Transcript What Is A Correlative Study? - Canadian Cancer Trials Group

Correlative Studies in Clinical Trials
Janet E. Dancey, MD
Objectives
• To define biomarker and correlative study
• To describe the types of biomarkers
• To understand the issues in designing
appropriate biomarker studies
• To understand the roles of biomarkers in
phase I, II and III studies
• To understand different biomarker trial
designs
What is a Biomarker?
• Biomarker
– 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 National Institutes of Health 2001
• Assay
– A method for determining the presence or quantity of a component
• Test
– A procedure that makes use of an assay for a particular purpose
Good biomarkers ≠ Good Assays ≠ Tests
What is a Correlative Study?
• A type of study that tests for a relationship
between a condition and a potential causal
factor of the condition
• Addresses the question:
– What is the correlation between the X and Y?
•
•
•
•
Cancer and obesity
QoL score and treatment
Response and Performance status
Prognosis and KRAS
plan2005.cancer.gov/glossary.html
Types of Clinical Biomarkers
Diagnosis
•Confirmation
•Staging
•Subtyping
Prediagnosis
Pretreatment
•Risk
•Screening
•Early
detection
•Prognostic
•Predictive
Intratreatment
Posttreatment
•Early
•Early
response
endpoint
or futility
•Recurrence/
•Toxicity
progression
monitoring monitoring
5
Biomarkers and Correlative Studies
• Biomarker is what is measured
• Correlative study is what is correlations are made
between the biomarker and something else
• In recent years many biospecimen based biomarkers
have been incorporated into cancer therapeutic trials
• Principles of design, measurement, data collection
and analysis are the same.
– Eg hypothesis, trial design, measurement, collection,
analysis
Case Study – A Novel Targeted Agent
You have been following the preclinical and clinical
research.
It is your opinion that the development of a class of
anticancer drugs could be facilitated by
incorporating laboratory and imaging correlates
early into drug development.
The sponsor wants investigators to develop a trial to
rapidly translate exciting results seen in animal
models into the clinic and that incorporates
laboratory and imaging studies that would define the
optimal dose and predict activity of this novel agent.
What do you do?
Get the Right Team
•
•
•
•
•
•
Scientist,
Clinician,
Pathologist,
Radiologist,
Trialist,
Statistician,
• Who is important?
–
–
–
–
–
CRAs
Treating nurses
Booking team
Pathology technicians
Scanning technicians
Who is critical? The patients!
Have the Right Core Facilities
•
•
•
•
•
Clinical trial intervention
Imaging
Pathology
Specimen management
Data management
Have the Right Tools
•
•
•
•
•
•
SOPs and training,
Protocol and consent forms,
Case report forms,
Reagents, technologies for analysis
Q/C
Etc, etc, etc
Know the limitations of Team, Cores and Tools!
Considerations for Biomarkers
• What is the question? (is it a good one?)
• Biomarker(s)
• Assay/technology
• Specimen
• Study/Trial Design
• Study Execution – Trial and Biomarker (s)
• Study Outcome – Trial and Biomarker(s)
• Likely Impact
• Economically and practically feasible
Biomarker, Assay, Specimen
Biomarker is relevant to
the intervention outcome
Assay reliably measures
biomarker in the specimen
Specimen is reasonable for
measuring the biomarker
Integrating Biomarkers into Clinical
Research: Biomarker
• Why this/these biomarker(s)?
– What is the question/purpose?
• (Is there one?)
• (Is it an important one?)
– Determining potential “value-added”
• Rationale and supporting data
– Laboratory experimentation
– relevance to disease and/or therapy
– Clinical evaluation
– prevalence and significance in normal and cancer patients
Integrating Biomarkers into Clinical
Research: Samples
• Will the Samples Be Adequate?
– Consider:
•
•
•
•
•
Type, Number, Timing, handling, shipping, storage
Demands on patients/clinic/laboratory staff
Technical requirements of the proposed analysis
Likely impact/plan for handling missing/inadequate specimens
Costs and resources
Specimen Characteristics
• Format
• Serum, FFPE or fresh/frozen tissue, etc.
• Biopsy, surgical specimen, etc.
• Collection
• Preservation
• Storage conditions & time prior to assay
16
Integrating Biomarkers into Clinical
Research: Assay
• Why this technology?
