March 25, 2020 - 5:52pm

Imagine you see a small light in the sky. Is it a star or a satellite? Without more information you can’t know. A huge, far-away object and a small, close object both provide the same amount of light.

Models of disease outbreaks suffer this problem. Imagine you have a novel disease that kills 10 people. You don’t know anything else about it. It could be that it’s infected 100,000 people and only kills one in 10,000; or that it kills every person who gets it, but only 10 people got it. Without more information you can’t know.

Of course you normally do have more information. With Covid-19 many people got sick or tested positive but didn’t die, so we know it doesn’t kill everyone. But we also know not everyone who gets the disease gets diagnosed, which means the death (and hospitalisation) rates are still unknown. The problem remains: the number of deaths we see is compatible with both “only thousands have it, quite a lot die” and “millions have it, not many die”.

A new model (available in preprint but not yet peer-reviewed) saying that it is plausible — even probable? — that the latter is true; that, in fact, as much as 68% of the UK population might already have or have had it. It’s received some press attention.

I’m sceptical. I have a heuristic: if a statistic is interesting, it’s probably not true. I’m sure the model’s maths are right but I doubt the assumptions, such as that just 0.1% of those infected are hospitalised. It may be true but there’s evidence it isn’t. And it would mean that the disease had been spreading person-to-person in the UK and Italy for a month before it was detected. Also, the study only looks at 15 days of data in each country – it’s a small dataset so susceptible to noise.

If true, it would be good news, since we’d be closer to the peak or even past it, and would utterly change our response. For the headline figure to be correct, the real number of cases would have to be about 5,000 times the UK confirmed cases. The herd immunity strategy would be back on. When serological tests that let us see who’s had the disease are widely available, I hope they show they’re right; I’ve just had a really interesting chat with a statistically minded scientist who thinks they could be. But as yet I think it’s unlikely.

There are other reasons to be optimistic if you want them: UK total death rates are still below the yearly average; this Italian doctor suggests the disease is less lethal than we think; Neil Ferguson, the Imperial College epidemiologist, now thinks that we’ve prepared sufficiently well that we won’t swamp ICU capacity.

But for now I think it’s probably unwise to decide that cold you had in February was really Covid-19. We still need to wash hands, still need to stay physically distant from each other. It’d be better to find out we did it but didn’t need to than that we needed to but didn’t do it. Most importantly: get those serological tests working ASAP.


Tom Chivers is a science writer. His second book, How to Read Numbers, is out now.

TomChivers