Age & sex factored mortality 1970 - 2021 (part 1)

England & Wales monthly mortality rate by quinary age band and sex: males under 45y

Background

We now get to the heart of the matter and the reason for my crunching population counts these last few days. I’ve been sitting on monthly counts of death by occurrence by quinary age band (18 bands) and sex for England & Wales for the period 1970 – 2021 that I’ve cobbled together from various FOI requests to the Office for National Statistics (ONS). This bit of veggie prep was summarised in my recent roundup newsletter.

It should be obvious to subscribers by now that although the ONS prefers to provide updated tabulations of deaths at registration on a weekly basis these figures are not much use when it comes to analysing trends owing to delays in processing that may be significant. In private correspondence with a group member an officer states…

Legally a person has 5 days to register a death with the General Register Office. These registrations come through to ONS on a daily basis where they are loaded on to our system for validation and coding. Some deaths are registered weeks or even months after the date of death; see our report on registration delays here which gives more detail on why these delays occur.

Bold text is my doing. What I find startling is that the ONS do not provide updated tabulations of deaths by date of occurrence as a matter of course - analysts have to rummage amongst FOI (user) requests. Such rummaging has provided me with one of the two key ingredients – the number deaths when they actually occurred.

Monthly estimates of population by quinary age band and sex for England & Wales for the period 1970 – 2021 are the second key ingredient and I’ve revealed a little of my statistical paddling below the surface in two recent newsletters in the series ‘Toward Age-Adjusted Mortality’.

Putting these two ingredients together means we can derive mortality rates per 100k of the same target population; that is to say I can divide the number of 50 – 54y male deaths by the estimated number of 50 – 54y living males to arrive at an age and sex-factored mortality rate based on actual date of death. This is like age-adjusted mortality but on steroids! Not only that but we can also crunch this calculation for every month running from January 1970 to December 2021. As cakes go this is pretty yummy.

With that background sketched out we can now turn to eyeballing a set of (hopefully) valuable slides, there being 36 in total (18 age bands by sex). I’ve split these into 4 groups of 9 to avoid hitting the substack newsletter limit. Get the kettle on, grab the biscuit tin and feast your eyes on this lot…

Time series for males 0 - 44y

We start with the youngest group of males (0 – 4y) and note a steady decline in mortality since 1970 from around 40 deaths per 100k 0 - 4y males to around 6 deaths per 100k. We may note that the vertical spread of points gradually lessens over time which means seasonal variability is in decline as mortality also declines, this being a fine example of something statisticians call heteroscedasticity. The wiggly green line is a Gaussian function that smooths seasonal variability and serves to guide the eye when scatter gets really wild. Disease outbreaks in a monthly time series like this are going to be marked by outlier points, though in this instance there are no extreme outliers to observe, the best example being the elevated mortality rate of 24.76 deaths per 100k 0 – 4y males for Dec 1989.

A similar steady decline is observed for 5 – 9y males, though please note the greatly reduced values on the y-axis which now expand to just 6.0 deaths per 100k 5 – 9y males instead of 50.0 deaths per 100k 0 – 4y males. Again we observe significant heteroscedasticity. This time there are a few choice outliers marking poor health in years 1971, 1981, 1988, 1989, 1994, 1997, 2009, 2011 and 2017, these likely being disease outbreaks. What intrigues me are outliers pointing to unusually low mortalities for certain months, for example a mortality of 2.09 deaths per 100k 5 – 9y males during February 1970 - I am sure we can learn just as much from these!

The steady decline in mortality for 10 – 14y males over the period 1970 - 2021 is almost perfectly linear with a great deal of seasonal scatter and modest heteroscedasticity. We observe an extraordinary outlier of 3.84 deaths per 100k 10 – 14y males for August 1984 which is most unusual and I wonder if excessive heat or conditions arising from an extended dry spell have played a role in this; for example, hot weather leading to an outbreak of pathogens in the food or water supply. The elevated mortality of 1.38 deaths per 100k 10 – 14y males for February 2019 is also worthy of note.

