The P value is for the entire OS (or PFS) data. It is not for the median/landmark/HR/restricted mean datapoint. If median. mean, 36 month and H/R OS are all presented, there is one P value for all.
Yes, the relative risk reduction is H/R in another flavor.
Regardless the analysis chosen, the comparator is the 100 randomized into the placebo arm. The single median, mean or landmark OS is not in itself useful for approval. In a randomized trial, it is that number compared to the other arm (even when using historical controls).
I have no problem with choosing an alternate metric such as 36 month OS or even restricted mean. I agree with some of your points that median can not be very reflective of true efficacy, and H/R is often misunderstood.
As far as some PRs just stating that they were met with no data. Can work for some companies but I do not think LP has garnered enough trust to get much PPS traction with it.
And "bog" is nothing I have ever heard of in this context.
News 
Market Data 
Discover 
AI
