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 235

Fig. 8.68 EDA plot with quadratic criteria between LPM / SPM and MMAD

8.5.3.3 Results for Stage Grouping

Four groups of stages were used for the grouped-stage analysis, which were
representative of FDA regulatory requirements. In each case, the sum of the mass
obtained on each of the stages in the NGI was presented as a percent of the nominal
strength of the product. Acceptance values were derived from the clinical batch
data, as presented in Table 8.11 .
Each of these groups was evaluated individually versus PCA as a means of
detecting change in APSD, and the combination of all four stage groups was also
evaluated (Table 8.11 ). Data for these stage groupings in combination gave 635
“passes” and 1,103 “failures.”

8.5.3.4 Comparison of EDA and Stage Grouping

Comparison of the EDA and the grouped-stage analysis was made with cross-
reference to the PCA as the control. PCA acts as a suitable control because it is
purely a statistical representation of the data. The results of this comparison are
presented in Table 8.12 . In both of these analyses, there is possibility of generating
an erroneous result with respect to the PCA work. These broadly fall into two cat-
egories and are labeled type I and type II errors. Type I errors represent an incorrect
rejection, i.e . , where the data represent a “good” batch but are rejected by the analy-
sis. In other words, in this context, a type I error occurred when the APSD profile
from a batch in the prediction set was deemed similar to the 252 clinically relevant
profiles from the PCA analysis but was classified as not similar by the other methods.
Type II errors represent an incorrect acceptance, i.e., where the data represent a