Transcript Biomarkers & Personalized Medicine

Biomarkers & Personalized Medicine:
Practical Considerations for Drug Development
Dominic G. Spinella, Ph.D.
Pfizer Oncology Translational Medicine Head
1
Some Definitions

Biological Marker (Biomarker): A characteristic that is
measured and evaluated as an indicator of normal biological
processes, pathogenic processes, or pharmacological responses
to a therapeutic intervention. Biomarkers may relate to efficacy,
safety, differentiation etc.

Diagnostic: A biomarker that has applicability in clinical use or
patient management (e.g. to identify a sub-population of patients
who would benefit most from a drug or suffer adverse events from
a drug).

Surrogate Endpoint: A biomarker accepted by regulatory
agencies as a substitute for a clinical endpoint (e.g. HIV load for
the stage of AIDS, LDL level for the risk of coronary artery
disease, blood pressure for the incidence of stroke & hemoglobin
A1C for the control of diabetes).
2
Some Definitions

Personalized Medicine: “Use of new methods of molecular
analysis to better manage a patient’s disease or predisposition
towards a disease. It aims to achieve optimal medical outcomes
by helping physicians and patients choose the disease
management approaches likely to work best in the context of that
patient’s genetic and environmental profile” (from the
Personalized Medicine Coalition)

Molecular signature: A constellation of several or many discrete
molecular characteristics (DNA, RNA or Protein) that collectively
constitute a biomarker of disease or drug response

Hypothesis-dependent marker/signature: A biomarker or
potential biomarker that is derived a priori from intrinsic
understanding of a disease or pathogenic process.

Hypothesis-independent marker/signature: A biomarker that is
derived post-hoc using –omics analysis of biologic samples from
patients with different phenotypes (e.g. drug responder vs. nonresponder)
3
The Hope (Hype?) of Personalized Medicine





Understanding the molecular basis of disease: Which therapy or
combination of therapies to use
Defining molecular changes or markers associated with disease
progression, response to treatment and relapse: When to treat with a
particular regime.
Identifying markers associated with safety & toleration: Choosing the
safest therapies and correct dose.
Identifying the right population for clinical trials
•
efficacy may only be evident in a subset of patients, rather than being
uniform across the whole population
Rescue a “failed” drug
•
Better understand the molecular characteristics of responsive vs.
non-responsive patients
4
Examples of PM using Biomarkers
in Current Drug Labels
Biomarker
Test
Drugs
CYP2C9
Recommended
Warfarin
EGFR
Required
Cetuximab
GPD6 deficiency
Recommended
Dapsone, Rasburicase
Her2Neu +ve
Required
Trastuzumab
TPMT variation
Recommended
Azathioprine, mercaptopurine,
thioguanine
UGT1A1
Recommended
Irinotecan
Urea cycle enzyme
deficiency
Recommended
Valproic acid
HLA-B*5701
Recommended
Abacavir
Adapted from: Frueh et al (2008) Pharmacogenomic biomarker information in drug labels approved by the
United States Food and Drug Administration: prevalence of related drug use. Pharmacother, 28:992-8
5
Most current PM Biomarkers are “obvious”





