Biomedical Engineering Reference
In-Depth Information
31]. When a phase II randomized comparison with a control therapy
is required, one can compare the treatment effect in the biomarker
positive and negative subgroups to get a suggestion as to whether
the biomarker is predictive, but a definitive analysis will require a
phase III biomarker-stratified design (Section 2.6.1).
2.5.2
Designs Involving Multiple Biomarkers
With each biomarker having its own associated targeted therapy, one
can assign patients to the targeted therapy that is appropriate for
them. If the setting is such that a control comparison is needed then
one would randomize patients with a particular biomarker to their
associated targeted therapy versus a control therapy. If a control
comparison is not required, e.g., seeking positive response rates with
single agents, then no randomization to control therapies would
be needed. An example of the latter situation is a National Cancer
Institute trial (NCT01306045) in advanced lung and thymic cancer,
where depending upon the patients' biomarkers, they are assigned
one of five targeted therapies [32]. In this setting, no standard active
therapy is available and thus observing any tumor responses would
be interesting so a control therapy is not required. Another example
was a trial [33] for patients with refractory metastatic cancer where
a panel of 64 biomarker targets is used to choose an individual
treatment targeted for each patient. The outcome was the number
of patients whose time to progression was >1.3 times longer than
their time to progression before entering the trial, a potentially
problematic outcome [34, 35].
When the multiple biomarkers do not each have an associated
targeted therapy, there are different approaches that can be used.
A retrospective approach examines a number of biomarkers to
see which, if any, are associated with good responses in a group of
patients given the same (or a set of) treatments. Care is required if
thousands of potential biomarkers are considered (e.g., expression
of many individual genes) in that an important biomarker could
easily be missed because of the background statistical noise of all
the biomarkers being considered [31]. A prospective approach
randomizes patients to different treatments and evaluates a (small)
set of biomarkers on the patients to look for associations. This type of
exploratory approach would be appropriate when one has little idea
about the relationships between the biomarkers and the therapies.