• Analytical/Laboratory Validation (Laboratory performance)
– How well are you measuring the measurand?
•
•
•
•
•
Precision / Reproducibility
Method Comparison
LoB, LoD, LoQ
Linearity
Stability
– Clinical Laboratory Standards Institute (CLSI) http://www.nccls.org/
– Is it fit for proposed purpose i.e. on the proposed samples/patients?
• Clinical Validation (“Clinical Qualification”)
– Does the test have clinical utility?
– Does it have added value over standard tests?
• FDA guidance : “Statistical Guidance on Reporting Results from Studies
Evaluating Diagnostic Tests” issued in final form in March, 2007,
http://www.fda.gov/cdrh/osb/guidance/1620.html
Assay Methods
• Detailed protocol
• Reagents or kits
• Quantitation method (e.g., manual, image analysis)
• Scoring & reporting
• QC procedures & reproducibility assessments
• Marker evaluations blinded to patient characteristics
and clinical endpoints
• Number and training of persons executing and
reading tests
Intratumoral heterogeneity of carbonic
anhydrase IX (CAIX)
Effect of distributional heterogeneity on the analysis of
tumor hypoxia based on carbonic anhydrase IX
VV Iakovlev, M Pintilie, A Morrison, et al
Laboratory Investigation (2007) 87, 1206–1217
a) Immunoperoxidase staining for CAIX in a
single tissue section. Analysis of the entire
section gave a value of 10.8% CAIX labeling.
The circles limit the analysis to 0.6 mm
simulated tissue microarray (TMA) cores, and
show a wide range in CAIX (for publication
purpose only, the image was digitally enhanced
to better visualize CAIX areas).
pSer473-Akt antibody in human gastroesophageal tumors
and HT-29 colon cancer xenografts measured by
immunohistochemical staining.
Baker, A. F. et al. Clin Cancer Res 2005;11:4338-4340
A, patient tumor samples. 1 and 2 are two surgically resected specimens and 3 and 4 are two biopsy specimens.
B, HT-29 human tumor xenografts excised from scid mice and kept at room temperature for the times shown. Each section
also includes in the upper right-hand quadrant an on-slide control of HT-29 colon cancer cells stained for pSer473-Akt.
Copyright ©2005 American Association for Cancer Research
Integrating Biomarkers into Clinical
Research: Samples
• Will the samples be adequate?
– Consider:
•
•
•
•
•
Type, Number, Timing, handling, shipping, storage
Demands on patients/clinic/laboratory staff
Technical requirements of the proposed analysis
Likely impact/plan for handling missing/inadequate specimens
Costs and resources
• Can the question still be answered?
Considerations
• Flow of patients through study
• Reasons for patient/specimen dropout
• Numbers and events
• Number of test outcomes at each analysis
stage and in each subgroup & missing values
Biomarkers in Clinical Research: Uses
• Phase I – proof of target inhibition after reaching biologically
active dose/concentration
• Phase II – predictive marker assessment after identifying
promising level of activity; studies should be larger and
randomized
• Phase III – prospective testing of biomarker and treatment
• Phase IV – prospective testing of biomarker and treatment
Markers and Endpoints for Trials of
Molecularly Targeted Agents
Changes in protein
function/phosphorylation
Biochemical Modulation
of Target Pathway
Protein
phosphorylation/function
Expression array
Markers of
Biological/Cellular
Response
Proliferation, Cell cycle phase,
Apoptosis, Angiogenesis,
Metabolism
Clinical Response
Objective Tumor Response
Prolonged Stable Disease
Clinical Benefit
Survival, quality of life
Phase 3
Biochemical Modulation
of Target
Phase 2
Blood and tissue levels
Phase 1
Pharmacokinetics
Phase 1 Trials: Considerations
• Primary goal: To identify an appropriate dose/schedule for
further evaluation
Small
patient
• Design principles:
numbers
– Maximize safety
– Minimize patients treated at biologically inactive doses
– Optimize efficiency
• Study population:
– Patients for whom no standard therapy
Heterogenous
Refractory
Tumours
Expect target modulation but not anti-tumour activity
Probability of Success in Assessing
Pharmacodynamic Effect in Biopsies
• It is certainly not 100%, and may be much lower
• Analogy to chemical synthesis: yield < 100% at each step
• Recognize potential problems and repair ASAP to maximize
probability of success.