A patient with a poorly active variant of a drug metabolizing enzyme will
have a different PK profile (and greater risk of over-exposure) to drugs
that are eliminated via pathways that employ that enzyme (e.g.
UGT1A1+ patients exposed to irinotecan).
Patients in which a drug target is poorly expressed (or not expressed at
all) in their disease will likely not respond to the drug (e.g. Her-2 negative
patients for trastuzumab, even ER- breast cancer patients exposed to
tamoxifen – probably the first example of PM).
Patients who have mutations in pathways downstream from the drug
target that render that target irrelevant, will not derive much benefit from
the drug (e.g. Kras mutant patients exposed to EGFR inhibitors).
DUH!!
Despite the “obviousness” of these biomarker hypotheses, establishing
and proving such associations is extraordinarily difficult. Why?
6
Challenges to the use of biomarker approaches
to Clinical Drug Development and PM
1.
We don’t know as much as we think we do!
Example: Sorafenib, a multi-tyrosine kinase inhibitor, was originally
developed as an inhibitor of Raf (an intermediate of the growth factor
signaling cascade that is mutated in some cancers). The “obvious”
approach of testing it in BRAF driven melanoma, would have led to
drug failure (it has no real benefit here). Only after clinical activity was
fortuitously discovered in a Phase 1 setting in RCC patients was it
recognized that its clinical benefit resulted from the fact that it is also
inhibits VEGF receptor tyrosine kinase.
7
We don’t know as much as we think we do!
8
We don’t know as much as we think we do!
9
We don’t know as much as we think we do!
10
Challenges to the use of biomarker approaches
to Clinical Drug Development and PM
1.
We don’t know as much as we think we do!
2.
Failure to distinguish between predictive and prognostic markers.
11
Prognostic vs. Predictive Factors
Prognostic Factor: Any measurement that is
associated with clinical outcome in the absence of
therapy, or with the application of a standard therapy
that all patients are likely to receive (a predictor of
the natural history of the disease).
Predictive Factor: Any measurement associated
with response or lack of response to a particular
therapy, where response can be defined using any
of the clinical endpoints commonly used in clinical
trials (eg, ER for patients with breast cancer).
Clark GM. Mol Oncol 2008, doi:10.1016/j.molonc.2007.12.001
12
Marker + Patients Treated with Experimental
Therapy
Median Survival = 12 months
• Cannot determine if Experimental Therapy confers meaningful benefit over
Standard Therapy
• Cannot evaluate prognostic value of Marker X
• Cannot assess predictive value of Marker X
13
Can we study stratified patients treated only with
the experimental therapy?
Marker + Patients
Marker - Patients
NO, because we can’t tell if marker+ patients would have done better than markerpatients regardless of treatment (i.e. it might be a prognostic marker)
We really need a control group!
14
Can we study an enriched population only?
Marker + Patients
Experimental
Therapy
Standard Therapy
NO, because even though marker+ patients do better on treatment relative to
standard therapy, we can’t tell if the treatment might have been equally better than
standard in marker– patients (i.e. it may not be marker of either prognosis or drug
effect). . . . only if we will be satisfied with half of an answer.
15
What we really need is a Biomarker-based
Study Design
Treatment A
Marker+
Randomize
Treatment B
Register
Stratify
Test Marker
Treatment A
Marker-
Randomize
Treatment B
16
Predictive vs. prognostic marker evaluation
Predictive but not prognostic
Marker +
T
S
T
S
Marker -
Marker is predictive (only patients who are marker+ show the treatment effect),
but not prognostic (marker- patients do the same as marker+ patients on
standard care).
17
Predictive vs. prognostic marker evaluation
Prognostic but not predictive
T
Marker +
S
T
S
Marker -
Marker is prognostic (marker+ patients do better than marker- patients on both
the treatment arms and the SOC arms), but not predictive (even though treated
patients do better than SOC all patients, i.e. there is a drug treatment effect,
the magnitude of the difference is the same in both marker- patients and
marker+ patients).
18
Predictive vs. prognostic marker evaluation
Marker +
T
S
T
S
Marker -
Marker is neither predictive nor prognostic (treatment is equivalently better
than SOC in both marker+ and marker- patients)
19
Predictive vs. prognostic marker evaluation
Predictive and prognostic
Marker +
T
S
T
S
Marker -
Marker is both predictive and prognostic (treatment is better than SOC in both
marker+ and marker- patients, but the magnitude of the treatment effect is
greater in the marker+ patients)
20
Challenges to the use of biomarker approaches
to Clinical Drug Development and PM
1.
We don’t know as much as we think we do!
2.
Failure to distinguish between predictive and prognostic markers.
3.
Data over-fitting and reliance on retrospective analyses.
The vast majority of retrospective analyses with hypothesis
independent (or other) approaches fail to be confirmed in prospective
studies. The statistics of determining the “significance” of differential
gene expression (for example) when there are tens of thousands of
analytes in only dozens of samples is extremely dicey.
21
Challenges to the use of biomarker approaches
to Clinical Drug Development and PM
1.
We don’t know as much as we think we do!
2.
Failure to distinguish between predictive and prognostic markers.
3.
Data over-fitting and reliance on retrospective analyses.
4.
Grafting a retrospective –omics analysis onto a failed Phase 3 study
will rarely “rescue” the drug – even if it successful in identifying the
molecular characteristics of a “responsive” subset.
22
Oncology Drug Development Timelines

