Urban75 Home About Offline BrixtonBuzz Contact

The nerdy amounts of pandemic detail thread

So does this imply that >35s are more likely to be doing social distancing properly (and it kind of works, even if you have quite a lot of contacts) whereas <35s tend not to be doing it properly, with unsurprising results?

I would want to examine that stuff more closely, eg the age of the social contacts they were having, since that could compound any issues in the younger groups. And the setting in which those contacts were made.

In other words, its not just whether people are 'doing it properly', its also about thing like the chances that the people they are having social contact with were infected at that stage. And so as things spread through the age groups, older groups may start to show levels more like those first seen in the young.

Plus those graphs dont give us any idea of how many people in the 35+ group actually have 21 or more contacts. In theory based on existing contact mixing pattern data, there will be less 35+ people in that category in the first place, so the problem isnt compounded in the same way it is for the young.

Plus there is a lot of modelling involved and assumptions in the modelling may influence this analysis in somewhat unfair ways.

And then there are the other factors such as the accommodation situation and job types many young people have, potentially increasing the base risk before they even start mixing.
 
I really wish that they would - for the purposes of covid stats - split "Northumberland" up. It is too diverse an area to be treated as one zone, and most of the hospital facilities are concentrated over by the urban area.
On the one hand you've got the old coal mining areas and on the other the rural hinterland of Kielder ...
(I know this map does what I mean, but that doesn't show the past changes in a particularly accessible way ... ArcGIS Web Application)
 
Too late, the seeding was done a while ago and by now its reasonable to assume infections are more generally spread.
tbh, I avoid members of that "group / category" anyway ! So, these stats sort of confirm my reasoning ...
 
One consequence of London being put on the local authority pandemic watchlist is that data from the different areas is presented in a weekly document.

An example from this weeks report (documents on the following site with titles like 'Contain framework lower tier local authority watchlist - maps by LSOA):


Screenshot 2020-10-02 at 19.54.16.png
Screenshot 2020-10-02 at 19.54.26.png
Screenshot 2020-10-02 at 19.54.48.png
 
Good article here about why focusing on clusters is so important. Basically concluding that the UK is not doing any of the right things (backwards contact tracing, rapid testing, targeting indoor poorly ventilated gatherings).

Yeah, I thought that article was really interesting. It's been clear for a while that indoor poorly ventilated gatherings are the main spreading point. The UK government is negligently and knowingly refusing to do the right things: pubs should not be open except for beer gardens (with headcount limits), university lecture theatres should not be in use, offices should be banned from being open unless they can show a public interest need to be. etc etc. It's no wonder the virus keeps spreading.
 
I've pointed this out on the main thread a couple of times but maybe it's more suited to this one. I'm still not satisfied with the reasons for the difference in estimates from official testing numbers and the Zoe project estimates.

I was initially comparing Lambeth and Highland. Thought I'd compare a few other regions too. I've made Lambeth the "index" for comparison cases. I'm not interested so much in the absolute numbers, just relative rates of infection. In other words, is somewhere else twice as bad as Lambeth, or half as bad, or whatever.


regiongov.uk "rolling rate"compared to index (lambeth)zoe est. "active cases per m"compared to index (lambeth)
Lambeth154.3190471
Highland15.70.168940.76
Stoke319.12.06134911.49
Rutland87.70.5742350.47
Edinburgh (city)90.30.59112101.24
Cornwall54.00.3538330.42

Of the three English regions there, the gov.uk and Zoe rates don't match exactly, but I wouldn't expect them to. They show the sort of variation I might expect when one's measuring test results and the other is an estimate based on self reported symptoms. And no doubt they are subject to different time lags. They very broadly agree with each other.

But Highlands still comes up with wildly different numbers (different by a factor of 7), and while Edinburgh is only out by a factor of 2, it's either doing a fair bit better than Lambeth or a fair bit worse.

So is it something to do with the way the Scottish data is processed compared to the english data? How do explain that factor of 7 difference? Are 7 times as many people being tested in Lambeth? Seems unlikely.
 
