Two cautionary tales about BA.2.86

Your crystal ball ain't so crystal clear

(This is a crosspost from @orbit_silver, which was my backup Twitter account, but is now going to have to be my main one)

The place with the longest series of BA.2.86 data is Denmark. We now have 10 weeks with BA.2.86 detected, so if we work out each week's market share and plot it on a logistic axis we should see a straight line that gives us the growth rate. But actually - not so much.

But it's tempting to look at the graph, try to build some narrative. Maybe there are oscillations caused by people isolating or waning of immunity or something. The eye of faith can pick out a four week cycle perhaps?

Except no. Although Denmark do a great job of sequencing relative to their population size, none of these points is based on more than 260 genomes in a week, so if you put confidence intervals on you can see that any particular point is very uncertain.

Whereas the UK gives us more useful data: we sequence many more genomes (up to 1600 per week), we are told that one week's data is anomalous because of the care home outbreak. So we get narrower confidence intervals, and a linear trend based on the six "normal" points fits very well.

So, first cautionary tale: don't look at the proportions without considering the sample sizes. Confidence intervals matter, particularly when you are trying to get juice out of not a huge amount of data.

Second cautionary tale: don't lose sight of the bigger picture. There's a been a huge amount of hype about BA.2.86, acres of media coverage, people arguing that it deserved a Greek letter, even the UK tweaking its vaccine strategy on the basis of it.

But as I pointed out when BA.2.86 emerged, the thing that mattered was always likely to be the growth rate. We can estimate that from the UK plot, and find about a 30% weekly advantage. That's not too bad, but on the other hand it's still exponential growth, right?

Yes, but the point is, that's a fixed growth rate against a constant background. Over 7 weeks, that's probably not a bad assumption, but it's worth remembering that the background itself is constantly evolving and becoming more fit as the better strains take over.

And we can even see this on the same plot. BA.2.86 isn't the only variant in town. One interesting comparator (HT: @OliasDave) is HK.3, which started around the same time as it, but seems to be growing faster, with more like an 82% weekly advantage.

So, actually maybe instead of worrying about BA.2.86 we should be worrying about HK.3 instead? Well, up to a point that's true. Based on its absolute numbers and growth rate, it seems like a decent candidate to start to affect hospital admissions in say 4-6 weeks' time.

But that's not really a surprise perhaps. We've been on a roller coaster of rising and falling admissions (albeit with a gradually falling amplitude lately) for a couple of years now, so it's not a huge surprise that this would continue, particularly over the winter.

On some level, it feels like if it wasn't going to be BA.2.86 or HK.3, it would have been something else. And BA.2.86 was interesting because it represented a jump into the unknown - where severity could potentially be higher for example.

Whereas the fact that we are talking about HK.3, or XBB.1.9.2.5.1.1.3 to give it its full name, might suggest that this is more of the same, that the next few months may be a continued diet of omicron soup rather than a sudden leap into the unknown. But of course, as ever there’s a fair amount of uncertainty, and probably more twists in the tale to come.