<What factors led to the relatively poor results of 9902A? How much of the shortfall in log rank p-value and HR is attributable to smaller size? trial imbalances favoring placebo, etc>
If the trial was run well, the small size would have had relatively little effect on the results while the sicker population and imbalances such as bone mets counted for more. However, a large factor might have been that there were more patients in D9902a than in D9901 who did not get full treatment (around 15% vs 5%). Per the patient distribution decision tree in the stat review, there was even one patient on the treatment arm of D9902a who did not get treated! Because of that, the small trial size might have played a larger role due to a reverse cross-over situation where a significant number patients on the treatment arm were acting like placebos. That might have had some effect on the poor TTP result in D9902a too.
<Why should the results of 9902B look much more like 9901 than 9902A at the interim and final?>
The larger enrollment number will make the trial more balanced across prognostic factors, both the ones accounted for and the ones unknown (the latter is really what people refer to when they say that a large trial is more robust than a small trial - we don't always know everything, e.g., genetic variations, etc.). In addition, stratifying on bone mets and Gleason scores should help. Last, per the CD54 data, the early enrollees on the treatment arm were compliant and hopefully the recent ones would be too due to better published trial data. So there should be less of the reverse cross-over issue above.
