Thread by @Jean__Fisch

I continue to see talks of high unexplained excess deaths in England and Wales in S1 2023

This is NOT confirmed by data by occurrence date and is an artefact resulting from
/ reporting by registration date
/ policy of only registering deaths once cause is assigned

THREAD

Just to make my point crystal clear

The S1 2023 deaths by registration date excl covid are POSITIVE and high which signals lots of unexplained deaths

Yet, when using deaths by occurrence date (as one should), the same total is NEGATIVE. So in fact no/few unexplained deaths

1/

If you think that this effect is just the result of the fact that I used the ONS baseline which is not a proper choice

The same conclusion holds true when using the approach that CMI proposes (ie based on 2019 mortality rates and adjusted for population development)

2/

Why is this so? Because of TWO main factors

Factor 1: The huge wave at the end of 2022 led to an unusually high amount of 2022 deaths being registered in 2023 (I estimate 7k more than in previous years)

So deaths by registration inflate the true death toll in 2023

3/

But that's not the end of the story: There is a 2nd more pernicious factor playing here

E+W is unique in that it only registers deaths once cause of death is assigned

Other countries register unclear deaths under a temporary ICD code and correct it once the cause is clear

4/

The weakness of the E+W system is that deaths that require to go through coroners (as is often the case in younger age bands) can take a year or more before they get registered

So the death data by occurrence date takes much longer to be complete

5/

Just to illustrate the magnitude of the issue: I found some data by age and occurrence for England

It shows how, at the end of May 2023

/ the last 18 months of the 15-44 data
/ the last 6-9 months of the 45-64 data

were essentially materially incomplete

6/

Here another way to show this based for the 15-44 age band

In 2022, ONS shared the same dataset by occurrence for deaths registered up to May 2022

If you compare it with the dataset with deaths registered up to May 2023, you see how the data diverges over the last 18 months

7/

The registration date for many deaths in younger age bands will be MUCH later (up to 18 months) than death occurrence

This means that the data by registration date will not reflect the timely pattern of the deaths in these age bands (compare blue and black line)

8/

But as, I explained yesterday, this has other consequences: If, as is likely in a pandemic, there are MORE deaths going through coroners, then X months later, deaths by registration will see a (fake) increase

And this is what I suspect is currently happening in 2023 as well

9/

Unlike previous years, there were much more deaths registered in May-Jun 2023 than occurred

(the deaths occured is the result of an estimate made by ONS which corrects for lag and is quite precise at top level)

10/

As I show here, this means that almost all the excess mortality seen in the data by registration date in May-Jun 2023 is the result of deaths that occurred before May

11/

https://twitter.com/Jean__Fisch/status/1681310266036989953

It took me quite a while to understand why deaths by registration were so off vs. the actual situation

What happens is that registration-based data is quite robust if one year is like another (as it was pre-pandemic)

12/

However, if, as is the case in a pandemic, there are "one-off" and lots of unusual events, constraints on registration capacity, etc., the data by registration date starts to be driven by noise instead of underlying death occurrences

And this what we are currently seeing

13/

So deaths by registration date are currently not suited for analysing excess mortality in England+Wales as it is affected by

/ much higher than usual spillover from 2022 to 2023

/ likely catch-up of past occurrences only registered now

/ timely noise in younger age bands

END

@AdeleGroyer (by significant, I mean 1-2% which is essentially within band of uncertainty)

That's why I was getting more and more concerned about the difference of message between registration date data and occurrence one

But happy to be proven wrong, as you know! :-)

@AdeleGroyer And of course, just to be clear: no topline excess does not mean no excess: It could well be that there is negative excess in older age bands and more positive in younger ones

Here the data from the 65-84 which is pretty complete up to Mar 23 and there is no excess before covid

NB: I put this thread on hold until I clarify seemingly contradictory elements
/ I find the ONS modelled deaths remarkably accurate
/ However, they imply an unrealistic shortening of the death registration lag

More to follow once I understand what's happening