balkanise the population into lots of islands with minimal bridges between them - the same household member going to the same shop every time, the same delivery people doing the same rounds to the same houses, the same food suppliers going to the same shops etc.   This creates islands with minimal chance of infection spreading between them and is a layer of structure on top of the current emergency measures.  

Here's a bit of support for your intuition, in a simple model from a former collaborator of mine - https://arxiv.org/abs/2003.08784

 Doug 28 Mar 2020
In reply to Richard J:

There was an article in Le Monde a day or so ago where Orange had used mobile phone data to show that about one million people had 'fled' Paris just before the restrictions on travel came into force & to show where they went. Obviously not everyone has a mobile phone but such data  must give a pretty good idea of population movements and maybe help

 Davidlees215 28 Mar 2020
In reply to TobyA:

Some of the best information comes from the diamond princess cruise ship where almost everyone on board was tested. Approximately a third had become infected over a few days. Half of those with positive results showed no symptoms, just over 1% died. 

There are some obvious restrictions to any analysis of this data. There were more older people on the ship than the general population, but although many had some underlying health conditions, few had very serious life limiting conditions. 

Given these demographics and the death rate, most dying within a couple of weeks of the ship docking, it would seem unlikely half the uk population have covid 19. But there's obviously a lot to learn and this study is probably looking at a number of possible scenarios with many being quite unlikely but possible. 

 HansStuttgart 28 Mar 2020
In reply to wintertree:

> My intuition is that you want to balkanise the population into lots of islands with minimal bridges between them - the same household member going to the same shop every time, the same delivery people doing the same rounds to the same houses, the same food suppliers going to the same shops etc.   This creates islands with minimal chance of infection spreading between them and is a layer of structure on top of the current emergency measures.  You can also then try and improve the training of the key "bridge" people in terms of protocols to avoid passing on infection [1]. This is how you keep a hot object from heating a cool object - minimise bridging structures.  The thermal analogy is by no means identical to a pandemic, but it's not a bad starting point.   I'm hoping make some basic models and also that the ICL one lands on GitHub soon so I can dive in to making a reference version.  

This is close to the Japanese approach. The main goal there was to prevent the spread from cluster to cluster.

The metro system is the big liability, in my opinion. As long as that is running, the amount of people in the parks becomes secondary.

See https://twitter.com/HironoriFunabi1/status/1241734715175993345 and https://twitter.com/HironoriFunabi1/status/1241734736336338944

 wintertree 28 Mar 2020
In reply to Richard J:

Thanks for the link.  Interesting reading and comforting to know there are potentially more ways out of this mess into something more functional than our current lockdown.

 wintertree 28 Mar 2020
In reply to HansStuttgart:

Thanks

 Wiley Coyote2 28 Mar 2020
In reply to TobyA:

No idea if this is tosh or not but either way an interesting contribution from a recently-retired pathologist and academic. The idea that UK mortality rate seems high because we are only testing worst cases makes sense but if it's not as lethal as figures suggest why are we seeing the horror shows in Italian hospitals?

https://www.spectator.co.uk/article/The-evidence-on-Covid-19-is-not-as-clea...

 girlymonkey 28 Mar 2020
In reply to Wiley Coyote2:

But the mortality rate isn't the only factor. It's the number of serious cases who need intensive hospital treatment and live. There needs to be hospital capacity for these people. If there isn't, the death rate will rocket and people who wouldn't have been about to die anyway will die. Which is why we need to slow it down!

 wintertree 28 Mar 2020
In reply to Wiley Coyote2:

He lost my respect by paragraph 3

> The simplest way to judge whether we have an exceptionally lethal disease is to look at the death rates. Are more people dying than we would expect to die anyway in a given week or month? Statistically, we would expect about 51,000 to die in Britain this month. At the time of writing, 422 deaths are linked to Covid-19 — so 0.8 per cent of that expected total. On a global basis, we’d expect 14 million to die over the first three months of the year. The world’s 18,944 coronavirus deaths represent 0.14 per cent of that total. These figures might shoot up but they are, right now, lower than other infectious diseases that we live with (such as flu). Not figures that would, in and of themselves, cause drastic global reactions.

