Theoretical Basis for the EDA Concept - Good Cascade Impactor Practices, AIM and EDA for Orally Inhaled Products - page 148

7.3 QC Testing: Purpose and Limitations

As an element of the quality control strategy, end-product testing schemes should
reflect an in-depth understanding of the required performance characteristics of the
product. There should be thoughtful selection of those critical tests that are best
performed at the end-product stage. It is important to avoid redundant testing of the
same quality attribute through multiple tests either at the end-product stage or in-
process. There should be an evaluation of the overall control scheme to avoid mul-
tiplicity issues. By “multiplicity issues,” we mean testing the same attribute multiple
times and not considering the statistical consequences in the setting of specification
limits [ 15 ]. Finally, statistical design and evaluation of the tests and acceptance
criteria should be an integral part of the development process [i.e., the end product
in process analytical technology (PAT)].
It is also important to understand the limitations of end-product QC testing. This
topic has been discussed in some detail by Tougas [ 15 ]. First and foremost of these
limitations is that the outcome of end-product testing is primarily limited to an
“accept” or “reject” decision. If a product exhibits substandard characteristics at the
end of the manufacturing process, there is typically little recourse, but to reject the
batch. This is particularly true of pharmaceutical manufacture where the impact is
related to human health. A further limitation arises if the end-product testing is
destructive in nature (common in pharmaceutical manufacture). Destructive testing
relies on testing a relatively small number of units that are representative of the batch
or lot. The results from such testing can be used to make inference about the batch
characteristics (mean or variability of measured characteristic), but are ineffective at
detecting low-frequency aberrant units that arise from an intermittent failure mode.
This is illustrated in Fig. 7.1 which depicts batch distributional characteristics that
are nominal, abnormal, and the case of intermittent failure modes.
This figure depicts four ways in which the frequency (number)-weighted distribu-
tion of a quality attribute related to the batch may lie with respect to quality limits, in
this example fixed at 98% and 102% of the nominal value. In the first instance, A, the
mean of the symmetrical distribution is centered within the quality limits, and the
variance is such that essentially no units are outside the quality limits. This represents
a batch with acceptable quality. Testing of this batch via representative samples
should result in a decision to accept this batch as being of suitable quality. Case B
illustrates a batch distribution where the variance is suitable, but the batch mean is
abnormal such that a significant portion of the units are outside the quality limits, and
therefore, QC testing should result in a decision that the batch is not of acceptable
quality. Case C illustrates a batch distribution with a nominal mean, but an abnormal
variance resulting in a significant fraction of units (both high and low) outside the
quality limits. As in case B, QC testing should result in a decision to not accept this
batch as being of suitable quality. In the final example, D, the overall distribution is
similar to case A, but the overall distribution contains additional modes such as might
arise from some intermittent failure mode. QC testing based on representative sam-
ples is unlikely to be effective at detecting these types of failures.