Biomedical Engineering Reference
In-Depth Information
for agents that may not be expected to shrink tumors but only delay
their growth (i.e., are cytostatic), and targeting benchmarks for
other endpoints (e.g., progression-free survival or overall survival)
may not be reliable [20] without an extensive meta-analysis of trial-
to-trial variability [21]. Another situation in which a single-arm trial
would be hard to interpret is when the new agent is given together
with an agent that has activity, making it difficult-to-impossible to
define a response rate target. In cases where single-arm trials would
be inappropriate, randomized phase II trial designs have been
proposed [22]. These trial designs are similar to definitive phase III
trials except that (a) the endpoint will typically not be a definitive
clinical benefit endpoint like overall survival but instead be an early
measure of clinical activity like response rates or progression-free
survival, and (b) the type 1 error rate (alpha level, significance level)
will typically be larger than the 0.05 used in a phase III trial.
There are different ways biomarkers can be used in phase II
trials. One possibility is to use a biomarker as the outcome rather
than tumor shrinkage or survival. However, one would need to be
convinced that sufficient “biomarker responses” in the patients
was indicative that the new therapy works (i.e., would demonstrate
clinical utility in a follow-up phase III trial) and insufficient biomarker
responses would be indicative that the new therapy does not work.
Biomarkers can seldom meet this standard. The use of levels and
changes in prostate-specific antigen (PSA) as a measure or response
in phase II prostate cancer trials illustrates some of the issues. It
has been discussed at length as to whether PSA should be used this
way [23, 24], especially because it is known that some agents can
affect PSA levels without affecting prostate tumors [25]. Even if one
has evidence that a biomarker is a reasonable surrogate for clinical
response for a set of treatments (perhaps from a meta-analysis [26]),
there will typically be an extrapolation to assume that it will be a
reasonable surrogate for a new agent, especially if the new agent has
a different mechanism of action than the previous set of treatments.
The most promising application of biomarkers in phase II trials
is to help define what population would be most appropriate to be
treated by a new agent. Two main goals for this application are (1)
to assess the new therapy in a biomarker-defined patient subset
where it is expected to work (an “enriched” population), and (2) to
determine if the new therapy should be developed for all patients or
an enriched population only. (Sometimes the definition of enrichment
