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 173

In addition to simplify the characterization of this approach, EDA performance
will be compared to the current regulatory data reduction procedure used for qual-
ity control (QC) purposes related to OIP aerosol APSD, which makes use of
grouped impactor stages.
As was mentioned in Chap. 6 , in the design and implementation of any measure-
ment system, it is important to consider the purpose of the measurement as this
informs both the design and evaluation of the effectiveness of the measurement and
associated system. For example, there are different considerations for a measure-
ment intended to characterize or describe an attribute of a particular object versus
one intended to make a decision about a batch of objects with respect to a particular
characteristic and based on representative samples. Wheeler [ 4 ] has described this
concept in more detail.
Measurements can be classified into four categories based on the general purpose
of measurement:
1. Description
2. Characterization
3. Representation
4. Prediction
Description refers to measurements that inform one about the attributes of the
item being measured. Characterization is similar to description except that it, in
addition, involves comparison of the measurements to some expectation for the par-
ticular object studied (i.e., a requirement or limit). Representation involves using
measurements on a representative sample to make inference about the population
the sample is intended to represent. The category of representation is in essence the
QC application where a batch is released or rejected on the basis of testing per-
formed on a sample(s) taken from the batch in question and comparing the measure-
ment results to some requirement. Finally, prediction is based on using measurements
of samples from current batches to predict the attributes of future batches (i.e., in
skip-lot testing). Since EDA is proposed for QC purposes, where batch disposition
is decided based on a representative sample, it is classified as a representation
measurement.
The adequacy of a particular measurement with respect to precision should con-
sider the variability of the measurement versus the variability of product being mea-
sured or the tolerances imposed on the product. This is the fundamental essence of
measurement system analysis (MSA) [ 5 ], which typically employs analysis of vari-
ance ( AnOvA ) designs to estimate measurement and product variances. These
types of designs are also collectively known as gage repeatability and reproducibil-
ity (Gage R&R) studies.
Besides an evaluation of the adequacy of measurements employed for QC, it
is important to consider the adequacy of the schema used to make the QC deci-
sion. In practice, measurements are made on some number of samples. The
results are then used in accordance with a previously established specification to
make an inference about the batch quality and decide on the disposition of the
batch (either release or reject). There are many possible “protocols” (schemas