The FDA views p-values as being necessary, but not sufficient, for the interpretation of study results. For example, they consider bias (in the statistical sense) as being at least as important. Sources of bias include study design, randomization strategy, the conduct (blinding, informative censoring, missing data), the analysis (e.g. changing pre-specified goals), reporting and interpretation (i.e. intention of protocols and amendments), and the like.
That is, there are many reasons why the FDA might consider the results of a study dead-on-arrival even though satisfactory cookbook p-values can be calculated