Transcript Slide 1

Drug/Diagnostic CoDevelopment
Paths to Getting the Right Drugs to
the Right Patients
Gracie Lieberman - Genentech
The 2011 Midwest Biopharmaceutical
Statistical Workshop
Personalized Health Care – Fantasy or Reality?
Here’s my tumor’s DNA
sequence
The Changing Landscape

The past decade has seen a great increase in the
understanding of cancer pathogenesis at a molecular
level.

Many signaling pathways have been identified and
implicated in cancer development.

Most of the today’s focus in oncology drug development
is on targeted therapies which are expected to be active
only in subsets of patients


The target of interest is often a key driver of tumor genesis and
therefore may serve as the basis for a candidate predictive
biomarker (e.g.. HER2-gene, mEGFR, mBRAF)
True targeted therapy requires not just a selective agent
but a means of identifying appropriate patients (i.e. a
diagnostic)
Implications for Development



Rational targeted drug development strategies aim to identify patient
populations likely to derive greatest clinical benefit from a new
molecular entity (NME).
Which target: selective or multiple
Which disease

Reliance on different nodes in the pathway (RTKs, PI3K mt, mTOR)
 Translatability of Proof Of Concept (POC) across diseases, line,
combinations

Which patient (role of diagnostics)


Putative Dx markers have strong links to tumor biology
Are pathway alterations predictive of target dependence



What is the size of the targeted population
Single agent vs. combination



Can we select patients based on any of them?
Chemotherapy, EGFR, MEK, VEGF…..
Appropriate comparator arm?
How hard to hit the target (dose and schedule)
Which Way to Rome?
Drug-Dx Co-development
Current
regulatory paradigm requires early biomarker discovery
Target
Discovery
Drug
Preclinical
Dev’t
Phase 1
Phase 2
Phase 3
FDA filing,
approval,
launch
+
Dx
Biomarker
discovery
Biomarker
Assay Dev’t
Clinical Validation
of biomarker hypothesis
PMA filing,
approval,
launch
FDA draft white paper on Drug-Dx co-development
To
maximize clinical benefit from our therapeutics:

Informed decision making around indication choice
 Enable patient selection
Predictive biomarker: a test that can be done before treatment to
predict whether a particular treatment is likely to be beneficial
Pharmacodynamic biomarker: a test that can be done pre- and posttreatment to confirm target modulation
Clinical Biomarker Strategy
“Follow the tumor genetics”

Focus on high prevalence oncogenic mutations,
amplification, tumor suppressors




Ensure assays are available for putative biomarkers
Path to Companion Diagnostics enabled
Biomarker testing incorporated in trial design
Mandatory tumor tissue
 Exploratory assays to evaluate other promising biomarkers
qRT-PCR gene signatures
Mutation and copy number profiling
 Multiple PD biomarkers to confirm pathway inhibition
PI3K/MEK Pathways: multiple nodes are candidates for molecules
targeting core oncogenic pathways; potential for broad application,
combinations, with strong biomarker hypotheses
GF
PTEN Prostate
RTK
P
P
GBM
P
P
PI3K
Ras
PTEN
S6K
Bladder
Gastric
AKT
mTOR
Scaffold
PDK1
Breast
Ovarian
HNSCC
HCC
Raf
MEK
ERK
Proliferation
Survival
Invasion
Melanoma
Pancreatic
NSCLC
Endometrial
CRC
PI3K
p110 oncogenic mutations in:
37% Endometrial
29% Breast
25% Colon
13% Bladder
PIK3CA amplified in 30% ovarian, lung
PTEN mutant/lost in:
breast, prostate, glioblastoma, melanoma,
pancreatic, endometrial, ovarian, lung,
head and neck, hepatocellular, thyroid
Ras
Preclinical Biomarker Discovery: Breast Cancer
Her2 Amplified
Trastuzumab, lapatinib
Basal-like
“Triple Negative”
No targeted therapy
Luminal
Hormonal therapies
Vimentin
(myoepithelial
basal)
E-cadherin
(epithelial)
Sorlie et al, PNAS, 2001
PHC assessment to guide development strategy
• Preclinical data in mBC suggests that cell lines
harboring PI3K/PTEN alterations are highly
sensitive to the pathway inhibitors
What does Dx hypothesis usually look like?

Scientific evidence based on
the Mechanism of Action (MOA)
 In-vitro and xenograft assays
 Availability of a well defined
biomarker hypothesis



One or at most two researchgrade assays
Single summary measure
Appropriate cutoff available or
to be derived for
ordinal/continuous biomarkers

Clinical evidence from our and
competitor’s trials (Phase I is
unlikely to yield useful efficacy
information)
 Prevalence of the proposed
biomarker (description of the
distribution if continuous) and
known prognostic characteristics
(given the line and indication)




Literature reports
Public databases
Should we consider a separate
tumor registry to address the
question?
If prognostic, is genomic drift a
potential issue?

