A thread on that German AZ efficacy for over-65s finding. Everyone is going to tell you "the confidence interval is too wide to infer anything", which is absolutely correct, but let's try to do a bit better than that with the intuitions.+ https://twitter.com/olivernmoody/status/1354781400071860230

+First let's note that the key thing for older people isn't really whether AZ stops you contracting CV. It's whether you get severely ill or die. That's why the UK is pursuing its "one dose" strategy despite limited evidence it cuts transmission: it does cut severity & death.+

The German data says that the AZ vaccine is *extremely* effective in preventing severe illness & hospitalisation. Over the whole sample, no-one who took the AZ vaccine got severely ill or hospitalised. https://twitter.com/olivernmoody/status/1354781400071860230

+Since no-one, across the whole sample got severely ill or died, no-one in the 65+ group got severely ill or died. The results are therefore consistent with (though as we shall see they prove nothing about) over-65s being protected from severe illness & hospitalisation.+

+The only data reported separately for over-65s are the number of infections. These say that in the group of over-65 that got the vaccine (341 people), 1 got CV. And in the group of over-65s that didn't get the vaccine (319 people) 1 got CV - the same number.+

+As I've already said, it wldn't much matter if there's no impact of the vaccine on cutting transmission amongst the over-65s if it does cut severe disease, hospitalisation & death. But these data can't even tell us anything meaningful about transmission.+

+If I have two groups of 300 people with something different about one group, & in each case a thing happens once, what does that mean? Obviously it means that thing different (getting the vaccine) doesn't totally guarantee the something doesn't happen (getting CV).+

+If there's some potential noise, so eg someone might register as getting CV who hadn't had it or someone might have an immunity deficiency problem that means a vaccine wldn't work anyway, your observation of 1 is very vulnerable to such noise. But there's something even worse.+

+Imagine that, instead of a sample of 300 we'd had 30,000. How many observations of the event should we expect, from the fact we observed 1 out of 300, excluding the sort of background noise mentioned above? Plausibly the answer is: anything from 1 to 199.+

+The most that an observation of 1 tells us (if it tells us anything) is that the answer is more likely to be 1 than to be 0 or 2. But in a larger sample that leave an awful lot of room for difference.+

+Suppose in a sample of 30,000 for each, we'd get 149 for the no-vaccine group & 51 in the vaccine group. That would seem like quite compelling evidence of a big transmission-reducing effect. But that'd be perfectly consistent with the numbers being 1 for both groups out of 300.+

+The reality is that the design of this study, given the low prevalence of the disease, was never going to be able to discriminate decisively effects from noise for the over-65s. It isn't that the vaccine failed the test: it just wasn't a test it was ever possible for it to pass.