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 233

Fig. 8.66 PCA scores plot of full data set; tPS[1] and tPS[2] are predicted principal component
scores (1 and 2) derived from overlaying the data onto the PCA model established using the 252
clinical measurements

to build the model, whereas a profile colored as “fail” means that the full-stage
APSD profile is not similar to the 252 APSD profiles used to build the model.
The scores plot with this coloration applied is shown in Fig. 8.66 . Of the 1,738
measurements, 934 fell within the Hotelling T 2 0.99 confidence limit of the clinically
relevant data set and could therefore be considered as “passing,” while 804 fell
outside and could therefore be considered as “failing.”
Next, EDA and grouped-stage approaches were evaluated using the same data
set, and the data generated by each was cross-referred to the scores plot for the full-
resolution NGI data.

8.5.3.2 Results for EDA Metric LPM / SPM

LPM / SPM ratio versus MMAD was compared using the same data set described in
the previous section. This plot was therefore based on the 252 clinically relevant CI
APSD profiles that formed the training data set of the PCA model. The MMAD was,
again, estimated from a logistic curve fit as described in Sect. 8.4.1 , in connection
with data preparation for MSA. Here, the LPM consisted of the sum of the material
deposited on S1-S3 of the NGI, and the SPM consisted of the sum of material