frrol, you are the first one to vocalize a concern I have with an interim peek. I appreciated your comment "we do a lot of subgrouping", and will expound on it below.
Regarding Hampel's peer reviewed AD A2-73 phase 2a publication. Recall the famous Figure 3a, showing the "Group" change in ADCS-ADL over 148 weeks.
Group2 seems to INCLUDE the super responders in Group1, yet Group2 is 15 points lower. The only difference appears to be that Group1 has APOE3 alleles. What that means is that all NON APOE3 alleled patients perform worse that the super responders in Group1. OK, so? Well if you solve for this unknown group, then these NON APOE3 alleled patients are really pulling the average down, and the drop of only 15 points is thanks to the super-responders keeping the average only down 15 pts!! That is my interpretation of the plot, and maybe where your comment leads.
Another concern about an interim look at the patient data, is the placebo patient data is mixed, or averaged in with all the other patients, just like we saw in the paragraph above. So the placebo patients will pull the scores down too.
Oh yes, then there are all the patients with the Sigmar1 variant and the COMT variant. They will likely be non responders and pull the scores down too!
Interim peeks do NOT unblind the patient data, so it all gets averaged together as described above.
Finally,the data analysis will take a lot of time. I strongly suspect that the trial would end, while they were still analyzing the "interim peek" data,...then they could stop the "peek" analysis and start on the actual unblinded final data analysis.
For these reasons (and MORE) I am opposed to an interim peek. Dr.M will need to have ALL the data AND unblinded, and then he will be able to present a clear (as you say) "subgrouped" explanation for the patients that are super responders, and maybe some "tweeners", and the non-responders. It will then be VERY CLEAR.
I know of one cancer therapy trial, that was unblinded and it shed no light on efficacy at all imho. That trial is what I would call "dirty" for various reasons, but the result is what I outlined above.
frrol, I am not sure if the above is what you were alluding to, so if you have additional light to shed on this, please do. I totally agree with all the points you made about not peeking!
News 
Market Data 
Discover 
AI
