SMART Designs for Developing
Dynamic Treatment Regimes
S.A. Murphy
Symposium on Causal Inference
Johns Hopkins, January, 2006
Outline
• Why Dynamic Treatment Regimes?
• Why SMART experimental designs?
• Design Principles and Analysis of Balanced
Designs
2
Dynamic treatment regimes are individually tailored
treatments, with treatment type and dosage changing
according to patient outcomes. Operationalize clinical
practice.
•Brooner et al. (2002) Treatment of Cocaine Addiction
•Breslin et al. (1999) Treatment of Alcohol Addiction
•Prokaska et al. (2001) Treatment of Tobacco Addiction
•Rush et al. (2003) Treatment of Depression
3
Why Dynamic Treatment Regimes?
• In Prevention
– Individuals are at risk of problem behaviors for
different reasons (multiple causes of the
problem behavior)
– Periods of high and low risk
– Periods of high and low need for treatment
– Some individuals may have already exhibited
early problems
4
Why Dynamic Treatment Regimes?
• In Treatment
– High heterogeneity in response to any one
treatment
• What works for one person may not work for
another
• What works now for a person may not work later
– Improvement often marred by relapse
– Intervals during which more intense treatment
is required alternate with intervals in which less
treatment is sufficient
– Co-occurring disorders may be common
5
Why not combine all possible efficacious therapies and
provide all of these to patient now and in the future?
•Treatment incurs side effects and substantial burden, particularly
over longer time periods.
•Problems with adherence:
•Variations of treatment or different delivery mechanisms
may increase adherence
•Excessive treatment may lead to non-adherence
•Treatment is costly (Would like to devote additional resources to
patients with more severe problems)
More is not always better!
6
Example of a Dynamic Treatment Regime
Treatment of alcohol dependence. Goal is to reduce drinking.
Following graduation from the intensive outpatient program the
patient is prescribed naltrexone. The patient is monitored weekly
over the next two months. If the patient experiences 2 or more
heavy drinking days during this period and is nonadherent then
the patient’s medication is augmented by CBI. If the patient
experiences 2 or more heavy drinking days during this period and
is adherent then the patient’s medication is switched to
acamprosate. If the patient is able to make the entire 2 months
with 1 or no heavy drinking days then the patient is continued on
naltrexone and the patient is provided telephone disease
management.
7
The Big Questions
•What is the best sequencing of treatments?
•What is the best timings of alterations in treatments?
•What information do we use to make these decisions?
8
Why SMART Trials?
What is a sequential multiple assignment randomized
trial (SMART)?
These are multi-stage trials; conceptually a
randomization takes place at each stage.
Goal is to inform the construction of a dynamic
treatment regime.
9
Sequential Multiple Assignment Randomization
Initial T xt
Intermediate Outcome
Secondary T xt
TDM +
Responder
R
counseling
TDM
Med B
Med A
Nonresponder
R
EM +
Med B+
Psychosocial
R
Responder
TDM +
R
counseling
TDM
Med A +
Psychosocial
Med B
Nonresponder
R
EM +
Med B+
10
Psychosocial
First Alternate Approach
• Why not use data from multiple trials to construct the
dynamic treatment regime?
• Choose the best initial treatment on the basis of a
randomized trial of initial treatments and choose the
best secondary treatment on the basis of a
randomized trial of secondary treatments.
11
Delayed Effects
Negative synergies: An initial treatment may produce a
higher proportion of responders but also result in side
effects that reduce the effectiveness of subsequent
treatments for those that do not respond. Or the
burden imposed by this initial treatment may be
sufficiently high so that nonresponders are less likely
to adhere to subsequent treatments.
12
Delayed Effects
Positive synergies: A treatment may not appear best
initially but may have enhanced long term
effectiveness when followed by a particular
maintenance treatment. Or the initial treatment may
lay the foundation for an enhanced effect of
subsequent treatments.
13
Summary:
When evaluating and comparing initial treatments we
need to take into account the effects of the secondary
treatments thus SMART
14
Second Alternate Approach to SMART
• Why not use data from multiple randomized trials to
construct the dynamic treatment regime?
• Use statistical methods that incorporate the potential
for delayed effects and are suited for combining data
from multiple randomized trials.
•Methods from Medical Decision Making
involving a variation of a Markovian assumption
15
Why statistical methods for combining
over multiple trials are not always the
answer
• Cohort Effects
• Causal effects of prior treatment and non-causal
correlations
16
Cohort Effects
Subjects who will enroll in, who remain in or who
are adherent in the trial of the initial treatments may
be quite different from the subjects in SMART.
17
Causal Effects of Initial Txt
TRIAL 1
TRIAL 2
Unknown
Causes
Treatment
one
Observed
Outcome 2
Treatment
two
Observed
Outcome 3
18
Non-causal Correlations
TRIAL 1
TRIAL 2
Unknown
Causes
Treatment
one
Observed
Outcome 2
Treatment
two
Observed
Outcome 3
19
Summary:
•Standard randomized trials may yield information
about different populations from SMART trials.
•Causal models are often incomplete resulting in
difficulties in comparing initial treatments that are to be
part of the dynamic treatment regime.
20
Third Alternate Approach to SMART
Why not use theory, clinical experience and expert
opinion to construct the dynamic treatment regime
and then compare this regime against an appropriate
alternative in a confirmatory randomized trial?
