is a known or suspected safety concern with the
drug or the class of drugs tested.
Interim analysis
Long-term clinical trials in life-threatening disease
areas or in diseases involving serious morbidities, or
in the study of drugs with possible serious toxicities,
it is imperative to monitor the data on an ongoing
basis and perform periodic interim analyses.
Interim analyses are performed for a variety of
reasons. Some of the main reasons are as follows:
(a) Stop the development of an ineffective
treatment.
(b) Stop the development of a toxic treatment.
(c) Terminate a trial in a life-threatening disease as
soon as enough evidence accumulates to con-
clude that one treatment is significantly more
efficacious than the other.
(d) Interim design adjustment (e.g. verification of
assumptions on variability, power recalcula-
tion and sample-size adjustment; verification
of assumptions on expected drug or control
group response rate).
(e) Plan additional trials.
(f) Plan for capital expenditures and product
launch.
(g) Make a regulatory submission for a short-term
portion of a long-term trial.
(h) Other regulatory reasons (e.g. opening the trial
to previously excluded high-risk subjects).
The first three reasons in the list include the possi-
bility of terminating the trial based on an interim
inferential analysis. The fourth reason can poten-
tially alter the trial’s conduct. The other reasons
should not, in principle, impact the trial.
Essentially, there are two separate issues
involved in performing an interim analysis: a
statistical issue and an administrative or trial man-
agement issue. The statistical issue is similar to the
multiplicity issue discussed in the previous para-
graph and applies to (a), (b) and (c) above. If we
perform an interim inferential test, the overall error
probability is inflated. Therefore, if one contem-
plates to perform an interim analysis with the
option of making inferences early and possibly
terminating the trial before its planned end, the
procedure used for making this determination
must be planned in advance and documented in
the study protocol just as any other inferential
procedure. As we discussed above, there will be a
statistical penalty in the sense that each of the
interim analyses and the final analysis will have
to be performed at a lower levelof significance than
the overall type I error rate. The statistical penalty
depends on the decision-making procedure to be
used.
Interim analysis for the purpose of reassessment
of the design assumptions and sample size recal-
culation has become a rather common place espe-
cially in large, long-term phase III trials. The
assumptions driving the design of these trials are
typically based on published on unpublished pre-
vious exploratory research or on extrapolations
from preclinical work. These assumptions often
involve a great deal of uncertainty. To reduce this
uncertainty, an interim analysis at some time point
early in the trial is planned, the sole purpose of
which is to use the data accumulated thus far and
estimate the parameters used to perform the power
calculations and make appropriate adjustments to
the trial design. Recalculation of the sample size is
the most typical purpose of such analysis. A num-
ber of procedures for an interim sample-size
adjustment were proposed in the statistical litera-
ture in recent years. One such approach is to cal-
culate the probability that the trial, when continued
as planned, will result in a significant outcome
conditioned on the accumulated data. When this
probability is calculated under the alternative
hypothesis, this (conditional) probability is called
theconditional power. If the conditional power is
equal or higher than the power used in the original
design of the trial, the trial will continue as
planned. If the conditional power is smaller, the
sample size will be increased. The increase of the25.11 ISSUES IN STATISTICAL TRIAL DESIGN 337