The multivariate analysis performed in the Phase II showed an HR of .5 when isolating the variable of control vs tx, compared with the overall HR of .61. I'm in the same camp as those here who aren't expecting a repeat of .61, but doesn't that multivariate isolation indicate that the selection bias was AGAINST tx?
They don't show the p-value but the top of the 95% CI interval for that factor was .87, so it would have been a rather small p-value, considering that p = .05 would be where the top of the CI interval was 1.00
This is in Table 3 on page 4252 of the 2010 Journal of Clinical Oncology article.
Selection bias I mentioned is something different.
If a co runs a number of Phase 2 trials for a new drug (or drugs) in different indications, some/most of the variations in the results would be due to randomness. However, the co would most probably select the indication where the drug seems the most active to move to Phase 3 and would hope to replicate the results of P2 in P3. However, you should expect the Phase 3 trial to show less effect than the one observed in P2.
For example, say the co runs 10 Phase 2 trials and gets HR=0.7 to HR=1.0, which is completely normal if say the true HR were 0.9. Then it's very normal for the phase 3 to regress to HR=0.9, i.e., show less effect than the one observed in P2.
This is the selection bias I was talking about. If the co had only one drug and one trial, you should still expect the P3 to show less effect than P2 due to a different bias (survival bias, as we only observe companies that move drugs to P3).