Transcript Study Design and Marker Validation

Efficient Designs for Phase II and
Phase III Trials
Jim Paul
CRUK Clinical Trials Unit Glasgow

Efficient
Maximising the probability of getting the right
answer as quickly as possible
 Finding effective treatments
 Dismissing ineffective
Very easy to get the wrong “answer”
 Miss effective treatments at phase II, because of poor study design
 Wasting resource on phase III studies of ineffective regimens

Getting it wrong:-
Comparison of Outcome in Phase II Studies and
Subsequent Randomized Control Studies Using
Identical Chemotherapeutic Regimens Zia et al JCO 2005
 Found 43 phase III studies fulfilling this criteria (19982003)
 Only 28% of subsequent phase III studies were
positive
 96% single arm phase II
 Mean size of phase II :- 52 patients
 Some evidence that bigger phase II studies
gave better predictions of phase III success
Efficient design for phase II trials

Phase II is the foundation on which we build –
needs to be sound
Reliably quantify/indicate the risk of success/failure
in a subsequent phase III study

Best way to do this is to randomise between the
experimental and control arms in the phase II setting

Randomisation in Phase II studies
•Van Glabbeke M, Steward W, Armand JP. Non-randomised phase II trials of drug
combinations: often meaningless, sometimes misleading. Are there alternative
strategies? EJC (2002)
•Recommended changes to oncology clinical trial design: Revolution or
evolution?, Ratain et al EJC, Jan 2008
“We strongly recommend that randomised comparative phase II trials
become a standard approach in oncology, especially for the development
of drug combinations”
•Optimising the design of phase II oncology trials: The importance of
randomisation. Mark Ratain and Dan Sargent EJC (2009)
“…fundamental need for randomisation in phase II oncology
trials….ideally with blinding and dose ranging.”

Need for randomisation: -
• Single arm studies prone to selection bias (clinician
selection/patient preference) – therefore comparisons
with historical benchmarks are intrinsically an unreliable
basis for decision making
• No real allowance for imprecision in the historical estimate
of response used as the basis for single arm phase II design
•Major problems in studies where the new drug is
combined with active standard treatment and the
additional benefit is likely to be “modest” and could
easily be lost in study bias or in the imprecision of the
estimated historical effect
Randomised Phase II Trial Design in
Ovarian Cancer

Role of randomisation
• What size of benefit can be expected from new agents in ovarian
cancer?
•Introduction of taxanes into first line (GOG111/OV10) – 25%
reduction in hazard for pfs
• Addition of taxane in relapse (ICON4) – 25% reduction in hazard
for pfs
• 25% reduction in hazard (HR=.75) corresponds to an
absolute increase in pfs of only approximately 10% at a given
time-point (e.g. 50% to 60% or 20% to 30%)
Randomised Phase II Trial Design

Screening design (Rubinstein J Clin Oncol 23:7199-7206 (2005))
• Use a permissive 1-sided significance level (10% or 20%) to
decide whether there is sufficient activity to go to phase III
• Set realistic bar for efficacy
Randomised Phase II Trial Design
Any way to reduce these numbers?
Early
stopping for futility in a randomised
phase II:Sin-Ho
Three
Jung Statist. Med. 2008; 27:568–583
outcome design:-
A Three Outcome Design for Randomised
Comparative phase II Clinical Trials. Hong and Wang.
Statistics in Medicine (2006)

Randomised Phase II Trial Design
 Three outcome design
Three
possible outcomes e.g.:-
Statistically significant at 10%:Clear indication from study data to proceed to phase III
Not statistically significant at 20%:Clear indication from study data NOT to proceed to phase III
Statistically significant at 20%, but not 10%:Uncertain outcome – decision based on other information
(changes in relevant biomarkers, clinical judgement)
Randomised Phase II Trial Design
 Three outcome design

Allows the sample size requirements for
screening designs to be lowered (20–30% less than
that of the corresponding two-outcome screening design)
The
price paid is the introduction of an area
of uncertainty
Avoids
the “discomfort” of a go/no-go
decision resting potentially on a single
response
A Better Estimate of Success at Phase III
1.2
1.2
1.1
1.1
1.0
0.9
0.8
0.7
0.6
0.5
0.4
Weighted average probability of
estimates of HR from phase II
II
Estimated
HazardHR
ratiofrom
(fromphase
phaseII)
phase II
HR from
Estimated Hazard
ratio
(Christy Chuang-Stein Sample size and the probability of a successful trial Pharmaceut. Statist. 2006; 5: 305–309)
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Probability of success in phase III
success over range of plausible
Moving from Phase II to Phase III
Use
these probabilities to prioritise your phase III effort
For
fixed phase III sample size estimated probability of
success depends on –
Estimated
phase II hazard ratio (bigger difference ->
bigger probability of success)
Precision
of estimate (higher precision of estimate ->
higher probability of success)
Seamless Phase II/III Trial Design

Reliable randomised Phase II designs tend to
be relatively large
- To make efficient use of the patients involved in
phase II - desirable to incorporate their information
into subsequent phase III – seamless phase II/III
- To make efficient use of control patients involved
- incorporate multiple experimental arms – not
all of which will go on to phase III stage
Seamless Phase II/III Trial Design
- Multi-Arm/Multi-Stage (MAMS) Designs –
Parmar et a JNCI 2008

GOG182 -ICON 5 prime example (4 experimental
arms/ 2 stages)
- Used intermediate phase II outcome (pfs) to decide
whether to continue experimental arms to phase III
Phase II bit
Phase III bit
Seamless Phase II/III Trial Design
- Multi-Arm/Multi-Stage (MAMS) Design –

MRC STAMPEDE study in prostate cancer
(1 control, 5 experimental arms – randomisation ratio (2:1:1:1:1:1)
Phase II bit
Phase III bit
Seamless Phase II/III Trial Design
- Multi-Arm/Multi-Stage (MAMS) Design –

Advantages
-Rapid and reliable way to assess a large number of promising new
combinations simultaneously – ICON5 (3.5 years to assess 4 new
treatments)

Disadvantages
-Time consuming and complex to set up
-Potentially very large studies requiring large inter-group
collaboration
Seamless Phase II/III Trial Design
- Sequential Bayesian Phase II/III Inoue, LYT; Thall, PF; Berry, DA BIOMETRICS (2002 )
- Randomised design
- Two arms only – experimental and contol
- Decision to switch from phase II to phase III can take place at several
times during a defined phase II period if there is enough favourable data
- Uses the impact of experimental treatment on the phase II end-point (6
months pfs) to predict impact of experimental treatment on phase III endpoint (survival)
6 months
pfs
Experimental
treatment
Predicted
Survival
Observed
Seamless Phase II/III Trial Design
- Sequential Bayesian Phase II/III Inoue, LYT; Thall, PF; Berry, DA BIOMETRICS (2002 )
- Sample size may be much smaller than “conventional”
sequential or multi-stage designs
-Smaller sample size due to use of phase
II end-point to predict effect of new
treatment on actual outcome (OS)
- Not tried in practice
- Issues of using phase II end-point
information in reaching definitive phase
III conclusions