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 232

Fig. 8.65 Loadings plot of clinically relevant measurements of APSD: S1 through S7 refer to NGI
stages, and MOC refers to the micro-orifice collector; PC[1] and PC[2] are principal component 1
and 2; the influence of each of the stages on PC[1] or PC[2] can be positive or negative—meaning,
for example, that high deposition on S2 would increase the PC[2] score but decrease the PC[1] score
a majority of the variation or change in the APSD profiles can be explained by the first
component, which applied positive loadings primarily to stages 5 and 6 and negative
loadings to stages 2 and 3 so a large percentage of the variation in APSD shifts in the
training data set can be explained by differences in shifts between those stages.
Using this model as the training data set, the remaining 1,738 measurements
acquired throughout the product development process were overlaid, and each data
point or batch indicated with green color for a “pass” (i.e., the data point lies within
the Hotelling T 2 ellipse of the model) and red for a “fail” (i.e., the data point lies
outside of the Hotelling T 2 ellipse of the model).
A data point in this plot represents a multistage CI APSD profile with loadings
applied to the data to reduce the dimensionality down to two components. The com-
ponent values or scores from these batches, the prediction set, are then plotted and
presented in this scores plot. The labels, tPS[1] and tPS[2] , represent the scores or
component values of the first two principal components of the prediction set. The
Hotelling T 2 0.99 confidence limit ellipse represents the space of the “typical”
APSD profiles from the training data set. Therefore, a profile colored as a “pass” in
this plot means that the profile is “similar” to the population of APSD profiles used