The 148 week data also raised doubt about the S1R variant playing a riole as it did not correlate to a worse response for MMSE performance.
A very important point.
Just expanding on what you already said...one would expect initially identified correlation to be maintained by the pure fact that once decline rates are established one would expect, at least on average, they'd be roughly maintained...at least relative to other subjects. That is purely by chance. If one adds a causative involvement (i.e. 2-73 "works" better in those with wtSIGMAR1) then maintenance of that correlation should persist.
Contrapositively, loss of this correlation over time argues against a causative role in the original correlation.