UKH

Just taking a peek at excess deaths data. Seem massively down over the last few weeks (more so than in previous periods of low prevalence).

If this carries on we'll be cancelling out peaks! Any ideas why so many fewer deaths compared to last summer?

https://www.travellingtabby.com/uk-coronavirus-tracker/deaths

We're in spring now, not summer.  It does not seem sensible to compare different times of year directly as your question does.  To start with the same time of year:

Why do we have fewer deaths than this time last year?

• Because we've got far less Covid around than this time last year.

Why do we have fewer deaths than this time of year in the period 2015-2019? I await informed analysis with interest.  If I had to place a bet on reasons...  

• Covid deaths are now very low, and we've hand a year of measures intended to lower the prevalence of transmissible diseases, and those measures are working - not just against Covid.  In a typical year (if there can be such as thing), there'd be about 2,000 weekly deaths from influenza and pneumonia at this time of year.   Deaths are currently about 2,000 / week lower than the average for 2015-2019.   So the numbers aren't wildly incompatible although I don't claim this is any kind of proof.  
• Something like 150,000 people have died before their time.   So they won't be contributing to death rates going forwards.  Given that on average people likely lost multiple years of future life, it's not enough alone to explain the difference.

In terms of trying to bridge between mid-spring and mid/late-summer to consider your question:

• Flu deaths are higher at this time of year than in late summer, and we've had control measures in place for longer which seems to have really hammered flu almost entirely out of circulation.
• A lot more people have died before their time since last summer so aren't dying any more.

They’re not down massively. They’ve been down about 500 a week for about 3 weeks now. At that rate it’ll take 3 years to cancel out the first peak. 

That 'Deaths from All Causes in the UK by Week Number', is a curious graph.

What happens around week 22 & 35/36 that causes the drop in all years?

And the blip after xmas. Is that people holding on, or just that deaths are recorded when staff go back?

You'd need to factor in accidents (on the road, in the workplace etc). Also people who couldn't get seen by a doc or get an operation for something preventable, will show up later. I wonder if the lower accidents vs not treating illness would cancel out.

Interesting data, even though I'm not certain what to make of it all.

When I say massively down I am referring in relation to the previous 5 year period (as per ONS definition of excess deaths). A fair point about comparison with last summer - I meant to make a comparison by prevalence.

The mean over the last 9 weeks is -1052/week (albeit with some fairly substantial deviation between weeks) i.e. -9500 total. That is substantial (~10% down in the context of total number of deaths over the period which is 10-12k/week).

Wintertree may be right about people dying before their time. If that is the case it will be interesting to see how that effect will play out over the next few months.

Road deaths ~1800/year according to RAC i.e. 34 a week (not sure how these are distributed) so cannot be a substantial contribution.

Workplace accidents I cannot imagine being a high number either in this modern age of health and safety? EDIT - slightly  >100/year total i.e. 2/week.

I think the xmas blip would be a recording matter probably?

Well that’s a lot of dislikes for me to collect without anyone saying what they disagree with...

Humans can be mental, so there you go.

If I had to guess I would probably find the squashing of influenza the most persuasive reason. Excess deaths were negative compared to last year in the autumn for a bit last year as well, before covid deaths started to ramp up again.

I suspect there is some "bringing forward deaths" effect, but its effect will be quite smeared out over time - as you say a lot of covid deaths will have been people who otherwise could have expected several more years. If you were going to see that effect as a measurable and short term impact, I would have assumed it would be most visible looking only at the very old, where residual life expectancies would have been least. But obviously those ages are also likely to be most benefitting from the squashing of influenza, so whether you could robustly unpick it is hard to say.

Looking at this year, it's May half term and August bank holiday. I would therefore lean towards reporting lag.

I don' think we typically see many flu deaths at this time of year its normally winter not spring.

I've heard several people say this type of pattern after a wave of covid is likely as many of the vulnerable were killed by covid.

There is a gaping discontinuity between the last point of 2020 and the first point of 2021 which makes me wonder a bit about the appropriateness of any filtering used.  

That contradicts wintertrees numbers above:

He doesn't mention his source, but is usually man to be relied on as far as numbers go.

So have I. But, as I mentioned above, I think the effect is going to be spread over time and difficult to unpick from other drivers. An awful lot of the people killed by covid would have an average life expectancy of more than 5 years (that's about the remaining life expectancy of someone aged 85), many more than 10 (75 - obviously the younger you are the younger your expected age of death is) and lots could in theory have expected more than 20 years. So over the next X years, where X is easily at least 20, there'll be 150,000 fewer deaths or whatever the toll is now.

I should have put it down as I can't now find the one I used.  Sloppy.

The ONS have plenty of sources, e.g. figure 2 here has similar data and is more official than the previous source I'd looked at - https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriage...

