1/ Had another look at the new Maccabi data. (tl;dr: things look good, but nothing beats a randomised trial) https://theo.io/post/2021-01-31-maccabi-2/

2/ I extracted the data from the "MHS" Kaplan-Meier plot in Maccabi's preprint ( https://www.medrxiv.org/content/10.1101/2021.01.27.21250612v1). This shows the cumulative total incidence of COVID that occurred by each of the days following vaccination.

3/ I converted this to daily incidence on each of the days following vaccination, which looks like this:

4/ Now to get some estimate of efficacy, we need to estimate what would have happened to an identical set of people who *didn't receive the vaccine*. For the first 7 days, this is relatively easy to guess, since we don't expect the vaccine to have much effect. It might look like

5/ But this is indicating that the genuine number of people getting SARS-CoV2 is changing during this period, likely due to the epidemic situation in Israel. After this point, where we expect the vaccine to have some effect, it becomes harder to estimate. Some possibilities:

6/ If we estimate vaccine efficacy around day 24 for each of those scenarios, we get 82% if the 'decreasing' scenario is true, 90% if the 'flat' scenario is true and '94%' if the 'growth' scenario is true.

7/ Now Maccabi used none of these scenarios. They used the period from day 1 to day 12, as the placebo-like control for days 13 to 24, like this:

8/ If you do that you get an efficacy of ~88% around day 24.

But they didn't calculate a point efficacy like that, they looked at the entire day 13 to 24 period, which gets an average efficacy of ~51% calculated in this way.

9/ Basically, the results you get are very dependent on the assumptions you make, some of which are pretty arbitrary (neither right nor wrong).

The Maccabi control is pretty conservative, partly in that the control includes day 10-12 period where vaccination may reduce cases.

10/ But it's pretty clear that daily incidence seems to fall dramatically over the course of the day 13-24 period, so 51% is likely to substantially underestimate the peak efficacy from the first dose