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 170

through “E” will be explored in turn, with the ultimate objective of finding a
definitive answer to the question posed concerning the relative discriminating abil-
ity of the two approaches. Importantly, these assessments have been made with real
OIP data, not computer-generated APSDs, so that the variability seen (Fig. 8.3 ) is
likely to be typical of the real world in terms of the quality of the data obtainable for
these products by the current compendial methods that involve full-resolution CIs.

8.3 Underlying Data for Metric Comparisons

The data sets for eight products analyzed using operating characteristic curves
(OCC) and measurement system analysis (MSA) came from the IPAC-RS APSD
database from OIP performance measurements that were collected in 2000. For that
database collection, pharmaceutical companies were asked to submit aerodynamic
particle size data for individual CI determinations, on a stage-by-stage basis, for as
many OIPs as possible, obtained from batch release testing and/or from real-time
stability studies. Data were presented as a percent of the amount of API stated in the
label claim (%LC). No other manipulations or normalizations were applied. To
avoid bias, it was recommended that companies submit either:

1. All available data for the product
2. Data for a random selection of batches
3. Data for all batches manufactured during a defined time span

The database does not have information about regulatory specifications applica-
ble to particular products; therefore, data could be included from batches that
“passed” as well as “failed” the regulatory APSD requirements. The data are there-
fore considered representative of real manufacturing capability and variability.
Companies were asked, however, not to include any batches that had clear quality
problems or manufacturing faults, as would have been detected by other (non-
APSD) tests and investigations. Each product was tested following its analytical
method.
The original database encompassed 34 OIPs [ 3 ]. For this analysis, eight products
were selected that had the largest amount of data per product. The selected OIPs
represent different product types (CFC suspension MDIs, HFA suspension MDIs,
an HFA solution MDI, and two DPIs) as well as having different APSD profiles, as
can be seen from the “spaghetti” plots of all available data points (Fig. 8.3a, b ).
The six MDI products were tested with ACI, and the two DPI products were
tested with a modified ACI. The number of CI runs available for each product is
listed in Table 8.2 .
The assessments based on principal component analysis (PCA) described
later in this chapter used a data set that was unrelated to the IPAC-RS database.
The PCA data set consisted of 1,990 individual NGI measurements for varia-
tions of a real (blinded) product throughout the development process, of which
252 measurements were derived from the dose-ranging clinical batches. The