I dont know as anyone is going to help you with that. I'm afraid I certainly cant, because I treat identified cases as a vague guide and I pay little attention to them when we are in periods where the serious implications of the pandemic show up in various bits of hospital data. If I were trying to compare those locations the first thing I'd try to do is use hospital data to build a more comprehensive picture, but then I'd also try to stick to England rather than comparing a location in England with one in Scotland, to rule out data, definition and reporting differences.
 
The following doesnt include Scotland and pertains to critical care admissions rather than hospital admissions as a whole. There is quite a lot of interesting data in there and much of it is separated into before and after September 1st. The following are for the critical care patients from September 1st onwards, ie those that will be seen as being part of the 2nd wave here.


Screenshot 2020-11-02 at 19.50.46.png
Screenshot 2020-11-02 at 19.52.59.png
Screenshot 2020-11-02 at 19.52.42.png
Screenshot 2020-11-02 at 19.51.20.png
 
I dont know exactly but I know there is a long list of caveats for that data, and also changes to the system over time. I havent paid much attention to that data throughout, although at least it did manage to send an appropriate signal in September. As for why it hasnt captured the picture much since then, perhaps its a regional issue, eg this from the caveats section of the page you linked to:

The North East, West Midlands, South East Coast, South Central, and Isle of Wight Ambulance Services use NHS Pathways to triage calls to 999. The North West, Yorkshire, East Midlands, East of England, London, and South Western Ambulance Services use another system to triage calls to 999. Therefore, for CCGs in those areas, data here will not include most 999 calls related to COVID-19.
 
I stuck this stuff in the main thread but I'm putting it here too for further discussion if required, and so it isnt lost in the fast moving main UK thread.

September modelling focussing on half-term circuit breakers, where the dashed lines are the no circuit breaker scenario and the others use various different growth rate values in the scenario of a circuit breaker: https://assets.publishing.service.g...tial__lockdown_for_2_weeks_over_Half-Term.pdf

Screenshot 2021-01-11 at 12.39.54.png

Other scenarios were modelled, including two circuit breakers:

Screenshot 2021-01-11 at 12.41.46.png

View attachment 248325

October modelling. I'm only showing the England graphs because otherwise my post will be way too long (well it already is too long but never mind).

From https://assets.publishing.service.g.../950631/S0821_SPI-M-O_Consensus_Statement.pdf

Medium term projections and R=0.6 scenario [.....] Orange shows the trajectory based on current trends and does not include the effect of future policy changes or past ones that have not yet been reflected in data. Blue shows a scenario in which a very stringent intervention is introduced on 26th October and maintained for the duration of the scenario. Both trajectories show interquartile ranges of model combinations. The dashed line reflects the current reasonable worst case scenario.

Screenshot 2021-01-11 at 12.33.21.png

Screenshot 2021-01-11 at 12.33.08.png

Late October evolution of the above modelling: https://assets.publishing.service.g...SAGE64_201028_SPI-M-O_Consensus_Statement.pdf

Screenshot 2021-01-11 at 12.47.58.png

Screenshot 2021-01-11 at 12.48.13.png

Early November version, where length of measures is 4 weeks instead of 6, and where a number of different R values are modelled.


Screenshot 2021-01-11 at 12.51.55.png
 
And another from the main UK thread:

One particular angle on how we got here.

Via SAGE modelling group consensus statements:

September 9th: https://assets.publishing.service.g...SAGE56_200909_SPI-M-O_Consensus_Statement.pdf

The current situation is in line with the latest reasonable worst-case scenario (RWCS), where incidence doubled once in August and once in the first two weeks of September, before re-imposed measures halt this growth. Under this scenario, there is an average incidence of approximately 11,900 infections per day for the second week of September in England. SPI-M-O’s estimated incidence range for England at present is 2,300 to 12,500 new infections per day, in line with estimates from ONS and REACT surveillance studies.

