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 185

Fig. 8.11 CI stage group 4 versus MMAD from individual CI APSD determinations for a CFC-
formulated MDI
w9jk01 ), there was a consistent pattern concerning groups 2 and 4. In general, group
2 increased, while group 4 decreased as a function of MMAD .
To perform an MSA, the prediction bands determined through the regression
analysis were in turn used to construct figures of merit that reflect the measurement
capabilities of the respective metrics. Prediction bands are an estimate of the inde-
terminate error associated with a regression model. In the present case, this is an
approximation of the measurement error associated with estimating MMAD by the
particular metric considered (i.e., LPM / SPM or stage grouping).
Calculations based on the prediction intervals, called figures of merit, were con-
structed from the regression models. These figures of merit reflect the measurement
capabilities and the measurement variability of the respective metrics, EDA, and
stage groupings. In the present case, based on inverse regression techniques, this is
an estimate of the measurement error associated with estimating MMAD by the
particular metric considered (i.e., LPM / SPM or stage grouping). The figures of
merit constructed from the regression models are for comparative purposes to give
an indication of the performance of EDA compared to grouped stages. These were
not intended to make statistically exact absolute statements about the population;
they were intended to be used in a relative sense to compare performance between
two different metrics.
One such figure of merit was constructed based on inverse regression. As an
example, from Fig. 8.9 , at the LPM / SPM ratio corresponding to the average MMAD ,
the limits of the prediction interval were projected onto the MMAD axis.