Whilst other data shows influenza to be almost gone by the end of 2020, there were still quite a few deaths which suggests many were due to pneumonia.  It's muddled though because flu is not a disease that people are routinely tested for, rather it's surveilled in a subset of places, and pneumonia is - as I understand it - a catch all term the effects of many difference causes, of which flu is one.  It's also notable that the "all deaths" winter spike is much larger than the "flu and pneumonia" spike, and that both spikes are still above baseline by May.    How much of each spike would there be without flu?  

https://travelhealthpro.org.uk/factsheet/51/seasonal-influenza

"Seasonal influenza occurs throughout the world. In temperate regions of the northern hemisphere most influenza activity is from November to March (between December and March in the United Kingdom). In the southern hemisphere most influenza activity occurs between April and September. In the tropics, influenza viruses can circulate throughout the year."

https://www.bupa.co.uk/health-information/childrens-health/flu

"Seasonal flu

Expert reviewer, Professor Robert Read, Professor of Infectious Diseases
Next review due June 2021

Seasonal flu (influenza) is an infection caused by a virus that can affect your nose, throat and lungs. People tend to get it around the same time every year. In the UK, people usually get seasonal flu between December and March, although outbreaks can happen as early as October and as late as May."

I don’t think that page can be right. 
 

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/...

I’ve got my idiot head in today and was reading “week no” as “month no” in that figure.  

This report gives a 5-year average for influenza and pneumonia in figure 2 and the number for deaths involving flu and pneumonia is ~1800/week by May.  The 2020 data is much higher, presumably for the obvious reasons...  Flu was almost eradicated this winter but is anyway low by May, but pneumonia kills all year round in a normal year.  Will be interesting to see what the data looks like for this year once it’s collated etc.

https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriage...

That site plots ONS data. Care to explain how it differs from your link? Looks similar to me.

You also have better outcomes for covid patients now due to things like dexamethasone which reduced deaths by 10% - I think I read that improved treatments have reduced the original covid death rate by 30%.  I think lockdowns will have reduced the other death rates too - fewer accidents at work, fewer RTAs, fewer sporting accidents, fewer activity accidents. Fishing is the most dangerous occupation and the boats haven't been going to sea as much due to Brexit. Flu levels have been decimated - we've not had a single flu death here last winter and the virus isn't even showing in the community (they routinely test for it when doing covid tests here).

Your 30% improvement in the covid death rate is an negligible contribution given the absolute numbers of covid deaths happening currently.

Contribution of workplace accidents, RTAs, sporting accidents are all total red herrings in the face of the absolute numbers (see by post above on the two former).

The potential contribution of flu/pneumonia reduction needs digging into - I don't have the time currently!

EDIT: See slide 10 here showing trends in influenza.

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/...

It seems to be that from this point on in until next winter flu is basically out of the picture. More interestingly, the trends in excess deaths are not following the historical flu prevalence rates (2018 & 19) over the preceding 9 weeks i.e. flu tends to follow fairly linear downwards trend with time while excess deaths do not.

Ah. I didn’t see the buttons for; all, 3mnth, 6mnth, 1yr etc. It defaults to 6mnths. Which is why it doesn’t look right. I was missing the entire 2020 epidemic. It’s also missing the last 4 years except where they’re overlaid on each other which makes it difficult to read. 
 

Excess deaths take into account normal deaths from the seasons so that 113,000 excess deaths is over and above any normal flu deaths.

Covid hasn’t just killed people a ‘few months early’.

You are correct that excess deaths is over a moving baseline (and thus over and above 'typical' flu deaths in any given week). Indeed, there is a case for more covid deaths occurring than the simple excess death tally if you consider the possible impact of any lockdown flu suppression. Of course statistically speaking there must be a substantial cross-population of those killed by covid who might otherwise have been killed by flu.

However, my point is more concerned with the current trend of negative excess deaths ongoing at the moment.

https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriage...

Nice link, annoyed I didn't find that one.

By far the biggest fractional reduction is to flu and pneumonia, but there is a slightly larger absolute reduction to dementia and Alzheimer's disease, and further smaller reductions elsewhere that add up.

I suspect it'll take a while to separate the causes from those who departed before their time, and from those who have benefited from the Covid measures in terms of respiratory disease etc.

With 60m people in the country you will have around 10k deaths from old age alone every week. 600k a year. Given we have only had 113k excess deaths it’s not going to impact that figure remotely. 
 

The concern has always been that the 600,000 all needed hospital care for 4 weeks in the same few months in a doomsday scenario. 

But the OP is not looking at "that figure" (the weekly deaths) but the difference from the average.  The difference from the pre-covid 5-year running average is currently around 2000/week (not 500/week as someone claimed up thread).   

If the excess deaths were people who would have died in a uniform distribution over the next year, it can fully explain the difference. If they are people who would have died in a uniform distribution over the next decade, it can explain 1/10th of the difference.  

I suspect that for care home residents in particular - hard hit in March/April 2020 - that the distribution was strongly biassed towards the short term.

So it seems clear to me that this is going to emerge as a contributory factor; not the main one but not negligible either. 

But really, everything is so different this year that trying to figure it out from the data available now is a tall order.

I suspect some of the lessons learnt will have consequences that extend beyond Covid in terms of life expectancy and care of the elderly.  

