Statistics and Process Validation: current Findings of the FDA
The "new" FDA's process validation guideline has been effective since January 2011. One considerable change was made to the original validation guideline from 1987 to put a significantly greater emphasis on statistics in the context of process validation. So far, relatively few inspection deficiencies had been observed by the FDA with regard to statistics. At a conference in September 2015 co-sponsored by the FDA, Grace McNally - Senior FDA official - reported about current "findings" in the 483s deficiency reports and in Establishment Inspection Reports (EIR). Now, deficiencies regarding statistical problematics can also be found here.
For example, it has been criticised that a (statistical) sampling plan had be misinterpreted. Wrong AQL values with regard to the number of samples have been noted based on MIL-STD-105D. Moreover, it has been criticised that the company didn't know the operation characteristics of its sampling plan.
Another criticised "finding" was that PPQ batches had been considered as "accepted" when all in-process controls and release specifications were met. It has also been criticised that no intra-batch variabilities have been examined. In addition, it has been noticed that there was no information available in the validation plan concerning the assessment of the process itself. There was also no indication about the objective of the determination of inter-batch variabilities.
Although OOS results had been found in 2 out of 4 PPQ batches, reduced IPC tests have been recommended in the PPQ report giving the justification that this was a standard procedure. Regarding this point, the FDA criticises the lack of scientific rationales for reduced sampling and monitoring. Interestingly, Grace McNally mentions possibilities for rationales of IPC sampling plans and the adaptation to a reduced size. In this context, she refers to the ANSI/ASQ Z1.4 norm and ISO 2859 whereby it is expressly pointed out that the ANSI norm recommends the production of at least 10 successful batches before reducing testing. According to the ISO norm even 15 successful batches are necessary.
The FDA notified a tablet process, criticising the fact that no rationales for warning and action limits were available. Furthermore, it has been criticised that no analyses on variabilities were available although they had been required internally and no capacity indices had been determined. There have been no analyses on the distribution of data, neither planned nor performed. The FDA also remarked that the calculation of variabilities is necessary to be able to make statements about process capacities.
Conclusion: Reinforcing the emphasis on statistics in the US FDA Process Validation Guideline from 2011 hasn't been really often addressed in the official deficiencies reports. This seems to be changing.