Transcript G4_Ostland_ Pharmacogenomics

Patient Selection Markers in
Drug Development Programs
Michael Ostland
Genentech
FDA/Industry Statistics Workshop:
Washington D.C., September 14 – 16, 2005
Outline
• Background
• Seven Questions from a Drug
Development POV
• Concluding Remarks
Background
• Most drugs benefit far less than 100% of the
patients who are treated.
• Patients who get no efficacy from a drug:
– Still run the risk of toxicity or side effects
– May miss out on a benefit they would have
received had they been treated with another
drug instead
– Add costs to the health care system
– Dilute efficacy estimates in clinical trials
Background (cont’d)
In drug development patient selection may:
• Enrich a population for patients who benefit,
thereby allowing a drug’s efficacy to be detected
in a smaller phase III trial. (see Maitournam and
Simon)
• Enrich a population for patients with a favorable
toxicity profile, thereby improving the benefit/risk
ratio.
Maitournam and Simon, Statist. Med. 2005; 24:329–339
Background (cont’d)
By “marker” we typically have a biomarker in
mind. Namely,
a characteristic that is objectively measured and evaluated as an
indicator of normal biologic processes, pathogenic processes, or
pharmacologic responses to a therapeutic intervention
(Biomarkers Definitions Working Group 2001)
In principle, any objectively measured baseline
characteristic (or completely specified
combination of multiple characteristics) could
form the basis for selecting patients to be
candidates to receive treatment.
Seven Basic Questions
1. What strategic imperative for patient selection?
2. What is the desired outcome from the
development program (“Target Product Profile”)?
3. What could phase III look like?
4. What should phase II look like?
5. How do results from phase II lead to decisions
about the design in phase III?
6. How many patients are needed in phase II to
ensure adequate decision making?
7. What marker and what threshold for “positive”?
Strategy for Patient Selection
Patient selection for clinical drug development can
proceed with one of several strategic imperatives:
• Efficacy:
– Include patients most likely to benefit
– Exclude patients least likely to benefit
• Safety:
– Include patients least likely to experience toxicity
– Exclude patients most likely to experience toxicity
Target Product Profile
Establish relationship between target efficacy/safety
and proportion of patients selected for treatment.
e.g., How much more
effective does a drug
need to be if only 40%
of the population can
be treated?
Other useful metrics
possible.
Phase III Designs with Patient
Selection
Option 1
Standard design, except only enroll patients
from marker selected population.
Question
What are the scientific and regulatory
implications of not performing a definitive
assessment of efficacy on unselected patients?
Phase III Designs (cont’d)
Option 2
Enroll all patients and assay for marker. Perform
two primary efficacy analyses while controlling
overall type I error rate:
(1) Efficacy among all patients
(2) Efficacy on marker selected patients
Question
How does efficacy on marker unselected patients
impact inference when (1) is positive?
Phase II Design
Usually best to consider a randomized trial:
• Allows assessment of whether the marker is
truly predictive of increased treatment benefit,
rather than simply prognostic for good
outcome.
• Assessment of safety with contemporaneous
control arm.
Ideally, one tests the marker prior to
randomization and stratifies, but this may not
be possible for logistic reasons.
Phase II Design (cont’d)
A randomized design with retrospective testing
assay
Treatment
Enrolled
Patients
Positive
A
Negative
B
Indeterminate
C
Positive
D
Negative
E
Indeterminate
F
randomize
assay
Control
Whether patients who test “indeterminate” ultimately get treated depends on
the selection strategy: exclude only those least likely to benefit (yes), or only
include only those most likely to benefit (no).
Phase II to III Decision Making
Broadly speaking, there are four possible
decisions after a phase II trial with a patient
selection marker:
1. Proceed to Ph III in marker+ subset only
2. Proceed to Ph III in all patients and perform
two tests: in all patients and in marker+ pts
3. Proceed to Ph III in all patients and ignore
marker
4. Do not proceed to Ph III at this time
Decision making (Cont’d)
Key efficacy comparisons:
• A vs. D: Treatment effect
among known marker
positive
• B vs. E: Trt. effect among
known marker negative
• (A+B+C) vs. (D+E+F):
Overall treatment effect
• (A vs. D) vs. (B vs. E):
Treatment effect by marker
interaction
Treatment
Control
Positive
A
Negative
B
Indeterminate
C
Positive
D
Negative
E
Indeterminate
F
Decision making (Cont’d)
• Present the key efficacy and safety comparisons
along with reasonable estimates of uncertainty
• Interpret results using Target Product Profile
• Take into account
– Asymmetry of the decision-making loss function
– Biologic plausibility
Hard and fast rules for all possibilities are hard
(impossible?) to come by.
Size of Phase II
• An area of great opportunity for statisticians
• Power is too rigid to be very useful
• Expected CI widths are hard to evaluate when
several parameters are of simultaneous interest
• Probably want to approach it from a decisiontheory point of view. But this is not trivial:
– The fore-mentioned lack of strictly defined decision
rules makes analysis impossible
– Quickly approach the sort of mind-numbing
complexity that confirms clinicians worst prejudices
about statisticians.
Marker Selection
• Best to have 1 – 3 candidate markers based on
biologic MOA and preclinical evidence. Then a
short phase II program can be used to
prospectively assess.
• Sometimes need to use part of phase II to screen
for candidate markers, and then a subsequent
clinical study (prospective or retrospective?) to
validate. This is lengthier and requires care
(multiplicity, cross-validation at proper level, etc.).
Fortunately, a lot of smart statisticians have made
good progress on these matters.
Similar points apply to establishing “positive” threshold
Concluding Remarks
• Phase II is a critical part of clinical development
when patient selection markers are considered
• Knowledge of the assay is helpful
• A clear Target Product Profile is critical
• Statisticians have an important role in planning
and decision making in this complex, uncertain
environment
• Planning for phase II in a way that can be usefully
communicated to decision makers is an open
question.
Acknowledgements
Alex Bajamonde
Cheryl Jones
Lee Kaiser
Gracie Lieberman
Ben Lyons
Howard Mackey
James Reimann
Julia Varshavsky
Xiaolin Wang