I recently asked the same questions of an oncologist with some background as a clinical investigator.His answers, while general, were along the following lines;
1. Clinical investigators with experience in a given life threatening disease on a Data Safety Monitoring Board would be expected to be sensitive to any trend in safety or clear lack of efficacy or exceptional evidence of efficacy, especially if a trial were unblinded. If any of those circumstances were to arise, they could be expected to bring it to the attention of the sponsor and the FDA. A close call on the efficacy question would be unlikely to give rise to early termination, but clearly poor or outstanding results may.
2. The second question would relate to two issues: both the size of the trial and whether or not there was any dosing and duration records on subsequent docetaxel treatment (The Provenge AC Briefing Materials suggested there were none, except a basic yes/no on subsequent chemo or taxane use). A retrospective subgroup analysis is hypothesis generating. However, retrospective subgroup analyses are expected to show comparable benefits among all subgroups (for example, that there are no well defined subgroups for which there are little or no treatment effects).In a relatively small trial subgroup testing becomes particularly difficult and relatively unreliable.That goes to a general bias for larger trials especially where there are other statistical issues.
While these answers seemed sensible, they are quite general and are not completely reducable to specific numbers. As I stated, I am neither a statistician, medical doctor or clinical investigator. Just a discussion, FWIW, that I thought interesting and somewhat relevant.
AI
