Below are some facts and opinions developed over the past few years watching DNDN and the progress of Provenge:

 

1. 2004: Even though Overall Survival was not a prespecified endpoint in D9901, a logrank analysis of survival data at 3 years showed stat sig p = 0.01.

 

2. 2004/2005: DNDN peeked in D9902a to see if TTP and/or OS was supportive. Both failed. However, a standard back-stepping Cox analysis algorithm when applied to D9901 and D9902a resulted in highly stat sig OS results for both trials.

 

4. July 2005: A pre-BLA meeting with the CBER division of the FDA was held to present all available analysis results at that time. Dr. Pazdur was at that meeting and supportive of a BLA for Provenge per the below Q&A with Dr. Gold at the UBS Conference in Sept 2005 as reported by an attendee (Ackabaker):

 

Mitch- We learned from our pre-BLA meeting that we had been overly conservative about the primary endpoint- that TTP should not have been a major focus- Pazdur sees survival as clearly the basis for approval- our TTP result with a .06 p-value is definitely "supportive"- though we never could have gotten full approval on it"

 

5. Nov 2006: The BLA was completed. For the first time, the FDA statisticians got to play with the submitted data. They found that the log-rank analysis for OS in D9901 was good. However, they also found a big hole in the Cox analysis. There were missing data leading to patients being excluded. Thus, the analysis for both D9901 and D9902a were on subgroups, not the entire ITT groups. Further, the excluded placebo patients appeared to live longer than the excluded treated ones. That meant that the findings of the Cox analysis might not be meaningful, or not meaningful enough to support licensure.

 

At this point, my guess was that Dr. Pazdur, without fully appreciating the intricacies of immunotherapy, likely felt that the statistical data fell far short of the acceptable standards that the FDA have been putting into place. Also, he might have felt that his previous flexibility in agreeing to let DNDN file a non-traditional BLA was misplaced. If it was up to him, he probably would have just issued a Refuse-To-File letter since without the Cox analysis, D9902a was not anywhere close to being supportive to D9901 and the OS logrank analysis for D9901 was both not pre-specified and post-hoc.

 

Again, I guess that CBER with their expertise in immunology trusted the data enough to argue for holding an Advisory Committee to discuss the data. There must have been lengthy discussions inside the FDA since the final decision to hold a meeting was not announced until very close to the date of the meeting itself in late March 2007.

 

6. Mar 2007: The AC was held. The safety vote passed 17-0. The efficacy vote started rocky and almost failed. Early voters stated that they thought the data had merit but the requirement of "Establishing efficacy" was too high to vote yes. Dr. Goodman, Director of CBER, stepped up and clarified that the requirement by law was only "Substantial Evidence of Efficacy". The voting was restarted, resulting in a 13-4 vote for efficacy.

 

7. Apr 2007: Drs Scher, Hussain and Flemming wrote letters to the FDA urging to delay approval. Dr. Scher's effort was aided by researchers at the NCI. Dr. Pazdur either started that effort or at least gave them the green light since Scher and Hussain were members of ODAC.

 

I wrote some early rebuttals to those letters on this board. Later, I helped a fellow called Wu, Rancherho and others to put together some of that contents and also contents from an independent letter by David Miller into a comprehensive letter that was sent to the FDA.

 

8. May 2007: The internal fighting at the FDA finished. Dr. Pazdur's side won the argument that the submitted data were sufficiently controversial that they should wait for a more definitive analysis from IMPACT. A CRL letter was issued asking for more efficacy data.

 

9. Mar 2008: The company, perhaps with help from the FDA, redesigned the IMPACT trials to change the alpha allocation method and the interim and final triggers for analysis. These changes were designed to both speed up the time line and to increase the chance of success for the interim look.

 

10. Oct 2008: The company announced that the interim look did not achieve stat sig but also gave the estimated HR at that time showing a trend of success for IMPACT. This board had many months of hand-wringing and bickering about how the final analysis will come out.

 

11. April 2009: The company announced that IMPACT had achieved stat sig. A new version of the BLA to be based on IMPACT as the pivotal trial will be filed in 4Q2009 with approval expected a maximum of 6 months afterward.

 

12. Some time in early 2010: Provenge approved. Well, ok, that's an opinion. Let's hope.

 

The delay decision, in hindsight with IMPACT data, is now known to be wrong but it was not unreasonable at the time given the nature of the data. From a data perspective, the faults could be laid at both DNDN and the FDA doors. In DNDN case, they were not careful enough with their data analysis, esp. the Cox analysis. In the FDA case, the decision makers did not have the wisdom and courage to follow the advice of the AC and their own internal people to approve.

 

No doubt a number of people involved in the delay of Provenge benefited handsomely from that. But we must remember that the opening for their success was the controversial quality of the data at that time. That was a hard and personal lesson for me. Even as I understood the science and trusted it, as observed at the ASM today, FDA approval is foremost about clean statistical data. That depends on factors beyond science, ie, trial engineering.

 

That's why, after the CRL, I spent much time learning to simulate IMPACT to see what factors could derail it or make it work. I shared some of that on the board as I was learning. I should also mention that P3analyze helped me much with certain aspects of statistics that I was not familiar with and firmed my thoughts on how the simulator should work.