Biogen's analysis, which showed a positive result in the first trial (EMERGE) in all of its primary and secondary cognitive endpoints, and a negative result in the second (ENGAGE), was done on the initial, pre-designated endpoints. It was only a "reanalysis" in the sense that it (rightly) used all of the data collected during the trial (through March 2019), as opposed to the futility analysis, which used data collected only up until December 2018, when that futility analysis began. Note that the futility analysis itself did not conclude that both trials were unlikely to reach positive results, merely that the second trial was (and as a result, both trials were terminated, since a positive result from both trials was the requirement for the standard FDA approval path).
Separate from this, Biogen *also* conducted a post-hoc analysis on a subgroup (participants enrolled after the fourth version of the trial protocol) which effectively received a higher dosage of the drug. In that post-hoc analysis, the reduction in the pace of cognitive decline was 30% (95% CI [1%, 60%]) in EMERGE and 27% (95% CI [-3%, 57%]) in ENGAGE. There are good reasons to be skeptical of doing this post-hoc analysis in the first place, but 1) it wasn't the analysis which led Biogen to claim a positive result in EMERGE, and 2) I would argue (to the extent we suspend our skepticism of its rationale) it's a more promising result than you've portrayed it as: the post-hoc analyses from both trials yielded very similar conclusions, both skirting the edge of statistical significance, and both with effect sizes of about a ~28.5% reduction in the pace of cognitive decline, which I'd consider more than a modest clinical benefit (that would buy 40% more time for someone).
See especially pages 17, 27, and 43 (I converted the raw CI's on page 43 into CI's in percentage decline, though upon reflection this was probably not completely kosher of me, since there's also a CI on the placebo).
I do think that, given previous negative results with anti-amyloid drug results, we should still be extra skeptical on Bayesian grounds.
Also, with the difficulty associated with running a confirmatory trial that's placebo-controlled (as opposed to a historical control or whatever), approval via AA was still a questionable idea.
Note that amyloid-targeting antibodies have had other successes, such as donanemab's recently concluded phase 2 trial [1] showing a 32% slowdown in cognitive decline. Furthermore, many of the nominal failures are actually modest evidence of efficacy when seen through a Bayesian lens: the phase 3 trial of solanezumab [2], for example, showed a slight cognitive benefit, which came close to but did not reach nominal statistical significance.
All of that said, it's hard to do a Bayesian update (other than to the need for earlier intervention) in the absence of a serious contender to the amyloid cascade hypothesis, which still has very strong evidence behind it. I discuss some of this evidence in [3].
Good points. i believe there are something like 10 anti-amyloid drugs that have been tried in humans. Not sure how many have gotten to phase 3.
I don't know the detailed results for each but my understanding is that while there's been occasional glimmers of success, the general trend is failure.
So 2 successes out of 10 trials is not strong evidence.
And again, if there was suggestive evidence it worked, it seems like the best step was to do another trial with better patient selection + higher dose, not do AA, where a good confirmatory trial is really tough.
All that said, you make the best case for it I've seen anywhere, and when time permits, you should post it somewhere.
You're making a better case than any doc I've seen on Reddit or Twitter!
I've heard from people in the field that Tau is becoming more of a focus. There was talk of herpes being involved a few yrs ago but iirc it didn't pan out.
Another point on the amyloid-β → phosphorylated tau causation: the aducanumab trials themselves showed strong and highly statistically significant reductions in p-tau, despite aducanumab only targeting amyloid (pp. 21, 31, and 32–34 in [1]).
So if you believe p-tau causes neurodegeneration (which pretty much everyone does these days), but you don't believe in the amyloid cascade hypothesis, you've got some serious explaining to do.
Tau is more proximately connected with neurodegeneration than amyloid-β, but tau pathology doesn't appear to spread from the medial temporal lobe until amyloid-β has spread throughout the brain. See the paragraph in [1] beginning, "several lines of evidence suggest Aβ deposition may be required for progression of tau pathology in AD". (This is one of the main reasons for believing that early intervention with amyloid-targeting therapy might be important.)
The connection to herpes (and other pathogens) seems to be real, in that amyloid-β appears to be an antimicrobial peptide. However, as argued by [2], this is nicely integrated into the amyloid cascade hypothesis: amyloid plaques form in response to microbial infection, but then fail to clear out even after they've done their job, leading to the cascade of events we call Alzheimer's Disease.
A correction on point 1 in your timeline:
Biogen's analysis, which showed a positive result in the first trial (EMERGE) in all of its primary and secondary cognitive endpoints, and a negative result in the second (ENGAGE), was done on the initial, pre-designated endpoints. It was only a "reanalysis" in the sense that it (rightly) used all of the data collected during the trial (through March 2019), as opposed to the futility analysis, which used data collected only up until December 2018, when that futility analysis began. Note that the futility analysis itself did not conclude that both trials were unlikely to reach positive results, merely that the second trial was (and as a result, both trials were terminated, since a positive result from both trials was the requirement for the standard FDA approval path).
