Transcript Slide

The Era of Personalised Healthcare: Designing
Clinical Studies with Biomarkers
Ugochi Emeribe, PhD
Outline
 Background
 concepts, and retrospective analyses of Gefitinib (Iressa) trial
 Classic Designs using Biomarkers
 Classifier, Prognostic and Predictive Biomarkers
 Sample Size Calculations
 Validating Biomarkers
 surrogate Biomarkers
 Conclusions
2
SPECIAL REPORT-Big Pharma's global guinea pigs
Chicago Tribune
Monday, May 9, 2011 8:13 AM CDT
As drug treatments become more targeted, scientists are
unraveling how small genetic variations may make one
medicine suitable for a particular group of people.
AstraZeneca's lung cancer drug Iressa, for example, failed to
help Western patients overall in tests but proved much more
effective in Asians -- a discovery that has shed valuable new
light on ways of tackling the disease worldwide.
“We are starting to understand ethnic differences through the
responses seen in global trials. By cherishing our genetic
diversity we can identify biomarkers like the one for Iressa.
That is really exciting.” says Dr. David Kerr, president of the European
Society for Medical Oncology.
What is Personalized Health Care?
Matching individual patient characteristics with drugs that
produce better outcomes for that patient
Perfect Medicine
• Effective in all patients!
• The same dose for every
patient!
• No side effects!
Real Medicines
• Effective only in some
patients
• Dose varies for
different patients
• Some patients may
develop adverse
events
 Herceptin is seen as the poster child for PHC
 But a classic example of a drug development that did not
start with PHC in mind is Gefitinib
Retrospective Analyses of Gefitinib Trials
 EGFR Mutation- first thought to be predictive was actually
prognostic
 EGFR Gene Amplification- first thought to be prognostic
was actually predictive
EGFR mutation status
1.0
Gefitinib 250/500mg and EGFR M+
(n = 14)
Gefitinib 250/500mg and EGFR M–
(n = 65)
Proportion event free
0.8
0.6
I
0.4
II
I
II
I
0.2
I
0.0
0
►
►
2
4
6
II
III I
II
IIIII
0.6
II
I
I
I II
I
0.4
II I I
III
II
0.2
I
I
Gefitinib 250/500mg and EGFR M+
(n = 14)
Gefitinib 250/500mg and EGFR M–
(n = 65)
I
0.8
Proportion event free
1.0
I
I
8
10
Progression free survival time (months)
12
0.0
0
2
4
6
8
10
Survival time (months)
Median TTP for EGFR mutation +ve cases was longer (116 days, range 25-171),
than that for mutation -ve cases (57 days, range 28-170)
There was no impact on OS
12
EGFR gene amplification
FISH: technique for measuring increased EGFR gene copy
Proportion
surviving
FISH +
FISH -
N=114, E=68
Cox HR=0.61 (0.36, 1.04)
p=0.07
N=256, E=157
Cox HR=1.16 (0.81, 1.64)
p=0.42
1.0
1.0
Gefitinib
Placebo
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.0
0.0
0
2
4
Gefitinib
Placebo
6 8 10 12 14 16
0 2 4 6 8 10 12 14
Time (months)
Time (months)
Interaction test: p=0.04
16
FISH positive status and clinical characteristics
60
% of
FISH 50
positive
40
patients
30
20
10
No.
patients 0
with
evaluable
samples:
Adeno Other
156
214
Histology
Never
Ever
48
322
Smoking
Asian
origin
11
Other
359
Ethnicity
Female Male
117
253
Gender
Conclusions from Gefitinib trials
 FISH+ status is the biomarker which is
the strongest predictor of Gefitinib benefit
on OS
 Patients who are FISH- are unlikely to
benefit from Gefitinib therapy.
 Therefore, EGFR amplification is a
predictive marker for benefit with Gefitinib
therapy.
Definitions
 Clinical Endpoint (or Outcome) : A characteristic or variable
that reflects how a patient feels, functions, or how long a
patient survives.
