The concern for the interim look is the large percentage (20%) of patients who enrolled in the final 18 months of the enrollment period. These patients, if still alive, will have to be censored at the time of the interim trigger in Sep/Oct 2008. If IMPACT goes like the other Phase III trials, there won't be much separation between the two arms for the first 18 months. The censored patients cloud up the p value.
DNDN has said that if they get better than 0.05 but worse than the interim alpha (probably around 0.01) using the primary analysis of Cox regression, they will submit to the FDA for approval again, but this submittal would be outside of the SPA. I can see the same elements who fought approval this spring fighting it again next year, especially if log rank is worse than 0.05 and Cox analysis drops it below 0.05. Of course, they are within the guidelines of the SPA if the Cox p value is better than the allocated alpha. Even so, longs will breathe easier if both log rank and Cox are better than the interim alpha. The best thing 9902B has going for it is that patients who live longer than 36 months get credit for those extra months. That's one of the factors that hurt the 9901 and 9902A p values. They had to stop at 36 months for the long-term survivors, and ~25% of the Provenge arm in 9901 probably lived for 65 to 84 months after randomization.
I'm not sure how they are going to run the Cox regression exercise. I had heard that they were only going to use the three stat sig prognostic factors that were common to both 9901 and 9902A (I believe bone mets, LDH, and PSA)...but I could be wrong. DNDN isn't saying anything.
