poster - BioStat Solutions Inc.

Who to Treat: A Molecular Signature Approach for Subgroup Identification
Scott L. Marshall, Lin Li, Tobias Guennel
BioStat Solutions, Inc., 5280 Corporate Drive, Suite C200, Frederick, MD 21703
Delivering actionable results through the power of cutting edge analytics
ABSTRACT
METHODS
The promise of personalized medicine is becoming a reality in the
space of oncology with a noticeable shift in the past decade from
targeting the largest possible population to targeting cancer
subtypes with smaller patient populations or biomarker-defined
subgroups of a patient population with enhanced response.
Approximately 60% of US drug sales related to anti-cancer drugs are
targeted therapies and it is estimated that a similar percentage of
drugs in development have a biomarker component. As the
emphasis on biomarker strategies across therapeutic areas is on the
rise, it is critical that consideration is given to advancing analytics
for the discovery and commercialization of biomarkers.
โ€ข
A novel approach for treatment-specific subgroup identification has
the ability to aggregate data across assay platforms and estimate
patient specific molecular signatures, which then serve as a
surrogate โ€œbiomarkerโ€ for membership in the unobserved
underlying treatment-specific subgroup, e.g. biomarker(+)
subgroup. This flexible strategy also provides for incorporation of
both scientific and business factors, such as confining the search
space to a subgroup size that is commercially viable, ultimately
resulting in actionable information for use in empirically based
decision making.
DISCUSSION
Consider a working model assuming that the biomarker profile ๐‘ด can serve
as a surrogate for the subgroup ๐’ฎ
Executive Summary
โ„Ž๐‘– ๐‘ก = โ„Ž0 ๐‘ก exp ๐‘๐‘– ๐›ฝ + ๐‘๐‘– ๐‘ด๐‘‡๐‘– ๐œฝ + ๐‘ฟ๐‘‡๐‘– ๐œถ0 + ๐‘ด๐‘‡๐‘– ๐œถ
Biomarkers
Stage 1:
โ€ข
Composite score is defined as ๐›พ๐‘– = โˆ’๐‘ด๐‘‡๐‘– ๐œฝ for patient ๐‘–
โ€ข
A cutoff ๐œ for the composite scores is identified through a grid search using a
max-chi square approach, and the significance is provided via a parametric
bootstrap approach
โ€ข
Subgroup can then be defined as all patients above a cutoff ๐œ,
i.e. ๐’ฎ๐œ = ๐‘–: ๐›พ๐‘– โ‰ฅ ๐œ
โ€ข
P-value is provided for test of enhanced treatment effect for identified
subgroup, and type I error is protected
Stage 1: Estimation of
Composite Score
Business Challenges
โ€ข Heterogeneous patient
populations -> heterogeneous
treatment response
โ€ข Optimizing the drug
development process
โ€ข Differentiation strategy
โ€ข Treatment response driven by
complex biological interactions
โ€ข Clinically relevant treatment
effect in the biomarker(+)
group
โ€ข Need to integrate data from
various biological assay
platforms
โ€ข Marketable subgroup size
Stage 2: Estimation of
Subgroup Defining Threshold
Subgroup is identified!
0.6
1.7
2.0
2.7
2.5
2.2
0.3
โ€ข
โ€ข
โ€ข
2.1
2.6
0.4
1.2
1.1
0.5
Overall Response Rate = 25%
Competitor Response Rate = 30%
Subgroup Response Rate = 50%
Subgroup Size = 20%
A subgroup with a competitive advantage exists
but it is desired to capture a
larger market share -> relax the threshold
Transform Challenges into Solutions
Patients with MMMS scores
Subgroup is identified!
0.6
Stage 2:
Subgroup
Identification
Of particular note, and a significant contribution to the literature, is
โ€ข A subgroup is estimated
to include any subjects
with scores above a
given threshold
Goal: evaluate type I error and power to detect a subgroup
with enhanced treatment effect via the MMMS approach
Biomarker data was simulated for 25 biomarkers
TTE outcomes were generated spiking in a subgroup with
enhanced treatment effect in presence/absence of main effect
Subgroup is defined by four continuous markers M1,โ€ฆ, M4:
๐‘†๐‘™๐‘–๐‘› = {๐‘–: ๐‘€๐‘–,1 ๐‘™๐‘œ๐‘”2 + ๐‘€๐‘–,2 ๐‘™๐‘œ๐‘”2 + ๐‘€๐‘–,3 ๐‘™๐‘œ๐‘”2 + ๐‘€๐‘–,4 ๐‘™๐‘œ๐‘”3 > ๐œ}
Subgroup hazard ratios: 0.3, 0.4, 0.5, 0.7, 1.0
Subgroup size=30%
Sample sizes: 300, 500, 750, 1000
RESULTS
Sample Size
300
300
500
500
750
750
1000
1000
Biomarker
Main Effect
Absent
Present
Absent
Present
Absent
Present
Absent
Present
Type I Error (%)
5%
10%
20%
3.9
8.2
19.1
3.7
8.4
18.2
4.3
9.3
19.8
5.1
9.5
19.0
5.4
10.1
19.9
4.6
7.5
16.7
4.6
10.1
18.6
4.1
8.6
18.1
โ€ข Control of type I error at the nominal level, even in the presence
of prognostic factors;
โ€ข Moderate power (i.e. 60%) is achieved in discovery oriented
scenarios, such as a phase 2 clinical trial with the alpha level
relaxed at 0.20;
โ€ข Capability to aggregate information across biological assays
Using an estimated patient-specific molecular signature results in
generating actionable information that addresses the scientific and
business challenges associated with subgroup identification and
subsequent patient stratification.
Practical Applications and Business Utility
โ€ข
โ€ข
โ€ข
โ€ข
Prospective integration of biomarkers into the study design
Stratification of biomarker(+) / biomarker(-) subgroups
Enrichment designs
Preliminary understanding of potential market share and
differentiation strategy
Current Extensions and Future Work
โ€ข
โ€ข
โ€ข
โ€ข
General linear model framework
Generalized linear model framework (e.g. logistic regression)
Estimation of confidence interval for subgroup defining threshold
Estimation and correction of bias for treatment effect in
biomarker(+) subgroup
REFERENCES
0.8
1.8
Li L, Guennel T, Marshall SL, Cheung LWK. (2014)
A multi-marker molecular signature approach for treatment-specific
subgroup identification with survival outcomes.
[Accepted; The Pharmacogenomics Journal]
1.7
2.0
2.7
2.5
2.2
โ€ข
โ€ข
โ€ข
0.8
1.8
Composite
Scores
Simulation Study
Example Application
Patients with MMMS scores
Scientific Challenges
โ€ข A score incorporating
information of all
biomarkers is derived
for each subject
(1) Estimates a multi-marker molecular signature (MMMS) via small
scale feature selection and
(2) Identifies and directly tests for a biomarker driven subgroup with
an enhanced treatment effect.
where ๐‘๐‘– ๐‘ด๐‘– ๐œฝ contains interaction terms of the treatment ๐‘๐‘– with each of
the biomarker values
โ€ข
Challenges in Patient Stratification
โ€ข Continuous biomarkers
โ€ข Categorical biomarker
This analytical framework provides a novel combination of elastic
net and the maximal chi-square approach in a flexible two stage
approach that:
0.3
2.1
2.6
0.4
1.1
1.2
0.5
Overall Response Rate = 25%
Competitor Response Rate = 30%
Subgroup Response Rate = 40%
Subgroup Size = 60%