Stratification studies
In pivotal studies, large numbers of patients are
studied so that their diverse clinical characteristics
can imitate better the ordinary patient population
than in earlier, more selective trials. When a variety
of concomitant factors (e.g. other diagnoses, wider
degree of disease severity, concomitant medica-
tions, etc.) are suspected, and may interact with
drug tolerability or efficacy, then patients may be
stratified into randomization groups according to
the presence or absence of such factors. For exam-
ple, patients with Crohn’s disease might be strati-
fied according to whether or not they also have
cutaneous manifestations, and each stratum then
randomized to active or placebo for a total of four
treatment groups, although with only two test treat-
ments. Separate statistical analyses for the strata
can then be planned, and the study size adjusted
accordingly. The efficacy of the new drug may be
found to be restricted to a (some) particular patient
subset(s). Regulatory authorities will often
approve indications with caveats based on such
subsets. For example, in the United States, one
indication for aprotonin is ‘...to reduce periopera-
tive blood loss...in selected cases of primary
coronary artery bypass graft surgery where the
risk of bleeding is especially high, for example
impaired hemostasis, presence of aspirin or coagu-
lopathy of other origin’. The risk of stratification
studies is that conservative regulatory authorities
will want to see statistical significance in all patient
subsets before allowing a short, broad indication in
labeling.
The ‘Large, simple study’ is a recently recog-
nized alternative to stratification, pioneered by
Peto. Large numbers of unselected patients are
subjected to a single randomization. If enough
patients are recruited, and if the randomization is
truly unbiased, then the large sample sizes will
allow all the potentially interacting variables (con-
comitant drugs, concomitant diseases, demo-
graphic variables, etc.) to balance out between
the treatment groups.
The ‘simple’ part of this approach is that, in
fundamental terms, the case report form can be
very short. There is no need to collect lots of
information about the patient’s clinical condition
because there is no use for these data. Trials of
cardiovascular drugs, on an almost epidemiologi-
cal scale, have been the most significant exampleof
this alternative approach. Literally, tens of thou-
sands of patients have been recruited under these
protocols with case report forms having fewer than
10 pages for each patient. Dr Robert Temple (1997;
Director of the Office of Drug Evaluation I, at
FDA) has commented that it may even be possible
to conduct large simple studies in treatment IND
situations, thus permitting the generation of effi-
cacy data outside of orthodox ‘phase III’ clinical
trial programs. However, in this case the end point
would have to be just as simple, for example,
survival or death of the patient, during a documen-
ted period of observation; Kaplan-Meier analysis
and other epidemiological approaches may also be
applied to such databases.
Although the conditions under which large sim-
ple trials can provide efficacy data are fairly well
worked out, it is important to consider whether (or
which) tolerability issues can be precisely
addressed in this way. If a tolerability factor
(adverse event) relates to the efficacy variable of
interest (e.g. a fatal adverse event in a patient
survival study), then a simple case report form
may provide relevant information. However, if
the adverse event type is rare or unanticipated
(e.g. the test drug causes unanticipated, significant
anaemia in 0.1 % of patients, and the protocol and
case report form do not collect hemoglobin values
before and after treatment), then it is very likely
that the adverse event will be missed. Large simple
studies can thus create undue confidence in product
tolerability (‘thousands of patients were exposed to
the agent during clinical trials’).9.12 Treatment withdrawal and
other specialized designsThere are rare cases where established treatments
are without strong evidence-based support. Two
good examples exist for digoxin: the treatment of
mild heart failure and the treatment of cardiac
asthenia, a diagnosis that is especially common in
Europe, and for which relatively small doses are
prescribed. When the effect of such treatments on112 CH9 PHASE II AND PHASE III CLINICAL STUDIES