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 220

Fig. 8.52 OIP product w9k001 : OC curve ( dashed green line ) and results for stage group 4; black
symbols are correct decisions; blue symbols are false rejections (type I errors). Outcomes and the
OC curve are based on the combined results of all three groups. The OC curve for LPM / SPM ratio
is superimposed for comparison
decisions. The LPM / SPM ratio exhibits a broad flat curve with relatively steep sides
compared to the grouped-stage OCC, which never attains 100% probability of
passing even in the center of the acceptance region. The false rejection outcomes
represented by the blue plot symbols occur consistently further away from the
boundaries of the acceptance region, and even through the middle of the region for
the grouped-stage approach than for the LPM / SPM ratio approach.
Another way of comparing the patterns of outcomes between the two approaches
is presented in Fig. 8.53 . All nine possible pairs of outcomes for the two approaches,
“G” for grouped stages and “R” for LPM / SPM , are plotted versus MMAD . Correct
decisions are represented by “0,” type I errors by “1” and type II errors by “2.”
In this case both approaches indicated the correct decision 52% of the time as shown
by the “G0/R0” category. This occurred in middle of the MMAD acceptance region
and also just outside the range and beyond. Both approaches indicated false
rejection (type I errors, represented by “G1/R1”) approximately 21% of the time,
which occurred just inside the limits of the acceptance region. However, approxi-
mately 27% of the time, the grouped-stage approach indicated false rejection (type
I errors) well inside the acceptance region, while LPM / SPM ratio indicated the
correct decision (represented by “G1/R0”). Increased occurrence of false rejection
away from the acceptance limits indicates less discrimination of the grouped-stage
approach to correctly predict MMAD .
This difference in discrimination capability is illustrated in Figs. 8.54 and 8.55 ,
which show the relative size of the uncertainty of the estimate of MMAD compared
to the range of acceptable values for stage group 2 and LPM / SPM, respectively.
A smaller amount of estimate uncertainty relative to MMAD acceptance limits indi-
cates a higher capability to detect differences in MMAD within the acceptance region.