Geography in COVID-19

This is a short piece that aims to explore the relationship between English and Welsh geography and COVID-19 deaths.

For my previous piece on excess non-covid mortality, see here

(All the data should be available for you to play around with - see the blurb at the bottom)

The last few weeks I have looked at non-excess deaths, now I want to look a bit broader: where is the mortality within the UK - and how does it map with COVID incidence. This has come about from conversations - largely with Danny Dorling and George Davey Smith - and thank them both for their thoughts. Errors and incorrect conclusions are entirely mine.

FIrstly, COVID-19 is an infectious disease (duh) and spreads like one - from person to person. So geography is important. It started in Wuhan, China, and then made its way to the UK, via lots of routes - we don’t know them all, yet, but likely through hundreds of separate, small introductions, from Italy, from Singapore, etc.

And the UK geography is at first, striking. I’m going to include largely death data here, as deaths follow cases, and the testing we have done in the UK has been largely hospital based, and so we have missed much of the community transmission. If you look at https://www.covidlive.co.uk/, you can drag and see the transmission data (it’s great), but that largely reflects admitted patients, so may not represent the degree of spread.

So first, let’s look a deaths in England and Wales related to COVID-19:

This graph splits each area into what’s called a MSOA, an area used by the ONS. There are around 7,000 in England and Wales, and all have a similar-ish number of people (around 10,000), although the population may well be very different.  This death data comes from this report from the ONS and includes all deaths from COVID between 1 Mar and 17 April 2020. The dot represents the centre  - "centroid" of each area.

As you can see, there is a huge diference between the area covered by each MSOA. Some of the rural ones cover miles and miles, the London ones are much tiny. So the map is squashed. I will, at some point, try to use a different style of presenting - i.e. a cartogram, but this will have to do for now.

Obvious points:

1. In most of central London, a significant proportion of the deaths are related to COVID

2. This is true in other large urban areas - and in some dotted rural areas.

3. It is actually - kind of everywhere. Remember, this is deaths up to April 17. Every parliamentary constituency has >1 death (data not shown - but I have attached in my GitHub)

The first question is - does this map look the same if we look at the total number of deaths? London is young (more on that later), so we might expect the total number of deaths in each region to actually not match the percentages?

Huh? It looks different! Firstly, this is a fault of this kind of mapping. You can't really see the London dots, they are falling over each other. But what you do appreciate, is that a small number of MSOA areas, there are no COVID-19 deaths, and then quite a few areas there are a few deaths, and then a minority have lots of deaths.

So, if we plot three separate colours: blue is no COVID, yellow is 1-3 deaths per MSOA, and red is more than 3 deaths (remember, each MSOA is ~10,000 people, so 100 deaths would be an approximate fatality rate of 0.1%), then we can clearly see that there are 'hotspots', but that most regions have a few deaths.

So what does that tell us? Well, firstly, it suggests that COVID-19 is widespread. Remember, these are deaths, so they represent cases from weeks earlier, and this data represents 1 Mar to 17 April. Many of these deaths will represent acquisition of this in February - prior to the lockdown. What this tells you, is that - by the time we shut down, COVID was pretty much everywhere. Perhaps some lucky regions escaped it - some areas of Wales, and perhaps East Anglia, but, nearly all regions had at least one COVID-19 death.

In fact, if you look at the number of deaths within each region, this becomes even more clear. I've limited this plot at 20 deaths to make it clearer, but you can see that the distribution of deaths looks almost Poisson (I have no way of proving this - may just look like that), but importantly, you can see that nearly all of these 7,000 regions have had a COVID-19 death.

So, I think we can say, given the available data, regardless of testing - there was community transmission COVID-19 in the vast majority of these small, 10,000 population areas within the UK in late February. Obviously, there was still movement then, and it is clear the original outbreak centred on London and other urban centres, but movement was already significantly reduced.

Is that interesting? Well, it seems to suggest that if the lockdown was an attempt to stop the spread of the infection around the UK, it was way too late. However, that is not to say it was pointless (I am just suggesting it did not fufil any role in stopping its movement within England and Wales).

There is a second, related question - about the increase in deaths noted in the 15-64 age group in England, that does not appear to be in any other European country (yet). See, for example, this report, where I have included their figure 2 below, which explores 'excess' mortality in the 15-64 group in England (which does not appear in the other areas of the UK).

You can see there is a clear spike in the England data. Can we explain this at all? Or is this England being bad? There are two points: this is early data - you can see there are limited data points to make any conclusions. The second is that the same policy (broadly speaking) was instituted in the devolved nations, at the same time, so it seems unlikely that this is policy related.

In fact, the data probably represents the fact that London is so young. Unlike other European countries, there is a huge age disparity between urban centres and the rural countryside. This is true of all places, but is marked in London, and also the outbreak was so London centric.

If we plot the MSOA areas and split them into 'old areas' and 'young areas' - by taking the number of more than 65 year old people as a percentage of the total population, then splitting half into the young areas, and half into the old areas, then we get this plot:

Nearly all the 'young' areas are in the urban centres, which happened to get hit first. Therefore, it is not too suprising - even given the increased mortality of age, that we should see an initially young shifted mortality in England. Let's look another way: If we plot the expected mortality from each MSOA (calculated using deaths from 2013-15 here), we see exactly the expected relationship. Remember, these MSOA are largely the same size, so differences in mortality are likely largely due to demographics. So, this is completely normal:

In areas with older populations, the average number of deaths we see over a time period like this is higher than in younger populations. I calculated these 'expected' deaths using the data linked above, but you can see its fairly accurate. Now, if we compare with the actual, real, non-COVID mortality over this period, we get this:

Pretty similar, really. Again, this is unexciting. We already know that older people are more likely to die. However, what does the COVID-19 mortality look like? We know that COVID-19 death is strongly linked to age, so we would expect the relationship between older areas and mortality to look the same..... but

The relationship really isn't there! But we know - well, we are really pretty sure - that older people are much more likely to die when they get COVID. How do we explain this? Well, we know that COVID hit the urban areas first, which are much younger, so it may be that we simply haven't seen the full effect of the older age groups, because we have had a predominately young population who have been exposed.

So, if we look at MSOA areas within 30 miles of either Birmingham, Manchester, or London (and call them urban, and the rest rural), then replot the same plot, we get the explanation:

Simply put, the COVID-19 burden of mortality has been in the urban areas, that are so much younger, that we have seen a 'young' spike in deaths. Within each classification (urban/rural), there is still a clear relationship with age, but this is hidden as there is so much more COVID mortality in the urban areas.

 Over the coming weeks, (I predict)we will see this plot slowly merge, and the spike in deaths will start to match other European countries.

Comments welcome!

Data availability:

All data is available on my GitHub.  It's not very neat (sorry, I'm new here). The key files are LA_data_blog2.R, and if you just want the raw CSV files, it is 

the MSOA_all_merged_with_distance.csv file. All the data comes from the relevant ONS page. It should be pretty obvious. The only data I generated which requires some explanation is the expected deaths data. It was generated from an ONS file that reported deaths from 2013-2015 for each MSOA. This allowed me to work out the 'normal' number of deaths in each MSOA over a 3 year period. I then divided that to fit the length of the recent ONS report (1 Mar to April 17), so I had an idea of how many deaths we would normally expect there. I normalised that to the 2015 deaths. That estimated around 65,000 deaths (the actual figure for 2017 for that period was ~70,000, so not bad). I increased each MSOA figure by a small amount to match that figure. Email or @ me if that doesn't make sense