On Calculating the True UK Covid-19 Death Toll

August 31, 2020

update Sept. 3

After recent revisions, I come back to the United Kingdom's disputed death toll from the SARS 2 / Covid-19 pandemic. It's a bit of a shell game to compare data, considering mixed standards and definitions between regions that change over time, adding up differently at different dates. I can only clarify some points and offer informed guesses for other parts in an effort to offer a concrete estimate of some value. 

An Error Corrected?

I'll start with the heralded revision of August 11, lowering the UK death toll by 5,377 in what was taken as an admission of undue inflation by deceptive health authorities. Peter Hitchens, Daily Mail: "the London Government was last week forced to admit that for some time its official death figures have been a wild overstatement of the facts." He called the revision "a huge admission" of "an error" he suspects was no accident. As Hitchens explains:

"They were forced into this by the brilliant forensic work of Professor Carl Heneghan and his brave colleagues at Oxford’s Centre for Evidence-Based Medicine, which showed that the previous figures were so loose that they could have included car-crash victims who once tested positive for Covid."  

Similarly, Kit Knightly noted at Off-Guardian: "Actually, over the last week the UK’s covid death count has reduced by over 5000, thanks to a review which removed duplicates and mistakes (which OffG predicted would happen months ago). The case count is bloated by at least 30,000 duplicates too."

I guess the case count WAS bloated - 30,000 cases were removed on July 2. Bing tracker, still seeming stumped how to plot the revised death toll.

As for "duplicates and mistakes," Knightly meant about what Hitchens, Heneghan, et al. suggest, but cites a BBC article mentioning neither: "Previously, people in England who died at any point following a positive test, regardless of cause, were counted in the figures," the article says, laying out two different issues: how and when people died. "But there will now be a cut-off of 28 days, providing a more accurate picture of the epidemic," it continues. That does suggest the prior view was less accurate, as "regardless of cause" would suggest, but that's actually debatable and/or might be left the same. Mistaken vs. inaccurate vs. debatably accurate ... And there's no hint I've seen, here or anywhere, of the death toll including duplicates. 

Nor is there any sign of a change to the cause aspect Hitchens noted, just a timeline change to 28-days after infection, as already done in the rest of the UK, and broadly elsewhere. Currently, PHE explains that even within their revised, inside-28-days death count for Covid-19 patients: "The actual cause of death may not be COVID-19 in all cases." 

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/908781/Technical_Summary_PHE_Data_Series_COVID-19_Deaths_20200812.pdf

But they must be hit by a bus, etc. WITHIN 28 days of a CONFIRMED infection? I have a query in about this point - just what types of deaths do they count - literally all, or just ones that at least might relate, or maybe something even more stringent? If I get any answer, I'll add it here. 

Update 9/3: I did. query to coronavirus-tracker@phe.gov.uk:  

Greetings. I'm Adam Larson, a researcher in the United States. I'm trying to follow the Covid-19 pandemic and sort out issues of confusion. One of these is the recent revision of PHE England tally of related fatalities. Some recent commentaries made it sound like a bunch of unrelated deaths - "even if they had a heart attack or were run over by a bus three months later" - were scrapped. But official sources and explanatory articles don't mention any change to the cause aspect, just a timeline change to 28-days after infection, as already done in the rest of the UK. Currently the website explains "The actual cause of death may not be COVID-19 in all cases" even inside that 28 days. Statistically, that should be the main killer in such a short span, but ... I didn't find a fuller explanation of what kinds of deaths are included. Is it, as understood, all causes, including car crash, etc.? Or does it only consider likely-related deaths like pulmonary complications, and maybe things like heart attack? 

answered:

It’s all deaths of people who have tested positive within the previous 28 days. So, yes, it does include people who died in a car crash but no Covid-victims who were diagnosed 29 days or longer ago.

Clarity pending, and as I read it, Hitchens is, It would be wrong to suggest the cause of death is much clearer now, and Hitchens makes no note of the 28-day cut-off that was actually changed, suggesting he doesn't even understand the issue. That does give less time for other things to kill people, but the degree is too unclear to presume it must be huge. 

I took a look at this "brilliant forensic work" Hitchens cited. 

Carl Heneghan and Yoon Loke, Center for Evidence-Based Medicine, July 16

"...it seems that (Public Health England) regularly looks for people on the NHS database who have ever tested positive, and simply checks to see if they are still alive or not. PHE does not appear to consider how long ago the COVID test result was, nor whether the person has been successfully treated in hospital and discharged to the community. Anyone who has tested COVID positive but subsequently died at a later date of any cause will be included on the PHE COVID death figures." 