It takes an average of 15 years for a new oncology drug to obtain FDA approval

Development costs escalating: >1.5B (including cost of failures)



Patient accrual is a rate limiting step in drug development. Oncology Ph 3 trials can take 3–4 years to
accrue the target number of patients (<5% of US adult cancer patients participate in clinical trials)
Average time from peak market sales to LOE is 4-6 years.
There is not enough time to LOE remaining after a failed Phase 3 study to complete a new
prospective registrational study incorporating a biomarker hypothesis generated from
retrospective analysis of a preciously failed Phase 3 to make this a viable strategy!
23
Challenges to the use of biomarker approaches
to Clinical Drug Development and PM
1.
We don’t know as much as we think we do!
2.
Failure to distinguish between predictive and prognostic markers.
3.
Data over-fitting and reliance on retrospective analyses.
4.
Grafting a retrospective –omics analysis onto a failed Phase 3 study
will rarely “rescue” the drug – even if it successful in identifying the
molecular characteristics of a “responsive” subset.
5.
Lack of incentives. Even successful drugs work only in a subset of
patients who receive them. Better defining that subset post-marketing
(where the N is large enough) may lead to better efficacy in that nowsmaller population (and a label restriction), but it will not lead to higher
reimbursement…
24
Challenges to the use of biomarker approaches
to Clinical Drug Development and PM
1.
We don’t know as much as we think we do!
2.
Failure to distinguish between predictive and prognostic markers.
3.
Data over-fitting and reliance on retrospective analyses.
4.
Grafting a retrospective –omics analysis onto a failed Phase 3 study
will rarely “rescue” the drug – even if it successful in identifying the
molecular characteristics of a “responsive” subset.
5.
Lack of incentives. Even successful drugs work only in a subset of
patients who receive them. Better defining that subset post-marketing
(where the N is large enough) may lead to better efficacy in that nowsmaller population -- and a label restriction -- but it will not lead to
higher reimbursement…
6.
Making binary decisions (“treat” / “don’t treat”) on continuous data.
25
Where do you draw the treat / don’t treat cut-off?
Number of patients
Intermediate risk
Low risk
Very low risk
High risk
Very high risk
Number of risk factors for disease X
26
Challenges to the use of biomarker approaches
to Clinical Drug Development and PM
1.
We don’t know as much as we think we do!
2.
Failure to distinguish between predictive and prognostic markers.
3.
Data over-fitting and reliance on retrospective analyses.
4.
Grafting a retrospective –omics analysis onto a failed Phase 3 study
will rarely “rescue” the drug – even if it successful in identifying the
molecular characteristics of a “responsive” subset.
5.
Lack of incentives. Even successful drugs work only in a subset of
patients who receive them. Better defining that subset post-marketing
(where the N is large enough) may lead to better efficacy in that nowsmaller population -- and a label restriction -- but it will not lead to
higher reimbursement…
6.
Making binary decisions (“treat” / “don’t treat”) on continuous data.
7.
Regulatory and partnering indemnification of co-diagnostics.
27
Personalized Medicine:
Some Conclusions and predictions





The march of science is inevitable and biomarker/PM
approaches to clinical care will continue to advance.
Given the current regulatory climate, initial focus will be
on prediction of adverse drug responses.
PM is expensive – at least in its initial stages. There will
be substantial resistance on the part of payors to
reimburse PM/BM/diagnostic tests unless true clinical
benefit can be definitively established.
Empirical approaches will trump molecular approaches
in all cases where the cost/risk : benefit ratio favors it.
Despite the hype, the impact of PM on medical practice
will be evolutionary rather than revolutionary.
28
29