When I arrived at university I wanted to specialise in genetics; by the time I left I wished I’d focussed on the history of science. Now I’m doing both!
We’re witnessing scientific history right now, of course, and the great thing is, much of it is online!
My research assistant 😉 drew my attention a couple of days ago to a Reuters Special Report: Johnson listened to his scientists about coronavirus – but they were slow to sound the alarm. Gold dust.
The Reuters report is based in part on minutes of the meetings of advisory committees to the UK government. I doubted the reporters had privileged access and, sure enough, if you look online you can find the minutes of meetings of the NERVTAG Committee.
Sounds like something from Spooks, doesn’t it? NERVTAG actually stands for New and Emerging Respiratory Virus Threats Advisory Group, so totally logical. No dramatic licence at all!
I thought I’d skim through the NERVTAG minutes. As you do.
And here’s something I found:
NF [Professor Neil Ferguson, Imperial College] commented that the Wuhan repatriation flights at the end of January suggested that there was about a 1% infection prevalence in the population of Wuhan. At that time, that can be compared with official case numbers and suggests that the cases [sic] numbers represented around 5% of all cases.Minute 3.5, NERVTAG Novel Coronavirus Seventh Meeting: 21 February 2020 (pdf)
Let’s recap before decontructing this historical artefact (as we say in the trade!).
As the Reuters article notes, the UK “prepared to fight the flu” and despite what we’d all witnessed in China, “assumed drastic [lockdown] actions would never be acceptable… in the UK” [my stress].
My thesis goes one step further: I suspect the boffs’ thinking was also hamstrung by a misunderstanding of the impact of a Covid-19 epidemic on the health service.
Exhibit A is the evidence given by the Chief Medical Adviser, Professor Chris Whitty – essentially a spokesperson for the UK epidemiological community – to the Health and Social Care Select Committee on 5th March. Whitty suggested 20% of the population of Wuhan (or Hubei province, it’s not clear which he was referring to, a point I’ll come back to) had been infected with the coronavirus.
NERVTAG Minute 3.5 is Exhibit B.
How do we get from 1% prevalence in Wuhan “at the end of January” to Professor Whitty’s 20%? Remember, Wuhan was locked down on 23rd January and the rest of Hubei a few days later, so the infections from that point would have occurred mainly in the household. By the end of January, nearly everyone in Wuhan (and the rest of Hubei) who was going to get infected would already have been infected. A 1% infection rate is therefore a reasonable ballpark for the infection rate of the population represented by the sample on the flights from Wuhan. Somehow this key fact has been forgotten.
Let’s detail what’s wrong with NERVTAG Minute 3.5:
1. I very much doubt that Professor Ferguson was referring to “repatriation” flights. Those wouldn’t really provide useful information about infection rates in Wuhan, since not only were Brits stuck in China not typical of the local population, you weren’t allowed on the flights if you showed any symptoms. No, he was talking about (generally earlier) international flights, which his group had just published a paper on, showing that many Covid-19 cases exported from China had been undetected.
2. Ferguson suggests a 1% prevalence in Wuhan at the end of January, but I’ve just noted that nearly everyone who was going to get infected in Wuhan had done so by that time. How, then, do I reconcile the 1% with the ~4% infection rate I’ve previously suggested, based on the mortality figures for Wuhan? (Actually, I might even want to revise that 4% down since a) there were ~2600 deaths in Wuhan, not 3,000 as I rounded; b) I took the population of Wuhan to be 10 million, a lowball estimate; and, c) I used an Infection Fatality Rate (IFR) of ~0.7% for Wuhan, based on that for the rest of China which, since it is lower than reported elsewhere, could potentially be distorted by under-reported deaths – but we’re talking ballpark here, so let’s leave it).
One possibility is that the Imperial research is based on flights from Wuhan International Airport. But that airport doesn’t just serve Wuhan itself. It’s the international (hub) airport for a larger area – roughly the whole of Hubei, a province of 60 million people. So, I ask myself, did Ferguson quote a figure for Hubei, or for flights from Wuhan International Airport, not “Wuhan”, as minuted?
Given that NERVTAG takes the population of Wuhan to be 19 million (according to the minutes of their meeting on 13th January – who’d have thought a major uncertainty in the estimation would be the population of Wuhan!!), about a third of that of Hubei, and the vast majority of cases occurred in Wuhan itself, that would imply an infection rate in Wuhan of around 3%.
3. Now we get to the bit that (almost) makes no sense whatsoever, and could have caused all the misunderstanding. I stress “could” – whilst I am highlighting genuine numerical discrepancies, I can only speculate on what the dramatis personae actually understood.
Ferguson “suggests that the cases [sic] numbers represented around 5% of all cases”. Why does this make no sense? Because he’s comparing big apples with little apples. The big apples are the cases detected on international flights, since we know these were infected. The little apples are the cases in Wuhan, because “[a]t that time” most of them had not yet developed symptoms so would not yet have been detected.
Based on cases on international flights, Ferguson thinks ~1% of the population of Wuhan was infected at the end of January. In round numbers, that’s about 200,000 people, taking the population of Wuhan to be near 20 million.
At the end of January, Wuhan was reporting around 10,000 cases, which is 5% of 200,000. This is the sort of calculation Ferguson or his team might have done to reach an estimate of 5%. But the 10,000 cases are not the full number actually infected by the end of January. Rather, it’s the number of infections detected by then, so represents a proportion (the detected proportion, that we’re trying to establish) of those who were actually infected maybe (on average) 10 days before.
But, as I’ve said before, in actual fact, most of those who were going to become infected in Wuhan already had been, by 31st January (if this point is not yet clear, see Tomas Pueyo’s Chart 11 for an illustration). We can do a sanity check. 31st January is a week after the lockdown in Wuhan. If infection rates double every 3 or 4 days, but only come to light when the serious cases turn up at hospital 10 days or so later, there’s easily time for case numbers to double 3 times, that is, increase by a factor of 8. So the figure that should have been used was not 10,000, but the full number of roughly 80,000 cases eventually detected in Wuhan.
4. Why does this matter?
It matters a lot, because I think the idea that China was only detecting 5% of cases in Wuhan gained traction. For example, on 9th March, I clearly thought only 5% of cases were serious (see points 5 and 6 in that previous post).
More importantly, ignoring the evidence for a much lower infection rate established by studies of cases on international flights, such as that by Imperial, Whitty claimed at least Wuhan had experienced 20% infection rates and if he believed that then he might have thought the NHS would be able to somehow cope.