September 23rd: https://assets.publishing.service.g...SAGE59_200923_SPI-M-O_Consensus_Statement.pdf

The epidemic is close to breaching the agreed Reasonable Worst Case Scenario on which NHS, DHSC and HMG contingency plans are based. As outlined by COVID-S, planning has followed a strategy under which action is taken in mid-September to halt epidemic growth. Unless the measures announced on 22nd September reduce R back below 1, it is likely that infection incidence and hospital admissions will exceed the planning levels.

Long-term management of the epidemic will be a balancing act between direct and indirect effects on health caused by COVID-19 and the economic and health disbenefits caused by intervention measures. There is great potential to use of data and modelling to inform the choice of policy measures that would meet the Government’s long-term strategic objectives. Such calculations would be complicated by uncertainty around if or when a highly effective vaccine or treatment will become available, but those difficulties are not insurmountable. SPI-M-O welcomes the clarity brought by stating the top strategic objectives are to protect the NHS and keep schools open. Further detail on specific objectives would allow advice to be issued on how policy objectives could be expected to achieve them.

October 21st: https://assets.publishing.service.g.../950631/S0821_SPI-M-O_Consensus_Statement.pdf

The number of infections, hospital admissions and deaths are exceeding those in the reasonable worst case planning scenario that is based on COVID-S’s winter planning strategy. This scenario assumed that decisive action would be taken in mid- September to halt an increase in transmission.

So the government totally failed to stick to the response the reasonable worst case planning for winter used. The modellers modelled various things at these stages, including various versions of a circuit breaker, and then once that opportunity was completely missed, measures in early November that continued for 6 weeks. And then finally measures in November for 4 weeks. The output of such modelling also enables me to see the curves they came up with both with and without such measures, and with a different range of R values. I will present these later.
 
No I wouldnt say that. The period of adjustment required to make the daily reported averages align well with what the actual number of deaths per day eventually settles at isnt very large. Especially not compared to the lengths of time we are talking about when describing the complex shape of this second wave.

If anything has made the link between the two numbers weaker in recent weeks, its been because of the Christmas and New Year holiday period. There were more reporting days that ended up equivalent to what we see at weekends, and also some of the four nations had even long gaps where they reported no deaths at all over non-working holiday periods. So the rolling average of deaths by date of reporting couldnt manage to be smoothed out as well as usual, and probably involves some trajectories that were too flat and then had to get steeper than normal for a little bit to compensate. ie normally a 7 day average is enough to smooth out weekend underreporting, but it was insufficient to deal with longer periods of underreporting. Deaths by date of death would have had the same issue of missing data for longer over Christmas, but with the advantage that when this data finally arrived, it was assigned to the right dates so now looks right when I graph it, as opposed to the blue line which has had to overcompensate later.

I do not routinely do my own comparisons of these two forms of data but since it came up I had a quick go, just crudely overlaying one graph on top of another and then moving one left or right a bit to compensate for timing differences. I was surprised how well things fitted apart from the period which I attempted to decribe in the previous paragraph, where we can see the blue line ends up sagging below the red one for a time.

The red line is deaths by actual date of death, so it falls off a cliff at the end because most data for those dates hasnt been reported yet. I am not saying that for sure those numbers will reach the exact level the blue line reached, a little over 900, at that stage, especially because the blue line was very steep due to playing catchup after getting behind. But the final number of deaths on those days certainly will have headed in that direction, further than is currently shown. I suppose this means I can also say that due to holiday delayed data the point you made has more validity to it in regards the recent numbers that started this bit of conversation the other day, than is normally the case. I suppose now I've done this I will revisit this comparison again a few times to see how things developed.

View attachment 248423
elbows I thought that a reply to this would be better placed on this thread, than the main one.

I guess that my basic question is: what's the advantage of using the by-date-of-report numbers instead of the by-date-of-death ones - when both are available? Does the former allow us to make a better guess at the current situation, and the direction of travel in the coming days?

Taking for example that information that is available today:
The 7-day average, by date of report, stands at 926. That's an average for reported deaths between 5 jan and 11 jan.
According to the gov.uk site, the latest day for which deaths-by-date-of-death is complete-ish, is the 6th Jan, and that stands at 787.