From the Quoted official explanation ("when a patient dies, the NHS central register of patients is notified (this is not limited to deaths in hospitals). The list of all lab-confirmed cases is checked against the NHS central register each day, to check if any of the patients have died.”) that might seem like a stretch. But it seems to be the case, as far as I can tell. They admit "The actual cause of death may not be COVID-19 in all cases."

The article muses "By this PHE definition, no one with COVID in England is allowed to ever recover from their illness." In fact the UK has always listed 0 people as recovered - I suppose it's some kind of definition issue, related to England's long-term view at the time.

"A patient who has tested positive, but successfully treated and discharged from hospital, will still be counted as a COVID death even if they had a heart attack or were run over by a bus three months later."

One of those causes is worth keeping track of - people who "recover" from covid, or get it with no one knowing, seem to die from strokes, heart attacks, diabetes, etc. at alarming rates within a short span of getting the virus. This might be why 'recovered' survivors are dying at 8x the usual rate.

"One reasonable approach would be to define community COVID-related deaths as those that occurred within 21 days of a COVID positive test result."

I'd suggest the 28-day cutoff they use is already too short to catch the full picture. What would be best is to filter by cause to eliminate unrelated deaths from possibly and clearly related ones, with a longer time limit, if any.  

"...there is a persistent pattern where ‘out of hospital’ data are contributing hundreds of additional deaths to the daily figures, and this shows no signs of letting up. Indeed, there were >100 non-hospital daily deaths recorded on 14 occasions over the last 30 days."

This isn't just a problem causing high numbers to appear - it also means many people are dying not long after their supposed recovery-discharge point. 

Data Review

What, when, how, and where do they count?

What: a death following a confirmed Covid-19 infection.

When: within 28 days, leaving covid survivor a pretty high-risk distinction. 

How (cause of death): pending clarity, this includes any kind, including totally unrelated. By the scale of things, the bulk of later deaths seem related, but not all of them. Clarity would be good.

Where: at one point, just patients dying in NHS hospitals were counted. In late April, a huge number of deaths in care homes were included, and new ones were counted from there. Two other categories: "home" and "other" were tallied by the Office for National Statistics, but less so by the regional Public Health services. It seems that currently in Wales, "hospitalised patients in Welsh Hospitals or care home residents" are counted, but not ones at home or other. In England and Scotland, deaths "in any setting" are counted. In Northern Ireland: unclear.)

Graphic: ONS count including all settings, as PHE and PHS claim to, had a count of 51,710 "deaths involving COVID-19" just in England and Wales, up to 31 July vs. 46,193 for all UK on 7/31 (basic PH number as listed on the Bing tracker). Diff = 5,517, plus Scotland and Northern Ireland difference. I'm still not sure what that reflects.

Official tally revised to 28-day cutoff, redone in shades of red  (source: https://coronavirus.data.gov.uk/deaths)

Now that overlaid on the ONS graph ... and it does seem the BBC plotting has the zero line drawn at more like 10 - it needed to be shifted up like that in order to fit, and the last tolls here are more like 10 than zero (shortly after this dropping to about 6 or 7 most days). This is a bit confusing, specially with two different but similar-looking 4-bar systems at work.

A simpler view with takeaway (and note the pink difference might also include some other difference, like location of death):

The lower and near-zero official death rate is lauded as a sign Covid-19 is no longer harmful, or that the UK is now "herd immune," or at least that they have the tricks down of "protecting the vulnerable." It's the latter, except it means vulnerable to dying VERY QUICKLY; a death 29 days or more after infection just doesn't count. But as we can see, the current bag of tricks is not doing as good a job of protecting those vulnerable to dying from it a bit slower.

Now for excess deaths - what may be quite a bit of the same thing, absent the verified infection. This issue is widely noted in the mainstream media, professional circles, and some governments as an indicator of the true death toll. Wherever Covid19 strikes, there's an unexplained surge of above-average deaths (causes often unclear but likely related when known).  Some places like Belgium and New York seem to have a fuller (if more speculative) counts that include likely/or presumed Covid deaths, while other places (especially densely-populated and poor cities) can't keep up with cases or deaths, and have several times what's suggested by their massive and still-incomplete excess fatalities - if they compile them at all. My main points and work on the subject.