Archival vs. Fresh Biopsies
Dedicated studies may be necessary to investigate assay or biomarker properties,
prevalence and prognostic significance
Underlying Assumption
Is there a strong Dx hypothesis?
yes
no
Is Dx to be used in the label?
no
yes
Is activity in Dx (-)
exceedingly unlikely?
yes
Patient
selection
To be used in the
indication
section of1the
label
Is Dx to be used in the label?
yes
WHY?
no
Dx will be a coprimary
or 2-ary endpoint
(testing prespecified
hypotheses)
no
Pause
Define path forward;
Consider: costs, risks
and timelines
Co-primary could
be used in the
indication or
clinical sections of
the label;
2-ary in the
clinical section of
the label
All comers
trial;
Exploratory
data mining
To be used for
publications,
hypothesis generation;
Rescue/retrofit Dx
PHC Assessment
Development Strategy
PHC Assessment
 Strong
Dx
hypothesis
 No activity in Dx-
 Strong
Dx
hypothesis
 Some activity in Dx-
 No
strong Dx
hypothesis
 Exploratory Stage
Development Strategy
 Patient
selection
through all
phases of
development
Selected
 Complex,
larger
phase IIs with
stratification
 Complex phase IIIs
Stratified
 No
selection or
stratification
 Data exploration
All Comers
Basic Principles
Prospectively Defined Dx markers are
required for any label enabling action.
 Dx markers must be defined prior to
pivotal trials and hence planned for in a
Clinical Development Plan (CDP).
 Dx. Marker evaluation often in phase II.
 Retrospective Analyses are only
considered “hypothesis” generating.

Incorporating Dx - Impact on components of CDP
 Target Product Profile (TPP)


Phase I trials


Selection for quick signal seeking
Phase II (POC) trials



Parallel development of companion diagnostic
Complex issues become more complex
More unknowns, more questions to answer
Phase III trials


Clinical Validation of Dx
Design depends on Phase II outcome

Selection, stratification or all-comers
Longer and Costlier but this is reality!
Phase II Considerations




Objective: simultaneous Rx/Dx evaluation
Scientific rationale and pre-clinical data - main
determinants of the scenario prior to Phase II
Strategic and operational considerations
Statistical considerations





Co-primary endpoints
Value added and feasibility of stratification
Defining cut-offs for continuous biomarker
Go/No Go decision algorithm
Dedicated efforts to investigate assay and/or
biomarker properties

Reproducibility, prevalence, prognostic value
Strategic considerations

How does the proposed POC trial fit into the clinical
development plan (CDP)?



The molecule CDP timeline and other ongoing/planned POC
trials




The TPP with the Dx component (clinical and commercial
considerations)
Rank speed, PTS and cost according to their relative importance
taking into account company portfolio and competitive landscape
The same Dx hypothesis may be addressed
Need to be informed with regards to the Dx by the proposed POC
trial
The follow up molecules which need to be informed by this
trial
In Stratified scenario, need to define and enable joint Rx/Dx
GO decision prior to Phase III
Operational Considerations


Population Dx prevalence
Tissue testing






Turn-around time
Assay-failure proportion
Enrollment rate
Number of qualified sites and the
ramp-up curve
Evaluation of sensitivity to the
assumptions
Protocol nuances…
Statistical Considerations
The design of the POC trial needs to enable a decision on the
population and co-primary endpoints in Phase III, i.e. all the Dx
subsets of primary interest need to be sufficiently populated




Single Arm vs. Randomized Controlled trial: in PHC setting the value of
randomized trial is higher than in All Comers development
Propose a decision algorithm based on operational characteristics and
evaluate its robustness to the departures from the assumptions under a variety
of underlying treatment/biomarker scenarios
Markers with pre-defined cutoffs: the proposed sample size in each biomarker
sub-group of primary interest should be roughly similar to the size of the All
Comers trial without the Dx
Significantly larger sample sizes may be required for a continuous biomarker,
possibly on the order of 100’s or 1000’s of events. As of today, the question of
identifying an appropriate cutoff for a truly continuous biomarker remains an
open problem from all, statistical, regulatory and clinical perspectives.
Selected Phase II Trial?

