Biomedical Engineering Reference
In-Depth Information
An example of this is the BATTLE trial, where chemorefractory
non-small cell lung cancer patients were randomly assigned to four
therapies and were simultaneously categorized into five biomarker
subgroups; the outcome was eight-week disease control [36]. Another
example is the I-SPY 2 trial, where locally advanced breast cancer
patients are randomly assigned neoadjuvant therapies and were
simultaneously categorized with multiple biomarkers; the outcome
was (modeled) pathological complete response rate [37]. Both the
BATTLE and I-SPY 2 trials use outcome-adaptive randomization,
in which treatment arms that are doing better receive a higher
proportion of patients randomized to them. However, outcome-
adaptive randomization leads to trials with larger required sample
sizes and a larger absolute number of patients having bad outcomes,
making its use controversial [38-41].
2.6
Biomarker Evaluations in Phase III Trials
For trials with a pre-specified binary biomarker, the three main
types of trial designs are biomarker-stratified designs, enrichment
designs, and biomarker strategy designs [42], although numerous
hybrid versions and extensions are possible. For trials in which
biomarker has not been completely specified (“locked down”) at the
time of the initiation of the trial, we briefly discuss some possible
phase III trial design approaches.
2.6.1
Biomarker-Stratified Designs
Biomarker-stratified designs allow for full information about the
relative treatment efficacy in each biomarker-defined subgroup
and in the overall patient population. When each patient is pre-
registered to the trial, the patient's biomarker status is assessed.
If a biomarker result is successfully obtained, the patient is then
offered full registration, and consenting patients are randomized to
the treatments within the appropriate biomarker stratum (possibly
stratified to ensure balance of treatment arms within each biomarker
subgroup). This design schema is presented in Fig. 2.2. If one of
the treatment arms is standard of care, the prognostic value of the
marker can be assessed in this design by comparing the standard-of-
care arms between the biomarker-positive and biomarker-negative
subgroups. Predictive value of the biomarker can be initially
