2023-10-09 11:37:28 ET
Summary
- Anavex Life Sciences Corp. presented follow-on data from its Alzheimer’s disease trial.
- This data uses a statistical methodology that is difficult to understand.
- It appears that the trial, despite the claims, was not successful.
The entire Alzheimer's disease ("AD") field got a boost on September 14 after beleaguered small cap biopharma Anavex Life Sciences Corp. (AVXL) said that blarcamesine demonstrated slowing of cognitive decline in a key phase 2b/3 study conducted in ~500 participants with early symptoms of Alzheimer's, including mild cognitive impairment and mild dementia. The study is not new, but the data is from a follow-on analysis - so it is kind of new.
I have been following AVXL for a while now, although I have never bought stock because it is just too controversial for my taste. When I say controversial, I mean this:
This comment is from Sept. 14, right after the follow-on data was published. This comment represents the bearish view that AVXL failed its Alzheimer's trial and also produced faulty Rett syndrome data. I discussed these questions in earlier articles. I referred to this here because this sort of constant controversy makes AVXL stock highly volatile, the kind of stock that risk-averse investors like myself like to avoid.
Going back to the phase 2a trial, these are the primary and secondary endpoints:
Primary (Current)
-
ADAS-Cog (Alzheimer Disease Assessment Scale-Cognition) :48 weeks
-
ADCS-ADL (Activities of Daily Living) :48 weeks.
Secondary (Current)
-
CDR-SB (Clinical Dementia Rating Scale Sum of Boxes) :Reduction in cognitive decline assessed from baseline over 48 weeks with ANAVEX2-73 compared with placebo using the Clinical Dementia Rating Scale Sum of Boxes (CDR-SB)
-
Number of participants with treatment-related adverse events as assessed by CTCAE v4.03 :Assess the safety and tolerability of ANAVEX2-73 compared to placebo.
So, we have two co-primary endpoints here, so the significance or alpha level will have to be calculated a little differently here in order to avoid a Type 1 error (avoid a false positive). However, the correction needs to be optimal as well, in order to avoid false negatives (Type II errors). This is called the Bonferroni correction, and the idea is to simply divide the alpha level you would normally select for a single primary endpoint for your desired stringency by the number of primary endpoints. Thus, if 0.05 is your desired alpha level, here, under the Bonferroni correction, it becomes 0.025 for each co-primary endpoint. Note that this is for each co-primary endpoint, not just for one, and this concept is called the familywise error rate ((FWER)), which basically addresses the overall probability of making at least one Type I error (false positive) in a set of multiple hypothesis tests or comparisons.
Keeping that discussion in mind, read what Anavex said in their data announcement :
The trial is successful in meeting the co-primary endpoints if the significance of each endpoint is P < 0.05, or if the significance of only one co-primary endpoint is P < 0.025. If only one primary endpoint is significant at an ? level of 0.025, then the secondary endpoint will be evaluated at the same level of 0.025. The trial was successful, since the differences in the least-squares mean ((LSM)) change from baseline to 48 weeks between the blarcamesine and placebo groups were ?1.783 [95% CI, ?3.314 to ?0.251]; (P = 0.0226) for ADAS-Cog13, and ?0.456 [95% CI, ?0.831 to ?0.080]; (P = 0.0175) for CDR-SB in patients with early Alzheimer's disease.
This is a very complicated data announcement, obfuscating the fact that data for the second co-primary endpoint, ADCS-ADL, has not been provided or considered. This design also does not appear to use the Bonferroni correction at all, despite seeming to do so, because note the very first sentence - "if the significance of each endpoint is P < 0.05" - then there is no Bonferroni correction applied. The second part really says, that supposing one primary endpoint does not show stat sig, then the second would have to have a much more stringent alpha level of 0.025; however, under the Bonferroni model, even if such a schema were allowed, it would have to be doubly stringent. Plus there is no schema where, if a co-primary endpoint fails to achieve stat sig, you take up a declared secondary endpoint and, simply by providing it with the same alpha level (see the highlighted selection), you can elevate it to some sort of primary endpoint and declare the trial successful under the Bonferroni model.
The clear takeaway from the data here is that blarcamesine failed to meet the ADCS-ADL co-primary endpoint, therefore, at least that part of the trial was a failure, and since the two co-primary endpoints cannot be independently assessed in a trial with both of them together, the entire trial, statistically, was a failure.
I do not have a PhD in Statistics, so I could be wrong in my analysis here. But after nearly two decades reading biopharma data press releases, I can tell you this - when a company has clear and strong data, their data press releases are never this convoluted. They make it very clear and easily understandable that they have a winner. If there is any doubt, do NOT give the company any benefit of the doubt. This is my guiding philosophy in investing in the murky if profitable world of biopharma.
For further details see:
Anavex: The September Data