C19’s excess death conundrum

The 2nd wave has seen a return to the argument C19 deaths are not real, reported within 28days of a test, are all “with” not “from”, and a total hysteria… and I’m left aghast at how dim you have to be to look at all the data and conclude that.

The reality of the ONS data is that to wk47 they have attributed 63k deaths to C19, but the data has 65.5k outstanding excess to trend deaths, with the militant argument being the 2.5k difference, and more, due to the labelling problems, having been caused by the lockdown, not C19. In fact, at the extreme, that the entire 2nd peak is fake.

Now, ignoring that this fails to identify the real cause, NHS disruption from:

1. planning for C19 leading to lower capacity
2. fear of attending
3. or, being unable to attend as planning has met with reality leading to cancelling other treatments

The whole idea that C19 labelled deaths need to be below excess deaths, even to allow for non-C19 deaths, seems to me to be a failure of reasoning.

How?! You might splutter. Surely there can’t be more deaths?

Well, since March excess deaths have just jumped up, then not really reverted, with the tally of all categories of death, including C19, not adding up to the excess we are seeing.

The argument that militants sceptics make is the C19 tally as it stands contains many mislabelled deaths. Yes, they are mislabelled, but in the same way that being knocked down by a car is a mislabelling of the heart attack that was going to kill you 2yrs later if you’d survived.

The timing of the death is crucial. None of the comorbidity tallies were dramtically lower in Apr, but since they have run lower, and cumulatively they will end the year down due to C19 having taken some chunk of them, but crucially it did so by accelerating their death. Some by only 1 week, others by up to 35 weeks to keep them form this year, and many far beyond that.

Worse cumulative excess deaths are lower than total accelerated deaths.

What does that mean? Well, here are ONS deaths for England & Wales as a cumulative deviation for each year from their 5yr avg, having normalised to per 1m of the population. The plot also includes recent/outlier years for context.

Essentially, the faster the line climbs each week the more unexpected deaths are being rapidly added to the tally, bringing more deaths that week than the previous 5yrs would have expected. If it drops it means we’ve had fewer, which probably means a recent surge in deaths is leading to an undertrack in subsequent weeks.

Excess to trend deaths peaked wk24, but 2019/20 was on a similar trend to last year with it being a mild year, and by wk24 the year ends up 65k above that trend, ~1090d/1m for population of 59m. To wk36 this then decays by only 3.5k, showing very few to be expected, and now it is climbing again back up to 65.5k by wk47.

So this means accelerated deaths are at least the 65k to wk24, plus the another 6k wks35–47 means 71k have been accelerated? More so, this is just a very broadstroke assessment of this, the same effect is happening and compounding each week, so this figure is actually bigger.

The core problem is deaths have been showing themselves to be expected all along, and we’ve been unable to count them. Hiding in plain sight in the data.

e.g. imagine we get 20 unexpected (or accelerated) deaths each wk for 6wks (dark blue). In wk1 we’ll notice all 20, but what about wk2? Well it all depends how many of these accelerated deaths were going to happen over the next few weeks. Lets say that 1/2 of them, 10, were accelerated 2 from each week over the subsequent 5 weeks. The reality will be these wks will see 2 fewer expected, but if we carry on getting unexpected deaths we won’t clearly see this. So in wk2, we’ll have another 20 unexpected, but we’ll only count 18.

Now wk2 is going to have the same effect, so in wk3 will have 20 unexpected deaths, but 2 fewer because of wk1, and 2 fewer due to wk2, meaning we’ll only see 16 excess deaths. By wk6 we’ll only see 10 excess deaths (for C19 this was already happening at the 1st peak in wk17 of 2020). If the unexpected deaths suddenly stop in wk7 we’ll appear to have 10 fewer than expected deaths.

By wk12 expected deaths will be matching the avg again, and we’ll have an excess cumulative of 60, but crucially, without a test telling you the true unexpected deaths, you can never measure that 120 deaths were accelerated. The best guess you could make would be 90 at the peak of the cumulative measure.

The reality of C19 has been the 1st peak was more blurred than this, by wk19 nonC19 deaths were undertracking heavily, but still containing many untest C19, so we can’t see that the true undertrack was. The 2nd peak has made this clearer, with nonC19 deaths far more stabily undertracking, with this effect building up over time. None of this is denying the very real possibility that excess deaths were induced by the panic or denial of treatment due to NHS pressures, etc. They will have fuelled this effect too.

The real question being is there some way to estimate from the data the degree to which this was happening, ideally distinguishing C19 and nonC19 deaths at the same time?

Short answer is, yes.

Fortunately from a statistical perspective C19 does something very unusual, it takes M:F deaths at a ratio of 58:42. As per this analysis I did earlier in the year, if you look at deaths in the last 5yrs this is a very unusual, as typically female deaths outstrip male deaths over winter due to there being more older women than men, never the other way round. Men tend to die more evenly spread over the year.

Here is an updated version of the key plot from the analysis. The blue lines show the 5yr avg proportion males deaths are at that time of year, with the dotted lines showing one stdev. Some observations:

1. There are more men than women, period.
2. Male deaths seem to happen more in the summer month, with female ones more in the winter. Given women live longer the latter is likely more flu/weather induced 90+ death.

What the plot shows that we are more interested in, is that it is very predictable, and that 2020, in red, has seen a higher track in male deaths. Likely due to the mild year initially, or possibly when flying blind a background rate of C19 starting from week 6 onwards before we were even really tracking it. After all, we know it entered the country in Jan.

Certainly by wks11–17 it becomes very clear. What follows is a brief tiny undertrack wks18–22, which the study discusses is probably due to “catchup” deaths — i.e. excess male deaths in wks11–17 will have taken short term male deaths leading to a female skew in subsequent weeks.

The analysis does some blue-sky thinking, going on to work out how many deaths would be needed with C19's 58:42 signature ratio to produce this skew, estimating a— min < likely < max — range for these.

As per this plot, by wk24, the likely scenario suggests up to ~72k C19 deaths, a further ~6k from non-C19 causes, but crucially with ~13k of those deaths having shown themselves to be expected already by the peak, leaving the ~65k outstanding excess deaths.

In wks 25–47 we’ve added 14.9k more recorded C19 deaths, but we already know we’ve undertested. Far greater testing now means this problem is far smaller, let’s be conservative and use 1/4 of the error to wk24 and call it 17k, which means seen 89k C19 accelerated deaths.

It is also likely that non-Covid mortality will have increased, let’s say to 8k, which means — 89 + 8 - 65.5 = 31.5k — of these combined extra deaths were expected so far this year, twice as many as by wk24.

C19 does not kill those furthest from death and ignore those closest. What the “from” and “with” argument entirely subverts is any intelligent debate about how the case counts we have don’t explain the data. Any death accelerated by more than 1wk is a “from” not a “with” death. Sure, a more marginal one than someone who had 10yrs left, but not really a distinction that is very useful or respectful to make.

Bottomline, if you’re after a display of real control over the effect C19 is having on the population, demand the government find ways of explaining more of our excess deaths, not fewer. What the data means is that for the NHS to notice you have it, and react to stop you dying, you have to be far further from death than we think, which should have you worried for all your old relatives.

Ironically, if my estimate is correct, it means spread is higher than we thought, which might explain Tcell levels being higher, so why militant sceptics aren’t all over this story beats me.