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 182

as a means of assessing near-infrared (NIR) spectroscopic data sets [ 12 ], in the
newspaper industry for process monitoring [ 13 ], as a means of assessing stationary
phase selectivity in HPLC analysis [ 14 ] and to classify beer taste profiles [ 15 ]. More
closely related to the subject of this topic is the work of Sandler and Wilson, where
PCA was used as a means of comparing particle size and shape data from laser dif-
fraction analysis [ 16 ]. In all of these cases, PCA was utilized in order to deal with a
large data set, where each sample consists of multiple variables, which would be
impossible to cross-correlate in a univariate fashion.
PCA generally consists of two types of graphical presentation: a scores plot and
a loadings plot:

• The scores plot is a Cartesian scatter plot where the abscissa axis contains a user-

selected principal component (PC). The ordinate axis contains another user-
selected PC. The plot contains points that represent the original samples projected
onto the user-selected PCs. By default, the scores plot shows data on the first two
PCs (PC1 versus PC2).

• The loadings plot is another Cartesian scatter plot that displays the individual

elements of the PCs. Since each PC is a vector, it has constituent elements which
are called the coefficients or loadings. By mathematical definition, the vector
length of each PC is 1. The loadings of a given PC represent the relative extent
to which the original “variables” influence the PC.

An APSD profile derived from either an ACI or NGI is very similar to the data
of Sandler and Wilson [ 16 ]. For the study described further on in this chapter, a
model was established that was based on 252 clinically relevant batch measure-
ments (described in Sect. 8.2 ). All other measurements were then assessed in
relation to this model. A comparison was subsequently made between the EDA and
grouped-stage approaches using the same data set. The number of errors was evalu-
ated using the PCA as the reference point.

8.5 Results and Assumptions from the Three Approaches

to the Evaluation of CI-Derived Metric Performance
in Assessment of OIP Quality

8.5.1 MSA Approach

A detailed measurement system analysis characterizing and comparing the perfor-
mance of the metric LPM / SPM to stage groupings was conducted using cascade
impactor results from IPAC-RS database. The MSA performed on the IPAC-RS
data set benefits from the nature of the data employed. This data set includes stage-
by-stage CI results for the eight products studied. Normally, for destructive testing,
as is the case with CI testing, the design would attempt to minimize sample-to-
sample variability and treat them as replicate measures. In this instance, the goal