Earlier quoted context omitted.
Nothing says p-hacking like a drug company canceling a trial and then, a few months later, re-analyzing the data and magically finding a benefit.
FDA usually puts some quite strict limits on the amount of p-hacking and general shenanigans they allow in clinical trials. But there's no approval yet, and no data to look at, so it's too early to judge. Anyway, usually Alzheimer trials are a jumble of noise. The cognitive measurements are imperfect and patients are improving and deteriorating randomly, taking a very different disease course than you'd expect from t…
> [W]hile the new larger EMERGE data do look like there’s a dose-responsive effect in the primary measurement (the CDR-SB scale) and three secondary dementia scores, the ENGAGE data actually show worse than placebo for CDR-SB at the higher dose (see slide 19 in Biogen’s presentation). They’re also worse than placebo in the MMSE scoring, but both the ADAS-Cog13 and the ADAS-ADL-MCI scales show apparent dose-responsive improvement. I have no idea why four different dementia rating systems should give such different results, and to a first approximation, no one else seems to have a good idea, either. So we’ve got the EMERGE/ENGAGE difference to explain (you’ve seen Biogen’s explanation, higher dosing later in the trial), and the variable results within the ENGAGE trial itself. And who are these high-dose patients in the ENGAGE study that Biogen says line up with the EMERGE trial? Ed Silverman at Stat quotes an analyst saying that it looks like fewer than 100 patients out of the whole set. That reminds me a bit of Biogen and Eisai’s tease of “We’re seeing positive data!” during the clinical trials, which also turned out to be based on a smaller number of patients than one would want when making a call on Alzheimer’s efficacy.[1]
He's written a fair amount about Alzheimer's,[2] and it's been an informative and useful commentary (at least for me).
[1] https://blogs.sciencemag.org/pipeline/archives/2019/10/23/th...
[2] https://search.sciencemag.org/?searchTerm=Lowe alzheimers&order=tfidf&limit=textFields&pageSize=10&&