Acquired on time?
Handled
Sample acquired?
Time documented?
properly?
Target
correct?
Assay
works?
Stored
properly?
Bad
karma?
P = 0.92 = 0.81, 0.93 = 0.72, 0.95 = 0.53 …
Shipped properly?
Biomarkers in Phase 1 Trials: Utility
• Proof of mechanism (drug hits proposed target)
• When toxicity may be insufficient to determine active
dose/schedule
– Unlikely to occur at dose/exposure that affects the target
– Due to off target effects and effects on target are uncertain
– To target a specific degree of target inhibition to avoid significant
toxicity
• When pharmacokinetics may be insufficient to determine
active dose/schedule
– Assay lacking
– Pharmacokinetics in plasma does not match effect in tissues
Phase 1 Trial Assessment of Target Effects Requirements
• Agent with acceptable preclinical activity, toxicology,
pharmacology
• Known association of target effect and tumour activity
• Well-characterized assay
•
•
•
•
•
•
% change in target or target level associated with efficacy
Concentration/exposure required for target effect
Time course for effect on target, duration, recovery
Threshold of detection and CV of target measurements
Target effect on tumour vs other tissues (eg PBMC, Skin, Buccal)
Collection, processing, shipping, storage effects known/optimized
• Usual target values and variability in human tissue known
• Patients’ tumours have relevant target
• Commitment from investigators/patients
• mandatory requirement (just like PK)
Biomarker Studies in Phase 1 Trials
• MGMT activity after O6-benzylguanine
• Friedman H et al J Clin Oncol 16:3570-5, 1998; Spiro et al. Cancer Res
59:2402-10 1999; Dolan et al Clin Cancer Res 8:2519-23, 2002
• 20S proteosome inhibition after bortezomib
• Lightcap E et al. Clin Chem 46:673-683, 2000; Adams J, Oncologist 1: 9-16,
2002;
• DCE-MRI after PTK787/valatanib
• Galbraith S et al NMR in Biomed 15:132-142, 2002; Morgan, B. et al. J Clin
Oncol; 21:3955-3964 2003;
• S6K inhibition after everolimus
• Tanaka C et al J Clin Oncol 26:1596-1602, 2008
• PARP Inhibition after ABT-888
• Kinders RJ, et al. Clin Cancer Res. 2008 Nov 1;14(21):6877-85
Biomarkers in Phase II/III
Prognostic versus Predictive Markers
• Prognostic markers provide information about the
patient’s cancer outcome
– e.g., cytogenetics in acute leukaemia; performance status
in lung cancer
• Predictive markers provide information regarding the
probability of benefit or risk from a specific therapy
• Markers may correlate with prognosis and treatment
outcome
– ER, HER2 in breast carcinoma
– KRAS in colorectal carcinoma
– MGMT in glioblastoma
Prognostic Marker
Measurement associated with clinical outcome in absence
of therapy or with application of standard therapy that all
patients are likely to receive.


Examples: Oncotype DX, uPA/PAI-1 by ELISA in breast cancer
Correlation with outcome not necessarily sufficient to impact clinical
decisions, may suggest targets for therapy
Good prognosis don’t need adjuv. Rx
Useful prognostic information?
Hazard ratio = .56
Hazard ratio = .18
31
Prognostic Marker Study Design
Marker +
All patients
Compare clinical
outcomes
Marker assay
Marker -
• Standard statistical methods such as log rank test or Cox PH
regression assume that study subjects constitute a random sample.
• If sampling is stratified or based on outcome, prognostic effect
estimates can be biased.
MGMT is Prognostic
•
KM Estimates of Overall Survival by MGMT Methylation Status
•
Hegi et al NEJM 352:997-1003 2005
MGMT is Prognostic
•
Hegi et al NEJM 352:997-1003 2005
Predictive Marker
Measurement associated with response or lack of response to a
particular therapy.


