Read on Twitter
(1/10) An updated thread on observed "epidemic burn-out", with some new links to interesting examples that I have come across. Tweet 4 contains my new favourite graph, so read on...
(2/10) To recap, I observed that case rates in London (see tweet 1) dropped below case rates in the rest of England very early, even though restrictions were identical and London is vastly more densely populated than the rest of England. Same with death rates:
(3/10) Despite some recent big increases in case rates in English regions, cases in London have remained relatively low:
(4/10) Analysis of every single Local Authority in England shows that areas of London (green) were hit hard and early but have shown minimal subsequent increases, whereas the current "hotspots" (orange) were generally less affected early on. NB. testing has increased over time:
(5/10) @Briggo0 produced this excellent pair of graphs showing a similar effect in south Wales. Newport peaked first then declined first, then Cardiff, then Swansea: https://twitter.com/Briggo0/status/1293524784731619328?s=20
(6/10) We can see a similar effect comparing Stockholm to the rest of Sweden: https://twitter.com/stevebrown2856/status/1283340056816689157?s=20
(7/10) @youyanggu demonstrated a great example comparing counties in Louisiana: https://twitter.com/youyanggu/status/1292898688273178625?s=20
(8/10) @youyanggu again, this time using data from Idaho: https://twitter.com/youyanggu/status/1294324998677639170?s=20
(9/10) It's fair to say I still don't really understand why this effect is observed at apparently low attack rates, but the pattern is consistent everywhere we look. Once a threshold of c. 20% is reached, cases drop away regardless of interventions.
(10/10) Modelling by @mgmgomes1 and @youyanggu suggests that R < 1 can occur at low rates due to heterogeneity of susceptibility, exposure and / or infectiousness. These models fit with the observed data.
