Page 199 - Assurance of Sterility for Sensitive Combination Products and Materials

Risk to the patient—Quantifying assurance of sterility 179

on/in a product can be treated as an attribute (zero or not-zero) but it can
also be treated as a sample from a distribution. Treating data as a sample from
a known distribution allows parametric statistical analysis. In this type of
analysis, the value (the magnitude) of the result is used in the computation.
Parametric data has an advantage in that more information is used from
each measurement. Given the same number of measurements, the confi-
dence bound calculated using parametric statistics will be smaller, closer
to the point estimate. The weakness of parametric analysis is the reliance
on knowing the underlying distribution. If the distribution is identified or
selected incorrectly, the results can be very misleading.

7.3 Statistical analysis of risk associated with packaging
and sterilization

7.3.1 Introduction to scenarios analyzed
Packaging to maintain sterility, aseptic processing, and terminal steriliza-
tion are controlled by a number of international standards, PDA docu-
ments, and Pharmacopeia monographs. A partial list of ISO standards to
be used for simplicity and consistency in this chapter is shown in Table 7.4.
Conformance to these consensus standards facilitates regulatory approvals.
The statistical claims associated with the quantifiable components of these
procedures will be examined in the sections below.
The analyses of packaging and aseptic processing are based on nonpara-
metric statistics. The use of nonparametric statistical analysis provides a sim-
pler starting point to calculate the point estimate and UCB of the PNSU*.
The analysis of terminal sterilization, a bioburden-based method, and an
overkill method are based on parametric statistics and are more complex.
The math that is used depends on assumptions about the distribution of
microbes and the statistics of half-cycle lethality, respectively. Industry has
been challenged to model the distribution of microbes; the Poisson distri-
bution is used despite this shortcoming. Specifically, the Poisson distribution
is discrete; its value is defined only for integers (zero microbes, one microbe,
etc.); and it requires that the average rate is constant and that each sample is
independent of the time or space.

7.3.2 Point estimates and confidence bounds—Packaging

An overview of packaging for terminally sterilized medical devices is cov-
ered in detail in Chapter 5. It includes descriptions of test methods that
may be used to establish that the packaging is compatible with the device,