White Doves
Last time, I shared Imperial's model of what a checked and unchecked COVID-19 pandemic would look like. It made a number of predictions:
1) We'll be in on/off social isolation mode for the next 18 months.
2) As soon as restrictions are relaxed, the number of cases of COVID-19 will peak again.
3) Mass gatherings (football matches, concerts..) have little effect on the spread COVID-19.
In a new Oxford paper however, they present a model which shows that over 68% of the population may already have been infected — the majority suffering a very mild illness.
“I am surprised that there has been such unqualified acceptance of the Imperial model”
Prof Gupta of the Oxford Study
If this Oxford assumptions is true — most of us have already been infected and we are well on our way to achieving herd immunity. The total pandemic would be expected to last 2–3 months.
What's wrong with Imperial's model?
The Imperial model is built using 13,000 lines of code. But this is where GIGO comes in. If any of their assumptions are wrong — the results of the model can be wildly inaccurate.
In computer science and statistics, garbage in, garbage out (GIGO) is the concept that flawed, or nonsense input data produces nonsense output or "garbage".
Some of their assumptions are probably inaccurate. In particular, it's difficult to know how much of the population has already been/is infected with a very mild disease without widespread testing.
Without this information, any projection into the future is limited.
1) As soon as restrictions are relaxed, the number of cases of COVID-19 will peak again.
Not necessarily. A paper from the author of The Black Swan suggests that after a few weeks of lockdown, almost all infected people are recognised, and they can isolate along with their family members. This has been seen in China, as it boasts no new domestic cases.
2) We'll be in on/off social isolation mode for the next 18 months.
A lot has been said about exponential growth of the virus, but the opposite is also true — social distancing and other measures will exponentially reduce the number of weekly cases (after the peak).
The Imperial model doesn't make a provision for aggressive testing and contact tracing post-lockdown either. With these interventions, it might be possible to relax restrictions in the immediate future.
What's Wrong with the Oxford Model? (and how it's been presented in the media)
The Oxford model hinges on a central assumption — that only 1/1000 people infected with COVID-19 will need hospitalisation. This is an unsubstantiated claim, and one that's made as part of some hypothetical modelling.
What the model actually tries to do—
Let's look at the number of COVID-19 deaths in the UK. They could either be due to:
1) Lots of people already being infected, but a low death rate (optimistic).
2) Few people being infected but a high death rate (pessimistic/Imperial model)
What would each scenario look like?
How it's been (mis)represented in media—
Over half of the UK have already been infected by COVID-19 and it's not that bad.
The Oxford paper is making a nuanced point. Due to our lack of testing, we don't know how much of the UK has been infected and recovered — it could be anywhere between 1.9% and 53%. Therefore, COVID-19 modelling is currently very limited — like the back-of-a-napkin business plan you make with a friend.
Granted, the Oxford paper is strange. This goal is never stated (although is later tweeted by one of the authors). The only contact details given are for a PR firm.
What's needed: testing of a random sample of the population at repeated intervals. Probably politically unsavoury since tests aren't even available for frontline staff.
New Paper
After becoming a regular at my SSC supervisor's first seizure clinic last year — I told him that I like Neurology and technology, and asked if we could write a paper together—
What I'm Reading
In The Rules of Contagion, epidemiologist Adam Kucharski explores topics including gun violence, online manipulation, and, of course, outbreaks of disease to show how much we get wrong about contagion...
So how many angels can dance on the head of a pin?
8.6766*10exp49 (ref).
All the best,
Mustafa