Why do COVID-19 growth rates vary across the UK?

A couple of weeks ago I wrote an article comparing the different growth rates of detected COVID-19 cases across the world and speculated that weather was playing an important role. We now have enough data from the UTLA (Upper Tier Local Authorities) to start comparing growth rates across the country. Whilst different countries have many other factors that might effect growth rate other than weather, such as testing regime, population density, demographic distribution, cultural behaviour and government policy, the UTLAs in the UK represent a set of regions where many of these factors are broadly similar. For example, government policy and testing regime should be fairly consistent across the UK and the weather has been largely consistent across the UK in the last month.

Below is an animation of the growth in cases of COVID-19 across the 10 regions of the UK normalised by the number of cases on 13 January.

And below is a chart showing the actual numbers. The "15 day growth" shows that actual growth over the last 15 days and the "7 day growth" shows that actual growth over the last 7 days. The final three columns show the resulting growth rates from fitting a linear least squares fit to the log of the cases values over the last 8 days.

It's clear that COVID-19 is growing at very different rates in different parts of the UK and even within different boroughs in London (see below). Taking into account the 10 main regions of the UK, the number of cases over 15 days since 13 March has increased 41.3x to 2,438 cases in the Midlands, but only 11.2x to 324 cases in Northern Ireland. If these rates were to continue for another 15 days to 12 April the number of cases int the Midlands will increase another 41.3x to 100,689 cases whereas the number of cases in Northern Ireland could only increase 11.2x to 3,629 cases.

The number of cases in London over 15 days since 13 March has increased 31.7x to 5,299 cases. However, unlike the other 9 regions that continue to have new cases at a constant rate, the growth in London has slowed over the last 7 days to 2.7x equivalent to 8.4x over 15 days. Coronavirus is thus growing slower now in London than in any of the other nine regions.If the growth in London continues at 8.4x over the next 15 days, then on 12 April there will be 44,511 cases in London.

However the growth rate varies greatly even in London. Over the last 7 days, the average has been 2.7x, but Waltham Forest has seen cases increase 4.1x (equivalent to 20x over 15 days) whilst Barnet has seen cases increase only 1.4x (equivalent to 2.0x over 15 days). At these rates Waltham will have 3,080 cases on 13 April, whereas Barnet will have only 240 cases. Bromley, Newham, Croydon, Havering, Redbridge and Waltham Forest have similarly high growth rates, whereas Westminster, Kensington and Chelsea and Barnet have similarly low growth rates.

It is a positive development that the growth rate in London has slowed, but troubling that the rate in the other nice regions does not seemed to have slowed yet at all. You can see this clearly in the following animation which shows straight lines on the logarithmic chart for all regions except for London, which is starting to curve downwards.

Analysing 60 UTLAs

The three tables below show the growth rate for the 60 UTLAs that had 40 or more cases on 24 March (that is UTLAs with enough cases that growth rates could be measured with some level of reliability). The actual "7 day growth" is listed, but they are ordered according to the 7 day growth rate calculated by doing a least squares fit to the log of the 8 days worth of data from 21 March to 28 March with these stats shown in the final three columns.

The highest growth rates for local areas across the whole UK over the last 7 days were in Sandwell (7.7x), Liverpool (7.3x), Warwickshire (6.6x), Birmingham (5.7x). The 7.7x increase over 7 days in Sandwell is equivalent to a 79x increase over 15 days which would mean there would 9,206 cases there by 12 April.

Why is this data useful? If we can identify what is causing the higher growth rates in some UTLAs by exploring influencing factors, then we may well be able to reduce the growth rates in these higher growth rate areas. If we were able to lower the growth rate to that of Barnet, we would be able to cope with COVID-19 fairly well. I haven't yet had time to explore this myself, but I hope that others are able to build on this analysis and help identify the causes.

It is interesting to look at the total number of infections per 1,000 inhabitants of a UTLA as well as the growth rate per se. One simple correlations seems, on the face of it quite clear. Southwark has the highest infection rate overall and with 319 cases and a population of 317,256, it is the first UTLA to pass the point of 1 in 1,000 inhabitants being infected. This is closely followed by Lambeth with 318 cases and a population of 325,917 such that 0.98 in 1,000 inhabitants are infected.

One theory to explain this is that infections of respiratory diseases such as flu and coronavirus are highest in boroughs with mainline stations. Indeed this paper from December 2018 performed a detailed analysis of flu infection rates in London Boroughs and found correlations:

• Analysing the link between public transport use and airborne transmission: mobility and contagion in the London underground

They concluded that this was due to people in these boroughs coming into contact with more people while traveling on the underground:

Specifically, we show that passengers departing from boroughs with higher ILI (flu) rates have higher number of contacts when travelling on the underground.

Southwark is served by London Bridge train station, Lambeth is served by Waterloo and Kensington and Chelsea and Westminster are served by London Victoria. These are three of the four busiest stations in London and, indeed, if you rewind a week back to 21 March, the highest number of cases per thousand population were in Kensington and Chelsea (0.48), Westminster (0.43), Southwark (0.42) and Lambeth (0.36). Surprisingly, however, Hackney and City of London, served by London's third busiest railway station had a relatively low number of infections per thousand population at 0.18, less than half that of the other four boroughs.

However, whilst proximity to a mainline station seems to have had a big affect on the early growth of COVID-19, this no longer appears to correlate with growth rates at this stage. Perhaps this is not surprising given many people are now in lock down at home and patterns of travel have changed in the last week in particular.

So what do you think could be causing the big differences? How can we lower the growth rates in all UTLAs? Please feel free to comment below.

Notes

If you would like to do your own analysis, you can find the raw case data in this Google spreadsheet. On the second sheet I've made a short tutorial on how to compute the numbers in the tables above:

• UK COVID-19 cases over time by UTLA

Credits

Xander Wiles for creating the video animations. This was done by exporting SVG files of graphs from Google Spreadsheets and the importing into Illustrator and Adobe After Effects.