Coronavirus - Modelling Aspects Only

[scotia]

Another week's data - with a bit of a surprise
A simple 2-parameter model is used where the Covid deaths, by registration date, are predicted as being 0.11 times the Covid hospital admissions of 10 days previous.
The latest Covid hospital admissions data is dated 17th October. The deaths by registration date will be subject to substantial updates (additions) for several days, so I have only included it up to the 10th October. All plotted points are 7-day summations to iron-out any weekly variations.



The surprise is the recent hesitation in the fall of admissions (and hence projected deaths). I had optimistically hoped for a continuing fall.
It looks like I should not have ignored the results from the sample-survey of the Office for National Statistics which had reported that for the week ending 3rd October, the infection rate had increased in England. I'll refrain from guessing what happens next

[scotia]

Another week's data .
A simple 2-parameter model is used where the Covid deaths, by registration date, are predicted as being 0.11 times the Covid hospital admissions of 10 days previous.
The latest Covid hospital admissions data is dated 24th October. The deaths by registration date will be subject to substantial updates (additions) for several days, so I have only included it up to the 17th October. All plotted points are 7-day summations to iron-out any weekly variations.



There is some good news. The "hesitation" in the fall of admissions (i.e. projected deaths) would appear to have passed, and the downward trend appears to have been re-asserted. Lets hope it continues. Next week should confirm whether or not the actual deaths have reached their peak as predicted.

[scotia]

Another week's data .
A simple 2-parameter model is used where the Covid deaths, by registration date, are predicted as being 0.11 times the Covid hospital admissions of 10 days previous.
The latest Covid hospital admissions data is dated 31st October. The deaths by registration date will be subject to substantial updates (additions) for several days, so I have only included it up to the 24th October. All plotted points are 7-day summations to iron-out any weekly variations.



Last week I hoped that the actual deaths would have reached their peak as predicted. It looks like they have. Hopefully next week we will definitely see a downward trend.

[scotia]

Another week's data .
A simple 2-parameter model is used where the Covid deaths, by registration date, are predicted as being 0.11 times the Covid hospital admissions of 10 days previous.
The latest Covid hospital admissions data is dated 7th November. The deaths by registration date will be subject to substantial updates (additions) for several days, so I have only included it up to the 31st October. All plotted points are 7-day summations to iron-out any weekly variations.



The good news is that, as expected, the actual deaths are definitely on a downward trend. The less good news is that the downward trend of the projected deaths is reducing, and we appear to be approaching a minimum. It also looks like the actual deaths are now sitting above the projections - rather than below as at the previous peak of four months earlier. Are we seeing a more virulent strain, or is it simply a feature of the onset of autumn weather, with the complication of other seasonal viral infections? Or?

[scotia]

Another week's data .
A simple 2-parameter model is used where the Covid deaths, by registration date, are predicted as being 0.11 times the Covid hospital admissions of 10 days previous.
The latest Covid hospital admissions data is dated 14th November. The deaths by registration date will be subject to substantial updates (additions) for several days, so I have only included it up to the 7th November. All plotted points are 7-day summations to iron-out any weekly variations.



Not much new to report - simply confirmation of the trends noticed last week. The registered deaths are continuing downwards, and the projected deaths (based on admissions) now appear to have levelled out. And, also as noted, the registered deaths curve is now sitting above the projected death curve - whereas in the previous peak it sat below the predictions.
We will soon be entering the party season - last year this coincided with a steep climb in admissions and deaths. Sorry to sound so pessimistic!
But it won't stop me from going out for a Christmas lunch with other (ancient) retired colleagues.

[scotia]

Another week's data .
A simple 2-parameter model is used where the Covid deaths, by registration date, are predicted as being 0.11 times the Covid hospital admissions of 10 days previous.
The latest Covid hospital admissions data is dated 21st November. The deaths by registration date will be subject to substantial updates (additions) for several days, so I have only included it up to the 14th of November. All plotted points are 7-day summations to iron-out any weekly variations.



Looking back at the summer peak, the projected deaths were lying above the actual deaths, while now the opposite is true. Its probably time to look at adjusting the model parameters to get a better fit to current data, hence I have increased the multiplier from 0.11 to 0.125 in the graph below



This is a better fit to the current data, but a significantly poorer fit to the summer peak. I have suggested that the reason for the apparent increased actual death rate is possibly seasonal - lots of other bugs going around, and more gatherings indoors.

[servodude]

I have suggested that the reason for the apparent increased actual death rate is possibly seasonal - lots of other bugs going around, and more gatherings indoors.

Sounds very plausible. The difference being that bit of the legacy seasonal bump who also had Covid.

As ever though when the underlying model parameters need a tweak it can also represent a change in the thing we're trying to isolate - doesn't sound like that would be indicating much to concern ourselves with at the moment

[9873210]

I have suggested that the reason for the apparent increased actual death rate is possibly seasonal - lots of other bugs going around, and more gatherings indoors.

