how much can we adapt? an eortc perspective saskia litière eortc - biostatistician
TRANSCRIPT
How much can we adapt?An EORTC perspective
Saskia Litière
EORTC - Biostatistician
I have no conflicts of interest
2
Adaptive designs• What?• Why?• The challenges• Examples
Currently part of EORTC portfolio Currently not (yet) part of EORTC portfolio
• Take home messages
Outline
3
“… a study that includes a prospectively planned opportunity for modification of one or more specified aspects of the study design and
hypotheses based on analysis of data (usually interim data) from subjects in the study. “
What is an adaptive design?
4
• They aim to make efficient use of patient and financial resources
• Allow for real-time learning during the course of a trial
• Relatively flexible: modifications possible in the course of trial which make the approach more robust to failure
• The drug development process is streamlined and optimized
Why use adaptive designs?
5
• To control the
operating characteristics
• To control the bias due to the adaptation Statistical Operational
• To guarantee that the results can
be interpreted and explained!
The challenges
6
• Early stopping for futility and/or efficacy
• Drop treatment arm(s) – also known as pick the winner designs
• Biomarker adaptive designs
• Sample size re-estimation• Adaptive randomization…
To name but a few …
Several possible approaches
7
Well-known
Less understood
Most of them come down to
8
Learn Confirm
One trial
Change H0? Change design parameters?
A few examples
9
EORTC 62012 in first line treatment of advanced, high grade STS
10
R
Doxorubicin
Doxorubicin + Ifosfamide In
terim
1: P
FS?
Group sequential design
Inte
rim 2
: OS?
Fina
l: O
S?
TRUSTS (EORTC 62091) in advanced or metastatic STS
11
R
Trabectedin 1.5 mg/m2 24-h
Doxorubicin 75 mg/m2
Doxo 75 mg/m2
T 3-h or 24-h
Sele
ct th
e be
stPF
S
Phase IIb3 x 40 pts
Phase III2 x 110 pts
Trabectedin 1.3 mg/m2 3-h
PFS?
Seamless phase II/III design
– Both steps are conducted independently and the results of both steps are combined in the end in an overall test result
– Shortens time and patient exposure– Relatively flexible – Efficient use of patient resources
– Complex design: statistics are difficult to explain– Gap in accrual between phase II and phase III– Logistically challenging– Difficult in studies with long-term endpoints
» Unless in combination with a short-term endpoint for the phase II part … another long and complex story on type I error and correlation
12
TRUSTS (EORTC 62091) in advanced or metastatic STS
13
Cytel Webinar for East®SurvAdapt, October 28, 2010
2-sided = a5%Power = 90%HR = 0.7
Sample size re-estimation
• May increase the risk of running an enlarged negative trial
• Possibility of second guessing A resampling decision can be easily interpreted
as “the treatment is not as efficient as expected”
→ Operational bias? Accrual?
→ May require extensive (expensive) logistics
Protection of study integrity is essential!
Sample size re-estimation
14
Battle Trial – Adaptive randomization
Lee et al.Zhou et al. CT 2008
Prior probability of
each treatment
success given marker
8-week outcome observed
Probabilities of treatment
success updated based
on observed results
Maximizes the chance that the patient receives the treatment that is most effective for him/her
Battle Trial – Adaptive randomization
Randomizeusing the weights given by prior prob
• Sample size?• Requires fast dataflow – logistically demanding
especially in large multicenter trials• Does not work for long-term endpoint. • Difficult to interpret results beyond estimation
Comparisons? Precision?
• Recruitment patterns can change during the course of the trial because of deduced knowledge of randomization probabilities
17
Adaptive randomization
• Simulations suggest very similar operational characteristics may be achieved if applying classical 2-stage designs with stopping rules Korn and Freidlin, JCO 2011 Yuan and Yin, JCO 2011
• Example of such an alternative: CREATE (EORTC 90101) A Simon 2-stage design is being used to assess
the activity of Crizotinib in each of 6 cohorts of patients (ALK/MET+)
18
Adaptive randomization
• The STBSG EORTC is more adaptive than you may have thought
• There are challenging times ahead, both for clinicians as well as statisticians Flexible design strategies More efficient use of resources
• While the sky seems to be the limit, experience teaches us to be wary and critical of solutions presented as ‘miracles’.
Conclusion
19
Stats colleagues at the EORTC, specifically
Acknowledgment
20
Laurence Collette Jan Bogaerts Murielle Mauer