So if we wanted to try and project forwards, to guess what the "real" number is for today, let's say we take a rate of increase for a 7-day period, the most recent period we can find using either method.

For the date-of-report numbers, we can compare the average for 5-11Jan (926) with the average for 29 Dec-4 Jan (617). That gives a rise of x1.50 or 50%.
For the date-of-death numbers we can compare the "actual" for 06 Jan (787) with the "actual" for 31 Dec (690) . That gives us a rise of x1.14 or 14%.

It's obvious that if we then tried to project those rates of increase forwards to the 12th Jan, we'd end up with very different numbers.

I may well have got some of my maths wrong, and I'm no statistician.

What I'm trying to understand is whether the rolling average of deaths by date reported somehow contains extra information, that makes it more useful.
 
I'm not an expert either, what I said about it last night was just stuff I deduced by observing the numbers I had for both sets of data. So I'm not going to test your maths or do any of my own.

I suppose I would say that the period you chose to do some maths on is far from ideal, because it contains the Christmas preorting periods that I suggested were responsible for a greater degree of divergence between the two sorts of numbers recently.

I guess I wouldnt say that averages by date reported contain more information, rather they present information differently, in a manner that may be more helpful to those looking to understand the recent trajectory. In other words, it doesnt suffer from what we've seen at times when I try to present deaths by date of death data, where people are tempted to misinterpret the droop and cliff fall that is always present at the end of that data due to reporting delays.

Both indicators are laggy, and the averaged out deaths by reporting death isnt really any quicker, there isnt less lag, its just that data is presented in a way that may direct peoples brains to a better understanding of how cases have risen in the last week, and that is the sort of thing people use to make assumptions about what may happen next. I doubt either form of data really provides any significant advantage when trying to 'see the peak coming', but there are times where it can be useful to base our impressions on of a mix of both, even if just to confirm what we think we're seeing and whether our minds are likely to be broadly accurately 'filling in the gaps' for recent periods where data is incomplete.

Really I suppose my main reason for getting involved in that conversation in the first place is more along the lines of 'looking at either will do, understanding the nature of whats being shown and the limitations of a particular version of the data is the key bit'. I'd also say that unlike the first wave, where there were some crucial periods where I didnt have useful cases or hospital data on a daily basis so deaths were my main guide to the state of play, this time there are other indicators I would use to see the peak emerging some days before a peak in deaths would be expected to show up in death data.
 
Taking for example that information that is available today:
The 7-day average, by date of report, stands at 926. That's an average for reported deaths between 5 jan and 11 jan.
According to the gov.uk site, the latest day for which deaths-by-date-of-death is complete-ish, is the 6th Jan, and that stands at 787.

I hope that you are noting that the UK figure by date of death for 6th Jan is now 832 and is currently 847 for the 7th.

Its not easy to make exact projections and the amount of data that lags behind and will be added later obviously grows when the deaths as a whole have grown.

The safe bet is to keep claims suitably vague at times like these. And not to bet that the difference between averages of reported deaths daily and actual deaths on those days will be as large as its tempting to believe it might. Which is not the same as saying they will be identical, its not like I have exact predictions for those dates.
 
I hope that you are noting that the UK figure by date of death for 6th Jan is now 832 and is currently 847 for the 7th.

Yes - so I see.

It would be interesting to see one of your multi-coloured graphs just now.

Edit - I see you just put one on the other thread.
 
Last edited:
Yes - so I see.

It would be interesting to see one of your multi-coloured graphs just now.

I've stuck one on the main thread. I dont think it really offers many clues. For example, despite picking at various details of your previous thoughts, I havent made claims about whether the peak this time will exceed the peak seen in the first wave. We are in the same realm but it could still end up falling some way short this time, or maybe it will still get there. I wont ultimately judge this and other first and second wave comparisons using 28-day limited figures alone anyway, I will do much of it with ONS data but I have to wait a long time to be able to do that properly.
 
