Thread by @VictimOfMaths

The more I think about this map, the more interesting I think it is.

So I thought I would do a little thread on it...

I'm interested in looking at geographical variation in COVID-19 deaths and trying to understand what patterns we might be able to see in them.

We can start by mapping overall COVID-19 death rates.

This fairly clearly shows higher rates in the North of England.

But hang on, London has seen a bigger increase in mortality during the pandemic than any other region in the UK, so why doesn't it show up in that map?

Well, a large part of the reason for that is that London is *young* and we know that the risk of COVID-19 mortality increases rapidly with age.

So maybe it's better instead to look at the % of all deaths which were from COVID. This picks out London much more clearly.

(It's worth saying here that there are other, more advanced ways of looking at this, for example calculating composite age-based risk scores for small areas, as in these great papers from @ikashnitsky @jm_aburto @drjenndowd @melindacmills and colleagues) ncbi.nlm.nih.gov/pmc/articles/P…

But anyway, the question that jumps out to me is to what extent COVID-19 deaths are associated with deprivation.

There are lots of ways of looking at this. The simplest is just to plot these things on a scatter plot...

If we do that, we can see that there is a correlation between higher deprivation and a greater % of deaths being from COVID-19. But there are still plenty of deprived areas with low COVID deaths and affluent areas with high COVID deaths.

There will be many factors contributing to this, including age, underlying health, ethnicity, ability to work from home etc.

But it's interesting to look for spatial patterns in the data. If we plot this on a map, what do we see?

Plotting two different things (COVID deaths and deprivation) on a map at the same time isn't trivial, but there one approach is to use a 'bivariate' approach where we use a carefully chosen colour scheme that can show two dimensions of colour at once.

So in this map, more pinkness = greater deprivation, more turquoiseness= more COVID-19 cases, and the combination of both colours (purple) means high levels of both.

These kinds of bivariate maps work well where either:

1) You have a strong correlation between the two dimensions, but you are interested in the outliers (the pink and turquoise areas)
2) You are looking for spatial patterns (e.g. are the pink areas clustered together)

Here's the bivariate map of deprivation and % of deaths from COVID.

Lots of Purple - as we would expect from the scatterplot. But this is definitely concentrated in urban areas.

Not so much white, but a clear tendency for this to be in rural areas like Surrey or Cambridgeshire.

So what about the pink areas? These are deprived areas which have seen relatively fewer COVID deaths.

There is a pretty clear trend for these pink areas to be rural areas further from major cities. Cornwall & Devon, Herefordshire, Norfolk, the Isle of Wight.

Finally, how about the turquoise areas? The affluent areas which have seen large numbers of COVID deaths.

These look to be fairly concentrated around the periphery of major cities - the outskirts of London, Cheshire, the leafy Western suburbs of Sheffield.

This isn't a rigorous statistical analysis, but the patterns look pretty clear to me. And there's no way you could have spotted these from a simple scatterplot. Or even by looking at separate maps showing deprivation and COVID-19 cases separately.

tl;dr MAPS FTW \o/

A much higher resolution version of the map can be found here:
github.com/VictimOfMaths/…
And the R code to generate it is here:
github.com/VictimOfMaths/…