John Roberts Profile picture
Dec 26, 2021 12 tweets 4 min read Read on X
We've recently seen a fall in COVID beds occupied in London where COVID is the primary cause of occupation, from around 85% to 75%. Is this a sign of Omicron being less serious? Or is it a result of a rapidly growing variant, and the lag before "With COVID" admissions rise?

1/ Image
Here's a very simple example, which starts at Day 0 with 100 From COVID adm'ns, and 20 With COVID (so 83% primary diagnosis).

I've modelled this assuming infections double every 3 days (up 25% per day) for just 10 days, level either side.

2/ Image
Now, people admitted who just happen to test positive ("With" COVID) should increase exactly in line with this curve. (Big simplification - I've not modelled age differences here.)

So the sharp-eyed will notice this graph is exactly the same, except for axes and titles.

2/ Image
However, for those admitted because of COVID, the upward curve is pushed to the right - I've assumed 7 days here, just for the purposes of this example.

3/ Image
The total admissions is the sum of the two, and we can see here how the "with COVID" starts building up first, followed by a much sharper increase once the "From COVID" adm'ns start rising, 7 days later.

4/ Image
In terms of the proportion of "From COVID" cases, the % drops as the "With COVID" cases start to increase first, only levels off once the "From COVID cases start to rise, and in this model, as I assume level from day 10, eventually it reverts to the original figure.

5/ Image
Thus, in a period of very rapid growth, it's not at all surprising that the proportion of "With COVID" cases has increased - indeed, I'd be more surprised if it hadn't.

But what, if seems the case, the driver of the rapid increase is a variant less likely to cause adm'n?

6/
We then decouple the change in "with COVID" adm'ns, which are still based on infection levels, and "From COVID" adm'ns, which will now be growing at a slower rate. I've assumed a rate which means by the end of 10 days there are only half as many From COVID adm'ns as before.

7/ Image
We then see a much lower total number by the end of the period (as expected), but the absolute contribution of the "With COVID" cases is as before.

7/ Image
And that means, that whilst the movement in the % falls in the same way at first, it doesn't recover to the original level and settles at a lower figure, as With COVID cases take a higher proportion than if severity was unchanged.

8/ Image
As noted before, this is a very simple example, and other factors, notably how any rise ripples through the age groups will add to the complexity.

But I hope it helps give some context to what we are seeing and might see in admission levels and the With/From split.

9/
And finally, it doesn't cover the important aspect of hospital derived infections - partly because I can't think of an obvious way to model it, and partly because this thread is long enough for Boxing Day!

10/10

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with John Roberts

John Roberts Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @john_actuary

Mar 15
A great thread on the state pension and National Insurance.

For me, the trailed abolition of NI and thus its replacement by general taxation in terms of funding state pension benefits will have a major generational redistribution of tax.

1/
It’s been the case that (in aggregate) at any one time the working generation funds the SPs of the retired generation above it.

2/
If the abolition of NI results in an increase (albeit smaller - else why bother) in income tax, whilst those in work will in total be better off, pensioners will be worse off.

3/
Read 9 tweets
Feb 7
The Pensions & Lifetime Savings Association has updated its guide to living costs in retirement. The full report is well worth a read, and goes into a lot of detail.

1/

retirementlivingstandards.org.uk/2023_research_…
Image
One key point is that it assumes that pensioners own their home outright - probably reasonable now, but the shift to renting means that in future years that may become increasingly questionable.

2/
It focuses on retirement income, but note that other sources may be used to fund retirement, whether it be income from savings/investments, or gradual withdrawal of capital. Much more likely to be relevant for those aspiring to a comfortable lifestyle of course.

3/
Read 7 tweets
Dec 21, 2023
UKHSA estimates that prevalence of COVID in England and Scotland has nearly tripled in the month since the ONS restarted its COVID infection surveillance.

1/

gov.uk/government/sta…
Image
Fortunately prevalence is lowest at the oldest, more vulnerable age groups, but is estimated at just under 6% in the 18 to 44 age groups.

2/ Image
Prevalence is estimated to be highest in the London area, at just over 6% across the population. Note though that confidence intervals are wider due to lower sample sizes than in previous studies.

3/ Image
Read 4 tweets
Sep 12, 2023
So with the news this morning that the earnings growth announced today means the state pension (SP) will very likely increase by another 8.5% next year, it's time to set out once again why the SP triple lock (TL) is such a bad idea.

1/ Image
It's all down to cherry-picking the best of the three rates each year. I did a thread nearly a year ago, that hopefully sets out clearly how the mechanism inevitably means that the SP will grow over time against both earnings (E) and prices (P).

2/
Visually, you can easily see how that cherry picking works in practice with this great chart from the FT.

The effect is greatest when one index lags the other by a large amount in one year, and then catches up in the next year.

3/ Image
Read 17 tweets
May 31, 2023
With BH's still distorting individual weeks' figures, the cumulative position gives a better view, with the latest CMI age-standardised analysis showing mortality 3.8% (of a full year's mortality) worse than its reference year of 2019.

1/
That 3.8% of a full year's mortality translates to an excess of 9.3% of YTD expected mortality, or over 21,000 excess deaths.

Bear in mind that pre-pandemic, you'd expect mortality 4 years later to be noticeably lower, due to the previous improving trend.

2/
The 9.3% excess is split 50/50 into "with COVID" deaths (where it was listed as a contributory factor on the death cert), and others.

Over 10,500 registrations in 2023 have had COVID listed as a cause - around two thirds of which it's put as the underlying factor.

3/
Read 5 tweets
May 13, 2023
A great example of how @ABridgen will RT any old rubbish, despite how absurd the numbers look, if they fit the anti-vax narrative he is now pursuing.

Even the original poster has now admitted that the figures were completely wrong and deleted the post.

1/
Here's the mea culpa - it was only wrong by a factor of 13, but at least the post has been deleted rather than just corrected and left up, when experience shows that only a fraction of the original audience will see the correction.

So what are the true numbers?

2/ Image
In E&W the peak week in 2020 was just under 9,000, and the second wave peak was pretty close to that number.

In total ONS has recorded 199,728 COVID related deaths in E&W since the pandemic started.

3/ Image
Read 8 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(