Excess Deaths
In our travels, we have been asked if we have been able to track excess deaths due to COVID-19. This is a very good question!
But, there’s a problem!
This is how WHO et al define Excess Deaths:
Or, if you’re looking at the figures at the end of May, then:
And this is what the ONS publishes as an Excess Deaths chart using this definition:
Can anyone spot the problem?
Now this may sound a bit callous, but we specialise in time-series analytics. The graph above causes us all to express concern that COVID-19 has caused significant excess deaths week on week from around week of 27 March compared with a smoothed 5-year average. See the problem?
In fact there are several problems with this:
1. You’re comparing week by week data (which implicitly contains variation) with 5-year smoothed data. To address this, we’ll use the consistency of time-series analysis for deaths over the past 5 years, as well as the deaths for 2020, and see what that tells us.
2. We can then see what the distribution looks like around the mean, and establish whether the data since 27 March is in fact a significant number of standard deviations from the mean
3. And now, the more controversial observation: We know that many (not all) people who died with COVID-19 had underlying health problems – who’s to say that they would not have passed away anyway from these problems – if not during the March – May period, then perhaps later this year. So, to address this, we should hold fire on making statements about excess deaths until at least the year end or possibly until March 2021? Then we can see if the death rate runs lower than the 5-year average (or against the mean if we use time-series analysis) after COVID-19 deaths, specifically, tend to 0 (possibly later in July?).
4. Looking carefully at the chart above, another question arises – why do Non-COVID-19 Deaths also rise during the same period – March – May?
In this article, we’ll address points 1 and 2, and in a later article we’ll look at the other points.
This is what the deaths data looks like over 5 plus years in England & Wales:
The blue points are the actual weekly results, the green line is the seasonal average calculated from all the points for the 5 years to January 2020, the upper and lower red (guide)lines give us a clue about the variation around the average (approx. 3 standard deviations), and the points in red, green and grey are points outside the norm (red being unusually high, green being unusually low, and grey being so far away from the mean that it is highly likely a major event happened).
Observations:
1. It’s remarkably stable – the pattern repeats consistently
2. More Deaths in Winter than Summer (not surprising, we kind of knew this anyway)
3. The extreme lows correspond to Public Holidays, so, no, “making it Xmas every day” won’t actually change anything!
4. The spike 23 December 2016 appears to be because Xmas Day fell on a Sunday (and not particularly bad weather at that time), whereas it was on a weekday in all other years (impacts registration of deaths).
5. The two red spikes in March 2018 were as a result of the so-called “Beast from the East”
6. Deaths kick off in anger (well above 3 standard deviations) around week 01 April 2020 and return around week 27 May 2020, largely due to COVID-19
Now we need to wait until December to see if the death rate will have run consistently below the seasonal average for 2020! Right now, we’re still above average – we’ll report back with more news later this year.