Statistical Issues in Incorporating and Testing Biomarkers in Phase III Clinical Trials

1
Statistical Issues in Incorporating and Testing
Biomarkers in Phase III Clinical Trials

• FDA/Industry Workshop September 29, 2006
• Daniel Sargent, PhD
• Sumithra Mandrekar, PhD
• Division of Biostatistics, Mayo Clinic
• L Collette, EORTC

2
What are we testing

• A (novel) therapeutic whose efficacy is predicted
  by a marker?
• A marker proposed to predict the efficacy of an
  (existing) therapeutic?

3
Preliminary information
Methods feasibility of measurement of the
marker in the target population
Specificity to the cancer of interest Cut point
for classification Prevalence of marker
expression in the target population
Properties as a prognostic marker (in absence of
treatment or With non targeted std agent)
Expected marker predictive effect Endpoint of
interest
4
Phase II/III Trials

• Patient Selection for targeted therapies
• Test the recommended dose on patients who are
  most likely to respond based on their molecular
  expression levels
• May result in a large savings of patients (Simon
  Maitournam, CCR 2004)

5
Trials in targeted populations

• Gains in efficiency depend on marker prevalence
  and relative efficacy in marker and marker -
  patients

(Simon Maitournam, CCR 2004)
6
Phase II/III Trials

• Designs for Targeted Trials
• May use standard approaches.
• Possible Issues
• Could lead to negative trials when the agent
  could have possible clinical benefit, since
  precise mechanism of action is unknown
• Could miss efficacy in other patients
• Inability to test association of the biologic
  endpoints with clinical outcomes in a Phase II
  setting

7
Targeted Trials

• Additional considerations
• Not always obvious as to who is likely to respond
  - often identified only after testing on all
  patients
• Slower accrual, and need to screen all patients
  anyway
• Need real time method for assessing patients who
  are / are not likely to respond

8
Example C-225 in colon cancer

• Early trials mandated EGRF expression
• (Saltz, JCO 2004, Cunningham, NEJM 2004)
• Response rate did not correlate with expression
  level (Cunningham, NEJM 2004)
• Faint RR 21
• Weak or Moderate RR 25
• Strong RR 23
• Case series demonstrates no correlation between
  expression and response
• (Chung, JCO 2005)
• Currently indicated only in patients with EGFR
  expressing tumors, but most current studies do
  not require EGFR expression

9
Design of Tumor Marker Studies

• Current staging and risk-stratification methods
  incompletely predict prognosis or treatment
  efficacy
• New therapeutic options emerging
• Optimizing and individualizing therapy is
  becoming increasingly desirable
• Very few potential biological markers are
  developed to the point of allowing reliable use
  in clinical practice

10
Prognostic Marker

• Single trait or signature of traits that
  separates different populations with respect to
  the risk of an outcome of interest in absence of
  treatment or despite non targeted standard
  treatment

Prognostic
No treatment or Standard, non targeted treatment
Marker Marker
11
Predictive Marker

• Single trait or signature of traits that
  separates different populations with respect to
  the outcome of interest in response to a
  particular (targeted) treatment

Predictive
No treatment or Standard
Targeted Treatment
Marker Marker
12
Validation
Prognostic marker
Series of patients with standard treatment
Designs?
Predictive Markers
Randomized Clinical Trials
13
Randomized Trials

• Trials to assess clinical usefulness of
  predictive markers i.e., does use of the marker
  result in a clinical benefit of a therapy
• Upfront stratification for the marker status
  before randomization
• Randomize and use a marker-based strategy to
  compare outcome between marker-based arm with
  non-marker based arm
• Sargent et al, JCO 2005

14
Design I upfront Stratification
Treatment A
Marker Level (-)
Randomize
Treatment B
Register
Test Marker
Treatment A
Marker Level ()
Randomize
Power trial separately within marker groups
Treatment B
Sargent et al., JCO 2005
15
Approach I Separate Tests
Treatment A (Std)
Statistical test With power
R
Marker -
Treatment B (New)
Test marker
Treatment A (Std)
Statistical test With power
R
Marker
Treatment B (New)
16
Approach II Interaction
Treatment A (Std)
Statistical test With power
R
Marker -
Treatment B (New)
Test marker
Treatment A (Std)
R
Marker
Treatment B (New)
17
Marker-based strategy design
M -
Marker- Based Strategy
Statistical Test with Power
Treatment A
M
Treatment B
Test marker
R
Non Marker Based Strategy
Treatment A
18
Design II Marker Based Strategy
Marker Level (-)
Treatment A
Marker Based Strategy
Marker Level ()
Treatment B
Register
Randomize
Test Marker
Treatment A
Non Marker Based Strategy
Randomize
Treatment B
Sargent et al., JCO 2005
19
Sample Size Interaction Design
HR 1.25
844
HR 0.69
1220
HR 0.86
1705
2223
2756
20
Sample size Strategy Design
TS -
Marker- Based Strategy
IFL (20 mo)
16.5 mo
TS
IO (14 mo)
HR 0.91
4629
R
Non Marker Based Strategy
IFL (15 mo)
R
15 mo
IO (15 mo)
21
Discussion

• Sample Size
• Typically large, especially if the marker effect
  size is modest
• Depends on many factors such as
• The marker prevalence in the target population
• The baseline risk in the unselected population
  receiving standard treatment
• The expected treatment difference in all marker
  groups

22
Conclusions

• The Marker Based Strategy design is preferable
  whenever more than one treatment are involved or
  when the treatment choice is based on a panel of
  markers
• That design generally requires more patients than
  the Interaction design
• The marker is also prognostic
• Dilution (marker patients receive the targeted
  therapy in the randomized non marker based group)

23
Conclusions

• In the case of a single marker and two
  treatments, Interaction Design preferable
• Separate Tests versus Interaction ?
• Depends on strength of evidence needed for the
  marker effect and sample size
• Whenever the interaction HR is larger than any of
  the treatment HRs (generally qualitative
  interaction) the interaction approach demands
  less patients
• A partial Separate Tests approach may be useful
  whenever no treatment difference is expected in
  one of the marker groups