There is at least one computational error in the Anavex presentation in the confidence intervals for the relative reduction in cognition (too narrow for a ratio of random variables with such standard deviations)
This is not material. What is material is how the p-values are calculated. This post replicates the p-values in the same way I and a few others have, that is using a one-tail test. Such a test is not the norm and goes against the guidance of FDA that the use of a two-tailed test is expected. In a nutshell, if they used a one-tail test, the reported p-values is half of what a two-tailed test would produce. If you double the p-values, you can see that the trial would fail to meet the original endpoints (change in the ADAS-Cog and CDR-SB scores) and presumably the unreported ADAS-ADL score change at the 5% significance level. At least for the pooled cohort.