December 1st: Visualising the step-change from vaccines

Here it is: the first entry into this year’s Data is Beautiful Advent Calendar. Inevitably, it involves COVID-19. It is not meant to be “deep” in any way, and I claim zero epidemiological expertise. But in the best tradition of my charts, it is, I hope, a) interesting; b) intriguing; c) potentially important; d) potentially insightful; and e) importantly, colourful.

What the chart shows is, for each OECD country, the evolution of their vaccination programme to date (X-axis, number of people fully vaccinated per a hundred people in the population, with the last data point referring to 29th November 2021) and the relative number of COVID-19 deaths each day compared to recorded new COVID-19 cases (Y-axis). Inevitably, there will be a lot of noise in the Y-axis variable, because the measurement and reporting of both deaths with COVID-19 and people testing positive is far from perfect.

In addition to measurement and reporting issues, there are of course a large number of other variables that would have influenced both the numerator (e.g., more deaths — all other things equal — would likely emerge if a country was running out of ICU capacity, or the infection was reaching a particularly vulnerable population) and the denominator (e.g., more people testing positive could simply be a function of the accuracy and prevalence of testing).

Nevertheless, unless I’m somehow fatally mistaken, it does seem that for several countries, the number of deaths per number of people testing positive was pretty flat — but vastly different between countries — until their vaccination programmes hit some kind of critical mass. At what level of vaccine penetration that tipping point was reached seems to have been different by country, no doubt because that ratio also depends on so many other factors (e.g., seasonality).

For example, in the UK (large-ish purple dots that start quite far up on the chart), the decline in death rates (measured in this simple way) really accelerated in June when the proportion of fully vaccinated people approached around 50%. In Germany (large-ish light green dots), it seemed to take until September and a 60% vacciation penetration before deaths per positive test started falling off more dramatically.

As with all of this data, I have more questions than answers. I also have a small worry that I’m missing some Bayesian logic here that would invalidate the entire approach. But I hope not, since we are looking at the number of people testing positive and dying in the entire population. (Therefore, any issues come back to the measurements and estimation techniques used.) Explanations, challenges, speculations, and requests for other ways of plotting the data are welcome!