On Wednesday, development-stage drugmaker Dendreon
The most significant news about Dendreon's revisions to the IMPACT study for its potential prostate cancer drug Provenge is that the study completion date has been moved up about a year from 2010 to the second half of 2009, and the odds of a successful clinical trial are essentially kept the same. There are some downsides, or rather missed opportunities, in such a move, though.
First, a little background
The way the clinical trial process works is that drugmakers are allowed to peek at the data their drugs are generating in a clinical trial before a study's completion, but at the cost of the FDA raising the bar for the study's success for every data peek a drugmaker makes. The penalty the FDA assesses for interim looks is to avoid drugmakers taking as many interim looks as possible, which, through randomness, can sometimes show a drug to be effective even if it isn't.
Drugmakers often accept that this trade-off of interim looks equals a higher approval hurdle because, for cancer treatments like Genentech's
For this reason, Dendreon had planned just such an interim efficacy look in the second half of 2008 with its IMPACT study.
So, what exactly happened with Dendreon?
What it sounds like Dendreon did with the IMPACT amendment last week was that it negotiated with the FDA a tougher efficacy hurdle (or p-value, in statistician's parlance) for its interim data peek in exchange for making the p-value that it has to achieve at the final data peek easier.
Instead of just accepting this interim data peek with a tougher p-value and lower odds for success (if Provenge truly does work), Dendreon at the same time increased the number of data points in its interim look. A higher number of data points (events) included in a sample has the effect of improving the odds for success of an interim (or final) data peek if a compound truly works.
Essentially, the more data points there are in a sample, the higher the odds are that a hypothesized effect of a drug will be visible and the less likely we will see a false negative, such as what many investors think happened with at least one of Dendreon's previous two phase 3 studies.
The two opposing effects of Dendreon increasing the number of events in the interim IMPACT analysis (lowering the hurdle for the interim look's success) but now having a lower p-value to overcome (raising the hurdle for the interim look's success) end up having the net effect of making the interim analysis' odds of successfully showing a true drug effect, if it exists (what we call the power of a study), "slightly" higher than it was previously for the IMPACT interim efficacy look.
One last note on the interim IMPACT data reveal: Normally, increasing the number of events in a clinical trial would have the effect of delaying a data reveal date, but, apparently, Dendreon was already expecting this amendment to its clinical trial to be accepted by the FDA. Its previous guidance for second half of 2008 interim IMPACT data was, in fact, based on the expectations of this revised clinical trial being approved by the FDA.
That's not the only change to IMPACT
It wasn't just its interim data reveal that got changed for Dendreon. In exchange for having a tougher interim p-value to overcome, Dendreon was granted an easier p-value to overcome in its final data reveal.
If Dendreon had kept its final data reveal at its 2010 date and kept the same number of data points for its final analysis, this easier p-value hurdle would have had the effect of increasing "the power" of the IMPACT study and, essentially, increasing the odds that the final IMPACT data would be successful if Provenge worked (i.e., the potential for a false negative result of the study would be reduced).
Instead, what Dendreon chose to do was the opposite of the way it amended the interim data reveal. Dendreon chose to reduce the number of events for the final data reveal. Reducing the number of events in a study has the effect of making the hurdle for success harder at a given drug's expected efficacy.
The net effect of fewer data points (raising the hurdle) for the final analysis but an easier p-value (lowering the hurdle) actually lowers the odds slightly (from 90% to 88%) that the final data reveal will be successful, even if Provenge truly works as Dendreon expects it to (which it may not).
In other words, the probability that Provenge works but that the IMPACT study throws out a bad outcome due to statistical chance has now gone up (from 10% previously to 12% now). This is why I've criticized the move by Dendreon to end the study sooner in my last article and on the Motley Fool's Dendreon message board.
Assuming that Dendreon did not change the number of events in the trial for the final data reveal, the effect of the easier p-value hurdle for its final data reveal would have meant a higher chance for IMPACT study success, rather than the lower chance IMPACT now has.
Reasonable people can disagree on the trade-off between a faster trial completion or improved odds for success. Dendreon obviously thinks the benefits of saving "about" a year for the final data is the better choice. My personal conservative statistical preference would have been for Dendreon to choose the option of improving IMPACT's odds for success rather than moving up the completion date, assuming the FDA allowed it.