Replies to post #2088 on Dendreon Corp fka DNDNQ

[gofishmarko]

>>> All that assures is that Cox methodology per se was mentioned in the statistical analysis plan. It does not imply that the way DNDN conducted the Cox analysis was pre-specified.<<<

My own reading of the document statement would be that the analysis was prespecified , meaning the complete method and details of the analysis as would be understandable to someone knowledgeable in the field.

Of course , I'm assuming that Fair Disclosure also means unambiguous disclosure. Yeah , right.

In this case I think the statement was carefully crafted so as to convey whatever definition they needed it to convey , then or in the future , depending on the circumstances.

[walldiver]

<<In the case of Provenge, DNDN did not start talking about Cox until well into 2005, when it was already clear to management that 9902a was coming up short.>>

Correct. Top-line survival results for the 9901 Phase III came back in October 2004, with the full log-rank survival data presented at the ASCO Prostate Symposium in Feb 2005. Since DNDN was considering filing the BLA on the two earlier Phase III trials (9901 and 9902A), they knew that the FDA biostatisticians would perform a Cox regression analysis on the trial data to see if the prognostic factors were biased in favor of the treatment arm. So, the lead investigator for 9901 performed the Cox regression analysis according to the widely used Halabi statistical nomogram in time to present at the main ASCO Conference in May 2005.

The exact methodology of the Cox analysis in the statistical analysis plan for 9901/9902A has not been widely publicized, but I'm assuming that the SAP specified it should be done according to the currently accepted practice...and since the Halabi nomogram has been the most widely used method for AIPC trials over the past several years and 9901's lead investigator was a collaborator on the original Halabi nomogram study, I would imagine that the Cox regression analysis for 9901 was on the up and up.

The fact that the 9902A Cox regression was done using the same five final prognostic factors that were stat sig for 9901, even though only two of them were stat sig individually, indicates to me that 9902A's Cox analysis was also on the up and up.