The UK is not unusual in having something like 50% more than predicted emerge on analysis (I had found around 70,000 total was likely, based on a graph at the Economist with a crude baseline adjustment (explained there, but even better below), compared to a smaller tally than I'm using now, to get about the same proportion anyway.) That issue is likely playing out in every hard-hit country, but maybe to a lesser degree in most places. An analysis by the Guardian found the UK had the biggest surge in excess deaths of 11 nations examined (10 European nations including Italy and Spain, and the United States) The UK surge was barely worse than Spain's AND included the unusual England count now under review, so that whole contest could be re-called. How DO Italy and Spain do their counts?

Consider this chart of 5-year fatality rates in UK (as seen in the weekly reports found here). 

The downward revision let me see an interesting pattern: covid survivors (past 28 days) have been dying at about 8x the usual rates for England. That highly suggests most of these past-28 deaths are from its damage, and supports the possibility of the same happening, at a lower rate, among the vast number of unconfirmed infections out there.

To establish the average death rate on the smallest scale possible, I found this government site allowed that. I grabbed the weekly reports for 2020 (which include evolving versions of that 5-year graphic), and for the same weeks from the previous three years for comparison (previous years are linked at the bottom). 

https://www.gov.uk/government/statistics/weekly-all-cause-mortality-surveillance-2019-to-2020

As an example of the given wording, the final week (Aug. 10-16) says "In week 33 2020, an estimated 9,392 all-cause deaths were registered in England and Wales," plus other details. For each week I compare that (notedly provisional) number, vs. a 3-year average for number of deaths in the same week, 2017-2019. For week 33, it's 9,074 (week 33 2019: 9,093 - 2018: 8,830 - 2017: 9,299). Comparing the expected average to what happened gives an excess of + 318. (then a daily portion, for plotting - total divided by 7 = +45). This is the first week in a while to be above average - likely due to a heat wave, but as I'll show, It might be a bit deadlier than the same heat wave if it had come last year, before Covid-19 and all the human wreckage it's caused.

My own three year average is probably a bit higher than a longer average would be, including the bad flu year 2017-18, with the year before also looking a bit high, the one after normal-low - see chart above. But for some weeks it may be low (2018 has some weeks missing, so it's a 2-year average there, skewed less deadly. In retrospect, a 5-year average would be better, but this is it for me for now).

Quickly, all my weekly excess vs. average findings, with yearly samples in just some cases (tedium), starting at week 8:

week 08 to 2/23: an estimated 10,841 all-cause deaths vs. 11,744 3 yr. av. (2019: 11,295 - 2018: 12,142 - 2017: 11,794) Excess: -1,263. Daily share: -180

week 9 (to 3/1): 10,816 vs. 10,985 3ya. Excess: -169  daily: -24

week 10 (to 3/8): 10,895 vs. 11,657 3ya. Excess: -762 daily: -109

week 11 to 3/15: 11,019 vs. 11,351 3ya. Excess: -332.  Daily: -47

week 12 to 3/23: 10,645 all-cause deaths vs. 10364 TWO year average (2018 missing). Excess: -190. Daily: -27

week 13: 11,141 vs. 10,243 2ya (2018 missing). Excess: +898. Daily: +128

week 14 (4/5): 16,387 vs. 10,033 2ya (2018 missing). Excess: + 6,137. Daily: 877

week 15 (4/12): 18,516 vs. 9,392 2ya (a diverse 10,291 in 2019 vs. 8,493 in week 15 2017) or better, 10,250 from another try with monthly totals 5 years back - April divided into 7/31). Excess: +8,838 (using the latter number). Daily: +1,263

week 16 (4/13-19): 22,351 vs. 9,640 2ya (2018 missing) excess: +12,711. Daily: +1,816

week 17 (to 4/26): 21,997 vs. 10,424 3ya (2019: 10,059 - 2018: 10,306 - 2017: 10,908) Excess: +11,573. Daily: +1,653 

week 18 (5/3): 17,953 vs. 10,141 3ya. Excess: +7,812. Daily: +1,116

week 19: 12,657 vs. 9,457 3ya (2019: 9,055 - 2018: 8,624 - 2017: 10,693). Excess: +3,200 - daily: +457