Examples: ER/PgR for endocrine Rx benefit in breast cancer
Statistical wisdom: Test for treatment by marker interaction
Qualitative interaction
• New drug better for M+ (h.r. = 0.44)
• Control drug better for M (h.r. = 1.31)
• Interaction = 0.44/1.31 = 0.33
Quantitative interaction
• New drug better for M+ (h.r. = 0.44)
• New drug better for M (h.r. = 0.76)
• Interaction = 0.44/0.76 = 0.58
MGMT is Predictive
•
KM Estimates of Overall Survival by MGMT Methylation Status and Treatment
Hegi et al NEJM 352:997-1003 2005 Hegi et al NEJM 352:997-1003 2005
Trials Designs: Prospective and Retrospective
Evaluation of Biomarkers
Prospective Study
Marker +
Rx
Outcome
Rx
Outcome
Histology
Marker -
Retrospective Study
Marker +
Good Outcome
Histology
Marker -
Rx
Marker +
Poor Outcome
Marker -
Biomarkers to Select Patients:
Prospective Evaluation
• Advantage
– Fewest numbers of patients
– Study design guaranteed to have sufficient power
to show treatment effect in marker group
• Disadvantage
– Must know marker to select patients
– Rapid turnaround to determine eligibility
Biomarkers to Select Patients:
Retrospective Evaluation
• Advantages
–
–
–
–
Maximize accrual
Need not know the right marker
Allows refinement of marker/assay while trial ongoing
Allows assessment in marker+/- groups
• Disadvantages
– Risk of insufficient numbers within marker group(s)
• Prevalence of different marker defined subgroups
– Collection of samples compromised
• Incomplete submission, suboptimal handling
– Results may not be generalizable
• Bias sampling
Retrospective Analysis with Incomplete
Specimen Collection
• RTOG study 86 -10. A randomized study of RT versus RT + TAB in
prostate cancer patients.
• To assess its prognostic value, p53 expression was assessed on
129 of 456 patients entered.
• Patients with p53 assessed had similar distributions of treatment
and prognostic factors as those who did not have p53 assessed.
• Statistically significant associations were found between the
presence of abnormal p53 protein expression and poor outcome.
Data from Grignon et al. J Natl Cancer Inst. 1997;89:158–165.
Pajak et al. Arch Pathol Lab Med. 2000;124:1011–1015
Retrospective Analysis with Incomplete
Specimen Collection
• The survival of those patients who had a p53 determined on their tumour
was statistically worse than those without a p53 determination (P = .03)
regardless of the actual p53 assay result.
Data from Grignon et al. J Natl Cancer Inst. 1997;89:158–165.
Figure from Pajak et al. Arch Pathol Lab Med. 2000;124:1011–1015
Biomarkers for Phase 2/3: Predictive markers
• Goal: identification of patients likely to benefit
(or elimination of those least likely to benefit)
• Considerations:
– Drug activity
– Treatment effect across patient subsets.
– Prevalence of the subset(s) of patients with “sensitive”
disease or at risk for toxicity.
– Assay performance i.e failure rate,
sensitivity/specificity/predictive value.
– Samples requirements
– Trial design to distinguish treatment and prognostic effects
Caveats for Assessing Treatment Effects in
Biomarker Defined Groups
• Biomarker may define a subgroup with a different prognosis
from historical outcome data from trials done in an
unselected group
• E.g. ER+ is both prognostic and predictive
– If the outcome with standard treatment is not well defined and/or the
outcome of interest is PFS/OS consider a randomized controlled phase
2 design
• If a trial is designed to assess treatment effects in Marker+
and Marker- groups
– False positives will dilute effect in marker+ group
– False negatives will dilute the apparent differences in treatment effect
between marker defined groups.
– Specimen loss or assay failure will increase the sample size
– Trial may be 2-4x size of a conventional study
Phase 2 or 3 Trial – Histologically Defined and
Biomarker Defined Patient Populations
Initial
Selection
Strata
Agent
Marker +
Histology
Stage
Randomize
Target
Tested
Marker -
Control
Agent
Outcome
Phase 3:
Survival
(Phase 2:
ORR, TTE)
Control
•
Trial is designed to assess treatment effects in Marker+ and Marker- groups
•
Marker assessment
•
If negative within marker groups,
– NB: need not be prospective stratification
– Assay failure increases number of patients screened
– False results will dilute differences in effects between marker groups
– Analyze between treatment groups
– Analyze for new markers, cut-points etc.