I've stuck one on the main thread. I dont think it really offers many clues. For example, despite picking at various details of your previous thoughts, I havent made claims about whether the peak this time will exceed the peak seen in the first wave. We are in the same realm but it could still end up falling some way short this time, or maybe it will still get there. I wont ultimately judge this and other first and second wave comparisons using 28-day limited figures alone anyway, I will do much of it with ONS data but I have to wait a long time to be able to do that properly.
Sure.
I would make no prediction about whether the first peak will be exceeded or not... my point really was that it's not obvious which way it'll go at the moment, whereas anyone looking only at the rolling average for by-day-reported could easily be led to believe that it already has, or is almost certainly going to.
 
Sure.
I would make no prediction about whether the first peak will be exceeded or not... my point really was that it's not obvious which way it'll go at the moment, whereas anyone looking only at the rolling average for by-day-reported could easily be led to believe that it already has, or is almost certainly going to.

Of course it is, there's over 60% more covid cases in hospital compared to the spring peak, and you don't think that will result in more deaths? :facepalm:
 
Of course it is, there's over 60% more covid cases in hospital compared to the spring peak, and you don't think that will result in more deaths? :facepalm:

I don't have the exact figures but the risk of dying in hospital has decreased by more than a third thanks to improved treatment.
 
And since there is more testing, we might expect a greater proportion of people in the hospital figures to be those who were in hospital for other reasons, but brought it in with them or caught it in hospital.

The high levels of hospitalisation and the much longer wave are reasons to expect the number of deaths in this wave to be higher than the first wave, probably very significantly larger given we have already reached the same sorts of totals since September 1st that we saw before September 1st, and we've still got the very peak and the downward slope to get through yet, which is a period where a lot of the deaths normally happen.

But there is a difference between making claims about the totals, and being able to make claims about the ultimate height of the very peak. I cannot say whether the very peak level this time will be higher than it was the first time.
 
Although frankly I dont think it would be a huge gamble to claim that the peak level will exceed that seen in the first wave. Its just I prefer to see for myself, its not the sort of prediction I'm focussed on. Whatever happens with the peak values, the totals are a disgrace that will capture most of my attention.
 
I don't have the exact figures but the risk of dying in hospital has decreased by more than a third thanks to improved treatment.
And you can see this looking at the first "hump" of this second wave, which for cases in hospital was not very far off the first peak, but the same hump in the deaths number was less than half of the first peak.
 
As an exercise I temporarily suspended my normal degrees of caution, not wanting to go very far out on a limb, data hesitancy etc, and saw where my mind would end up without these restraints. A horrible peak level of death well beyond that seen in the first wave is what my mind settled on under those conditions. Now I have to wait and see quite how far off the mark I ended up without my usual restraints. It will take a while to find out, especially since I tend to focus on deaths by actual date of death rather than by reporting date.
 
Unpleasant detail:


Researchers looked at more than 4,000 patients who were admitted to intensive care units in 114 hospital trusts in England between April and June last year.

They found the risk of dying was almost a fifth higher in ICUs where more than 85% of beds were occupied, than in those running at between 45% and 85% capacity.

That meant a 60-year-old being treated in one of these units had the same risk of dying as a 70-year-old on a quieter ward.

The Royal College of Emergency Medicine sets 85% as the maximum safe level of bed occupancy.

However, the team found there was no tipping point after which deaths rose - instead, survival rates fell consistently as bed-occupancy increased.

This suggests "a lot of harm is occurring before you get to 85%".

Patients admitted to ICUs that were less than 45% full were 25% less likely to die than average.

Even pre-Covid, data suggests larger ICUs had lower death rates - with a 25% increase in bed numbers linked to a corresponding 25% fall in mortality.

And the findings are supported by a wealth of evidence from before the pandemic and from around the world.

I havent read the paper that prompted that news article yet. https://www.medrxiv.org/content/10.1101/2021.01.11.21249461v1.full.pdf
 
Hospital infections (nosocomial infections), one of the subjects I try to focus on, got covered by the December 16th document from SAGES pandemic modelling group, which was published on January 15th:


Screenshot 2021-01-17 at 14.58.18.png
 
Last edited:
Back
Top Bottom