I get a p-value of 0.503 for Cog. That's very minor. If you use slightly lower n-values (excluding dropouts), say n=161 placebo and n=301 treatment, then you get 0.494.
So, I wouldn't worry about it.
However, AVXL did NOT use this method to compute p-value. They are getting 0.03 (which likely comes from evanmiller.org like calculation and one-tailed calculation).