>My understanding was that in practice (i.e., FDA eyes) the 2-sided P value would be rejected for showing a benefit if the curves were not visually clear wrt the treatment arm being on the correct side throughout.<
Do you have a citation for this? My impression is that early curve crossing in cancer trials is quite common, and I have not heard of any cases where log-rank analysis was rejected on this basis alone. T.i.a.
>Oh yes, it is possible the curves have only some minor early crossing, gets much better at 12-18 months, and a continuation of the 12-18 month effect could provide a winner. And it's possible that biologists will discover a species of airborn pigs.<
Market Data 
Markets 
AI
