Coronavirus - Modelling Aspects Only

[scotia]

Correction

Moderator Message:
Now corrected in the original

[scotia]

Back on the Coronavirus -General Chat - No Statistics board, Redsturgeon posted on the 26th July

https://www.lemonfool.co.uk/viewtopic.php?p=517289#p517289

Which reported "I heard today that covid deaths are up in the UK".

I responded that some of the published data may be out of date - so I have come over to this board to provide my reasoning. I'm getting back to my simple model which proposes that the deaths are proportional to the hospital admissions a period of days before. In order to get the most up to date death data , I previously used the Covid deaths by publish date. However this data became erratic as to when it was published, and now only seems to appear around once per week - so I gave up on this model. The alternative was to use the Covid deaths by registration date - but this has the disadvantage that it is not static - with updates (additions) appearing many days after the most recent values. However I thought that an entry might be reasonably static about a week after its first publication.
So looking at the most recent data extracted this evening (27/7/22) for England, there is data on both admissions and deaths by registration date up to 25/7/22. To make allowance for later updates to the deaths data, I have assumed that it would be reasonable static at 18/7/22, so I have only included deaths up to that date. And for the model I have used a multiplicative factor as 0.11, and the delay between admissions and deaths as 10 days. This fit was achieved by looking at the earlier peak around 13th April. All of the data plotted are 7-day totals - to iron out any weekly dependencies,



We see that the plotted deaths are continuing to rise - but the last (blue) data point represents the week ending 18/7/22. Looking at the (red) projected deaths, based on admissions, we see that a peak is around five days later - around 23/7/22. So if we believe the model, we are probably past the peak of deaths. Now I don't expect the simple model to give precise results - given that its parameter fit is based on data around April-May, however i would be (unpleasantly) surprised if we do not see that the registered death rate has peaked around now. But given that we need to wait for around a week for registration date deaths to settle down, we probably will need to wait for at least a further week to see if my hypothesis is correct. Optimistically - I hope I am

I have not included Scotland in this model, since no Scottish Covid deaths data has been published since the 2nd of June.

[funduffer]

Scotia,

I have followed your modelling of deaths v hospital admission for some time, and it is interesting to me that the correlating factor has changed over time.

You are now using a factor of 0.11, which I think is lower than what you started with.

I wondered if you had a record of how this factor has changed during the pandemic? My impression is it has reduced over time, suggesting that immunity due to vaccination or prior infection has blunted the deadliness of the virus, and/ or treatments have improved.

Any thoughts?

FD

[scotia]

Scotia,

I have followed your modelling of deaths v hospital admission for some time, and it is interesting to me that the correlating factor has changed over time.

You are now using a factor of 0.11, which I think is lower than what you started with.

I wondered if you had a record of how this factor has changed during the pandemic? My impression is it has reduced over time, suggesting that immunity due to vaccination or prior infection has blunted the deadliness of the virus, and/ or treatments have improved.

Any thoughts?

FD

Yes - when the factor being used clearly needs updated, it signals that there has been some change in the effect of the Covid virus. My original graphs (early Dec 2020) used past data to estimate the factors - and 0.265 with a delay of 13 days seemed to fit. But as we moved from late December into January, it became obvious that these these numbers were too optimistic for the recent data, and 0.3 with an 11 day delay became a better fit. This coincided with the PM announcing that the latest variant had a higher mortality rate. However vaccines were on the way - and hopefully this would bring substantial changes - and they did! By the end of February 2021 I revised the multiplicative factor down to 0.26 to achieve a better fit to the most recent data, then down to 0.2 in early March, and further down to 0.175 by early April. By late April I was struggling to find a fit to the ever-reducing numbers, and I had moved the multiplicative figure down to 0.075 with a time slip of 23 days. The success of the vaccines had been spectacular. But moving on to May/June, the numbers began to rise again - probably due to the new Delta variant which was causing some concern. By September 2021 the mutiplicative factor had risen to 0.14 with a time slip of 11 days. In November the factor was increased to 0.15. I continued with these parameters - however by February 2022 the fit became poor - with the shapes of the death curves not following those of the admissions. By April 2022 the deaths by published dates became obviously erratic - and by the end of May I felt that there was little point in attempting to match them to the admissions data. The Scottish Covid deaths data ceased on 2nd June. The English Covid deaths by published date has become even more erratic - although it looks like it may be settling down to a once-a-week number. The English Covid deaths by registration date has continued - but the numbers it contains are subject to adjustment for a considerable time after they are first published. Hence my (previous) reluctance to use it. I'll continue to keep a watch.

