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 175

can be modeled as the sum of variances associated with the units (i.e., product related)
and the variance associated with the measurement system. The latter can, in principle,
be partitioned among components of the measurement system (e.g., according to
components outlined in the AIAG definition). An MSA for this situation might consist
of an experiment designed to allow estimation of these variances with the goal of
understanding if the measurement system is appropriate. For the QC situation, “appro-
priate” means that the measurements could be reliably used as the basis for QC deci-
sions. This definition implies that a product deemed unacceptable could be detected
through the measurements. It also follows that some type of comparison of the esti-
mated variance of the measurement system is made to either the variance of the prod-
uct or to some limits that distinguish between acceptable and unacceptable product.
The ideal situation for MSA is where measurements can be made repeatedly on
the same object. This provides a simple direct means of distinguishing variability
due to the measurement system from the variability of the actual objects being mea-
sured. This situation is possible when testing is nondestructive but is problematic
when samples are destroyed by the measurement process. The latter is typical in
chemical analysis and is the case in the specific situation associated with APSD
determinations obtained by cascade impaction.
Destructive measurements are very common in pharmaceutical analysis, where
often a sample must be transformed (e.g., dissolved, diluted, reacted, etc.) to make
a chemical or physical measurement. The normal recourse for an MSA is to seek an
experimental design that minimizes the influence of sample-to-sample variability.
For example, a larger composite sample might be generated to enable drawing mul-
tiple measurement samples from a homogeneous blend, or product is sampled from
a discrete portion of the batch in which production variation is minimized.
The data, analyzed in the MSA section, represent an example of destructive test-
ing. OIPs generate an aerosol that is intended to be inhaled by the patient. Each
actuation of a given inhaler results in a unique aerosol influenced by both the formu-
lation and the delivery device (e.g., MDI, DPI, or other inhalation device). CI test-
ing consists of actuating the OIP, size fractionating and collecting the resulting
aerosol particles on different stages of the apparatus, and subsequently performing
a chemical assay of the mass deposited on these stages and related accessories.
The cascade impaction data for the MSA were not generated from a planned
experimental design, as might be typical for an MSA design. The data represent
historical cascade impaction results from the IPAC-RS database, and thus, the anal-
ysis could be considered a historical MSA. The statistical approaches for this sec-
tion are intended to characterize, using regression techniques, the variability in the
EDA measurement and stage grouping measurement relative to the MMAD values
from the full-resolution CI results. Hence, even though there was only one physical
process of performing the testing that resulted in the original measurements from
individual stages, four additional metrics—the values of MMAD , the LPM / SPM
ratio, ISM , and the stage groupings—can be calculated from those original
measurements. MMAD in this instance was determined according to the approach
described by Christopher et al. via construction of the cumulative mass-weighted
APSD from the individual stage data [ 8 ].