Before I tried to explain an apparent increase in the death rate (meaning your multiplier) I would want to make sure there was an actual increase.

You discussed some changes in the data collection in July. It is plausible that formal and informal changes in the way the data was gathered is entirely responsible for the change in the multiplier. It should remain plausible until somebody does enough work to rule it out (which IMHO would be a major bit of work)

[scotia]

I have suggested that the reason for the apparent increased actual death rate is possibly seasonal - lots of other bugs going around, and more gatherings indoors.

Before I tried to explain an apparent increase in the death rate (meaning your multiplier) I would want to make sure there was an actual increase.

You discussed some changes in the data collection in July. It is plausible that formal and informal changes in the way the data was gathered is entirely responsible for the change in the multiplier. It should remain plausible until somebody does enough work to rule it out (which IMHO would be a major bit of work)

Yes - there were significant changes. These particularly affected the deaths at the date which notification arrived at the registrar - i.e. not the actual date of the death. I had to switch over to actual death dates - see
viewtopic.php?f=98&t=22737&start=1300#p517776
All of my plotted graphs since then have used the same dataset. But is it consistently the same data? Has it changed at all? It is difficult to know. For example, back in March I noticed the previously published admissions data for Scotland had changed - and I discovered that re-admissions
were being counted, and the data was back-corrected. It was stated that this had already been carried out for the English data - but I never discovered when.
I think the publication of Covid-related data is now a very low priority, and it is doubtful whether or not we will get any clear indication of any changes to its collection. In Scotland, the deaths data ceased in June, and the admissions data ceased in September.
In England the Covid data is published weekly on Thursday evenings - with the latest Admissions and Deaths numbers from the preceding Monday.
Now looking at my recent graphs, if the deaths or admissions data had undergone a step change in the manner in which it was collected and published, I would have hoped to have noticed a step - but rather it looks more like a gradual change - which, in my view, is more supportive of a real effect.

[9873210]

I think the publication of Covid-related data is now a very low priority, and it is doubtful whether or not we will get any clear indication of any changes to its collection. In Scotland, the deaths data ceased in June, and the admissions data ceased in September.
In England the Covid data is published weekly on Thursday evenings - with the latest Admissions and Deaths numbers from the preceding Monday.
Now looking at my recent graphs, if the deaths or admissions data had undergone a step change in the manner in which it was collected and published, I would have hoped to have noticed a step - but rather it looks more like a gradual change - which, in my view, is more supportive of a real effect.

Data gathering becoming a low priority isn't an announce policy. It happens over time.

The data originates from hundreds, perhaps thousands, of civil servants. Once data becomes a low priority, it will be gathered with varying levels of diligence. I'd expect this to evolve over different time scales in different places as people get replaced, get new high priority task, or gradually realize that they can get away with less diligence. I would not expect a step function, but a change over a few months.

I would certainly not claim my interpretation is correct, but it is plausible, as is your interpretation. Once the data integrity degrades lots of things become plausible and it's hard to tell what actually happened.

[Julian]

…
Looking back at the summer peak, the projected deaths were lying above the actual deaths, while now the opposite is true. Its probably time to look at adjusting the model parameters to get a better fit to current data, hence I have increased the multiplier from 0.11 to 0.125 in the graph below
…
This is a better fit to the current data, but a significantly poorer fit to the summer peak. I have suggested that the reason for the apparent increased actual death rate is possibly seasonal - lots of other bugs going around, and more gatherings indoors.

With the recently raised caveat, if the increase is real and not an artefact of changes in data collection methodology/accuracy, another possible explanation might be some hopefully minor reduction in the various vaccine’s protection against severe disease after 1, 2 and 3 dose regimens.

I haven’t been following stuff as closely as I did, partly because I haven’t been in the UK for almost 2 months now and won’t be back until just after Christmas, but my impression from what I have read so far is that the uptake of the bivalent booster is lower than for the initial mostly (exc J&J) 2-dose cycle and is lower than the uptake for the first booster. The bivalent booster would be the 4th dose for most people who are not in a higher risk group (which includes over a certain age but can’t remember what that cutoff is) so for those who have not had it yet their last dose, their first and so far only booster, would have been back in November last year so approximately a year ago. Although from what I have read it is pretty widely accepted that vaccine protection against severe disease & death is far more durable than protection against basic symptomatic infection I can imagine that on a scale of the UK (or England) population even a few percentage points of waning could create an observable effect.

But then the above ignores the possible counter effect of Omicron having caused so many more natural infections over the last year which might act to boost protection against severe disease & death in survivors. There are just so many factors at play, and now in a heavily “tainted” population where “clean” control groups are almost, perhaps literally, impossible to find and confounding factors all over the place that I would think are becoming ever harder to control for. I fear that there is a lot of causality around this pandemic that we will simply never be certain about and what we do begin to understand might take us many years to get to that understanding.