Separate from this, Biogen *also* conducted a post-hoc analysis on a subgroup (participants enrolled after the fourth version of the trial protocol) which effectively received a higher dosage of the drug. In that post-hoc analysis, the reduction in the pace of cognitive decline was 30% (95% CI [1%, 60%]) in EMERGE and 27% (95% CI [-3%, 57%]) in ENGAGE. There are good reasons to be skeptical of doing this post-hoc analysis in the first place, but 1) it wasn't the analysis which led Biogen to claim a positive result in EMERGE, and 2) I would argue (to the extent we suspend our skepticism of its rationale) it's a more promising result than you've portrayed it as: the post-hoc analyses from both trials yielded very similar conclusions, both skirting the edge of statistical significance, and both with effect sizes of about a ~28.5% reduction in the pace of cognitive decline, which I'd consider more than a modest clinical benefit (that would buy 40% more time for someone).
Thanks for those insights. I'll definitely add that in. Do you have an article you can link to read up on those issues?
Most of the data I referenced comes from https://investors.biogen.com/static-files/f91e95d9-2fce-46ce-9115-0628cfe96e83.
See especially pages 17, 27, and 43 (I converted the raw CI's on page 43 into CI's in percentage decline, though upon reflection this was probably not completely kosher of me, since there's also a CI on the placebo).
Thanks. I'll update the post once I read through that in-depth.
I do think that, given previous negative results with anti-amyloid drug results, we should still be extra skeptical on Bayesian grounds.
Also, with the difficulty associated with running a confirmatory trial that's placebo-controlled (as opposed to a historical control or whatever), approval via AA was still a questionable idea.
Note that amyloid-targeting antibodies have had other successes, such as donanemab's recently concluded phase 2 trial [1] showing a 32% slowdown in cognitive decline. Furthermore, many of the nominal failures are actually modest evidence of efficacy when seen through a Bayesian lens: the phase 3 trial of solanezumab [2], for example, showed a slight cognitive benefit, which came close to but did not reach nominal statistical significance.
All of that said, it's hard to do a Bayesian update (other than to the need for earlier intervention) in the absence of a serious contender to the amyloid cascade hypothesis, which still has very strong evidence behind it. I discuss some of this evidence in [3].
[1] Mintun et al (2021). Donanemab in Early Alzheimer’s Disease. https://www.nejm.org/doi/full/10.1056/NEJMoa2100708
[2] Honig et al (2018). Trial of Solanezumab for Mild Dementia Due to Alzheimer’s Disease. https://www.nejm.org/doi/full/10.1056/NEJMoa1705971
[3] https://astralcodexten.substack.com/p/highlights-from-the-comments-on-aducanumab/comments#comment-2667148
Good points. i believe there are something like 10 anti-amyloid drugs that have been tried in humans. Not sure how many have gotten to phase 3.
I don't know the detailed results for each but my understanding is that while there's been occasional glimmers of success, the general trend is failure.
So 2 successes out of 10 trials is not strong evidence.
And again, if there was suggestive evidence it worked, it seems like the best step was to do another trial with better patient selection + higher dose, not do AA, where a good confirmatory trial is really tough.
All that said, you make the best case for it I've seen anywhere, and when time permits, you should post it somewhere.
You're making a better case than any doc I've seen on Reddit or Twitter!
I've heard from people in the field that Tau is becoming more of a focus. There was talk of herpes being involved a few yrs ago but iirc it didn't pan out.
Another point on the amyloid-β → phosphorylated tau causation: the aducanumab trials themselves showed strong and highly statistically significant reductions in p-tau, despite aducanumab only targeting amyloid (pp. 21, 31, and 32–34 in [1]).
So if you believe p-tau causes neurodegeneration (which pretty much everyone does these days), but you don't believe in the amyloid cascade hypothesis, you've got some serious explaining to do.
[1] https://investors.biogen.com/static-files/f91e95d9-2fce-46ce-9115-0628cfe96e83
Tau is more proximately connected with neurodegeneration than amyloid-β, but tau pathology doesn't appear to spread from the medial temporal lobe until amyloid-β has spread throughout the brain. See the paragraph in [1] beginning, "several lines of evidence suggest Aβ deposition may be required for progression of tau pathology in AD". (This is one of the main reasons for believing that early intervention with amyloid-targeting therapy might be important.)
The connection to herpes (and other pathogens) seems to be real, in that amyloid-β appears to be an antimicrobial peptide. However, as argued by [2], this is nicely integrated into the amyloid cascade hypothesis: amyloid plaques form in response to microbial infection, but then fail to clear out even after they've done their job, leading to the cascade of events we call Alzheimer's Disease.
[1] Long and Holtzman (2019). Alzheimer Disease: An Update on Pathobiology and Treatment Strategies. https://www.sciencedirect.com/science/article/pii/S0092867419310074
[2] Eimer et al (2018). Alzheimer’s Disease-Associated β-Amyloid Is Rapidly Seeded by Herpesviridae to Protect against Brain Infection. https://www.cell.com/neuron/fulltext/S0896-6273(18)30526-9