Balancing the equation

Too many problems, oh why am I here?

My last piece finished with some examples of places where the UK Government (in my view correctly) resisted the temptation to react during the COVID pandemic. Of course, that doesn’t imply that I think they got everything right, and in particular that they didn’t under-react on many occasions.

I really didn’t say that because firstly I think the failures are more or less taken as read, and secondly there’s no shortage of people with big platforms in politics, science, press and social media who will consistently point these things out, so I tend to feel like that side of things is covered already.

However, since it’s been suggested to me that this made my last piece slanted in one way, I will demonstrate my impeccable centrist credentials with some examples where I think the official response missed out in the opposite direction. Note that none of these are the usual suspects - so no mention of PPE, care homes, the Cheltenham Festival, trying to save Christmas, Test and Trace and so on. Again, I’m not saying all these don’t matter, I’m just assuming everyone has views about them already.

But here, in roughly chronological order, are five data-based times when I think we could have done better, and where the public inquiry could ask some interesting questions:

1. March 2020 growth rates. Why as late as 16th March 2020 was the Prime Minister talking about a 5-6 day doubling time for infections, when the true figure was likely closer to 3 days? In his press conference that day he said

   It looks as though we’re now approaching the fast growth part of the upward curve. And without drastic action, cases could double every 5 or 6 days.

   But actually, even in data available at the time, it’s clear that the growth rate was significantly faster than that. Here’s a case ratio plot based on reporting date (so in real-time as it appeared). At the time, cases were roughly five times higher than they had been the previous week, corresponding to 3 day doubling.

   

   Now, of course, testing was ramping up at the same time, which meant that case growth might over-state true infection growth. But even so, positivity had been growing fast throughout early March (this regression line fitted to early points gives about 3.3 day doubling, albeit with a slowdown as measures started to take effect):

   

   The effect of any mis-estimation of growth rates is very serious: if you think it’s 5 day doubling when it is actually taking 3 days, then in 15 days time the epidemic will be 4 times bigger than you expected (five doublings not three). Given that the strategy appeared to be to gain some level of infection immunity before shutting down the country to protect NHS capacity, timing of the lockdown decision was crucial. So, why did the Prime Minister quote this particular figure, and was 5 to 6 day doubling really plausible?

2. September 2020 death rates On 21st September, Whitty and Vallance’s infamous press briefing gave an illustrative scenario of weekly doubling that would lead to 50,000 cases a day by the end of October. This drew a lot of attention (and indeed this case figure was not reached that autumn), but there was less focus on their death prediction:

   50,000 cases per day would be expected to lead a month later, so the middle of November say, to 200 plus deaths per day

   But this was almost certainly an underestimate, and based on far too optimistic case fatality rates. Even though English cases reached around 20,000 per day before the second lockdown at the end of October, we actually saw over 400 deaths by November 19th. This might suggest that if we really had reached 50,000 cases, the death figure would have been nearer 1,000. Why was this death estimate low-balled?

3. November to December 2020 variant spotting The terrible January 2021 peak in deaths was largely due to the arrival of the alpha variant, spreading out from Kent and reaching London in large numbers before Christmas. Obviously it’s easy to look back with perfect hindsight, and think that it should have been spotted sooner. But for example, it is clear looking now at the S-gene failure data from the ONS survey that logistic growth (straight line growth on logistic axes of the kind that you would expect from a new variant) was going on in the South East in November 2020.

   

   But could this have been spotted earlier? It may be impossible to know for sure, but it’s interesting to think whether some degree of automated sense-checking of incoming data could have spotted this (or indeed the Immensa test failures), particularly taking into account the geographically concentrated effects. As a “lesson learned” it seems like a worthy goal to think how to develop such sense-checking procedures now, in anticipation of any future pandemic.

4. September 2020 to July 2021: axes to grind Of course, you’d expect me to say this, but I still believe that the September 2020 and June 2021 returns to case growth were masked by the presentation of official data on linear, rather than logarithmic scales. This meant that the early growth appeared insignificant compared with previous peaks, making things seem less bad than they were. For example, see this from the 27th May briefing, where the latest increases seem insignificant because the height of the January peak compresses the axis down:

   

   Whereas at a regional level it was very clear using a log scale that cases in the North West (where delta first became established) had been growing exponentially for a while:

   

   Even then, there were no further press briefings until 14th June, at which stage the original plans for “freedom day” on 21st June were pushed back by 4 weeks, based on graphs like this:

   

   So I think it’s reasonable to ask whether better data representations could have led to the public and politicians being more on board with the nature of the delta threat sooner, and working even harder to pull forward vaccination programmes and so on.

   Of course, logarithmic scales may not seem natural at first, and might have taken some getting used to for the general public. But I firmly believe that this could have been done via gentle reminders at the daily press conferences, and via their introduction on the dashboard sooner than July 2021. If so, the public could have understood earlier the seriousness of the situation and adapted their behaviour accordingly.

5. Autumn 2021 schools vaccine rollout. After a careful analysis of the benefits and issues, the JCVI recommended COVID vaccinations for 12-15 year olds at the beginning of September 2021. However, there was not a “spade-ready” programme available at the start of school term: for example by 13th October there had still been only 210k vaccinations out of 3 million in that age group (7% of that population). Partly as a result, case numbers rose very high within that age group that autumn (see ONS survey):

   

   It feels to me worth asking whether a vaccine programme that had been able to hit the ground running straight after the JCVI’s decision could have prevented this, and saved many children from illness and isolation?

Housekeeping note: I am off Twitter at the moment, so if you found this interesting, do feel free to share it there or elsewhere.