Puma Biotechnology reporting its latest adaptive phase II trial data using Bayesian analysis caught my interest. It seems controversy has erupted from the fact tha Puma did not report its trial results using the usual PFS measures but focused more on use of Bayesian staistical analysis to predict their success in a phase III trial with patients of certain characteristics. In this analysis, they concluded they are more than 70% likely to better results clnical results using a drug cocktail and their drug neratinib versus the same cocktail and Herceptin (current standard). Then the controversy explosion ensued.
Ok. Puma has clearly use non-traditional statistical methods to assess ther trial data. Does that mean they ae wrong? Or are they possibly advancing the statistical treatment one can use for this type of assessment? Afterall, Puma is also using this method to establish which subsets of cancer patients, defined by specific biomarkers, are likely to respond most positively to their drug. Isn't that a goal in assessing cancer treatments now? Get the drug to the right patients who have the best chance of responding positvely and showing the drug actually works? Isn't that what biomarkers are all about?
I don't see the big problem here. The Puma drug is NOT geting approved on this basis. It is being moved on to phase III studies where their statistical model canbe further assessed for accuracy. What's wrong with that? There is data to support th3e position may work. Bayesian statistics are a legitimate field of statistical analysis. The science of the phase III study will speak for itself. The drugs will work as predicted or they won't. But there seems to me to be ample evidence to proceed and to then evaluate the drug based on the actual results in phase III. Again, this statistical treatment may turn out to be an advancement. Status quo is not necessarily best -- nor does it imply it can never be improved upon. See Fierce Biotech.
Posted by Bruce Lehr Dec 5th 2013.