Using ARIMA To Investigate COVID Death (part 1)

Cranking the handle on the latest daily data sitting in the UK GOV coronavirus dashboard

Anybody wanting to undertake this analysis for themselves can do so by taking advantage of the splendid download section of the UK GOV coronavirus dashboard, which may be found here. I’m sticking with the nation of England as usual because it’s easier to obtain data, there also being more variables available for this nation via the dashboard and other sources.

To business…

This is how things stood in the dashboard archive in terms of certified COVID deaths and 28-day COVID deaths as at 27 July 2022: we have two distinct peaks and a strange, drawn-out rumble. Certification counts exceeded 28-day counts in the first wave because physicians did not need a PCR test result to back their opinion as to cause of death. The recent decline in certified deaths compared to 28-day counts is rather interesting and I’m not sure which series is the closest approximation to the truth! To kick things off I decided to go for certified COVID death as my dependent variable

The Slicer

Trying to model this entire series as one lump is almost certainly doomed to failure for many things changed in the course of the pandemic from mutations to definitions to patient management, to procedures to protocols and back. Then, of course, we have the vaccines doing their thing (or not). I thus reached for the slicer and sliced this series into three juicy segments using the point of minimal daily death counts as my guide.

Each slice may well require a different ARIMA model structure and this will provide useful insight. Obviously I can’t assess the impact of vaccines in the first segment and neither can I assess use of lateral flow devices but what I can usefully do is assess daily case count as an independent predictor of COVID death.