CORONAVIRUS 2020: RHYTHM OF THE DEATH THE SECOND

[Abnormal Palm]

Zetetic

Cheers. I totally get what you're saying.

Not to argue against that, and perhaps this is from a more personal perspective, I do think it's necessary to take some broad strokes decisions, and then deal with/learn from the consequences/mistakes on a more local scale, which then inform the next broad strokes decisions. If you don't do that, and this is not meant to be at all pointed, you are stuck with speculation and identifying only the problems, while potentially disregarding the more significant benefits of the actions taken.

This might just come down to a difference in approach. My focus is always to learn as much as possible as quickly as possible, and to have my initial conclusions challenged at different levels, but I'm also very conscious that a decision must be made - and at the most effective moment. I will assume that those who have devised these models are very aware of their limitations but without attempting to put them into practice and then listening and learning quickly about the consequences, there's little point in creating the model. It's a facile point but almost everywhere in the world, we're still at the 'suck it and see' phase.

One difficulty with this situation is that it seems that we can't even learn a great deal from those who have been through this a couple of months earlier than us because as a timeframe to gather data, there's still not enough to draw any significant conclusions and as you point out, it's pretty hard to draw accurate parallels between countries for a variety of reasons. All we can do with much confidence is look back and say that track and trace and earlier lockdown would have bought us some more time and NHS capacity to decide what to do next. I think much of what we're going through would have still been in place and many of the problems would be the same, an unavoidable consequence of lockdown in a global pandemic.

Obviously, there have been plenty of individual smaller scale fuck ups regarding PPE and missed email lies and reagents and testing sites and so on that we will need to hammer them on. I may be too generous about the potential errors of the broader lockdown strategy, and the on/off measures, but I think it's a starting point, basically.

[greencalx]

I don't think so, unfortunately - it's meant to be taken seriously as a UK-wide model, and part of the point is that these models aren't scale-free.

I'm not sure which model you're talking about, but I'm guessing it's something like an SIR model with the assumption of a well-mixed population. If you're looking at a large enough population, I think well-mixing is actually a pretty reasonable approximation. Human social contact networks are very efficient at mixing (in this case, infections) across the entire population, thanks to the "six degrees of separation" business. Preventing people from travelling should in principle make the contact network more spatial, at which point real distances between people start to matter. However, it's well known for example from population genetics that you only need a very low level of contact between different geographical regions to homogenise a population. If we're wanting to look at random effects, I'd start by looking at the variation in the number of contacts that people have while infected.

But I'm not an epidemic modeller, but, I suspect, neither are you.

[imitationleather]

NEW THREAD:
https://www.cookdandbombd.co.uk/forums/index.php?topic=79369

NEW THREAD:
https://www.cookdandbombd.co.uk/forums/index.php?topic=79369

NEW THREAD:
https://www.cookdandbombd.co.uk/forums/index.php?topic=79369

[Cerys]

Oh - is there a new thread?

[Zetetic]

However, it's well known for example from population genetics that you only need a very low level of contact between different geographical regions to homogenise a population.

Have London and South West England[nb]The English NHS regions.[/nb] shown similar rates of outbreak?

On the 9th March, London had 61 confirmed cases. Two weeks later they had 2,400. Four weeks later they had 12,000.
On the 15th March, South West England had 61 confirmed cases. Two weeks later, they had 740. Four weeks later they had 3,000.

I note:
- These are both actually quite big regions, with a fair amount of interconnection pre-lockdown.
- Population of London is quite a bit bigger.[nb]About 9 million vs 5.5 million, I think[/nb]. Population density and multiple outbreaks is probably part of the point.[nb]Noting your point about "number of contacts that people have while infected". There's stuff from avian flu work about the limited number of (human) contacts that farmers seem to have a lot of the time...[/nb]
- Lockdown started on the 24th March. Perhaps that's part of it. Perhaps some of the noises before the 24th are part of it.
- Testing is also probably part of this, and I don't want to overstate the precision of those numbers in representing the underlying outbreak. But I don't think they're meaningless either.

So far South West England is doing relatively okay - from what I hear - hospital and critical care demand-wise, and that's despite having a hideously vulnerable population and starting with one of the worst capacity situations. (I'd be really interested to know if this is nonsense.)

(One thing that might actually help Whitehall make more careful decisions is that nowhere is worse off than London in many ways. But that might not last.)

But I'm not an epidemic modeller, but, I suspect, neither are you.