[funduffer]

Scotia,

I have followed your modelling of deaths v hospital admission for some time, and it is interesting to me that the correlating factor has changed over time.

You are now using a factor of 0.11, which I think is lower than what you started with.

I wondered if you had a record of how this factor has changed during the pandemic? My impression is it has reduced over time, suggesting that immunity due to vaccination or prior infection has blunted the deadliness of the virus, and/ or treatments have improved.

Any thoughts?

FD

Yes - when the factor being used clearly needs updated, it signals that there has been some change in the effect of the Covid virus. My original graphs (early Dec 2020) used past data to estimate the factors - and 0.265 with a delay of 13 days seemed to fit. But as we moved from late December into January, it became obvious that these these numbers were too optimistic for the recent data, and 0.3 with an 11 day delay became a better fit. This coincided with the PM announcing that the latest variant had a higher mortality rate. However vaccines were on the way - and hopefully this would bring substantial changes - and they did! By the end of February 2021 I revised the multiplicative factor down to 0.26 to achieve a better fit to the most recent data, then down to 0.2 in early March, and further down to 0.175 by early April. By late April I was struggling to find a fit to the ever-reducing numbers, and I had moved the multiplicative figure down to 0.075 with a time slip of 23 days. The success of the vaccines had been spectacular. But moving on to May/June, the numbers began to rise again - probably due to the new Delta variant which was causing some concern. By September 2021 the mutiplicative factor had risen to 0.14 with a time slip of 11 days. In November the factor was increased to 0.15. I continued with these parameters - however by February 2022 the fit became poor - with the shapes of the death curves not following those of the admissions. By April 2022 the deaths by published dates became obviously erratic - and by the end of May I felt that there was little point in attempting to match them to the admissions data. The Scottish Covid deaths data ceased on 2nd June. The English Covid deaths by published date has become even more erratic - although it looks like it may be settling down to a once-a-week number. The English Covid deaths by registration date has continued - but the numbers it contains are subject to adjustment for a considerable time after they are first published. Hence my (previous) reluctance to use it. I'll continue to keep a watch.

Fascinating. In effect, this one parameter has charted the course of the entire pandemic, covering the effects of new variants and vaccines. I hadn’t realised it has been such a roller-coaster ride!

[scotia]

I'm following up my previous graph with a further week of English data (extracted this evening from coronavirus.data.gov.uk) to see if last week's tentative prediction is confirmed - that the English Covid deaths (by registration date) should have peaked around 23rd July and, and are now falling. I use a simple 2-parameter model 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 1st August, as is the Covid deaths by registration date. However the deaths by registration date will be subject to updates (additions) for several days, so I have only included it up to a week previous to the latest data - i.e. to 25th July. All plotted points are 7-day summations to iron-out any weekly variations.



And yes - the actual Covid deaths by registration date have peaked - around the 20-21st July. And the projected deaths (from the admissions) have continued to fall - all good news. However I was a bit puzzled by the statistically significant jump in deaths a few days before the peak. So I had a look at the daily deaths by registration date for the week 16th to 22nd July. These were 157, 189, 156, 293, 205, 145, 137. The anomalously large 293 was on 19th July - when England experienced its highest ever temperature in modern times.

[scotia]

A further week of English data (extracted this evening from coronavirus.data.gov.uk). I use a simple 2-parameter model 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 8th August, as is the Covid deaths by registration date. However the deaths by registration date will be subject to updates (additions) for several days, so I have only included it up to a week previous to the latest data - i.e. to 1st August. All plotted points are 7-day summations to iron-out any weekly variations.



I think it is now clear that the registered deaths, and their prediction from hospital admissions are now falling in synchronism from a peak around 20-21 July. The anomalously large jump in registered deaths a few days before the peak was due to the significantly large daily deaths (293) on July 19th (when England experienced its highest ever temperature in modern times). The downward step after the peak is the result of this 19th July figure dropping out of the 7 day total.
Currently all good news. I'll keep a watch, and if there is any upturn in the data, I'll publish further plots.