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 212

some range of values (upper and lower acceptance limits for MMAD ). Three
grouped-stage test statistics—group 2, group 3, and group 4 mass depositions (anal-
ogous to sample statistics such as the mean or a distribution percentile)—are calcu-
lated from the CI sample data. If all three test statistics lie within some range
(upper and lower acceptance limits for the respective groupings), we fail to reject
the null hypothesis (i.e., there is no evidence that the MMAD is outside the range of
acceptable values).
If any one of the three grouped-stage test statistics falls outside acceptance limits,
the null hypothesis is rejected (i.e., there is evidence that the MMAD is outside the
range of acceptable values). In the same way, the null hypothesis is tested using the
LPM / SPM test statistic. If the ratio falls within acceptance limits for ratio, we fail to
reject the null hypothesis that the MMAD is outside an acceptable range of values.
If the ratio falls outside its acceptance limits, the null hypothesis is rejected, indicat-
ing evidence that the MMAD is outside an acceptable range of values.
As described previously, for the grouped-stage approach, the value of ISM is
completely confounded with shifts in APSD. Therefore, ISM is used as an addi-
tional test statistic in assessing the performance of both EDA and stage groupings.
For the grouped-stage approach, all three stage grouping test statistics and ISM must
be within their respective acceptance limits to indicate no evidence that the value of
MMAD is outside an acceptable range. Similarly, both LPM / SPM ratio and ISM
must be within acceptance limits to fail to reject the null hypothesis.
In this case it is not feasible to determine the distributions of the test statistics, so
a closed-form approach, such as for a test statistic assumed to follow a t-distribution,
is not used. Instead, the method used here is similar to other non-closed-form
approaches such as bootstrapping.
The assessment process used to compare the two approaches is illustrated here as
an example using data for a specific product in the IPAC-RS APSD database (prod-
uct w9k001 ). The basis for establishing comparable acceptance limits for EDA
LPM / SPM ratio and grouped stages for a common range of acceptable MMAD val-
ues is illustrated in Fig. 8.41 .
Acceptance limits for each of the stage groupings were established by calculat-
ing distribution percentiles for each of the grouped-stage results and then selecting
percentiles at the extremes of the distributions (e.g., 1st and 99th percentiles), which
were consistent with typical FDA limits [ 2 ]. Regression analysis was then used to
determine the relationship between values of mass on each grouped stage and the
corresponding MMAD , with a prediction interval, representing some level of uncer-
tainty, around the regression line. The intersection of the prediction interval with the
upper and lower limits for the grouped stages was used to establish, for the purpose
of this assessment, a range considered to represent acceptable values of MMAD , as
illustrated in Figs. 8.42 , 8.43 , and 8.44 .
Group 3, which represents the middle portion of the APSD, showed poor correla-
tion with MMAD ( R 2 = 0.009). Groups 2 and 4, representing the “tails” of the APSD,
showed similar, strong correlation with MMAD ( R 2 = 0.58 and 0.59, respectively) and
resulted in similar limits for the range of MMAD (3.02-4.04 μm and 3.00-3.98 μm).