The companys stock had surged on the announcement of topline data from this trial last October, but no detailed analysis was offered then, and little more is forthcoming now (AB Sciences dog drug data raise more questions than answers, October 31, 2012). Trial failed Although the trial had unequivocally missed its primary endpoint, what had caused the excitement was the revelation that two subgroups patients with pain above a certain threshold and those with a particular genetic marker had done particularly well on masitinib, boosting median overall survival versus standalone gemcitabine with barely believable p values of 0.010 and 0.000038 respectively. Fridays presentation did reveal the metrics behind the pain grouping: patients had been categorised into three subgroups based on their pain score on a visual analogue scale (VAS). 137 patients had a VAS score above 20mm, corresponding to a high-pain group particularly poorly responsive to gemcitabine, and these had yielded the first positive subgroup result.
However, that was where the clarity ended. AB said secondary endpoints had included some kind of pain intensity evaluation, but it would not say whether this meant that the precise VAS-based analysis had been specified prospectively in the trials protocol. Even less was revealed about the analysis around the biomarker, whose identity AB says remains confidential. According to the ASCO-GI poster 66% of the patients for whom genetic data were available had had the particular marker, but this hardly seems relevant given that there is no way of knowing how many patients this amounts to. Questions regarding the nature of the biomarker, patient numbers and the design of the study would now be made available later in a future publication, Alain Moussy, ABs CEO, told EP Vantage.
Meaningless at worst Clearly, for any subgroup analysis to hold water scientifically it should be pre-specified in the study protocol, and any p values need to be adjusted for multiple analyses of the same data. If especially strong activity is seen in patients carrying a particular biomarker, prior scientific literature should be available to back up the particular significance of that biomarker. With no evidence to the contrary, the strong inference is that AB has simply mined the results from the failed study until it hit something significant. If this is the case then the touted positive results are merely informative at best, and statistically meaningless at worst.
Remarkably, though, masitinib has already been filed in the EU for approval in pancreatic cancer, with reference made in the filing to the biomarker in question. AB has said it intends to develop a companion diagnostic test, strongly implying that it intends to target masitinib at patients with the marker. The EMA told EP Vantage that it would not reveal details of any filing beyond confirming that masitinibs was in oncology. In December, Societe Generale analysts who have a buy rating on ABs stock wrote that they were not aware of any cancer drugs that had failed phase III but were subsequently approved based on post hoc subgroup data.