Excess Death Figures: Further Considerations (part 2)
Excess death has become a popular yardstick for assessing the impact of COVID, government policies and COVID therapies but this method is brimming with issues. In this article I try a different tack
In my 5-part series entitled Excess Deaths by Cause, England 2020/w1 – 2022/w46 I provided a series of slides of weekly excess death by ICD-10 chapter by age band for the top five frequent causes of death, which together accounted for 79% of all deaths over the period 2014 – 2022. These were derived from a rather handy date-of-death dataset obtained under FOIA from the Office for National Statistics (ONS) by Joel Smalley. Despite my grumbling I used the method adopted by the ONS in which a baseline estimate for weekly deaths is calculated from prior 5-year means, the exception being use of 2015 – 2019 weekly means for derivation of excess deaths for 2021 and 2022 as well as for 2020 in order to minimise bias that would be introduced by using the pandemic period as a reference for ‘normality’.
In my article entitled Excess Death Figures: Further Considerations (part 1) I tackled just one of the issues plaguing derivation of excess deaths by deriving standardised weekly counts based on the age profile of the population of England using 2019 as the reference year. In doing so we saw a reduction in excess death for some age groups and conditions with little to zero change for others, depending on whether the sub-population was increasing in number or changing very little over time.
Plat Du Jour
Today I want to tackle another issue and that is use of a 5-year mean to represent a baseline value. This makes sense if things are not changing over time but this is rarely the case with disease and healthcare provision for the treatment of that disease. For example, if death from diseases of the circulatory system has been declining over time owing to better diagnostics and improved therapies then a 5-year mean is going to yield an inflated baseline, thus reducing estimates of excess death. Equally, if death from diseases of the circulatory system has been increasing over time owing to closure of services and a rise in obesity then a 5-year mean is going to yield a conservative baseline, thus increasing estimates of excess death.
Projections, My Dear Watson, Projections!
Whilst all this sounds like a right pain what we can do is dump the rather inadequate 5-year mean approach and have a go at using projections instead. Yes, dear reader, this means statistical modelling. Yes, it means we can make a pig’s ear of matters and obfuscate if we want to. Then again, with the wind in the right direction, we can make a jolly good go of things and produce estimates that have more meaning; estimates that give a nod to all those awkwardly changing factors. Yaroo and break out the buns!
In preparation for this I decided to hang a right and zip round the corner. I sought an intuitive graphical representation of what fancy time series/linear regression modelling would bring to the table, and so I invented something called mean mortality.
Mean Mortality
Mean mortality is simply the arithmetical mean of all the mortalities derived for the ten age groups separately (under 1, 1 – 17, 18 – 29, 30 – 39, 40 – 49, 50 - 59, 60 – 69, 70 - 79, 80 – 89, 90+). This strange and wonderful calculation implies a theoretical population – an idealised population, if you will – that consists of equal numbers of age each band that is invariant over time. Another way to look at this is that we’ve standardised the data such that any trends we see must be due to changes in everything other than population dynamics.
So let’s have a look at mean mortality for the five most frequently recorded causes of death for the period 2014/w23 – 2022/w46: