Spectral Analysis Of Certified COVID Death (part 1)

I utilise an engineering tool to investigate periodicity for certified daily COVID death within England for the period 30 January 2020 – 6 October 2023

For my final fiddle with spectral analysis I fancied pulling down new daily certified COVID deaths for the nation of England from the UK GOV coronavirus dashboard. You can obtain the data yourself by working through the menus or you can take advantage of the API programming language that is used for the download section. Here’s the API code I used to shovel the latest data to disc with a minimum of fuss that you can copy into your browser:

https://api.coronavirus.data.gov.uk/v2/data?areaType=nation&areaCode=E92000001&metric=newDailyNsoDeathsByDeathDate&format=csv&release=2023-11-02

I’ve covered what this count means in several articles so will just quickly mention that we’re talking about daily deaths by date of death for those folk with mention of COVID-19 somewhere on the death certificate (section I or section II). These are thus counts of deaths with COVID as well as deaths that have been declared to be due to COVID. Though rather dubious PCR test results will have dominated matters these will have not been the only consideration for certifying physicians. We should note that a positive test result is not required for the purposes of certification since ICD10 emergency code U07.2 (originally defined as COVID-19, virus unidentified) can be invoked for suspected cases, this being a catch-all code to cover a clinical diagnosis (as opposed to a lab diagnosis). Surreal as it may seem we need to think of these counts in terms of deaths with suspected COVID (whatever this means). Coming from a clinical audit background I am tempted to utter some rude words.

That’s my Monday morning rant done so I guess I better make a start by getting the kettle on and plotting these surreptitious counts out as a time series:

Loud and not very proud is that first sizzingly sharp peak in spring of 2020 when everybody seemed to die synchronously regardless of age, sex, medical diagnosis, immune function, SARS-COV-2 infection, occupation and location – a strange fact that bothers me and certainly bothers the bods at PANDA, with this article being their latest thoughts on the subject.

Then we’ve got that monster 2020/21 seasonal peak during which influenza and several other bugs magically disappeared because of something called viral dominance that comes across to me like an exercise in hand-waving. But let us suppose all strains of influenza were indeed erased from the population in some way – how can they then leap back from non-existence if they rely on human-to-human transmission to remain viable? This is an awkward question with sharp edges that puzzled Hope-Simpson in his seminal The Transmission Of Epidemic Influenza. He tried to answer it by assuming humans are perpetually carrying the full complement of possible flu bugs tucked away in our being somewhere and that we have done so since early on in our evolution. A brave attempt at salvaging a grand virology puzzle, methinks!

After those twin peaks of incredibly uncertain death we’ve got some sort of viral shuffle that hints at periodicity on a scale we wouldn’t expect. Did SARS-COV-2 really do a regular 14-week smack-the-humans dance for a while or was something else going on?  All rather queer if you ask me so I guess I better run these daily counts through spectral analysis without further ado:

We are looking at a wall of spectral white noise, which is what we expect if deaths were occurring randomly over time. For sure we’ll get more deaths whenever disease prevalence kicks-up or when a rather nasty mutation takes the reins but these events are not going to impact on the date of death in periodic fashion. Those two grey dashed lines mark f = 0.071 (14-day periodicity) and f = 0.143 (7-day periodicity); absence of any spectral peaks reassures us that people were not being bumped-off to create beds or save cash.

Some readers might be wondering what spectral analysis yields for that viral shuffle period that runs from 2021/w21 to 2023/w37 or thereabouts. Try this:

There’s something going on down at the low frequency end that I’ve marked by a dashed line whose frequency fetches-up at f = 0.0112. Our trusty hand-held calculators will tell us that this translates into a regular periodicity of 89.3 days or 12.8 weeks. IMHO that’s mighty strange indeed, but we’ve already spotted the bumps and guessed 14-week periodicity using eyeballs alone; but 13 weeks it seems to be. This is not just strange but mighty peculiar for 13-weeks feels very ‘administrative’ – are we looking at evidence of deaths to order or something else that drove the daily count in a periodic manner for a while?

I’m sure that some readers will suggest periodicity arising from vaccination campaigns and/or delays between doses but there may well be other more prosaic reasons such as patient management (including end-of-life protocols) and hospital throughput. Ideally I’d now dial up various historic (pre-pandemic) daily series for cancer, cardiac and respiratory deaths just to get a feel for what dynamics are buried down in counts prior to the fascist shenanigans but for the time being that must remain a pipe dream.

Kettle On!