Absolutely not. But I do have some experience with healthcare capacity and demand, and getting to grips with range of predictable variation that you need to plan services on, and how that relates to population sizes and incidence rates.

Drawing some exceptionally pretty charts showing variation within the UK means that as of last week I'm in the faintly ridiculous situation of being a) invited to our national COVID-19 epidemiological modelling group[nb]Although I bet there's more than one that think it's the national group.[/nb] and b) being asked by people far wiser than me[nb]With doctorates that involve epidemiological and healthcare modelling, if that helps.[/nb] to nudge that group into ... better considering local variation so that bits of the NHS actually get something they can make decisions on[nb]Noting the echo of TBC's post and BlodwynPig's article. Of course, in the meantime, people with even less epidemiological understanding than are cobbling together models so that they have something to take to (virtual) meetings.[/nb]. (My plan is to get myself replaced with a colleague who is both more knowledgeable and better-tempered than myself.)

This is just in the tiny corner of the UK that I'm professionally attached to, where it's so much easier for people to just talk to one another - at any level - than in England (in my experience). How do we think it's working at levels above the devolved nations?

I don't think the problem is that epidemiologists are thick, but I do think that their interests are naturally narrow and focused at a different level to regional healthcare capacity. There's only so much time in the day, and they have to decide what to spend it on.

I will assume that those who have devised these models are very aware of their limitations

Having said the above, in some cases I think you might be appalled at the level of ignorance of or simply lack of interest in the shortcomings of their assumptions. Perhaps I'm overgeneralising from the worst of my experiences of people who come to play in healthcare - perhaps you've had more positive ones on the whole.

but without attempting to put them into practice and then listening and learning quickly about the consequences, there's little point in creating the model.

I entirely agree with this. Given the lack of regional power, even really of the devolved nations' governments[nb]We've had police forces that didn't understand that our laws for COVID-19 FPNs were different to the English ones...[/nb], this is going to rely on excellent communication to-and-from the centre.

My concern is that while Ferguson's work (etc.) has helped provoke some positive action, it may also support reversing it inappropriately. If the English regions and devolved nations are well-represented in Whitehall, then that risk is probably averted.[nb]Well.[/nb]

[Abnormal Palm]

Great post, plenty to chew on and very best of luck with it! I'm sure that's going to be a hell of an eye-opener.

[greencalx]

I think you (Z) might be asking different questions to the ones that the models can answer.

I thought there was an objection to models based on small populations being scaled up to large populations. Actually, that's the easy part, because you pretty much know the infection is going to spread everywhere, and you can happily average over that. What you don't know is where it's going to hit first. Also differences in things like transmission rates are going to be less relevant at this scale.

But if you really do want to make day-by-day predictions on a region-by-region basis, then you are going to need a lot of data about how people interact with each other in each of those regions. I think we can all speculate as to why London would see a bigger rate of increase than the SW. I suspect that by the time you've amassed that data, it's probably too late to be useful to you, alas.

My impression is that the Imperial lot (and therefore the Govt) have mostly been looking at national-scale modelling, and to the extent that I have knowledge of it (i.e., not a lot), I don't see a huge amount to complain about because details tend not to matter at larger scales.

[Zetetic]

I thought there was an objection to models based on small populations being scaled up to large populations.

No, the opposite.

My impression is that the Imperial lot (and therefore the Govt) have mostly been looking at [UK]-scale modelling

The problem is that critical care capacity doesn't exist as a unit at that level. A spare staffed bed in Bangor doesn't help you if you live in Glasgow or Hackney.

[Zetetic]

But if you really do want to make day-by-day predictions on a region-by-region basis, then you are going to need a lot of data about how people interact with each other in each of those regions. ...  I suspect that by the time you've amassed that data, it's probably too late to be useful to you, alas.

Mmm.  But at what level do you need to understand hospital bed and critical care demand? (If you're a manager in an hospital, or if you're in Whitehall deciding whether to flick the restrictions switch?)

Vaguely relevant: I've heard there's a lot of interest in Google's community mobility. (I had a quick look at my town, tried to work out how to get relevant data out of the PDF without losing my mind and forgot about it.)

[BlodwynPig]

Loving this thread now. Perhaps a "data/model/technical" side thread?

[greencalx]

Ah sorry I misunderstood.

