1. That's a simple way to make a correspondence between the number of events in a trial with 225 enrollees to a trial with 512 enrollees. From their, we can do a sanity check if they overpowered or underpowered the trial back in 2005 under "their" then assumption of proportional hazard because we know the data was NPH. With current data, we know they did underpower the trial somewhat since the computable HR was about 1.38 for 360 and not 1.45. That made the reduction of 56 events a bit harsh.
2. The final HR for 304 should be slightly higher than 1.30 so you are in the right range.
3. For accuracy, the conditional probability must be computed based on all known conditions and not just the blind fact that the interim missed. If you take into account the announced HR and its CI, then my estimation is that the conditional probablility would be in the same range of 60% as the power.
As I said in an earlier post, if the final misses, it won't miss by much. A treatment with plausible efficacy data and with little safety issue will not be discarded just because the trial was slightly underpowered. In fact, at that point because the data will be unblinded, it will be trivial to run the kind of sims that I am running to compute much more exactly what the conditional probability of success will be for an event count higher than 304. Both the FDA and the company will find a way to continue testing the drug while making it available in some form (probably via extending the ProACT trial) if that comes to pass. I can't recall off-hand but P3analyze posted an example cancer drug approved on that scenario some time ago.