The data clearly shows death rates from Coronavirus are not "steady state" unlike his 51,000 per month number, but are rising rapidly - indeed near exponentially in many Western European countries.  We haven't yet see the virus burn through the exponential phase in any country other than some in East Asia with far more "proactive test/trace" and "responsive lockdown" approaches than us, so the "early exponential phase" numbers he compares are sadly already well out of date and are likely to continue closing the gap to 51,000 rapidly.

His core thesis might be right, but "softening the reader up to it" by playing silly buggers with numbers when he should know better is highly disingenuous.

Speaking of which:

> At the time of writing, the UK’s 422 deaths and 8,077 known cases give an apparent death rate of 5 per cent.

This is another case of playing fast and loose; at the time almost all of this 8077 known cases were "active" and not "recovered or dead" so the death rate is a minimum bound for those 8077 as some of them just hadn't died yet.  We know there's a multi-day lag between detected infection and average death so comparing instantaneous "infected" and "dead" in an exponentially growth phase is highly disingenuous.  You need to analyse the numbers over all cases with a final outcome; the UK does not have enough data to do that meaningfully yet.

He further neglects that flu season is spread out like wildfire but this infection is burning outwards like wildfire with massive consequences for the safety of all the medical workers and potentially limiting our ability to give live-saving care to the victims.  Ask yourself - does Northern Italy now look anything like flu season?

None of this is to say that his core hypothesis - we are massively under detecting - is wrong; and the CMO has openly acknowledged this as a likely possibility.    If he's right, about 0.8% of the population is infected.  That means ~ 6 doublings of case numbers to go until we reach 50% infection where herd immunity starts to kick in.   I'd estimate order of 100,000 deaths in that case, and if lockdown hadn't happened, we'd be there is ~ 18 days.  

But I find the article offensive in its deliberate misuse of numbers to support a theory - valid or not.  

Post edited at 16:50
 wintertree 28 Mar 2020
In reply to Thread:

A chance for several people on this thread to contribute - https://epcced.github.io/ramp/

 Richard J 28 Mar 2020
In reply to Doug:

There's a really interesting article in this week's Economist (free with registration) on "the Coronavirus panopticon" - different countries' approaches to using mobile phone data for documenting the pandemic, inputting data for modelling, and for contact tracing - https://www.economist.com/briefing/2020/03/26/countries-are-using-apps-and-...

On the use of covert surveillance data:

 "The use of data becomes most fraught when it moves beyond modelling and informing policy to the direct tracking of individuals in order to see from whom they got the disease. Such contact-tracing can be an important public-health tool. It also has a resemblance to modern counter-terrorism tactics. “The technology to track and trace already exists and is being used by governments all around the world,” says Mike Bracken, a partner at Public Digital, a consultancy, and former boss of the British government’s digital services. To what extent those capabilities are now part of the fight against covid-19, no one will say."

On Google:

"Google says that, having heard epidemiologists make such points, it is not planning to use the location data it collects to do contact tracing. The data-collection mechanisms built into products like Android or Google Maps are “not designed to provide robust or high-confidence records for medical purposes and the data cannot be adapted to this goal”, the company says. Facebook says something similar. Both companies can be assumed to think that talking explicitly about how well they might be able to do such things would raise concerns about privacy."

 Offwidth 29 Mar 2020
In reply to wintertree:

Agreed with all of that. I felt the Oxford modelling draft paper got a bit of an unfair panning (due to sloppy 'conclusions' that peer reviewing would have removed ). This article from Prof Lee is amongst the most insensitive I have ever seen in STEM, in terms of completely understating the implications of overwhemed health systems. He says "The scenes from the Italian hospitals have been shocking, and make for grim television. But television is not science." this is so crass that I think he needs to apologise to those Italian health staff. The scientific evidence is even more grim than the news coverage.  It's a shame, as I think his points on how we assign cause of deaths are important.

Post edited at 11:26

New Topic
This topic has been archived, and won't accept reply postings.
Loading Notifications...