Lightnews — Scholar-powered news

Prof. Michael Fuhrer

@michaelsfuhrer.bsky.social

2.6K followers 1K following 3K posts

Epidemic epistemic trespassing. Knows a lot about graphene. Monash Uni. Directed fleet.org.au. Fellow @scienceacademy.bsky.social. Plays bass for www.instagram.com/push_the_trigger Bird photos at www.flickr.com/photos/off-world

Posts Media Videos Starter Packs

Pinned

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · Aug 27

Part II of this deeply unpopular series! 😴💤

Which artists are making how much from streaming?

And just how unequal is the music streaming industry, anyway?

1/

2 4 14

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 17h

I think the big change is that we test a lot more for flu now, and probably detect almost twice as many hospitalizations and deaths.

Still, 2024-25 was worse than 2019-20 and probably had 2/3 as many deaths as 2017-18.

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

But generation time is very short (~2 days), so when it gets going it takes off fast.

They're really different viruses.

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

That's more a product of "waning" immunity which is largely immune escape due to genetic drift. Studies comparing time to 1st infection in children to time between infections have concluded that R0 is fairly low for flu (2.5 or so).

1 1

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

Thanks!

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

Acute covid really did cause a lot of deaths in 2020-2022 (5-10X average flu season!)

But in 2025 acute covid is causing far fewer deaths. It seems highly implausible that covid is associated with more excess deaths than flu in 2025.

36/36

2 1 4

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

There are other plausible stories out there, I am sure, and plenty of implausible ones too - looking at excess all-cause deaths can rule out some of the highly implausible scenarios.

35/

1 1

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

Is it "right"? Who knows! It is *impossible* to verify that excess deaths are "correct" as they involve measurement against a counterfactual world which doesn't exist and can't be verified. But it tells a story that is plausible.

34/

1 1

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

In the end, I think the answer depends more strongly on how you set up the model than whether you’ve weighted the data correctly. Hence I also don’t try to estimate uncertainties - the actual uncertainty is systematic and hard to account for because the baseline is unknowable!

33/

1 1

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

I did a simple least-squares regression. I didn’t properly weight the data by Poissonian uncertainty so take it with a grain of salt.

32/

1 1

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

N.B. the UKHSA model use far more parameters with 3 flu parameters for each season. I’d be a little concerned this might tend to over-attribute *any* winter excess to flu.

31/
khub.net/documents/13...

1 1

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

I found a slightly better fit if I used the average of one week and the following week’s registered deaths as the explanatory variable (that is, registered deaths ½ week delayed compared to all-cause excess). I didn’t try hard to see if that could be improved by adding more parameters.

30/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

I also allowed for an unknown linear background (2 more parameters) in the post-2020 fit period, so that a bad baseline wouldn’t skew the flu/covid results.

29/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

I then fit weekly deaths from March 2020 through 2025 using the same seasonal baseline, plus an additional flu parameter for the covid era (post-Feb 2020), and 2 covid parameters (pre- and post-July 2022).

28/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

I fit 2015-2019 monthly deaths (because only monthly available from CDC WONDER pre-2018) to a seasonal baseline of 6 parameters (intercept and linear slope, and amplitude and phase of two sine terms) and a single flu parameter.

27/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

N.B. UKHSA used registered covid deaths as an explanatory variable, but for flu used a proxy for prevalence calculated from outpatient visits and positivity. Probably because there are so few registered flu deaths in UK.

My model is more symmetric, treating covid and flu on equal footing.

26/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

Model details:

For explanatory variables, I used registered deaths with covid (ICD-10 U07.1) or flu (ICD-10 J09-11) listed as a contributing cause. That is, I assumed that the actual excess is proportional to registered deaths (multiplied by some factor).

25/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

For example, Canada is an outlier, with the largest negative unexplained effect on LE, but relatively small effect of covid.

Analysing Canada's all-cause deaths to try to determine covid excess would give a very misleading conclusion.

24/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

Here’s a paper which examined changes in life expectancy due to various causes of death in 24 countries.

The “unexplained” changes varied greatly, and appear poorly correlated with changes due to covid.

23/
academic.oup.com/pnasnexus/ar...

Validate User

academic.oup.com

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

This isn’t to say that we can’t probe post-acute covid and flu deaths using this type of analysis. But looking at all-cause deaths probably doesn’t tell us anything with any certainty.

22/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

But, if we have a particular cause of death in mind, we can get a fairly precise answer!

For example, deaths from alcoholic liver disease show a clear association with the onset of the pandemic, which is highly suggestive of a causal relationship.

21/

bsky.app/profile/mich...

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · May 30

Focusing on alcoholic liver disease only (ICD-10 code K70) shows an even more dramatic increase during the pandemic, and more than half the liver disease excess deaths are due to alcoholic liver disease.

4/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

If the hypothesis is: post-acute covid causes some deaths, with an unknown distribution of causes, ages, and time since acute covid, then we are unlikely to get an answer with any meaningful precision as to how many deaths are associated with post-acute covid.

20/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

If the hypothesis is that excess deaths happen in proportion to acute covid and flu deaths, then yes, we can get a reasonably precise answer as to how many of those deaths there are.

19/

1 1

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

This is important. Excess all-cause deaths is a very blunt measure. Without a hypothesis to narrow the search, excess deaths can’t tell us much.

18/

1 1

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

Importantly, the uncertainty in determining the size of these excesses is very large; a slightly different baseline would change their size by a lot.

17/

Prof. Michael Fuhrer @michaelsfuhrer.bsky.social · 1d

Can we determine how many excess deaths are associated with post-acute covid and flu?

My model and the UKHSA model both have unexplained excesses (see below). They’re small (smaller than the acute covid excesses) and vary from year to year.

16/

2 1