As to when, it would have been when the unblinded data was delivered to them after the decision to stop the trial, probably within a few hours to a few days after the decision. Part of the enrollment process was to identify 'most likely clinical diagnosis' (ACAD's words from the pres). So the full dataset would haave had all the info into ACAD then (approximately early Sep 2019).
As to why, my guess is that is the normal thing that scientists do. I doubt they had any thoughts of downside given they had FDA agreement on trial design, stopping criteria, etc. And remember the data pres was at CTAD = "Clinical Trials for Alz Disease", with an audience of scientists So detailed trial analysis was the point of the conf. As I relook at the pres, the data on subtype was not the focus of the pres. There were 38 slides, with 3 (slides 26-28) focused on "Exploratory Analyses by Most Likely Clinical Diagnosis" (their section title) - 1 section title slide, 1 on open label results by subtype (graph showing 73% to 83% reduction in SAPS-H+D with line for each subtype starting at 0 and ending at wk 12 in range -18 to -23), and 1 on blinded outcomes (relapse rate by subtype pima v pbo). The relapse by subtype numbers were in a footnote on the 3rd of those slides. So the subtype aspect was not the focus of the pres. The other 35 slides were focused to combined results, as the trial was intended to be evaluated.
So overall I think no subterfuge by ACAD. Rather, FDA (unfairly, wrongheadedly, in contravention of their own foundational criteria) changed the evaluation criteria.