Read on Twitter
I am deeply skeptical about the headline conclusions from this new Oxford uni study. A more considered response will follow but brief thoughts now on why I don’t believe the conclusion that over half the UK has had the virus. 1/9 https://twitter.com/PickardJE/status/1242489840068894720">https://twitter.com/PickardJE...
The basic modelling approach is uncontentious, an S/I/R model (similar to what I’ve previously described). Core parameters like R0 seem reasonable.
They model death in two stages;
a) size of population vulnerable to becoming seriously ill
b) death rates once seriously ill. 2/9
They model death in two stages;
a) size of population vulnerable to becoming seriously ill
b) death rates once seriously ill. 2/9
They calibrate b) based on existing studies. It doesn’t look unreasonable.
For a) though they seem to pluck values of out thin air. They consider values from 10% to 0.1%.
They then solve for whatever the infected population needs to be to explain the observed deaths. 3/9
For a) though they seem to pluck values of out thin air. They consider values from 10% to 0.1%.
They then solve for whatever the infected population needs to be to explain the observed deaths. 3/9
The curves at the top of figure 3 indicate what the infected population needs to be to explain the deaths based on their assumption about the size of the vulnerable population. 4/9
If 10% of the population is vulnerable to becoming seriously ill then only a small fraction of the UK (by eye, about 5%) has had it (dark red line).
If 0.1% of the population is vulnerable then a majority of the UK (about 70%) has had it (yellow line). 5/9
If 0.1% of the population is vulnerable then a majority of the UK (about 70%) has had it (yellow line). 5/9
So fundamentally all they are saying is that, with a fixed observed number of deaths, there is a strong inverse relationship between the rate of becoming seriously ill and the number of people who have been infected. That’s pretty obvious! 6/9
So why don’t I believe their headline-grabbing scenario with a low risk of getting seriously ill and therefore a large infected population?
For one thing, 5% of the UK is over 80. More still have relevant co-morbidities. Why would only 0.1% be vulnerable? 7/9
For one thing, 5% of the UK is over 80. More still have relevant co-morbidities. Why would only 0.1% be vulnerable? 7/9
Secondly, fitting the data. They have only considered the first 15 days of deaths in each country (just 144 deaths in the UK). That’s going to make it a volatile fit. By eye, it doesn’t look like their model explains the more recent death data well, especially Italy. 8/9
That’s it for now. In short I don’t believe this is robust and I certainly don’t believe that a majority of the UK had been infected by 19 March. Nowhere near that!
Please don’t ignore public health messages to self-isolate based on this report! 9/9
Please don’t ignore public health messages to self-isolate based on this report! 9/9
Here’s a link to the paper because every time I number a thread there’s always at least one more tweet. 10/9 https://www.dropbox.com/s/oxmu2rwsnhi9j9c/Draft-COVID-19-Model%20%2813%29.pdf?dl=0">https://www.dropbox.com/s/oxmu2rw...
It looks like Adam and I are in full agreement with out more-or-less simultaneously posted responses! https://twitter.com/adamjkucharski/status/1242569554171179008?s=21">https://twitter.com/adamjkuch...
https://twitter.com/actuarybyday/status/1243122456732282881?s=21">https://twitter.com/actuaryby... https://twitter.com/ActuaryByDay/status/1243122456732282881">https://twitter.com/ActuaryBy...
