Replies to post #444411 on Anavex Life Sciences Corp (AVXL)

[Doc328]

For comparison, I just looked at the two PDFs you posted, and I couldn't find any hierarchy of alternative statistical tests for any single endpoint in the lecanemab SAP

They are there but not presented in an obvious manner. For lecanemab, the primary endpoint and how the statistics are performed are clearly described in 5.1.1 (p. 442) detailed in 5.4.1 (p. 449). They had only the one primary endpoint (CDR-SB change from baseline, LSM using MMRM). Supplementary analysis of the CDR-SB outcome included effect of ApoE status on the CDR-SB, use of AChEI (i.e. aricept) on the CDR-SB, PP analysis instead of MMRM (p. 451). Additionally several sensitivity analyses (p 449-450) with different imputation methods for CDR-SB are described. Then exploratory supplemental analysis (p. 455-456) looks at CDR-SB rate of change, and time to worsening . Thus there is the primary analysis of CDR-SB (highest in the statistical hierarchy, followed by supplemental and then exploratory --- one outcome is analyze 8 different ways but just one 'counts' as THE primary endpoint.

The aducanumab SAP spells this out more clearly on page 19 and 20. (In the table primary analysis, sensitivity analysis and supplemental analysis is nicely presented - note logistic regression/odds ratio is supplemental)
https://classic.clinicaltrials.gov/ProvidedDocs/00/NCT02477800/SAP_001.pdf

Likewise for trofinetide. THE co-primary endpoints are RSBQ change from baseline to 12 weeks and CGI-I change from baseline to 12 weeks (p. 34) with the analysis set as defined a few pages earlier and imputation with MMRM/MAR. Sensitivity analysis uses different analysis sets and different imputation methods of the same endpoint. Supplemental analysis includes responder analysis of CGI-I and RSBQ (this would use logistic regression and allow calculation of an Odds Ratio). Note the responder analysis is lower on the statistical hierarchy -- it is an alternative way to look at CGI and RSBQ and is present in the SAP. Despite using the same outcome (CGI or RSBQ) it is not the primary endpoint and Acadia would have been laughed at if they tried to claim success with the responder analysis if the primary analysis had failed.

As can be seen there are many analyses in the SAP for each outcome but only one is at the top of the hierarchy in regards to method, analysis set and imputation.

How this relates to Anavex: Similar to Schrodinger's cat, while the SAP is hidden from the public, it is theoretically possible for BOTH the Odds Ratio/Logistic regression AND the difference in means to be THE primary analysis at the top of the hierarchy, once the box is open only one of these methods can be at the top. Either Missling lied in December 22 about Odds Ratio or he lied in June about LSM being at the top (and was this the primary imputation or part of sensitivity analysis). Your knowledgeable friend would likely agree with the above.