Predictive Marker Study Design
Marker-guided Vs. Randomized Design
Randomize To Use Of Marker Versus No Marker Evaluation
Marker Determined
Treatment
All Patients
M+
New Drug
M−
Control
New Drug
Randomize Treatment
Control
• Provides direct measure of patient willingness to follow marker-assigned therapy
• Marker guided treatment may be attractive to patients or clinicians
• Inefficient compared to completely randomized or randomized block design
Predictive Marker Study Design
• The randomized designs offer the greatest flexibility
to examine multiple markers.
• Require no a priori assumptions, and are more
efficient than marker-guided designs.
• False positive, false negatives will make treatment
outcomes in marker defined groups appear alike.
• Larger studies
Sample Size Considerations for Predictive
Marker Studies
• Testing for an interaction between marker status and
treatment
• Power depends on
– Number of EVENTS
– AND their distribution into marker-by-treatment categories
– Magnitude of effect (e.g., ratio of hazard ratios or difference of
survival differences)
– Significance level (e.g., usually 0.05)
• Test of interaction typically requires 2-4 times as many events
as test for treatment main effect
Power Problems
• Embedding prognostic and predictive questions in
“large” treatment trials
– Events sufficient for answering treatment question, may be
insufficient for prognostic or predictive questions in
marker defined subgroups
– Specimen retrieval and assay failures exacerbate sample
size problem
• If initial trial does not provide definitive answer, may
not be able to prospectively test marker question
Overcoming Sample Size Limitations
• Combine over multiple small studies
– Patients and assays comparable?
– Identify relevant studies?
• Publication bias
• Description limited
• REMARK reporting guidelines (McShane et al, 2005: BJC, EJC, JCO, JNCI,
NCPO) might help facilitate pooled analyses
• Some large prospective marker trials will be needed
– MINDACT
– TAILORx
– N0723
• Reduce noise in marker measurements to lessen attenuation
of marker effects
– Assay improvements
Statistical Analysis Methods
• Methods for calculating or comparing measures
of diagnostic accuracy
–
–
–
–
Sensitivity/specificity
Likelihood ratios
Diagnostic odds ratio
Area under ROC curve
• Methods used to quantify uncertainty in accuracy
measures (e.g., 95%)
• Methods for calculating test reproducibility, if any
50
Statistical analysis methods
•
•
•
•
Model building & assumptions
Variable selection
Missing data handling
Coding of marker values in analyses (e.g.,
continuous vs. categorized)
• Internal or external validation
51
Barriers to successful inclusion of
biomarker studies
• Often: add-on studies that aren’t necessary for achieving the
(primary) objectives of the clinical trial
– Clinical trials are hard to do well
– If something is not a primary or secondary objective it is even harder to
do it well
• Clinical trials are not designed to answer the biomarker
question
• Biomarker studies may use assays that are not well established
– It is almost impossible to learn to do an assay while studying a novel
treatment
• “normal” range, relevant cut-points, reproducibility
– Too many variables
Biomarkers in Clinical Research:
Implications
•
•
•
•
Compelling BUT
Increased burden on patients/clinical staff
Significant coordination effort required
Significant increased cost over ‘traditional’ treatment trials
– Numbers of patients may be increased and cost/patient will increase
• Qualification, standardization and QA takes time, energy and
money
• Have a good question, team of experts, assays, specimens,
trial design and analysis
• Consider the feasibility, economics and likely impact
Think about how you will report the
study before you start
• STARD: STAndards for Reporting of
Diagnostic accuracy (Bossuyt et al.)
• Published (2003): Clin Chem, Ann Intern Med, Radiol, BMJ,
AJCP, Clin BioChem, Clin Chem Lab Med
• Re-published (2003-4): 8 journals
• REMARK: REporting recommendations for
tumor MARKer prognostic studies (McShane
et al.)
• Published (2005): BJC, EJC, JCO, JNCI, NCPO
• Re-published (2006): BCRT, Exp Oncol
Additional Designs and Issues
• Sargent D, et al. Clinical Trial Designs for
Predictive Marker Validation in Cancer
Treatment Trials. J Clin Oncol 23:2020-2027,
2005
• Simon R, Maitournam A. Evaluating the
efficiency of targeted designs for randomized
clinical trials. Clin Cancer Res; 10: 6759–6763,
2004