The alternative may be the same regime but with one
component altered.
21
Problems with the two group trials (or repeated cycles
of randomized two group trials)
•Dynamic treatment regimes are multi-component
treatments:
• when to start treatment?, when to alter treatment?,
which treatment is best next?, what information to
use to make each of the above decisions?
•We are not opening the black box— we don’t know
why we get or do not get significance and
•Heavy reliance on expert opinion or best guesses -to choose not only the components but the level of
22
these components.
Problems with the two group comparison (or repeated
cycles of randomized two group trials)
•Results may not replicate well --miss interactions,
•Takes a long time to “optimize” the multicomponent treatment --method depends on no
interactions and
•Some components are costly --retain costly, inactive
components
23
Sequential Multiple Assignment Randomization
Initial T xt
Intermediate Outcome
Secondary T xt
TDM+
Responder
R
counseling
TDM
Med B
Med A
Nonresponder
R
EM +
Med B+
Psychosocial
R
Responder
TDM+
R
counseling
TDM
Med A +
Psychosocial
Med B
Nonresponder
R
EM +
Med B+
24
Psychosocial
Examples of SMART designs:
•CATIE (2001) Treatment of Psychosis in Alzheimer’s
Patients
•CATIE (2001) Treatment of Psychosis in
Schizophrenia
•STAR*D (2003) Treatment of Depression
•Thall et al. (2000) Treatment of Prostate Cancer
•Oslin (on-going) Treatment of Alcohol Dependence
25
SMART Designing Principles
26
SMART Designing Principles
•KEEP IT SIMPLE: At each decision point, restrict
class of treatments only by ethical, feasibility or strong
scientific considerations. Use a low dimension
summary (responder status) instead of all intermediate
outcomes (time until nonresponse, adherence, burden,
stress level, etc.) to restrict class of next treatments.
•Collect intermediate outcomes that might be useful in
ascertaining for whom each treatment works best;
information that might enter into the dynamic treatment
regime.
27
SMART Designing Principles
•Choose primary hypotheses that are both scientifically
important and aid in developing the dynamic treatment
regime.
•Power trial to address these hypotheses.
•Choose secondary hypotheses that further develop the
dynamic treatment regime and use the randomization to
eliminate confounding.
•Trial is not necessarily powered to address these
hypotheses.
28
SMART Designing Principles:
Primary Hypothesis
•EXAMPLE 1: (sample size is highly constrained):
Hypothesize that given the secondary treatments
provided, the initial treatment Med A + psychosocial
counseling leads to lower drinking than the initial
treatment Med A alone.
•EXAMPLE 2: (sample size is less constrained):
Hypothesize that a particular dynamic treatment regime
beginning with Med A+ psychosocial counseling results
in lower drinking than a particular dynamic treatment
regime beginning with Med A.
29
Initial T xt
Intermediate Outcome
Responder
Secondary T xt
TDM +
counseling
TDM
Med B
Med A
Nonresponder
EM +
Med B+
Psychosocial
Intensive Outpatient
Program
Responder
TDM +
counseling
TDM
Med A +
Psychosocial
Med B
Nonresponder
EM +
Med B+
Psychosocial
30
Initial T xt
Intermediate Outcome
Responder
Secondary T xt
TDM +
counseling
TDM
Med B
Med A
Nonresponder
EM +
Med B+
Psychosocial
Intensive Outpatient
Program
Responder
TDM +
counseling
TDM
Med A +
Psychosocial
Med B
Nonresponder
EM +
Med B+
Psychosocial
31
An analysis that is less useful in the
development of dynamic treatment
regimes:
Decide whether med A is better than med A +
psychosocial counseling by comparing intermediate
outcomes (proportion of immediate responders).
32
SMART Designing Principles
•Choose secondary hypotheses that further develop the
dynamic treatment regime and use the randomization
to eliminate confounding.
•EXAMPLE: Hypothesize that non-adhering nonresponders will have lower drinking if provided a
change in medication + EM+ counseling as compared to
a change in medication only.
33
Initial T xt
Intermediate Outcome
Responder
Secondary T xt
TDM +
counseling
TDM
Med B
Med A
Nonresponder
EM +
Med B+
Psychosocial
Intensive Outpatient
Program
Responder
TDM +
counseling
TDM
Med A +
Psychosocial
Med B
Nonresponder
EM +
Med B+
Psychosocial
34
Discussion
• Secondary analyses can use pretreatment
variables and outcomes to provide evidence
for a more sophisticated dynamic treatment
regime. (when and for whom?)
• SMART design and analyses targeted at
scientific goal of informing the construction
of a high quality dynamic treatment regime
• Adherence is not a nuisance; adherence
indicates need to tailor treatment.
35
This seminar can be found at:
http://www.stat.lsa.umich.edu/~samurphy/
seminars/Hopkins0106.ppt
This seminar is partially based on a paper with Kevin
Lynch, Jim McKay, David Oslin and Tom Ten Have.
Email me with questions or if you would like a
copy:
samurphy@umich.edu
36
Conceptual Structure
Unknown
Causes
Observed
Variables
Unknown
Causes
Treatment 1
Observed
Outcomes
Time 2
Treatment 2
Observed
Outcomes
Time 3
37