The half dose was an 'experimental endpoint'. Always planned but not seen as critical. FDA rules now stipulate you can only register on primary endpoints declared before the trial starts. I suspect AZ have argued that the accelerated nature of the process should take this into account.
I don't think that gives an accurate picture of what happened. Even AstraZeneca themselves are calling the half-dosage a mistake. Their US trial of that vaccine does not currently contain a half-dose arm [1], though it likely will be amended to do so. As best I can read it, the dosage amount was planned but it was intended to represent the "full" dosage and not to be a half-dosage, but was updated when they noticed unusually mild physical symptoms in people receiving the vaccine. [2] I haven't seen anything to support the idea that a half-dosage was an intended endpoint, whether secondary or otherwise. Do you have a source for that?
No, the results are not valid, or at least not in a useful way. The group who have received the wrong dose isn't a random sample from the treatment group, so we don't know whether it's more effective or not. It could even be less so.
I don't know what the regulator will do about this situation; it's clear that the vaccine is effective in at least one of the dosing regimes administered, but being entirely certain as to which is better is probably impossible to determine from the data available.
My understanding is that the results are still useful and can be include in the statistical analysis. This is what I heard form the BBC's Newscast podcast
Professor Jennifer Rogers (clinical trial statistician):
> "There was some planned dosing differences anyway but this one happened by accident. Now that doesn't mean these results are completely invalid - doesn't mean that at all. You can make changes to your protocol and you can make changes to what you're gonna analyse all the way up until you actually see your data.... If you haven't seen what the data looks like, you are allowed to make changes to your protocols and it is quite common, it does happen that people make changes as to what they're going to analyse.
> "So this change was carried out with discussion with the regulators so it was all fine..."
Now I tried to find the same information reported online and I found this from the same Professor:
> It is perfectly acceptable to make changes to the protocol prior to database lock, so the protocol could have been updated to include this additional analysis (the point of closing a database is to ensure a trial can remain blind, meaning researchers can’t carry out ad-hoc analysis or potentially selectively report results before proper analysis takes place). However, according to version 14 of the protocol, dated 9 November 2020, the primary analysis was set to be the efficacy of two doses of vaccine (across both half and full dose), with secondary analyses being the efficacy of at least one full dose and efficacy of two full doses of vaccine. Efficacy of half dose with a full dose booster was not considered as a secondary analysis in the protocol and so could be an ad-hoc analysis post database lock.
So I'm not actually sure whether AstraZeneca announced the change to their analysis before or after they started looking at their results. If they announced they were going to include the half-doses before database lock then they can use them as valid results.