Issues to consider:


Strength of available scientific evidence (what is the
PHC assessment?)
Timelines (selected vs. stratified)



Regulatory considerations



Operational considerations
Number of sites
Would we have to propose Stratified Phase III?
What do we not learn by restricting POC to selected
population and when may it be justifiable?
Safety and Ethical concerns
Adaptive Phase II Design considerations


One-arm response driven trials: modification of Simon’s
two stage design to include evaluation of Dx subsets
Randomized time-to-event trials: dynamic patient
allocation ratio adjustment based on early readout






In theory allows for arbitrary number of Dx subgroups and active
treatments
Generally involves a Bayesian approach with continuous updating
Relies on the relevance of the early end point to the clinical
outcome of interest
Increased operational complexity
In practice, sample size requirements or operational complexity
can make complex adaptive designs intractable for the initial proof
of concept evaluation.
Understanding Dx sub-groups is key
Phase III Decision Criteria Based on Phase II Result
Phase II Results
HR is large in AC
HR is small in AC
HR is small in Dx+
Selected
AC if HR in Dx+ and Dxis similar
Co-primary (Dx+, AC) if
HR in Dx+ << HR in DxSelected if HR is large
in Dx-
HR is large in Dx+
Stop
AC
Phase III Considerations

Study Objective
 Establish
risk/benefit
 Clinical Validation of Dx

Implementation issues
 Analytically
validate Dx assay before applying it to
specimens in pivotal trials
 Accruing / prospective stratification based on Dx assay

Analysis approaches
 Test two hypothesis,
 All comers
 Dx positive subgroup
 Appropriately control for type I error
 Clearly define your decision tree –
“freebies”
there are no
End of Phase III Decision Criteria
Phase III
outcome
Not statistically
significant in all Comers
Statistically significant in all
Comers
All Comers claim if no diff.
b/w Dx- & Dx+ groups
Statistically
significant in Dx+
group
SELECTION CLAIM
Greater benefit claim if
clinically meaningful diff. b/w
Dx- & Dx+
Selection Claim if no
improvement in Dx- group
“Rescue” or “Retrofit” Diagnostics
J. Woodcock, 2010
Old Drugs – New Tests




Biomarker not known at the time of study
initiation
New scientific advancements/new technologies
Data not analyzed with that biomarker as part of
the hypothesis
Biomarker discovery – generation of new
hypotheses
 Inform
future molecules
Retrofit Dx/Exploratory Analysis
Prospective/Retrospective Study

Completed or post-interim-analysis trial
 Clinical
outcomes data unblinded and analyzed
without the biomarker data
 Overall treatment effect on the primary outcome in the
intent to treat population (everyone randomized)

Statistically persuasive result, usually p value < 0.025 onesided
 Patient
samples collected prior to treatment initiation
 Diagnostic hypothesis/analysis plan - prospectively
specified
 Analysis is retrospective

If no statistical significance on primary endpoint everything further is exploratory
R. T. O’Neill; FDA; Dec 16, 2008
Components of good biomarker analysis plan


Role of randomization - fairness of comparison
Marker availability – impact of convenience samples


Marker performance



How many hypothesis
How many outcomes
Model selection


Marker performance and prevalence may explain study to study
heterogeneity
Statistical control of false positive conclusions –


Bias due to missing data
Over-fitting can lead to bias
Validation methods

Data to generate the hypothesis vs. data to confirm the
hypothesis
Cutoff Selection for Continuous Biomarkers



Sparse literature and examples; no FDA guidance
Ideal approach: biologically or clinically meaningful cutoff;
multi-modal distribution due to underlying biology.
Common approaches (data driven): good for exploration,
but not good enough for implementation in subsequent
confirmatory trials




Percentiles: use medians, quartiles, etc.
Optimization: e.g. find cutoffs that maximize
– treatment effect differences in marker+/-;
– marker+ subset that meets a pre-specified treatment effect.
multiplicity issues
Explore effect vs. cutoff profiles
Developing a Complex Predictor

Uncharted area for predictive application:



Multi-step processes involving many decisions & tuning:





No HA-approved examples of complex predictor-based companion
diagnostics predictive of a specific drug
Only related examples: tests for prognosis of breast tumor
recurrence: Oncotype DX (21-gene qPCR), Mamma Print (70gene array)
Feature selection, model building, performance evaluation
Choice of options & algorithms, especially the performance metric,
should be guided by the clinical objective and the nature of dataset
Major concern: high variance & over fitting due to large p
(features), small n (sample size)
Validation strategy important
Reality check: Set realistic expectation on the feasibility
given the sample size, data quality and potential impact
Summary
“Biomarker qualification is a promising new route for
establishing novel diagnostics in drug development.
The controversy over clinical utility of diagnostics in
drug therapy is a reflection of the underlying
progress in understanding the basis of variability in
human responses to interventions; in this regard, it
is good news.
Most scientific progress is ushered in by disputes and
disagreements; hopefully, these will not cause us to
lose sight of the promise of safer or more effective
drugs in the near future.”
Janet Woodcock; Assessing the Clinical Utility of Diagnostics Used in Drug Therapy.
Clinical Pharmacology &Therapeutics; Volume 88 Number 6; December 2010
Acknowledgements





Cheryl Jones
Jane Fridlyand
Ru-Fang Yeh
Mark Lackner
Biostatistics Predictive Biomarker Guidance
2010 Team
Thank You!