Biomedical Engineering Reference
In-Depth Information
that a rawmaterial lot will perform acceptably in a manufacturing plant. A biological use
test is considered the gold standard because it matches the real conditions as closely as
possible. Because a use test consists of a scaled-down version of the actual conditions
expected in the manufacturing plant, it provides assurance that no unforeseen variable or
component in the material will adversely affect the process. Although much weight is
given to results of a use test, it still suffers from run-to-run variability, takes approxi-
mately 2 weeks to conduct, is much more complex than more analytical raw material
tests, and still does not provide understanding about factors that may underlie the
performance. Analytical tests are preferable to use tests because they can generate
information about underlying factors, but analytical tests also present risks. Since any
analytical test will only be able to measure certain aspects about a sample, there is a risk
of missing something important. Selection of an analytical strategy demands careful
consideration to minimize the risk from unforeseen compounds entering the process as
well as the risk from operational complexity.
As the experience with a raw material accumulates, these data factor into the
probability that a false-negative or false-positive result will occur. Could an analytical
testing scheme provide both types of information—intentionally robust and insensitive
as well as intentionally “over” sensitive and informativewith a definite strategy built in to
transition between the types of information needed as the product development timeline
moves forward? Bayes' theorem [4] can be used to understand the balance of these risks
with time. Before exploring how Bayes' theorem can be applied, we need to understand
more about the types of analytical tests that are available and the relativemerits. Also, we
will need some data to work with.
Analytical methods differ in many ways from each other, but one of the key
differences has to do with their relative sensitivity and specificity. Bulk measurements
like osmolality or pH are not capable of tracking the individual components that make up
a complex raw material. A material could easily have the same pH or osmolality while
having a different underlying composition than previous lots. These tests may not detect
important composition changes and, therefore, be classified as allowing a high percent-
age of false negatives—a material would pass even if a minor component were not
suitable.
NMR and MS, in contrast, provide unparalleled information about the chemical
content in a mixture, but may be so sensitive, that even unimportant differences between
batches of raw material could be detected. These tests would be classified as having a
high type I error rate, but also high information content. Table 11.1 provides general
classification for selected analytical methods based on their perceived error type.
MS and NMR can have such complex signals that multivariate analytical methods
are used to interpret the data. This involves obtaining signals for a set of samples that will
be used to “train” the method. As new samples get tested, they can be incorporated into
the training set to improve the robustness of the model. During the initial phase, it is more
likely that such a multivariate method will fail simply because the method is sensitive to
factors that are not important to the outcome of the manufacturing process. Building a
manufacturing process that has a tolerance for the time and temporary uncertainties
associated with development of such multivariate methods might be one valuable
outcome of a generalized “QbD” for raw materials strategy.