- Julian

[scotia]

There is not much to report on the model data which is different from the trends shown in last week's graph. I'll wait until next week before using up more storage and bandwidth

[scotia]

An update of the graphs from two weeks ago. I am now using the updated parameter numbers which are optimised around the most recent peak. So we now have a simple 2-parameter model where the Covid deaths, by registration date, are predicted as being 0.125 times the Covid hospital admissions of 10 days previous.
The latest Covid hospital admissions data is dated 5th December. The deaths by registration date will be subject to substantial updates (additions) for several days, so I have only included it up to the 28th of November. All plotted points are 7-day summations to iron-out any weekly variations.



There are reasons for optimism. The actual deaths are still reducing towards the projected minimum, and the projected deaths (based on admissions) have only started to slowly increase. But I fear that the party season may change this picture.

[swill453]

It's really interesting how COVID in this country has pretty much dropped out of the news. According to the Zoe COVID app we're in the "up" phase of another wave, with estimated 215,000 new cases per day.

Meanwhile according to Google, China cases peaked below 50,000 new cases per day and are now heading down again.

Scott.

[scotia]

An update of the graphs from two weeks ago. I am now using the updated parameter numbers which are optimised around the most recent peak. So we now have a simple 2-parameter model where the Covid deaths, by registration date, are predicted as being 0.125 times the Covid hospital admissions of 10 days previous.
The latest Covid hospital admissions data is dated 19th December. The deaths by registration date will be subject to substantial updates (additions) for several days, so I have only included it up to the 12th of December. All plotted points are 7-day summations to iron-out any weekly variations.



After following a broad valley, the projected deaths (based on admissions) are now climbing, and the actual deaths are now starting to follow.

[Itsallaguess]

Dec 23 (Reuters) - Nearly 37 million people in China may have been infected with COVID-19 on a single day this week, Bloomberg News reported on Friday, citing estimates from the government's top health authority.

About 248 million people, which is nearly 18% of the population, are likely to have contracted the virus in the first 20 days of December, the report said, citing minutes from an internal meeting of China's National Health Commission held on Wednesday.

https://www.reuters.com/business/healthcare-pharmaceuticals/china-estimates-covid-surge-is-infecting-37-million-people-day-bloomberg-news-2022-12-23/

Cheers,

Itsallaguess

[servodude]

Dec 23 (Reuters) - Nearly 37 million people in China may have been infected with COVID-19 on a single day this week, Bloomberg News reported on Friday, citing estimates from the government's top health authority.

About 248 million people, which is nearly 18% of the population, are likely to have contracted the virus in the first 20 days of December, the report said, citing minutes from an internal meeting of China's National Health Commission held on Wednesday.

https://www.reuters.com/business/healthcare-pharmaceuticals/china-estimates-covid-surge-is-infecting-37-million-people-day-bloomberg-news-2022-12-23/

Cheers,

Itsallaguess

They're really quite staggering numbers on one level - and a real world lily pad thought experiment on the other.

[Itsallaguess]

Dec 23 (Reuters) - Nearly 37 million people in China may have been infected with COVID-19 on a single day this week, Bloomberg News reported on Friday, citing estimates from the government's top health authority.

About 248 million people, which is nearly 18% of the population, are likely to have contracted the virus in the first 20 days of December, the report said, citing minutes from an internal meeting of China's National Health Commission held on Wednesday.

https://www.reuters.com/business/healthcare-pharmaceuticals/china-estimates-covid-surge-is-infecting-37-million-people-day-bloomberg-news-2022-12-23/

They're really quite staggering numbers on one level - and a real world lily pad thought experiment on the other.

On one level, it's almost a shame it's happening in such a secretive society, because they're basically now running a live 'let-it-rip' experiment, and the medical community really would benefit from having access to robust and reliable data if it were ever to be made available, and if it could be trusted, of course...

Given the granular and layered levels of secrecy throughout though, I'm not even sure they'll learn anything themselves about the torrent they're now likely to be living through, never mind allowing such data into the wider scientific community.

One thing that's worth remembering is that even though they were taking a long-term 'zero COVID' approach earlier, the country has been rolling out a vaccine programme that's got what will hopefully prove to be 'useful' coverage of the most vulnerable in their society -



Source - https://www.bbc.co.uk/news/63798484


Of course, the above numbers could be much better at this point, where they're now simply opening the flood-gates, but when reading the above BBC report on China vaccinations, I was actually initially surprised by how good it was, but then I considered the recent widespread protests and perhaps understood them a little better, because ultimately a society has to ask itself what the point of getting vaccinated multiple times is, if highly restrictive lock-downs and case-management is also then still seen to be required, and I think that political question has now been answered by the rapid broader removal of the earlier restrictions...

Cheers,

Itsallaguess

[Dod101]

Meanwhile the headlines in the FT this morning

'China battles 250 million Covid cases

More than half of Beijing infected Surge follows lifting of the curbs Health system overrun '

Dod

[servodude]

On one level, it's almost a shame it's happening in such a secretive society

It certainly doesn't help US try to make sense of it!

I'll try and remember to come back with thoughts after the holidays
If I forget... I hope you have a great time
- stay well

-sd