Some have written to me asking about vaccine assumptions, and how much they affect the projections. I assume a 40% overall reduction in transmission per dose - i.e. an 80% reduction for someone fully vaccinated. How much would it change things if that were different? 1/🧵
TL;DR, this much. Here's the current NSW projection with 70%, 80% and 88% reduction in spread assumed for fully vaccinated people (and half that for single-dosed). More on those numbers below. 2/🧵
The overall reduction in transmission from a vaccine comes from it preventing some fraction of infections—this is called E_i, and also preventing some fraction of onward transmission in the infections it doesn't prevent—this is called E_t. 3/🧵
The overall reduction is then 1 - (1 - E_i)×(1 - E_t). So a vaccine that prevents 60% of infections, and 50% of transmission, reduces overall transmission by 80%. 4/🧵
See this pastebin for some arguments and references for why I think those numbers are reasonable: pastebin.com/raw/66C0ExvZ. See the Oxford study in particular (ndm.ox.ac.uk/files/coronavi…), which measures E_i of the AZ vaccines against delta to be about 60%. 5/🧵
But there's a confidence interval there. Maybe the vaccine only prevents 50% of infections. Maybe it prevents 70%. We don't have much data on E_t for delta, but typically for other variants it's about 50%. But let's consider a range: 40% to 60%. 6/🧵
Taking the lower and upper ends of both ranges for E_i and E_t and feeding them to our formula for the overall reduction in transmission gives the range 70% to 88%. Roughly speaking, this is a range in which the true efficacy of the vaccines against transmission might lie. 7/🧵
And how much does that affect our expectations for the outbreak in NSW? Back to the gif again, this much. The peak could be a bit higher or lower, and a bit earlier or later. 8/🧵
So the exact efficacy of the vaccines against spread doesn't affect very much whether or when the outbreak in NSW will peak. What it will matter more for is how much restrictions can be loosened after the peak without cases growing again. 9/🧵
The final level of coverage will also matter a lot at that point. We will still be very interested in exactly how good the vaccines are at what they do - including of course, prevention of severe disease. 10/🧵
But over the next few weeks, our expectations of how things will play out in NSW are probably not that sensitive to the exact efficacy of the vaccines against spread. Unless we're grossly wrong about the vaccines in this respect, NSW's next few weeks seem fairly clear. 11/11
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Since the modelling the NSW gov has released is in terms of the LGAs of concern only, here's my model re-run with only the LGAs of concern considered instead of the whole of NSW. Result is extremely similar to that of the gov's modelling.
The smoothing for mine is centred though, meaning to compare with their 7-day moving average, you'll wanna push my projection to the right ~3.5 days. Taking that into account, it looks like my model is predicting a peak about ~1 week later than them, but of the same height.
And the difference in timing is pretty much within the uncertainty range of both projections, so they basically agree to within the accuracy claimed by each model
Animated version of my NSW projections over time. *not 100% identical to actual past projections because I'm using my current code running on old data and there have been some minor changes, but very close.
And the version with cases on a log scale
You can see that we did not see the projected effect of vaccinations for some time. If they were working as expected, something else must have been increasing R_eff at the same time. If you squint, maybe the decrease in R since mid-August has been in line with expectations.