Conclusions: Neither study D9901 nor study D9902A met any study objectives A review and analyses of the data submitted, including sensitivity analyses and review of death events, supported the finding of an increase in the median survival reported by the sponsor in APC8015 arm compared with the APC-Placebo in study D9901. However, the lack of a pre-specified primary method for survival analyses renders it difficult to estimate the type I error of this survival analysis. In addition, the six month difference in median survival times between D9901 and D9902A despite similar study design, inclusion criteria, and baseline characteristics, suggest that the eligibility criteria did not define a homogeneous population in these small studies. These observations increase the possibility that the survival difference in D9901 might be attributable to chance.
Safety and tolerability: APC8015 was generally well tolerated; approximately 90% of subjects in the two studies received the 3 infusions specified by the protocol. The most frequently reported adverse events in APC8015 treated subjects were transient chills, fatigue, and pyrexia. However, the increased CVA frequency reported in subjects treated with APC8015 constitutes a potential safety concern.
Let's first address the efficacy. Provenge has failed ALL primary and secondary endpoints in both their trials. Survival was not even an endpoint in the trials. Below is a summary of the secondary endpoints found in the briefing documents:
Note the 0% response rate as well as the complete and utter failure to meet any of these endpoints. And now on to the primary endpoint, time to disease progression. A picture says a thousand words so here is the graph of TTP:
In other words, there is basically no treatment effect on tumor size or tumor progression or even tumor related pain. The only thing positive that ever came out of the 9901 trial was the survival benefit. But as the FDA statistician explains, as the analysis was not pre-specified it makes sense to look at several to see how the data would have come out. Take a look at the table below. In 3 of the 6 models, they actually miss significance, and in one that was a hit, they just barely hit with a p=0.048. Also note that in the test where they saw a p=0.002, there were a whopping 10 datapoints missing. And if these patients were on drug, they tended to have reacted worse to treatment. If they were on placebo they were reacting better. In other words, those 10 missing data points helped the Provenge arm look much better than it was.
Treatment effect on overall survival using Cox model with different set of covariates (total subjects: 127)
*: The sponsor’s analyses and confirmed by the reviewer
~: Reviewer’s analysis
So, in conclusion, Provenge missed all their endpoints and even their survival signal is in question. Now let's move onto safety. People have been arguing that with a survival benefit and a clean safety profile, Provenge should get approved. Apparently, it might not be so safe. There appears to be an increase in the number of cerebrovascular accidents, with 5 people dying in the Provenge arm compared to 0 in placebo. The key conclusion here is "although these differences did not reach statistical significance, the increased CVA frequency is a potential safety signal."
With these briefing documents in hand I am definitely looking forward to the actual panel meeting and vote on Provenge.
Disclosure: Author has a short position in DNDN