 Biomarker (or Biological marker): A characteristic objectively
measured as an indicator of normal biologic or pathogenic
process, or pharmacologic responses to a therapeutic
intervention.
 measured once before treatment
 Types of Biomarkers
 Prognostic
 Predictive
10
Prognostic vs Predictive
 Prognostic markers indicate that clinical outcome is
independent of treatment.
 Stage of disease is a prognostic marker for survival outcome.
 Predictive biomarkers show treatment effect on the clinical
endpoint.
 High Her-2 gene copy number in advanced breast cancer is predictive
for the effect of Herceptin.
 Statistically, a predictive marker is a marker that interacts with
treatment “significantly.”
…and will adversely impact on power
…and will adversely impact on power
•…therefore it is obvious that a biomarker
targeted approach to drug development will
lead to smaller, more secure and more
successful developments
•losers will be dropped early and winners
taken forward, resulting in more successful
drug development…
How about this?
 100 treatment resistant patients are
offered a new drug
 70 respond and 30 do not.
 How do we interpret this experiment?
Which is the correct interpretation?
 A - Treatment works for 70% of
patients 100% of the time and for
30% of patients 0% of the time.
 Or…
 B - Treatment works in 100% of
patients 70% of the time.
Which is the correct interpretation?
 A - No within patient variability –patients
are deterministically responders or non
responders
 B - Within patient variability –drug has
some effect in all patients, but patients
vary in their response –sometimes they
respond, sometimes they don’t
What does this mean?
•In most situations, it is impossible to know if patients respond
deterministically
•To know for sure requires repeat administration of drug (and
control) in within-patient crossover trials
However, such trials are impossible in many settings,
especially oncology, so that there is little choice but to assume
interpretation A.
% surviving or progression-free
Suppose biomarker target identified in
patients treated with drug, show target
+vepatients do better than target –vepatients
Time
% surviving or progression-free
…suppose the same is true for patients
treated with control, target +vepatients do
better than target –vepatients
Time
This is an example of a prognostic
biomarker
•Patients with the biomarker do better than
those without it irrespective of the treatment
they receive
This biomarker is not predictive for the
effect of drug over control
•Using this biomarker as a basis for patient
selection is unlikely to result in a positive
outcome for drug
Predictive vs. Prognostic
Predictive
Prognostic
% surviving or progression-free
Biomarker +vepatients treated with drug
do better than biomarker +vepatients
treated with control
Time
•This is an example of a predictive
biomarker biomarker
•+vepatients do better when treated with
drug than when treated with control
•biomarker –vepatients do less well on
both drug and control
So, we need to stratify on receptor status
and then randomize to drug and no drug to
assess the true potential of a drug
Therefore, data from a properly designed Phase II trial
could be used to assess the true value of receptor status
How about designing late Phase trials?
Just an example
C
E
Effect
+ve (25%)
6 months
12 months
0.50
–ve (75%)
6 months
6 months
1.0
All patients
6 months
7.5 months
0.80
No. required
to screen
All patients
+ve (25%)
No. required
to enroll
1000
117
1median follow-up of 18 months assumed
468
To validate biomarkers……
Sensitivity
Pr(test +ve/true +ve)
Specificity
Pr(test –ve/true –ve)
Positive Predictive Value
Pr(true +ve/test +ve)
An imperfect test lessens the advantage
of a biomarker strategy
Sens, Spec
C
E
Effect
size
No. Required No. required
to enroll
to screen
100%, 100%
6
12
0.50
117
468
95%, 75%
6
9.4
0.64
260
613
75%, 95%
6
11
0.55
149
663
75%, 75%
6
9
0.68
317
845
Anyway, assume we have the
perfect test, what happens if there
is some modest effect in –ve pts?
Is a selected design still best?
