Read on Twitter
Well, dear readers, my take on "the winning formula" for Oxford/AstraZeneca did not age well. According to the Indian manufacturing partner, it isn't the "low"-dose-1st subgroup. It's another subgroup, perhaps a subgroup of a subgroup... https://www.livemint.com/science/health/efficacy-of-oxford-jabs-at-95-if-given-3-months-apart-sii-11609181643451.html HT @wogboy11 ..1/6
..The sweet spot he says is "a 2-to-3 months’ gap between dose 1 & dose 2". First let's recap their original data. They said intervals didn't make a difference. For the "standard"-"standard" group >8 wks apart vaccine efficacy was 65·6% (24·5 to 84·4). Ditto 6 weeks cutoff ..2/6
...The protocol doesn't limit the possible exploratory subgroup analyses 
: so many ways they could have found a subgroup that hit this mark. Eg half the already-small "low"-dose-1st group had >12 wks apart: sliced off those perhaps? But I don't think that's the main point ..3/6
: so many ways they could have found a subgroup that hit this mark. Eg half the already-small "low"-dose-1st group had >12 wks apart: sliced off those perhaps? But I don't think that's the main point ..3/6
...If the proposed regimen is 2-3 mths apart, then we need to know the vaccine efficacy rate not just 2 weeks after the 2nd dose: but the vaccine efficacy since the 1st dose, because that's a long time to leave people half-ish-immunized & there's still no certainty on dose ...4/6
...Consider Tozinameran (BNT/Pfizer), the only vaccine that's given us that calculation so far. Although 95% for that primary endpoint after 2nd dose immunity kicked in, it was 82% (76-87%) if you also counted the people who got sick in the 3 weeks between dose 1 & dose 2 ...5/6
...and the range of uncertainty as subgroups get smaller & risk of multiple testing is going to matter a lot. We can't afford to throw a baby out with the bathwater. But there's an inherent risk of error in exploratory subgroup analysis too. 6/6 https://statistically-funny.blogspot.com/2014/03/if-at-first-you-dont-succeed.html