This slide for 15 – 19y males over the period 1970 – 2021 holds great interest for this is the age band where the impact of employment, driving, drinking, riding and other risk-laden activities first comes into play. We observe a steady decline with heteroscedasticity as before, but we also see two wiggles centred on 1986 and 1994 when mortality took a broad dip. I am wondering if these are related to legislative changes more so than healthcare policies or disease dynamic because of the broadness of the effect. In the same vein there is a broad kink around 2012 that serves to halt what appears to be a steady decline in mortality; this kink pushing savings into stagnation at around 2.5 deaths per 100k 15 – 19y males. I am certain this is the sorry impact of the Health & Social Care Act (2012). The outlier of 8.87 deaths per 100k 15 – 19y males for August 1987 stands out like a sore thumb, and once more we have this curious height of summer death blip, which could be as simple as an increase in motorcycle accidents as young lads rip up the hot tarmac on a glorious day. Then there’s the issue of drowning and risky activities where lads can come a cropper. A rather curious feature of this series is the sharp decline in mortality during 2021. Is this a genuine and sudden decline in mortality for 15 – 19y males or are we missing some deaths for this age category?

This is where things start to get really interesting! Gone is the steady decline over the period 1970 – 2021 and in its place is pretty much a plateau from 1970 to 1998, after which we observe a rather dramatic decline. I can only imagine this is due to broad legislative or policy changes that impact particularly on this age group, and I’m thinking along the lines of knife laws, licensing laws and laws pertaining to road vehicle use. The wiggles around 1983 and 1994 would seem to correspond to the wiggles for 15 – 19y males, which suggests broad reforms of one sort or another. Policies put in place after 1998 certainly lessened risk of death for this age group right up to the kink of 2012 where we find evidence of an upturn in mortality following introduction of the Health & Social Care Act (2012). Particularly noteworthy is the dramatic decline in mortality in the latter months of 2021. This looks artefactual and I can only conclude that we are missing youthful deaths that are still to be counted. Also noteworthy is the lack of outliers during the pandemic period.

This group of males is only 5 years older than the previous group yet we see something very different happening - mortality increases after 1984 to a peak around 1998 instead of wiggling. These are broad changes and could be changes that are indicative of poor policy making, closure of healthcare services, increased likelihood of risky behaviours and changes in various laws. Somehow the light dawned on somebody after 1998 and 25 – 29y males enjoyed improving mortality right up until 2012. There are no outliers marking the passing of the pandemic peak and the artificial decline in mortality for the latter months of 2021 is quite pronounced – just how many youthful deaths are missing from ONS figures?

Mortality for 30 – 34y males pretty much follows the previous pattern of sweeping reforms that make things better then worse then better until the NHS catastrophe of 2012. This time we get to see two notable outliers during the pandemic, these being a mortality of 8.97 deaths per 100k 30 – 34y males for March 2020 and a mortality of 8.85 deaths per 100k for January 2021. That being said anybody would be hard pushed to eyeball this data series and spot the pandemic if they arrived from Mars this morning – my best guess would be 2018 if I didn’t know any better! The tell-tale tail for 2021 once again reminds us that the ONS haven’t processed all youthful deaths as yet.

We arrive at the penultimate slide in this newsletter find a somewhat compressed data series for 35 – 39y males. Whatever those sweeping changes between 1984 and 2012 were for the younger age groups they clearly did not impact so much on 35 – 39y males. The 2012 catastrophic kink in service provision is very much evident as is the hole in data processing for male deaths during 2021. Yes, we have a couple of pandemic outliers during 2020 these being a mortality of 12.13 deaths per 100k 35 – 39y males for April 2020 and a mortality of 12.37 deaths per 100k for January 2021. What we should not ignore is the third outlier of 11.93 deaths per 100k for December 2020 - just when was the pandemic supposed to start?

We arrive at the last slide for this newsletter to discover yet more compression for 40 – 44y males. Gone are the mid-series kinks and wiggles of the younger generations, which suggest these are a peculiarity of the youthful male. These may have arisen through changes in behaviour, circumstance, law, policy and health service provision; we have no idea. The pandemic now starts to shine with two notable outliers of 20.30 deaths per 100k 40 – 44y males for April 2020 and a mortality of 21.03 deaths per 100k for January 2021, these peak periods being consistent with those for 35 – 39y males. My eyeballs suggest that there are holes in the ONS data but perhaps not as many – a touch of ‘hole’ modelling may be in order once all four parts have been published.