Biomedical Engineering Reference
In-Depth Information
Table 8.3 Differences between the construction and meaning of OCCs in this chapter examples
and in the customary QC use
OCC in this chapter examples
OCC in customary use
Demonstrates and compares the performance
of different metrics ( LPM / SPM and grouped-
stage depositions) to detect changes in APSD
by the ability to predict MMAD
Demonstrates and compares performance
of a test (i.e., a combination of metrics,
decision-making rules, sampling plan,
and acceptance criteria) or different
tests to make correct decisions
The goal is to characterize the metric's ability
to make correct decisions regarding shifts
in MMAD
The goal is to characterize the ability of
the entire test(s) to make correct QC
decisions about batch disposition
False rejections and false acceptances are
“misclassifications” by the metric
False rejections and false acceptances are
errors in decision-making
“Limits” for the metric's values are chosen to
enable a consistent and fair comparison
“Limits” or “acceptance criteria” are given
as part of the test that is being studied
evaluation of different metrics using the same specification criteria applied to the
same data set. By contrast, customarily, OCCs are used to evaluate a different speci-
fication criteria or different test designs using a range of simulated data sets
(Table 8.3 ). The metrics being compared here are the LPM / SPM ratio and the
grouped-stage API mass depositions in the specified groupings, both of which
reflect a breakdown of the mass of API into their specified sub-fractions.
8.4.4
PCA Approach: Definitions and Basic Concepts
Both QC testing and in vitro equivalence testing of APSD share the essential goal
of comparing a range of numbers (i.e., the APSD profile) against another range of
numbers and determining their “sameness.” For QC, that “another” range of num-
bers represents APSD of clinical pivotal batches. For equivalence testing, it is the
original product or initial version of a product. Comparing APSD profiles could be
done simplistically by overlaying the APSD profiles and “eyeballing” them to deter-
mine similarity or differences; however, devising a more objective means of deter-
mining whether a difference in the shape of a profile is statistically relevant or not
presents a more complex challenge. PCA is one way of accomplishing such a
comparison.
Principal component analysis is mathematically defined as an orthogonal linear
transformation that transforms the data to a new coordinate system such that the
greatest variance by any projection of the data comes to lie on the first coordinate
(called the first principal component), the second greatest variance on the second
coordinate, and so on [ 10 ].
PCA operates on the principle of reducing the dimensionality of a given data set
containing a large number of variables. PCA has been used to deal with distribu-
tional data successfully in a number of fields [ 11 ]. For example, PCA has been used
Search WWH ::




Custom Search