week 20 (to 5/17): 14,573 vs. 10,234 3ya. Excess: +4,339 - daily average: +620

week 21: 12,288 vs. 9,987 3ya. Excess: +2,301 - daily: +329

week 22: (5/30): 9,824 vs. 3ya 8,246 (8,260 in week 22 2019 - 8,147 in 2018 - 8,332 in 2017.) Excess: +1,578 - daily: +225

week 23: 10,709 vs. 9,952 3ya. Excess: +757 - daily: +108

week 24 (6/14): 10,709 vs. 9,385 3ya. Excess: +1,324 - daily: +189

week 25 (6/21): 9,339 vs. 9,447 3ya. Excess: -108 - daily: -15

week 26 to 6/28, 8,979 vs. 9,352 3ya. Excess: -373 - daily: -53

week 27 to 7/5: 9,140 in vs. 9,194 3ya (2019: 9,062 - 2018: 9,258 - 2017, an estimated 9,263). Excess: -54. Daily: -8

week 28 to 7/13, 8,690 vs. 9,283 3ya - Excess: -104 - daily: -15

week 29 to 7/20: 8,823 vs. 9,107 3ya (2019: 9,080 - 2018: 9,127 - 2017: 9,113) Excess: -284. Daily: -40

week 30: 8,891  vs. 9,045 3ya. Excess: -154. Daily: -22

week 31 to (to 8/2) 8,946 vs. 9,124 3ya Excess: -178. Daily: -25

week 32 (to 8/9), an estimated 8,945 vs. 9,160 3ya. Excess: -215. Daily: -31

I plotted these in lavender-blue behind the same confirmed curves and to the same scale. That's based on daily tallies, so I take the weekly excess divided by 7 and projected evenly over a 7-day block for comparison. And in the backdrop I show confirmed cases (orange) and possible unconfirmed (gold). Pale, blurred, and not to vertical scale, these are just for a basic idea on the timeline why there might be unknown covid deaths this far back - starting from two confirmed cases on 2 February that had been listed at the Bing tracker 'til mid-April, but don't don't appear now. We'll come back to this shortly.

These  weekly totals can't be read too literally, especially with unexplained dramatic variance, they may reflect lags in reporting where one week is short and the next catches up. Especially at the peak, where the system would be overwhelmed, the greater number probably died in week 15, but far more come out under week 16. But there is also a rise of excess deaths then, and the past-28 day covid deaths are proportionally higher in that same time, the week or so after the main hospital-based peak. So maybe it did add up to more dead in week 16. Unless week 15 carried over to 17 as well, it seems there was a lot of ongoing death in all three weeks, much of it not captured in the official toll.

Again, these totals refer just to England (pop. 55.9m) and Wales (pop. 3.23m), so missing a similar portion for the smaller populaces in Scotland (5.5m) and Northern Ireland (1.908m). I find Scotland has 9% of the combined E+W pop. and NI has 3% E+W. That's 12% between them, but both likely have lower rates of this problem than either England or Wales, so I'll say add a combined 10% addition is a fair estimate, wherever that comes up.

All UK est. excess. from 2/24 (all + added and - subtracted) = 58,863 (E+W) + 10% (est. S+NI) = 64,749 above-average deaths, likely vast-majority covid-caused. This basic number has been widely reached and reported on as the best benchmark for how many the virus has actually killed in the UK. Excess deaths past 65,000 was in fact widely reported on June 23, while I only get that as of August 16. I think they start counting later, include less of that pre-infection low I include, since I'm also considering a "baseline adjustment" that plays into.

Baseline Adjustment: Some have noted a likely or actual added drop in non-covid deaths, and I think this is worth considering - the applicable baseline for non-covid deaths isn't just average but apparently well below that. See UK 5-years mortality graph above, where the past 4 years had NO Covid-19 deaths, as it didn't exist. Note the huge spike this year is preceded and followed by dips below average, the former being quite sharp, and the latter managing this despite the ongoing covid deaths, so all other causes combined must be well below the 3-year norm. (The same dips can be seen in more detail in my graphic.) Consider the yearly hump in winter is mainly down to influenza, etc. and these deaths tend to fall as soon as people start distancing (+ usual infection-death time lag). Deaths from traffic accidents and violence will be down when people stay home. Some other classes will fall and others will rise, on balance coming out below average.   

So if the applicable baseline is below average/zero, just counting the deaths above that line will miss the others between it and the actual baseline.

