Thread by @JamesWard73

Another quick ‘model update’ thread: it’s good news on vaccine uptake, but I also take a look at the Warwick & Imperial models, and explore a scenario with higher R0 that generates a (small-ish) fourth wave next winter. 1/n

Starting with the good news, the real-life experience and polling data on vaccine update keeps getting better, so I’ve updated my model to reflect this: now assuming 95% take-up in all high-risk categories (JCVI 1-9) and 90% in other adults. 2/n

(note I’m still assuming 5% dropout on the second dose, which may be wrong – I’d welcome any insights or links to data on how this is going – but to be honest, it’s not that material to the outcomes) 3/n

And where does this get us to? Well, as you might expect it nudges vaccine-acquired immunity a bit higher, so we’re at ~70% in mid-June, and ~73% in October (after vaccinating teenagers). So we’re not predicting any kind of ‘exit wave’ in that scenario 4/n

I also spent a bit of time at the weekend looking through the latest published versions of the Warwick and Imperial models that feed into SPI-M and SAGE. See detailed papers here: gov.uk/government/col… 5/n

I won’t do a full review here, but as a quick summary: the models are definitely getting closer to more up-to-date assumptions, and have taken note of previous criticisms e.g. they do now explore the impact of seasonality (with similar conclusions to mine) 6/n

However I think they are still very cautious in one key area: the assumption of vaccine impact on severe disease, hospitalisation, and death. The Warwick model has no separate assumption for severe disease impacts, so is just using VE estimates for mild disease, 7/n

…based on the Phase 3 trial data (70% after 1 dose and 88% after 2 doses). The Imperial model does have a separate assumption, but for AZ it’s set to 80% after 2 doses (vs. 98% after 2 doses of Pfizer) which feels pessimistic based on what we know today. 8/n

So I think this is the key reason why the Warwick/Imperial models are still producing estimates in their central cases of around 50-60k deaths in an exit wave, whereas my model (when it was predicting an exit wave) was looking at more like 20-30k. 9/n

[Incidentally, I do have strong sympathy for the modellers, as it’s very easy for me to receive new data and re-publish on Twitter within the hour, whereas their publication cycle (via SPI-M and SAGE) is at least a week and often longer – so it’s very easy for us 10/n

to criticise them for having out-of-date assumptions, when the data is moving so quickly. I expect we’ll see some further revision to those numbers, and hence to the models’ projections for deaths and hospitalisations, on their next update] 11/n

The other thing I picked up by looking at their models was that we have a different view on R0, and on the impact of baseline controls. Specifically, the Warwick/Imperial models appear to be assuming R0 for the new variant is in the order of 4.5 (vs. my 3.6)... 12/n

…but whereas I’m assuming a relatively small impact from ‘baseline controls’ such as TTI and masks, they’re assuming a much bigger benefit. The net effect of this is almost zero in terms of our predictions for what happens on the unlocking journey, … 13/n

…unless you start to remove the ‘baseline controls’. So I ran a couple of scenarios in my model to explore what would happen if R0 were higher (4 or 4.5), and with the baseline controls removed 14/n

In the case of R0=4, it doesn’t make a big difference: even if you remove the baseline controls, the ‘exit wave’ is hardly material: 15/n

If I move R0 up to 4.5, and remove the baseline controls, then you do get back to an exit wave – but it’s not that extreme: c. 15k additional deaths, and looks to be within NHS capacity on hospitalisations – this might feel a bit like a “bad winter” for flu. 16/n

It would still be good to avoid that if we can; it won’t stop us unlocking, but it may be an argument for maintaining some light-touch baseline controls through next winter. By that stage we’ll know a lot more and will be able to judge better what, if anything, is needed. 17/n

All the usual caveats apply – in particular I am not (yet) modelling any changes to immunity over time, or any possibility of re-infection, either with current variant or potential new variants. I’m hoping to work on that in the next week or so. /end