ARIMA modelling of age & sex standardised monthly non-COVID mortality 1970 – 2021
Things that go bump in the night (rev 1.0)
Back on 6th June in this newsletter I mentioned my rummaging about in ONS’ pantry for data that enabled me to calculate non-COVID age & sex adjusted mortality (a.k.a. AGAM). A couple of slides were produced and we compared this time series with that for all cause death to reveal the impact of COVID-19 in Apr 2020 and Jan 2021. This morning in an idle moment I ran an ARIMA model over the non-COVID time series, the results of which left my eyebrows on the ceiling.
Whilst we may expect the outlier months of Apr 2020 and Jan 2021 as folk fell foul of SARS-COV-2 peak infection we don’t expect outlier months for folk not falling foul of SARS-COV-2; it should be the sorry business of death pretty much as usual, and especially so if we assume lockdown and other factors (like lack of routine healthcare services and closure of primary care) didn’t really kill people.
Well that ain’t necessarily so, sang Sportin’ Life, the drug dealer in Gershwin’s Porgy and Bess.
I say this because in that idle moment I quasi-accidentally ran an ARIMA model for non-COVID mortality that revealed an unexpected increase of 13.7% (p=0.006, 95% C.I.[3.7%; 24.6%]) for Mar 2020 and an unexpected increase of 27.1% (p<0.001, 95% C.I.[14.6%; 40.9%]) for Apr 2020. In plain English folk other than COVID-19 cases were suddenly dying in bigger numbers during these two months over and above that which may be expected.
I attach a slide from the accidental gretl session file so you may inspect these oddities for yourself – there’s a hump where there shouldn’t be one! Just after 2020 you’ll see a modest blue hump for what was anticipated and a larger red hump for what was observed (that I’ve marked with a dirty great arrow). So what was suddenly killing all these folk in Mar – Apr 2020 if it wasn’t COVID?
Eagle-eyed subscribers will also have spotted the discrepancy between forecast mortality (green blob with 95%C.I.) and observed mortality for Dec 2021 (red dangler). This is almost certainly due to delays in processing death certificates, but what puzzles me is the deaths were counted by ONS on 29th April 2022, which surely permits sufficient time for them to catch up on paperwork. I have therefore submitted a pleasantly-worded FOI request to the ONS for the latest death count by quinary age group and sex.
Paid-up subscribers will be treated to the revised gretl session file so they may inspect the latest model, slides and data for themselves.
Kettle on!