sample size may not depend solely on statistical
considerations but also on budgetary or other con-
siderations. Although there is no reason in princi-
ple why one would not allow reducing the sample
size if it turns out that the trial is overpowered, all
the methods for sample-size adjustment allow only
for increasing the size of the trial. The reason is a
practical one. As the trial sponsor has already
committed the resources to conduct the trial as
designed, they would rather used them all and
have a more powerful trial rather than saving and
risking that the trial may be underpowered. Clearly,
if one incorporates the option of an interim sample-
size adjustment, at the outset, the expected sample
size of such a trial will be larger than a trial
designed without such an option using the same
design assumptions. Therefore, if the criterion for
statistical significance at the end of the trial is the
same, such a design will result in a more powerful
trial than if this option was not available. This gain
comes, of course, with a price tag: the type I error
probability is increased as well. Therefore, the
inclusion of an interim redesign must be planned
as an integral part of the trial design and proper
adjustments must be made to make sure that the
resulting decision procedure has the desired power
while still properly controlling the type I error rate.
The timing of an interim analysis for reassessment
of the sample size is very important. One would
want to conduct this analysis after sufficient
amount of data have accumulated so that the esti-
mated design parameters are stable and reliable.
However, one would also not want to wait too long
either. For a typical large trial, when a sample-size
adjustment procedure is planned after 30–50% of
the data are available, the estimates are reasonably
stable and the statistical penalty involved in the
p-value adjustment is relatively small. Also, some
procedures leading to an early decision to stop the
trial for lack of efficacy, or afutility analysisalso
involve the calculation of conditional power. The
same calculations involved in sample-size adjust-
ment can be used to assess futility. If the condi-
tional power is very low (which would lead to a
substantial increase in the sample size, should the
trial continue), the sponsor may want to consider
terminating the trial and reallocating the unused
resources to more promising investigations.
The administrative issue involves the poten-
tial for the introduction of bias. Any interim
analysis, regardless of whether or not it is
done with the intention to affect the ongoing
trial, involves the possibility of introducing
bias if the analysis requires the breaking of the
blind. The FDA is particularly weary about
these types of analyses because even if every
step and decision is well documented, it is
impossible to anticipate the impact on the trial
that a partial, even preliminary, knowledge of
the efficacy results might have. For this reason,
it is imperative that such analyses, regardless of
their declared purpose, will be performed with
strict guidelines as to who will be unblinded,
and how the results will be disseminated. It is
important to make sure that individuals directly
involved in the trial conduct and management,
such as investigators, monitors and other project
personnel, should remain blinded to the data and
the results of the interim analysis. It has become
standard practice in the pharmaceutical industry
to appoint a Data Monitoring Board consisting
entirely of people uninvolved in the trial con-
duct to review the interim data and analyses and
make recommendations. The Pharmaceutical
Manufacturers Association published a position
paper discussing in details the various aspects of
theissueasitrelatetothespecificcircum-
stances of new drug development (PMA Biosta-
tistics and Medical Ad-Hoc Committee on
Interim Analysis, 1993). The ICH guidelines
address this issue as well.The execution of an interim analysis should be a
completely confidential process because unblinded
data and results are potentially involved. All staff
involved in the conduct of the trial should remain
blind to the results of such analyses, because of the
possibility that their attitudes to the trial will be
modified and cause changes in the characteristics of
patients to be recruited or biases in treatment com-
parisons. This principle may be applied to all inves-
tigator staff and to staff employed by the sponsor
except for those who are directly involved in the
execution of the interim analysis. Investigators
should be informed only about the decision to con-
tinue or to discontinue the trial, or to implement
modifications to trial procedures.338 CH25 STATISTICAL PRINCIPLES AND APPLICATION IN BIOPHARMACEUTICAL RESEARCH