The Oxford/AstraZeneca Vaccine Efficacy Data

 By Derek Lowe

We now have a complete writeup of the efficacy data from the Oxford/AZ vaccine effort in The Lancet, and I’m glad to see it. There have been a number of questions about this candidate and its effects in the clinic, so the chance to get a closer look is welcome.

This is of course the “ChAdOx” vaccine, with the first part of that acronym standing for “chimpanzee adenovirus”. And it’s the first viral-vector candidate that has reported out. J&J is coming soon with their adenovirus-26 vector, perhaps even sooner than expected due to the terrible number of coronavirus cases in the US, and Russia’s Gamelaya Institute has their one-shot-of-adenovirus-5 and one-shot-of-adenovirus-26 vaccine as well, for which data are unfortunately very sparse. This paper is from the UK, South Africa, and Brazil trials (others are ongoing), with a total of over 23,000 participants. The portioning of those into treatment and control groups, though, is where some of the arguing starts. These trials are notable for using an approved meningitis vaccine in the control group in the UK trial, rather than saline (as indeed the US trial of this vaccine is doing). The Brazil arm did that for the first shot, with saline for the second, and the South African trial used only saline controls. Famously (by now) a subset of patients in the UK trial got a half dose of vaccine first, followed by a full dose as booster, which seems to have been an out-and-out mistake in handling and formulating the vaccine. Update: here’s the New York Times on the various mistakes that have been associated with this one. Among those is a further difficulty with the half-dose vaccine group – some of them got their booster shots much later than others, for unexplained reasons.

In the larger cohort who got the two regular doses, vaccine efficacy was 62% (with a 95% confidence interval of 41% to 75%). There were 27 cases in the 4440 treatment patients, compared to 71 cases out of 4455 patients in the controls. Meanwhile, in the half-dose-first group, efficacy comes out at 90% (with a 95% confidence interval of 67% to 97%). Both of those 95% CIs are a bit of a spread. In this group, 3 out of 1367 patients in the treatment group came down with coronavirus, compared to 30 out of 1374 in the controls. That control-group infection rate is definitely higher than what was seen in the two-full-dose group, which makes you wonder if it’s running randomly high (which would make the half-dose group look better than it really is) or if the larger full-dose group was running randomly low (which would make it look worse than it really is, but remember, with its larger sample size there’s correspondingly somewhat less room to believe that it was that far out of whack). We need to add in to these calculations the news that the half-dose group included no patients older than 55, and to wonder what effect that had on the numbers, too. The paper reports a combined overall efficacy of 70.45 (95% CI between 54.8% and 80.6%), but how much you trust that one comes down to how different you think these two groups are.

Basically, the more you believe the 90% number is a statistical fluke, the more you then believe that the real efficacy of the vaccine is probably in the 60-per-cent range. On the other hand, if you buy into the idea that there’s something really different and useful about giving everyone a half dose at first, you can be somewhat more optimistic. That belief is optimistic in itself, but it’s not completely nuts, either. Remember, one of the things about a viral vector is that you raise antibodies not only to the protein that whose genetic message you’re delivering, but to the viral vector itself. It’s not impossible that a lower dose the first time made subsequent antibody neutralization of the second dose less of a problem. But you’d need to prove that. And to prove, with intention, that a half-first-dose really does work better to start with.

A bright spot in the data is that (counting from 21 days after the first shot) there were overall ten hospitalizations for coronavirus (two severe, one leading to death), but none at all in any of the treatment groups. This would be a good point to note that the coronavirus cases totaled in all these trials were based on symptoms, which were then confirmed by RT-PCR swab assays. The UK trial also had regular swab assays taken to try to get data on asymptomatic infections, which is something that wasn’t done in either the Pfizer/BionTech or Moderna trials. The numbers aren’t huge, and they aren’t tight, but they’re all we have. There were 24 asymptomatic patients in the low-dose-first group, for a vaccine efficacy of 58.9% (95% confidence interval from 1 to 82.9%, and that’s what I mean by “not tight”), and 45 patients in the two-standard-dose group (efficacy of 3.8%, unfortunately). So at least for this vaccine, the efficacy at preventing asymptomatic cases is notably lower than that seem for symptomatic ones, and that may well be true for all of them. The differences in those asymptomatic numbers, though, also argue that the low-dose vaccination regimen was indeed different from the two-standard-dose one, don’t they?

Now to safety. The top line numbers look reasonable, with the usual reactogenic effects in the treatment groups, and with very similar numbers of adverse events in the treatment and control groups. The part that makes you wonder a bit, though, is the case of transverse myelitis (which made quite a bit of news at the time, at least once the details rather grudgingly emerged). This was two weeks after the second vaccination, and considered by the monitoring committee as likely to be related to the vaccine. It turns out that there were two more, but that news isn’t as bad as it sounds. One of these was in a participant who (as it turned out) had undiagnosed multiple sclerosis, so that’s less likely to be related to the vaccination. The other was in one of the meningitis control arms. It sounds like there was some arguing about whether that was vaccine-related or not (it was 68 days after the shot), but it’s certainly not related to the coronavirus vaccine, anyway. There still is that one case, and transverse myelitis is one of those neurological syndromes that can indeed be associated with autoimmune function (and thus possibly with vaccination). So this will bear close watching as this vaccine gets rolled out.

Which brings up the question of when that might be. As I understand it AstraZeneca is even now wrestling with the problem of whether to run another trial using that low/high dosing protocol. If the different results seen here are real, that could be a significant improvement, both in overall efficacy and in using up less vaccine, but there’s also a real chance of chasing after a mirage. If you go with the numbers as they are, though, it would seem that this vaccine is likely to be provably inferior to the Pfizer/BioNTech and Moderna mRNA ones. That doesn’t mean that it’s not good enough to be useful in the pandemic – just that there are others that could be even more useful, if you could deploy them instead. As it stands, there’s a real risk of having some patients (or indeed whole countries) feel as if they’re being dealt a second-class vaccine if this one is their only option.

Naturally, the same questions apply here as were mentioned in the Pfizer article that I posted earlier today – duration of action, effect on viral transmission, and so on. We’re going to have to wait and collect more data to be able to say anything about these, for any of the vaccines. But the data today make me quite curious about the coming J&J data. That’s another adenovirus vector, as mentioned, and there’s at least a possibility that we will then have a good head-to-head comparison between mRNA vaccines and adenovirus vector vaccines for the same pathogen. That’s a pretty strange situation – when did we ever have multiple simultaneous vaccines for the same disease? And there’s the Novavax recombinant protein one coming after that. No, this is a unique situation. Let’s hope it stays unique!

Update: it looks like India’s regulatory authorities have already declined to issue an accelerated approval for the Oxford/AZ candidate, which may be a sign of trouble to come. . .

https://blogs.sciencemag.org/pipeline/archives/2020/12/09/the-oxford-astrazeneca-vaccine-efficacy-data