Even a small effect in biomarker –vepts erodes the
advantage of a biomarker strategy
C
E
Effect
+ve (25%)
6 months
12 months
0.50
–ve (75%)
6 months
7.5 months
0.80
All patients
6 months
8.7 months
0.69
No. required
to screen
All patients
+ve (25%)
No. required
to enroll
384
117
Effect in –vepts = 1/3 effect in +vepts
468
Likely the relationship between treatment effect and biomarker
level is continuous, reflecting underlying biology
In a biomarker strategy we would
need to be very confident that
(i) we had a very good test
(ii) the biomarker ‘-ve’ population
achieved no or very little benefit
from treatment
In late phase development, testing across the
population offers some advantages
Power the trial for an “interaction”
 “Is there any statistical evidence that the treatment
effect in +vepts is different to the treatment effect in
-ve pts?”
 If “Yes”, valid to look at +ve and -ve groups separately.
 Possibility of labeling in (+) pts.
 If “No”, then there is no statistical rationale for
looking at +ve and –ve patients separately.
 Compare treatments in the overall population, irrespective of
biomarker status.
An example
 Treatment A is either better or worse than treatment
B (qualitative interaction)
 Treatment HR = 0.74
 Interaction HR = HR(+vepts) / HR(-vepts)
= 0.48 / 2.85 = 0.17
 If interaction effect size is better that treatment effect
size
 Than interaction is highly significant
Power of Interaction Test
 Interaction test has very low power
 So, validation of predictive biomarker is more complicated --due to limited power of the interaction test
 It is known that as inflation factor for total sample size
decreases, so does interaction effect size in relation to
overall treatment effect size.
 Therefore, inflation factor is required to increase the sample
size to ensure interaction test has the same power as the
original sample size calculated for overall treatment effect.
Design can provide confirmatory evidence in either all
patients or the subset of biomarker +ve patients
Alternatively, patient selection adaptive designs can
identify those patients most likely to benefit
Randomize
Interim
Continuance
The need for surrogate endpoints
 In many settings, the primary clinical endpoint
takes large, long term trials
 Breast cancer recurrence, cardiac events, osteoporotic fracture,
death from prostate cancer
 To reduce time and expense and to bring effective
medicines to patients quickly requires use of
surrogate endpoints
Statistical definition of Surrogacy
 “A response variable for which the test of
the null hypothesis of no relationship to
the treatment groups under comparison
is also a valid test of the corresponding
hypothesis based on the true endpoint.”
by Prentice, (1989)
The Problem with Prentice
Criteria based on Prentice’s definition are problematic
 Cannot prove the null
 The ‘%’ effect retained is not a true percentage, and
CI for the ‘%’ effect retained is usually very wide
 Cannot realistically expect 100% of a drug effect on
OS to be explained by a direct effect on the disease
itself.
Newer Approaches to Surrogacy
 That reliably predict drug effect on a later clinical
outcome (e.g. OS or PFS) given the effect of drug
on some earlier endpoint.
 Buyse and Molenbergs (2000, 2002) provided a
meta-analytic methodology for doing just this.
 Unlike Prentice, this approach does not require
proving the null nor the presence of a significant
treatment effect.
Strong evidence of surrogacy: Relation between tumor response
to first-line chemotherapy and survival in advanced colorectal
cancer: a meta-analysis
R2=0.97
Using methodology to quantitate uncertainty in
prediction Ovarian cancer
Conclusions
 There should be an assurance that selected test for
biomarker is correct. So, validation of biomarkers in early
in drug development is imperative.
 Treatment interaction effect has to be factored in sample
size calculation for late phase studies.
Backup Slides
FISH: technique for measuring increased EGFR gene copy
Fluorescent In Situ Hybridisation (FISH) is a technique for
measuring increased EGFR gene copy number
A piece of synthetic DNA labelled
with a fluorescent tag binds to the
EGFR gene. A probe to the gene
CEP7 labelled with a second probe
acts as a reference
The normal situati
‘balance disomy’
EGFR probe
Control probe
‘balanced polysom
‘gene amplificatio
Cappuzzo et al 2005
EGFR gene copy number (FISH) in ISEL trial
Pattern
Patients
n=370
(%)
Disomy
15.7
Low trisomy
24.1
High trisomy
2.2
Low polysomy
27.3
High polysomy
17.0
Gene amplification
13.8
Note: Categories in blue above represent those considered FISH+