So now the meta-analysis of phase 2/3 data from 4 of the trials of the Oxford/AstraZeneca vaccine is out @TheLancet. Because of course it is. 🤦♀️ marlin-prod.literatumonline.com/pb-assets/Lanc… Background: wired.com/story/the-astr… 1/n
...Mystery #1 - why Indian phase 3 trial isn't included - isn't solved. They report only that they have trials in UK, Brazil, SA & Kenya (Kenyan data not included). Trials in the US, Japan & Russia also not mentioned... 2/n 
...Ah...in the statistical analysis plan, they describe the 4 studies as Oxford-sponsored. The trials that aren't mentioned presumably aren't. Given there was a serious adverse event in the Indian trial in October, this raises issues... 3/n marlin-prod.literatumonline.com/pb-assets/Lanc…
...A major issue is the regulators: authors report that MHRA (UK regulator) & EMA (Euro regulator) signed off on this pooled analysis approach. Presumably considering authorization on a partial set of trials. (Best guess, it's 2/3 of total people in ph 3 trials for the vax)...4/n
...There's less standardization within these trials than there appeared to be from previously available info. Eg for UK & Brazil phase 3, less than half of people had injections more than 6 weeks apart (up to over 12 weeks): that low dose/standard dose subgroup: all >6 wks... 5/n
...Starting with safety again. There is *no* data on mild to severe reactions. None. 😱 There is data on a category that is variously called severe/serious, but it is only serious adverse events (SAEs) & AESI (special interest)...6/n
...Rate of SAEs is balanced across the groups. However, 1 person had transverse myelitis 14 days after vax, determined to be possibly vaccine-related (recovered). There is another potentially vaccine-related SAE (fever), but not yet known if they are in vax group (recovered) 7/n
...1 other person had transverse myelitis 10 days after the covid vaccine: the person had pre-existing, previously undiagnosed multiple sclerosis & independent committee of neurologists determined it unlikely to be related... 8/n
...SD=standard dose, LD=low dose (half-ish). Vaccine efficacy was 14 days after 2 doses. For SD/SD in UK trial, wouldn't reach FDA threshold, because lower bound of confidence interval under 30%: 60·3% (28·0 to 78·2). In Brazil, would've squeaked in: 64·2% (30·7 to 81·5).. 9/n
...Meta-analysis of the 2 SD/SD groups is that 62% that was in the press release [62·1% (41·0 to 75·7)]. The LD/SD was 90·0% (67·4 to 97·0) & higher rate wasn't related to age or other factors they analyzed.
If you combine SD/SD & LD/SD you get 70·4% (54·8 to 80·6)...10/n
If you combine SD/SD & LD/SD you get 70·4% (54·8 to 80·6)...10/n
...Efficacy against asymptomatic infection not shown for SD/SD: 3·8% (−72·4 to 46·3). Efficacy in the LD/SD group was 58.9% (1·0 to 82·9) (not adjusted for age). It was 29 vs 40 events, so more data needed. (Not broken down by trial) ...11/n
...The long interval between 1st & 2nd injections is because they decided on 2nd injection later in the trials & it needed to be manufactured. Complicates other efficacy analysis. We need the US trial, test for that LD/SD hypothesis & see safety data from all trials. 12/fin
...0 severe Covid-19 events in covid vax group; 2 in control (not enough to calculate efficacy).
21% of UK trial were over 55; 11% Brazil. Too few, with too few events, to calculate efficacy
A mystery solved: number of people with neither SD/SD or LD/SD was negligible (<100)
21% of UK trial were over 55; 11% Brazil. Too few, with too few events, to calculate efficacy
A mystery solved: number of people with neither SD/SD or LD/SD was negligible (<100)
⬆️ Cont'd: Clearing up some questions/contradictions about the Oxford/AstraZeneca vaccine trials. Yes, the low dose was definitely a mistake, so we can ignore alternate stories. Here's the explanation in the Lancet paper & the SAP also shows it (Statistical Analysis Plan)... 14/n
Re a point I made at WIRED: wired.com/story/the-astr…
The SAP makes it clear that low dose (LD) & standard dose (SD) were not precise or consistent, & the highest LD gets close to the lowest possible SD. (I added the green for LD & blue on SDs) ... 15/n
The SAP makes it clear that low dose (LD) & standard dose (SD) were not precise or consistent, & the highest LD gets close to the lowest possible SD. (I added the green for LD & blue on SDs) ... 15/n
...If you want to read more, there's more on the differences in concentrations among vaccine lots within the types of doses - here for example in protocol for the Brazilian trial & there's discussion... 16/n
See Table S2 in App 1 for details of who's not included in the primary efficacy analysis (2 injections + >14 days): 5,807 getting covid vax in the primary efficacy analysis; by my count, 3,776 who aren't
See Table S5 ⬇️ for intervals between injections...17/ends
See Table S5 ⬇️ for intervals between injections...17/ends
• • •
Missing some Tweet in this thread? You can try to
force a refresh
