Performance Characterization of EDA and Its Potential to Improve Decision Making in Product Batch Release - Good Cascade Impactor Practices, AIM and EDA for Orally Inhaled Products - page 177

minimize these errors. One could either improve the fundamental measurement by
reducing the variability of the measurement or increase the number of measure-
ments in order to improve the precision of the estimate of the quality attribute.
As discussed earlier, in the context of Wheeler's classification of measurements,
the QC testing discussed here corresponds to “characterization” and “representa-
tion.” It is important, however, to make the distinction between these two categories.
In the former case, the decision is strictly with respect to the measured entity. Thus,
in a practical sense, characterization implies both nondestructive testing and 100%
inspection. The much more common situation for pharmaceutical products is repre-
sentation. The typical batch release process involves testing a relatively small num-
ber of samples drawn from the batch and then releasing or rejecting the batch on the
basis of the measurement results based on some a priori sampling plan and decision-
making process.
While MSA focuses on characterizing the precision of the measurement, an
OCC considers and characterizes the entire schema used to make the QC decision,
including the sampling plan and the rules used to arrive at the final decision. The
OCC approach recognizes that the quality of decisions made based on testing is a
function of not only the measurement characteristics (precision and accuracy) but
also the sampling plan and decision rules. The construction of this decision-making
process (i.e., “design of the test”) allows one to adjust the relative anticipated rates
of type I (false rejection) versus type II (false acceptance) error, although there is
always some trade-off between the two types of error. This may be important
depending on relative consequences of making either of these errors. In the pharma-
ceutical world, there is in general more emphasis on minimizing type II errors at the
expense of type I errors.
Simply stated, an OCC is the determination of the probability of acceptance (or
rejection) as a function of the true value of the quality attribute being estimated. In
essence, an OCC is a transfer function that relates the probability of a particular
decision (accept or reject) to true values of the quality attribute being evaluated. The
“true” value of a population parameter can never be known in practice because even
a 100% inspection would yield only an estimate, whose accuracy and precision is
influenced by the analytical uncertainty and other limitations of the measurement
technique. The OCC approach requires simulating, or modeling, of the underlying
data distribution (so that “true” attributes and parameters are known exactly) and
evaluating the ability of a test (i.e., a combination of limits and a specified schema)
to make correct reject/accept decisions, as a function of the (assumed) quality of the
material under consideration.
Consider the hypothetical case of a tablet assay where the desired absolute batch
mean is between 90% and 110% LC. The ideal situation would be where the true
value of the assay for any batch (all tablets) is known with certainty, with no measure-
ment error and consequently no type I or II errors. The resulting OCC would consist
of a step function where the probability of acceptance would be zero for all assay
values above 110% and below 90% and unity for all values between 90% and 110%.
Now consider the impact of measurement error. For batches where the true assay
was well below 90% or well above 110%, the probability of acceptance would