and, of course,
you would like a survival scatter plot to show that
the placebo ( at least the true placebos) lie
pretty much on a line observed=predicted+small_delta to
make up for historical calibration( gleason inflation if
applicable, better nutrition, pomegranate juice consumption,
etc). If the placebo predicted vs survival curve
doesn't have some decent fit, the "corrected" provenge
results may be suspect. All the fit quality parameters
need to be examined but you have to be sceptical if
there are gross problems in the simple stuff- see my earlier post on the incorrect published structure due to a software bug- if it doesn't look right you need to be sceptical.
You can quote a median survival difference and a p value but looking at the graph - which is less susceptible to processing errors- can add or reduce confidence in the derived numbers.