PS - Also note that if you found the survival curves for the Zytiga 302 trial disturbing because the curves didn't separate until 20% of the population had died then the ARRY OS data should be even more disturbing because, as their PR indicates in statisticalese, they probably have placebo patients doing better than treated for a substantial fraction of the population.
Note: It is amusing to me that our positions on Zytiga and ARRY are flipped - so, FWIW, my criteria are:
a) As long as the curves are never inverted then I do not have a problem with saying the treatment is good.
b) Even if they are substantially inverted I still think it should be approved if every effort was made to predict who benefits and who doesn't AND, in the event it isn't possible to predict who benefits and who is hurt, then there is a black box or other similar warning to patients that it is a roll of the dice as to whether they will be hurt or helped.