Page 188 - Assurance of Sterility for Sensitive Combination Products and Materials

170 Assurance of sterility for sensitive combination products and materials

regulators ask for statistical evidence. Regulations commonly state that the
sampling plans must be statistically valid. In the simplest sense, ‘statistically
valid’ means a conclusion is an accurate representation of the truth. In a
stricter and statistically testable sense, ‘statistically valid’ means the sample
sizes can be traced to a predefined statistical objective. This statistical objec-
tive is often that the worst-case performance exceeds the minimum value
that has been deemed to be acceptable, after accounting for any statistical
uncertainty.
This section is aimed at the non-statistician. The level of detail is
intended to support the concepts and conclusions without overwhelm-
ing the reader. It explains statistics as applied to sterility assurance but
does not provide detailed derivations of the statistical concepts or the
calculations.
The central statistical concept in this section regards uncertainty. The
concept will be introduced using a point estimate and a two-sided confi-
dence interval associated with that point estimate. Intuitive examples are
provided to help the reader understand the concepts. The concept of two-
sided confidence intervals will be simplified into a single-sided confidence
bound. As this chapter is focused on the risk associated with a non-sterile
unit, only the upper boundary of the risk is of interest. This is called the
upper confidence bound (UCB) of the risk, specifically the UCB of the
PNSU*.
After these concepts are developed, point estimates and confidence
bounds will be applied to several representative processes that are required
to obtain and maintain sterility (see Section 7.3).

7.2.1 Point estimates

The simplest statistical analysis provides the best estimate of the answer.
Averages of groups and proportions are examples of this kind of statistical
estimate. The value that is calculated is called a point estimate. The point
estimate accurately represents the result. If the inference space of the analysis
is limited to the data that was used in the calculation, the analysis can stop
here. Most often the data that is available is a portion of some larger popula-
tion, that is, the data is a sample. The intended use of the analysis defines the
larger population. Sterilization studies are used to make a statement about
the effectiveness of a process that includes manufacturing and sterilization.
The data is a sample because the conclusion will be generalized and applied