Covid-19 and purity spirals

But yes it's difficult to determine who would have progressed to needing full ventilation.

I think giving oxygen at an earlier stage is better than ventilation. Many more people with more basic breathing support survive, according to the critical care audit report on covid-19. Vulnerable and older people included. The 111 line is apparently in many cases telling people with already severe symptoms to stay at home until they are becoming unable to speak/breathe etc.

There seems to be increasing evidence (maybe evidence is too strong, discussion perhaps?) that full ventilation isn't all that successful for Covid-19

That’s funny because I thought it was the opposite. That if you were sick enough to be ventilated then doing it early was better. The issue being that if you’re being fully ventilated then you’re very sick as a starting point. I’m open to new information though!

Oh agreed, that was very much an example of the challenge in that for this model you don't even have a definitime time zero value.

The risk factor breakdown is very interesting and makes it hard to compare between countries (and therefore to extrapolate their info to our situation)

The number of asymptomatic people is a huge factor IMO. It potentially means that the initial r0 was way off as well as meaning current assumptions (which were in turn extrapolated based on the previous estimate and the reduction in contact) are off. Many more may have been infected by each infected person than was every caught by testing. But that then also means that the intervention rates and the survival rates are potentially much lower/higher than originally thought.

There seems to be increasing evidence (maybe evidence is too strong, discussion perhaps?) that full ventilation isn't all that successful for Covid-19

The geekery is both interesting and welcome! Also the fact that the stats provided by the Chinese government are likely to be not so much unreliable as deliberately falsified isn't helping.

We don't know how many infected people there are...so trying to model how it evolves is very challenging.

I don’t think any of the factors that could be plugged into a model are known or reliably constant. rO will vary a lot between individual people based on community contact but also probably based on viral load. Intuitively you’d also say that symptomatic people are more likely to pass it on (more respiratory symptoms = more droplets) bit that may not be true either. I’m really interested in delving into the individual risk factors and separating out the confounders. Ie is it being male in itself that’s risky, or just the male cardiovascular risk factors or that overweight men are more likely to be centrally obese, which makes ventilation even harder. Apparently with COVID ventilation is much harder, and much higher pressures are used at baseline.

Those were the days! Compared to this it seems almost fun (I know it was horrid at the time btw, I do remember)

Love a bit of geek.

And yes, indyref

Sorry, I just realise I typed a whole load of geek there...

Expert judgement - which is basically intuition and experience - is a huge component of these sort of complex models. It has to be as the current data can only tell you so much.

Even where the current data is solidly factual - e.g. a current market price of an actively trading stock - then it only gives you data for there and then. You've then got to model the evolution which requires a whole series of assumptions. Most of the time they will be based upon data; some will be based on a historical time series for example. But there's always an overlay of judgement to this - do we use the full data set, as some of it corrupot, how should we weight it - but then also the consideration of whether the past is a good representation of the future. Then you've got to consider correlations - how does X change when Y changes, is the relationship constant and so on.

When we're talking about modelling C19 the starting numbers aren't even solidly factual. Look at the range of possible estimated for death stats alone - ranging from well under 1% up to about 13% as overall average case fatality rates. We don't know how many infected people there are...so trying to model how it evolves is very challenging.

It's easy to snipe at models - and I was guilty of it myself earlier - but the complexity is not easy and the level of uncertainty when you're dealing with such a new virus makes it challenging. They'll be using stochastic models to understand a range of possible outcomes but I'd expect the ranges to be pretty wide.

I'm not sure I really understand this assumption that we will be able to do something.

I do think there are actions governments and people can take that will help and we should try to manage it without avoidable loss of life. But it could easily be the case that what we can do is limited and will never be all that successful. It's a natural disease and like many natural diseases it will run its course and infect and kill people.

There are many many diseases like this that kill people every year, and there is nothing we can do. Having made progress with many does not mean we will always make the progress we would like in the future. When we fail, we don't shut down life indefinitely, we learn to live with it. We could quarantine society forever and I imagine many scourges would go down in frequency, from flu to lice to chlamydia. But there is no way people are going to stop going about the business of life in the long term to achieve that. People as a whole accept infestation, infection, and death every day. In many ways I think they find it easier than accepting society shutting itself up.

Sorry, intuition and experience!

fivethirtyeight.com/features/best-case-and-worst-case-coronavirus-forecasts-are-very-far-apart/

This article was interesting statistically, pointed out that intuition and intuition was relied heavily on for modelling predictions.

Indyrefer Dances?

I'll admit I find the modelling problem that this poses quite interesting (I do similar work but in a different field)

Liking your numbers Statistically
Always loved your analyses. Glad to see you here

Hooves, I actually believe getting the numbers lower to allow for contact tracing could be a feasible strategy but one question I have not yet seen you answer is: How do you want to get the number low enough for that to become a possibility while borders/non-essential businesses remain open? No policing of people's shopping habits is ever going to compensate for the mixing that's still happening at the airports, Amazon warehouses etc.

Every model is being slated from what I've seen

Actually, I might be misinterpreting but I think the figure is total first wave deaths - including those who have died already. So a further roughly 27k.

No you’re right, I misread. It’s an “as of through to” number, not a “from” number.

A further 27 thousand doesn’t seem that extreme then.

The IHME were saying 66,000 at the beginning of last week and have almost halved it since. Interestingly, they're also saying that Italy will have 20,000 in total in this first wave. The numbers in Italy are dropping but they're currently at 19,650 with many people still in hospital so that seems a little off as well.

That modelling is being slated though so wouldn't hold much store by it

Actually, I might be misinterpreting but I think the figure is total first wave deaths - including those who have died already. So a further roughly 27k.

These numbers aren't very believable though; Italy is close to that number already and whilst they are declining it seems unlikely they'll hit that target. I do note however that it's specifically called first wave deaths - I don't know when they might be predicting wave 2 for Italy.

I think it tries to incorporate lockdown (or not) population density, hospital and intensive care/ventilator capacity.

It would be interesting to know as over that period they will HAVE to have built in assumptions about lockdown.

mobile.twitter.com/natesilver538/status/1249373172690046977/photo/1

FWIW the IHME is predicting another 37 000 deaths in the U.K. in the next (just under) 4 months. Which seems high, wonder what their modelling is based on.

Just clocked something in the BBC description of the app

"At present, the idea is that people who have self-diagnosed as having coronavirus will be able to declare their status in the app.

The software will then send a "yellow alert" to any other users who they have recently been close to for an extended period of time.

If a medical test confirms that the original user is indeed infected, then a "red alert" will be sent instead, signalling that the other users should go into quarantine."

So you won't be told to go to quarantine unless it's confirmed, but will be notified if it's suspected. I could see that changing if the testing volumes aren't ramped up drastically