Page 134 - Assurance of Sterility for Sensitive Combination Products and Materials

116 Assurance of sterility for sensitive combination products and materials

in this section. In layman’s terms, power can be defined simply as the abil-
ity to detect an effect if there truly is one. For example, power is a mea-
sure of the ability to demonstrate a process change lowered package seal
strength if the change truly did lower seal strength. In statistical terms power
is P(Rejecting H 0 |H 1 is true), where H 0 is the null hypothesis and H 1 is the
alternative. Please refer to Moore et al. [91] for a more thorough treatment
of hypothesis testing.
A definition of the elements of power and sample size calculations is in
Table 5.7.
Typically, an alpha value is set at either 5% or 1% depending on the impact
of falsely detecting an effect. The effect size is set from technical expectations
or industry benchmarks. Power is typically set at 80% or 90% for the mini-
mum effect size. Sample size is often limited due to the experimental budget
in the form of cost and/or time. Care should be taken to obtain an estimate of
the standard deviation from a random, representative historical sample.

Table 5.7 Elements of a power calculation.
Element Description and examples
α (alpha) or Type I Error • P(Rejecting H 0 |H 0 is true)
Rate • α can often be described as the chance of falsely
detecting an effect.
• There is a 5% chance of falsely detecting the
process change reduced seal strength
Δ or Effect size • The minimum expected change
• The new package design should increase yield
from 93% to 97%, an effect size of 4%.
Power • P(Rejecting H 0 |H 1 is true)
• Power can be described as the ability to detect an
effect if there truly is one.
• There is an 80% chance of detecting a 2% increase
in dye penetration from the old configuration to
the new configuration.
n or Sample size • The number of items to be tested.
• Note that many statistical software packages
provide sample size in number per group versus
the total sample size. Before implementing a test
plan, verify whether the software supplies the total
sample size or the sample size per group.
Standard deviation • A measure of the spread of the data, typically
obtained from historical experimentation or
process operations.