The exact "actual baseline" can't be told without more detailed data, but I tried a reasonable guess, connecting the lows before and after the spike with a little inference. The early dip in late January may be related; it seems like flu deaths were slightly above average, but just as news of the spreading virus in China came in at least, may have the most vulnerable groups being extra careful and already preventing many flu deaths. A sharp drop is noted the week Jan 20-26, just as Chinese authorities admitted what many suspected - Covid-19 was communicable and might have spread anywhere by now. Actual cases soon appear in the UK (the redacted ones of Feb. 2 would be reported - looking it up, on January 31 in a mother and son from China were confirmed and noted on January 31, and a woman in Nottingham "may have caught it as early as February 9" (community transmission) (DailyMail) A Wikipedia article notes earlier likely appearances in mid-December (via China) and mid-January (via Austria). 

This might be enough news to sharpen the same point and explain a month of lows with a big dip in week 8 (Feb. 18-23) to more than 1,200 deaths below average, quite possibly despite a few early covid deaths. It's even possible these were happening earlier yet, in that apparent spike of flu deaths already going as the year started. But seeing it dip even lower when thing WERE heating up, then seeing it quickly pushed up from week 9 forward, as if in an immediate prelude to the coming mega-spike - well, that seems like a good spot to start counting likely covid deaths above that low baseline, which I set by roughly averaging weeks 7 and 8 (noting that weekly variance). 

The pink baseline I drew in here may start too low, but upon lockdown declared on March 23, following distancing advice by March 16 and general warning preceding that, deaths should reach a low similar to what's shown - around 160/day less than average. In many places, the shock of the orders makes for wide compliance for about a week. From there it would creep up as people realize the world wasn't obviously ending and start moving about more. Some get the idea and find a safe balance, while others increasingly question, protest, and defy mitigation efforts. Some of them will die, but at a much, much lower rate than they were about to.

Also in play are the negative effects of lockdown - lost wages, suicides and overdoses, canceled medical care, etc. will have death tolls that variously rise and fall over time. But considering how it all comes out below average so far, these don't seem to add up to very much yet. I don't know how much to add or where, except mainly it's not much, and takes a while to emerge - "lockdown" won't explain the sudden and massive spikes of death just as the virus was taking off and only starting to come under control. That will be the virus, people.

Anyway, after drawing in the new guessed baseline, I would draw a light blue box just a bit smaller than the distance to the 0 line or below-average total. Since most tallying here is done above zero/average, I add these to the top of the blue bar to clarify it's a proposed addition. In context, the additions seem reasonable -we already have so many deaths above average, it shouldn't be too surprising there are even more. I wouldn't be surprised if my adjustments were too big, or too small. But until such knowledge exists, this seems like a good guess, and either way, the bulk of it, some 65,000 deaths above regular average, still have no better explanation.

In essence, just as the confirmed cases yield these past-28 days fast-dying survivors, and the bulk of other excess deaths are the equivalent from the vast number unconfirmed cases. Those will tend to be milder cases than the known ones (largely found in hospital upon admission for extreme symptoms). But it's only natural the same kind of damage occurs with the unknown cases, probably in a much lower proportion, but one that still adds up. 

Estimate: ~80,000 Total Deaths

Adding all the plusses and subtracting the minuses for weekly deaths: 

All UK est. excess. from 2/24 to 8/16 = 58,863 (E+W) + 10% (S+NI) = 64,749 most of them likely caused by Covid-19. 

Baseline adjustment all UK, 2/24-8/16: 18,249 (E+W) + 10% (S+NI) = 20,074  total deaths *below the zero line*, most of them likely caused by Covid-19. 

Total est. excess, all UK 2/24-8/16: (64,749+20,074) = 84,823

Besides the actual baseline, the actual Covid-19 portion of that huge total is harder to call. I mean "most," and I'm including deaths the virus only had a direct hand in. I'd say a conservative minimum is 80%, and I suspect 90% is more like it, and will give a broad range of 80-96%, personally favoring the upper half of this range:

if 80% = 67,858

if 90% = 76,341

if 96% =  81,430

This allows me to offer a rough estimate of around 80,000 people actually killed by the virus - one way or another, with or without the help of CO-morbidities (which had been just conditions until the virus transformed them). That's across the U.K. and all settings, but only so far. That's at least 64% and as much as 97% above what's now counted (41,410 at 8/16). It comes to at least 1,005 dead per one million populace (or 0.1% dead) and as much as 1,185/m, which would be one of the highest national death rates in the world.