I understand your desire to match capacity to demand, but it's going to be really difficult. If it were up to me (and thank goodness it isn't), I'd probably try and measure the early-time growth rate in a given region, as then you don't have to quantify all the local behavioural factors that will contribute to this. I'd then make the rash assumption that the recovery-rate is a function of the disease, rather than behavioural aspects, which means you can then look at curves from other outbreaks to project forwards. This is essentially how they used to do weather forecasts. It's a bit shit, but probably all that can be reasonably done on the available 2-3 week timescale.

[greencalx]

Blodders - maybe, though I've probably exhausted my knowledge of spreading processes across networks...

[BlodwynPig]

Blodders - maybe, though I've probably exhausted my knowledge of spreading processes across networks...

It all sounded about right, averaging, scaling - same with population based modelling.

Will read tomorrow properly if you could create a thread as im on phone in bed

[Zetetic]

I understand your desire to match capacity to demand, but it's going to be really difficult.

Or demand to capacity, as the suggestion is.

My point is that it's "really difficult", and the idea that we can toggle restrictions on-and-off to control excess demand at the level where capacity is found is being oversold.

[Zetetic]

I'd probably try and measure the early-time growth rate in a given region

Yes.

[greencalx]

Is that an instruction?

[Zetetic]

Have edited, with an example.

(Not the only chart you might use.)

[Zetetic]

Or to try to show change in rate of growth. (Which probably shows impact of low-volume of testing in large part, unfortunately, but hopefully not entirely and reflects on hospitalisation rates at least.)

Both cribbed off the Burn-Murdoch/FT styles, of course.

[BlodwynPig]

Or demand to capacity, as the suggestion is.

My point is that it's "really difficult", and the idea that we can toggle restrictions on-and-off to control excess demand at the level where capacity is found is being oversold.

The thing with model predictive control is that your model and subsequent control actions can be adapted as new data becomes available, you can also apply constraints - its not simple on/off control or other low-level control.

[greencalx]

Apart from London, they all look by eye to be peaking. (Apparently dividing new cases by total cases helps you see more easily when the exp phase is ending). Problem solved in terms of number of beds required? Answer = the number you've got now.

My ignorance of disease mechanics is causing me to puzzle the following. Given the lack of immunity in the population and the fact this seems to spread quite easily, you'd expect the number of cases to keep growing until pretty much everyone is infected. Social distancing can only slow things down. It can't change the carrying capacity, unless the virus goes extinct which hasn't happened. So why is there a peak so soon?

[Zetetic]

Thinking testing is real issue - we might just be seeing regional health systems reaching their limits for hospitalisations-per-week (as you sort of suggest).

We are still seeing hundreds or thousands of new cases per week on each of those regions though. R0 ≈ 1? That's probably very optimistic...

[Dr Rock]

How are soaps and that going to deal with Coronavirus? If the country ends up being in lockdown for months how come everyone in Albert Square are immune? I fear it may undermine the show's realism.

Same thing with Holby City or whatever.

[greencalx]

I'd be amazed if R0 was anywhere close to 1... and it'll shoot back up again if we're let out of our houses.

Are you saying the levelling off is an artefact of hitting capacity, rather than a real levelling off? I have to say I find it difficult to interpret number of cases statistics because unless you're testing people randomly I don't see how it gives you any information about people you haven't tested.

[Zetetic]

The thing with model predictive control is that your model and subsequent control actions can be adapted as new data becomes available, you can also apply constraints - its not simple on/off control or other low-level control.

What constraints are you thinking of in this case? Noting you've got to convince other people to enact them every day, by law or otherwise.

Think it's worth emphasising that you've got to know two-to-four weeks in advance - at a minimum - of cases actually hitting hospitals that you've reached a point that need to close the tap a bit.

[Zetetic]

Are you saying the levelling off is an artefact of hitting capacity, rather than a real levelling off?

I really don't know, but I do wonder.

There isn't public data for deaths at the same level - but, for nations, in the same style: 1 and 2. These are the 'daily reported' mostly in-hospital post-positive-test figures.[nb]So Scotland, for example, might be doing better at keeping people at home to die or might be seeing a real levelling off - haven't looked into other sources for them. Edit: And I vaguely remember some odd discrepancies between what PHE put on their dashboard and figures that Scotland announces which I never tried to understand properly.[/nb]

I'd be amazed if R0 was anywhere close to 1... and it'll shoot back up again if we're let out of our houses.

Yep. Better get some testing and contact-tracing going at scale, I guess.