OK, I already explained this on my blog (in French), but let me say it once more here in English. About the Imperial College paper by Neil Ferguson &al imperial.ac.uk/media/imperial… which is scaring the shit out of everybody: let me put this in perspective. •1/13
Yes, the paper DOES mention an 81% attack rate for R₀=2.4 and this translates to 510 000 deaths in the UK for an unmititgated epidemic. BUT the point of the paper is NOT to compute this. Their point is to compute the effect ON this rate of various mitigation measures. •2/13
So they did not use a particularly sophisticated model to compute the attack rate because it was NOT their goal. If we want to see how much school closures and so on affect the attack rate, such a model is not necessarily useful. •3/13
AFAICT, they don't provide a link to the actual model so I could run the computations myself (which really shameful, btw). But I suspect it's a compartmental model with age compartments but no finer structure of the social graph. This explains the ~80% attack rate: … •4/13
… ~87% is what you get when you plug in R₀=2.4 in the formula for the attack rate in the SIR model. But flu epidemics have R₀ close to this, and with no real prior immunity (1918, 1957), they only infect ~30% of the population! •5/13
Why is this? My intuition is that the reason real epidemics have a much lower attack rate is that the fine structure of the social graph matters a lot: see what I write here (+next tweet): •6/13
Or here (+next couple tweets): •7/13
THIS IS NOT TO SAY that the Imperial team doesn't know what they're talking about, merely that their goal was NOT to compute the attack rate. They don't even mention it in the abstract. Their model is used to COMPARE scenarios. •8/13
Also worth noting is that they're using a fairly pessimistic infection fatality rate of 0.9% (assuming health system is not overrun), again, because their goal is not to compute number of deaths but to compare scenarios. Korea seems to have a lower CASE fatality rate. •9/13
There are a number of signs that we're missing a huge number of infections: e.g., science.sciencemag.org/content/early/… — on an anecdotal level, an amazingly large number of my friends (with no physical contact between them) are reporting mild flu-like symptoms. •10/13
Finally, one thing to note about the Imperial paper is that they show the number of critical care beds, in their graphs, at their surge capacity (red line). But there are a number of ways to increase availability of such beds. Total number of beds would be more evocative. •11/13
The main point of the paper is to make clear the dilemma which I was explaining in this thread: threadreaderapp.com/thread/1237372… — the “suppression” (①) versus “mitigation” (②) dilemma. Both have huge problems, and there is no easy choice! •12/13
But let's not use the paper to draw overly dramatic conclusions which were not the authors' intention: it was clear from the start that an epidemic would not easily stop before attacking around 20% or 30%, be it 80%, of the population, and an IFR of 1% is a big deal! •13/13
