Transcript SMART Experimental Designs for Developing Adaptive

Sizing a Trial for the
Development of
Adaptive Treatment Strategies
Alena I. Oetting
The Society for Clinical Trials, 29th Annual Meeting
St. Louis, MO
May 2008
joint work with Janet Levy, Roger Weiss, and Susan Murphy
Multi-stage decision making
• Problem: Making multiple decisions across
time in order to optimize outcome of interest
– diseases which require many treatment decisions
• heterogeneity of patient response
• variable nature of disease: symptoms wax & wane
• high relapse rate
• Research: Sample size formulae for answering
questions about treatment decision rules
(adaptive treatment strategies) using new trial
design (SMART)
– current work: binary final outcome
Adaptive Treatment Strategy (ATS)
• Set of decision rules that tailor a sequence of
treatments to an individual patient
– input: patient characteristics, response to
treatment
– output: how/when to change treatment
• Questions to address in constructing ATS
– Best sequencing of treatments?
– Best timings of treatment transitions? (more to
less; switch; augment)
– What information to use to make decisions?
Example of ATS:
Prescription Opiod Dependence
Rules adapt to
previous response
Bup/Nx +
Nonresponse:
Any opiod use
during first 4 wks
Buprenorphine/Naloxone
(Bup/Nx) +
Medical Management (MM) +
MM +
CognitiveBehavioral
Therapy (CBT)
Final response
(binary): abstinence
Individual Drug Counseling
(IDC)
for 4 wks
Response:
No opiod use
during first 4 wks
Relapse
Prevention
Therapy (RPT)
* Example modeled after a trial currently in progress within Clinical Trials Network of NIDA
A SMART Design
• Sequential Multiple Assignment Randomized Trial
• A multi-stage trial; at each stage, subject
randomized to treatment decision
• SMART- precursors already conducted:
– CATIE (2001): psychosis in Alzheimer’s patients
– STAR*D (2003): depression
• Focus: Sample sizes for SMART design w/
– two decision points
– arbitrary number of options at each decision point
A SMART Design: Example
Bup/Nx + MM + CBT
nonresponse
R
Bup/Nx + MM + IDC
Bup/Nx + MM
response
RPT
response
RPT
R
Bup/Nx + MM +
IDC
Bup/Nx + MM + CBT
nonresponse
R = randomization
R
Bup/Nx + MM + IDC
Sample Size Formulae
Many potential primary research questions
COMPONENTS
1. Is there effect on final outcome btwn initial txts?
2. Is there effect on final outcome btwn txts for
nonresponders?
WHOLE STRATEGIES
1. Is there effect on final outcome btwn two
(prespecified) txt strategies?
2. What is best (highest response rate) txt strategy?
Sample Size Formulae:
Is continuity correction needed?
• Used normal approximation to binomial,
with & without continuity correction
• Sample sizes with continuity correction are
larger than those without.
• Simulations show both versions (with and
without continuity correction) give
conservative sample sizes; power
estimates not significantly different
Sample Sizes for SMART Example
response testing: main
rate after effect of 1st
initial
stage txt
treatment
testing:
2nd stage
txt effect
(nonresponders)
testing:
difference
between two
strategies
choosing
best strategy
(not hypothesis test)
0.1
103
114
206
53
0.3
103
147
206
53
0.5
103
206
206
53
• 2 initial txt options; 2 options for nonresponders, 1 for
responders
• Powered to detect difference in proportions of 0.2 w/
probability 0.8; for hypothesis tests, α = 0.1
Summary
• Sample size formulae & corresponding test
statistics developed for using SMART to construct
adaptive treatment strategies.
– continuous & binary final outcomes; two decision points,
arbitrary number of treatment options at each decision
– survival time as final outcome (Wahed and Feng)
• Binary final outcome: no advantage to using
continuity correction in sizing trial
• Sample size calculators available on web (already
available for continuous; binary in June 2008)
Questions?
This talk is based on joint work with
Janet Levy (Duke School of Nursing),
Roger Weiss (Harvard Medical School),
Susan Murphy (University of Michigan)
Email me with questions or if you would like
a copy of our paper (continuous),
technical report (binary):
[email protected]