Having had a nearly non-existent flu season is something of a concern for winter 21/22.  I haven't been following the specialist news on the antipodean flu strain forecast, but their low numbers pose another concern to our coming winter.

I claimed that. From the ONS gov website.  Which is why I questioned the websites figures. Or at least the way they’re being presented. 
 

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/...

Nice link. Looks like we should see negative excess (surplus?!) deaths coming down I suppose as respiratory viruses drop off. I would imagine dementia and covid have a very high cross-population.

So, the graph presented in the ONS report has deaths currently 2,000/week below the baseline, not 500/week.

I took a screenshot of the plot.  I drew a blue rectangle on the right down vertically from the baseline to the current value.  I then made an identical copy of the rectangle and dragged it over to the y-axis on the left hand side.  It's height is - near as damnit - 2,000/week.

Mind you, the recent drop down is quite stark, and perhaps it will be reversed soon - the data series of full of brief sojourns up and down; if not it certainly will be remarkable.

Yes. The yellow area is never correct. It just uses the currently collected data. It does say that at the bottom. 

Perhaps that is the disparity, but before the most provisional part the difference from their baseline had fallen to over 1,000 / day, not 500/day.  I don't 

The two data sets don't seem to agree on absolute numbers either, and the one you linked doesn't define what its "baseline" is...

The whole set is here. The blue line on the graph is the 5 year average. 
 

https://www.gov.uk/government/statistics/weekly-all-cause-mortality-surveil...

Is it?  I haven't actually managed to find a definition of the blue "baseline"; perhaps it's one one of the older documents.

• It certainly isn't a running 5-year average because 2021 doesn't change to reflect the Covid spike of 2020.
• It certainly isn't a by-week 5-year average, as the signifiant time uncorrelated variance in the data series would not reduce to such a smooth curve over just 5 years.
• It looks to me like a pretty simple parametric fit to the data, whose peak is reducing very slightly year-on-year.  

Super-crude point below - I've drawn 2021 ~ week 13 as a pink line on the right, and projected the weekly count at that point left.  I then identify week 13 on the black curve for the 4 previous years.  Measuring the line protrusions and calibrating to the y-axis suggests that the current weekly death at week 13 is ~1,500/week less than a running average of the previous 4 years (this is the last week it's reasonably valid to include the previous year, before Covid took over.)

So comparing the OPs link - using a 5-year pre-covid average - with this undefined "baseline" is comparing apples with an unknown mystery fruit. This data itself supports a claim of ~1,000/week below its mysterious "baseline" and ~1,500/week below the 4-year weekly average at the end of March.  The OPs link is valid to May.   I assume the actual data source is online somewhere.

Meanwhile, not sure what exactly PHE are doing for their baseline.

The plot is for England. The numbers are for England and Wales. 

That doesn't explain what the "baseline" is, nor address my point that the baseline in no way resembles an average of the last 4 years; it seems reasonable to assume the baseline also therefore doesn't resemble the 5-year running average in the OPs link.  Doing a 4-year running average on this data suggests numbers way closer to those the OP gave than the 500/week suggested here.

Public Holidays. Registrars are closed so you can’t register a death. This means you get fewer deaths registered in a week with holidays and these get ‘caught up’ in the following week. Leading to these weird blips. In the data by date of registration. You don’t see them in the data by the date of occurrence, since I’m obviously people still die on public holidays.

IIRC this is the same (weird) baseline that EuroMOMO use. It’s a model (a Serfling model I think) fitted on data from Spring and Autumn only. The bits in between are interpolated under some assumptions of sinusoidalness. The idea of this is that it gives you a kind of ‘ideal’ baseline in the absence of summer heatwave deaths or seasonal winter flu deaths (though still allowing for seasonal variation on total mortality rates). So it tells you more clearly how bad a flu season you’re having, for example.

It’s not very useful for the present situation though, except to give you a clue how many flu deaths have been averted this year.

Excellent, thank you.  I’d not heard of it before but I can see what it’s intended to achieve.  I wonder how useful the model is after such an exceptional year?  I’m quite uncomfortable with a data derived model that doesn’t preserve area under the curve from the fitting period of the data, but if I understand correctly that’s the entire point.  So, it’s very different to the 5-year pre-covid average in the OPs link.  Looking at some reference texts, with enough years in a graph it has a remarkable similarly to the shape of the sunspot count with solar cycles (not implying causation! - different frequencies - but just a similar look).

I’m getting a sense from the various mortality and flu surveillance statistics that what’s going on is a lot more complex than “flu season” - a lot of seasonality is perhaps lumped under “flu and pneumonia” with most of the causes of pneumonia not surveilled or tested.  If control measures this winter turn out to have significantly lowered pneumonia deaths as well, that opens up a whole host of questions about future practice; social questions as well as medical.

That is indeed the point. It’s for answering questions like ‘was 2018 a worse flu year than 2017’ or ‘did the UK have a worse 2018 for excess winter deaths than Denmark’. Both PHE and EuroMOMO’s use of it long predates COVID.

It’s not useless, but it rather changes the sort of questions you can answer with it.