Testing subgroups requires some method of defining alpha allocation else something will almost always stick by pure luck. Trials have dozens of p[refined subgroups.
What if the trial had worked in high risk only? In men? I those with stage 1? In US patients? In those over 65?
There are plenty of ways to do this. But they must defined in advance.
Just looking at all the subgroups and claiming victory if one works is what DD was talking about with his M&Ms.
CVM themselves said they powered the trial at 0.05 for OS in the entire population. That is end of story.