(1/13) The curious case of COVID-19 trends in London vs the rest of England, and the case for resistance in the population: A thread. The graph below shows the trend of COVID cases in London, March to May. The orange line is a centred 7-day moving average. So far, so normal:

(2/13) We can estimate a crude R(t), by comparing the change in the moving average over 4-day periods. Note that, by this crude measure, R(t) was already falling by the beginning of March:

(3/13) Here’s the same graph for the North West region. For the benefit of those reading from outside the UK, social distancing measures were implemented on 16th March and a moderate lockdown from 23rd March. The measures were uniform across England:

(4/13) Let’s compare the early trends, London vs. North West, normalising using cases / 1000 population:

(5/13) With a relatively small percentage of the population affected, what trend might we expect to see with uniform lockdown measures? Something like this, maybe?

(6/13) Let’s look at the actual trends. Now this is odd. What’s going on here then? Testing capacity increased enormously through April, which explains the apparent flat-lining in the North West, but what’s going on in London?

(7/13) Ok, maybe the North West is unusual. What about the other regions? Wowser, this is getting really odd, London peaks first but then falls away (again, note the effect of increased testing):

(8/13) If we normalise against the peak in each region, the trend is even more stark. London was hit early and hard, but then cases plummet while others regions catch up:

(9/13) So, once London was hit early on (probably with lots of undetected cases), the epidemic faded, even though social distancing is much more difficult in a huge city and with many key workers still using public transport. Pretty much all over by the end of May:

(10/13) So, something else was in play. Cambridge University has estimated the attack rate (% of people infected) in London at 18% ( https://www.mrc-bsu.cam.ac.uk/now-casting/ ) and this is supported by seroprevalence data:

(11/13) Intriguingly, this is a similar level of prevalence to not-locked-down Stockholm, where the outbreak is also fading fast, https://www.folkhalsomyndigheten.se/contentassets/e1702f53eea144cdb1ca2ef854b45c35/estimates-peak-day-infected-during-covid-19-outbreak-20103.pdf and New York City https://www.6sqft.com/new-york-covid-antibody-test-preliminary-results/

(12/13) So, we can see from the data that, once prevalence approaches 20%, the epidemic seems to burn out. Otherwise, why would cases fall in London before anywhere else in England? Even though R would naturally tend to be higher in a huge city?

(13/13) Why? I’m not a biologist, but if, say, 50% of the population is *not susceptible* to COVID-19, and a further 20% has at least temporary immunity through exposure, then R(t) is forced below 1. The data suggest significant inherent non-susceptibility in the population. Ends

Addendum 1: Timeline of deaths in each region, demonstrating how London was hit hardest and earliest:

Addendum 2: Death rate normalised as % of maximum, demonstrating that (again) London declines quicker than the other regions, having been hit harder:

Addendum 3: Cumulative death rates:

Addendum 4: Seroprevalence survey data (using not particularly sensitive Euroimmun test but adjusted for sensitivity), showing broad correlation with attack rates, from https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/897199/Weekly_COVID19_Surveillance_report_-_week_27.pdf

Addendum 5: Appreciate not everyone will be familiar with the population distribution of England, so here's a regional summary graph, demonstrating just how densely populated is London. Bear in mind that England as a whole is probably the most densely populated country in Europe:

Addendum 6: Crude Rt on 1st April 2020, estimated using the Robert Koch Institute method based deaths by date of death in NHS hospitals in England using a 20 day lag. Rt lower in London that any other region:

Addendum 7: Crude Rt time series, mid March to end April for English regions: