Getting SMART about Developing Individualized Sequences of ...dalmiral/slides/IMPACT_NC_2nov2012.pdf · Getting SMART about Developing Individualized Sequences of Health Interventions

Getting SMART about DevelopingIndividualized Sequences of Health

Interventions

Daniel Almirall1,2 Linda M Collins2,3,5

Susan A Murphy1,2,4

1Institute for Social Research, University of Michigan2The Methodology Center, Penn State University

3Human Development and Family Studies, Pennsylvania State University4Department of Statistics, University of Michigan

5Department of Statistics, Pennsylvania State University

IMPACT Meeting - North Carolina - November 2, 2012Almirall, Collins, Murphy Building Adaptive Interventions 1 / 44

Adaptive InterventionsEvaluating versus Building an Adaptive Intervention?

Sequential Multiple Assignment Randomized Trial (SMART)SMART Design Principles

Discussion

Outline

Adaptive InterventionsWhat? Why?

Evaluating versus Building an Adaptive Intervention?

Sequential Multiple Assignment Randomized Trial (SMART)What are SMARTs?

SMART Design PrinciplesKeep it SimpleChoosing Primary and Secondary Hypotheses

Discussion

Almirall, Collins, Murphy Building Adaptive Interventions 2 / 44



Discussion

What? Why?

ADAPTIVE INTERVENTIONS




Discussion

What? Why?

Definition: An Adaptive Intervention is

I a sequence of individually tailored decision rulesI that specify whether, how, and/or whenI to alter the intensity, type, dosage, or delivery of treatmentI at critical decision points in the course of care.

Adaptive Interventions operationalize sequential decisionmaking with the aim of improving clinical practice.




Discussion

What? Why?

Concrete Example of an Adaptive InterventionPediatric Anxiety Example (SAD, GAD, SoP)

Maintain: CBT

CBT

Add Treatment: CBT + MED

Responder

s

Non-Responders

Tailoring Variable First-line Txt Second-line Txt

I Goal is to minimize the child’s symptom profile/trajectory.




Discussion

What? Why?

What makes up an Adaptive Intervention?1. Critical decisions: treatment options and more2. Tailoring variables: to decide how to adapt treatment3. Decision rules: inputs tailoring variable, outputs one or

more recommended treatments

Maintain: CBT

CBT


Responder

s

Non-Responders

Tailoring Variable First-line Txt Second-line Txt

Adaptive interventions AKA: dynamic treatment regimes, adaptive treatmentstrategies, treatment algorithms, structured treatment interruptions, practiceparameters, ASAM criteria...


An Adaptive Intervention in Obesity

Maintain: GBI Refreshers

GBI Augment: GBI + Personal Health

Coach

Responder

Non‐Responder

Tailoring Variable: RESP if

4mo Wgt Change: >10% #Family Exprmts: avg>2/q Dietary intke: avg>3fv/wk

First‐line Txt:Months 0‐4

Second‐line Txt: Months 4‐12

Step Down: PHC‐lite (phone) GBI +

Personal Health Coach

Augment: GBI + Personal Health

Coach + Medication

Responder

Non‐Responder

Obesity: BMI ≤ 97%ile

Obesity: BMI > 97%ile

Tailoring Variable at Intake



Discussion

What? Why?

Why Adaptive Interventions?Necessary because...

I Nature of chronic disorders (substance use, mental health,diabetes, cancer)

I Waxing and waning course (multiple relapse, recurrence)I Genetic and non-genetic factors influence courseI Co-occuring disorders may arise

I High patient heterogeneity in response to treatment

I Within person (over time) differential response to treatmentI Between person differential response to treatment

All require sequences of treatment decisions.




Discussion

GENERATING HYPOTHESES vs BUILDING vs EVALUATING

ADAPTIVE INTERVENTIONS?




Discussion

3 Different Research Questions/Aims= 3 Different Research Designs

I Aim 1: When generating hypotheses to build anAdaptive Intervention: e.g., Does augmenting txt (asobserved in a previous trial) for non-responders correlatewith better outcomes?

I Aim 2: When building an Adaptive Intervention: e.g, Whatare the best tailoring variables and/or decision rules?

I Aim 3: When evaluating a particular AdaptiveIntervention: e.g. Does the AI have a statisticallysignificant effect as compared to control intervention?


3 Different Research Questions/Aims= 3 Different Research Designs

Ex. Q1: Does augmenting txt for non-responders (as observedin a previous trial) correlate with better outcomes?

Ex. Q2: What are the best tailoring variables or decision rules?

Ex. Q3: Does the Adaptive Intervention have a statisticallysignificant effect as compared to control intervention?

Observational ExperimentalStudies Studies

e.g., Analysis ofQuestion Aim Previous RCT SMART RCT

1 Hypothesis Gen. YES ≈ ∼2 Building ≈ YES ≈3 Evaluating ∼ ≈ YES



Discussion

What are SMARTs?

SEQUENTIAL MULTIPLE ASSIGNMENTRANDOMIZED TRIALS (SMARTs)




Discussion

What are SMARTs?

What is a Sequential Multiple AssignmentRandomized Trial (SMART)?

I Multi-stage trials; same participants throughoutI Each stage corresponds to a critical decision pointI At each stage, subjects randomized to set of treatment

optionsI The goal of a SMART is to inform the development of

adaptive interventions.

I will give you an example SMART, but first...




Discussion

What are SMARTs?

Motivation for an Example SMARTChild-Adolescent Anxiety Multi-modal Study (CAMS)

I CAMS: acute-phase, efficacy, RCT for child anxiety

I CBT+MED > MED ≈ CBT > Placebo

I However, some families and clinicians remain concernedabout the use of MED in this population

I So an important next question for clinical practice is“Can we delay the use of MED?””If so, for whom?”

I Some children may do fine w/ CBT only and not need MED.


Concrete Example of a SMART: Pediatric AnxietyCourtesy of Scott N Compton, Duke University Medical Center

Add Treatment: CBT + MED + FT Non-Responders

CBT + MED Maintain: CBT + MED

Step Down: CBT Only

R Maintain:

CBT

CBT Add Treatment: CBT + MED

Switch Treatment: MED

Responders R

Responders

Non-Responders R

O2 + Primary

Tailoring Variable First-line Txt Second-line Txt Y O1

One Adaptive Intervention Within the SMART



Step Down: CBT Only R Maintain:

CBT



Responders R

Responders

Non-Responders R

O2 + Primary


Another Adaptive Intervention Within the SMART



Step Down: CBT Only

R Maintain:

CBT



Responders R

Responder

s

Non-Responders R

O2 + Primary


4 Embedded Adaptive Interventions in this SMART

Add Treatment:CBT + MED + FTNon‐Responders

CBT + MED Step Down:CBT Boost

Responders

Add Treatment:CBT + MED + FT Non‐Responders

CBT + MED Maintain:CBT + MED

Responders

Maintain:CBT BoostResponders

CBT Add Treatment:

CBT + MEDNon‐Responders


CBT Switch Treatment:

MEDNon‐Responders

AI 1

AI 2

AI 3

AI 4



Discussion

Keep it SimpleChoosing Primary and Secondary Hypotheses

SMART DESIGN PRINCIPLES




Discussion


SMART Design Principles

I KISS Principle: Keep It Simple, Straightforward

I Power for simple important primary hypotheses

I Take Appropriate steps to develop a moredeeply-individualized Adaptive Intervention




Discussion


Keep It Simple, StraightforwardOverarching Principle

At each stage, or critical decision point,...I Use low dimensional summary to restrict subsequent

treatmentsI Use binary responder statusI Should be easy to use in actual clinical practice

I Restrict class of treatment options only by ethical,feasibility, or strong scientific considerations

I Collect additional, auxiliary time-varying measuresI To develop a more deeply-tailored Adaptive InterventionI Think time-varying effect moderators




Discussion


SMART Design: Primary Aims

Choose a simple primary aim/question that aids developmentof an adaptive intervention.

Power the SMART to test this hypothesis.


Primary Aim Example 1What is the main effect of initial CBT vs initial CBT+MED on longitudinal outcomes?



Step Down: CBT Only R Maintain:

CBT


Switch Treatmnt: MED

Responders R

Responders

Non-Responders R

O2 + Primary


PowerES N0.8 340.5 830.2 505ρ = 0.60α = 0.05β = 0.20

Primary Aim Example 2Is there a difference between two of the embedded adaptive interventions?

Add Treatment:CBT + MED + FTNon‐Responders

CBT + MED Step Down:CBT Boost

Responders

Add Treatment:CBT + MED + FT Non‐Responders

CBT + MED Maintain:CBT + MED

Responders


CBT Add Treatment:

CBT + MEDNon‐Responders


CBT Switch Treatment:

MEDNon‐Responders

AI 1

AI 2

AI 3

AI 4 Sample size calculators exist for this; see Oetting, Levy, Weiss,

and Murphy 2011.



Discussion


SMART Design: Secondary Aims

Choose secondary aims/questions that further develop theAdaptive Intervention and take advantage of sequentialrandomization to eliminate confounding.




Discussion


Secondary Aim Examples 1 and 2Best second-line treatment and second-line treatment tailoring aim.

O2 = CBT adherence, time to non-response, allegiance with therapist, changes in home environment



Non-Responders R

O2 + Primary

Tailoring Variable First-line Txt Second-line Txt Y

CBT


Secondary Aim Example 3Build a more deeply tailored adaptive intervention.



Step Down: CBT Only

R Maintain:

CBT



Responders R

Responders

Non-Responders R

O2 + Primary


O1 = demographics, genetics, sub-diagnoses, co-morbidities, etc…

O2 = adherence, time to NR, changes at home, etc…



Discussion

DISCUSSION




Discussion

Adaptive Interventions vs Adaptive ExperimentalDesigns?

I These ideas are not (necessarily) related. Confusing!

I Adaptive interventions are a type of intervention designI Adaptive experimental designs are particular type of

experimental designI SMARTs are not Adaptive Experimental DesignsI SMARTs do inform development of Adaptive Interventions




Discussion

Take Home the FollowingI Adaptive Interventions individualize treatment up-front and

throughout

I Adaptive Interventions are guides for clinical practice

I SMARTs are used to build better Adaptive InterventionsI Next trial: SMART-optimized Adaptive Intervention vs.

state-of-the-art treatment

I SMARTs do not necessarily require larger sample sizes

I Existing RCTs can be used to begin to learn aboutadaptive interventions

I Observational study


Interventions for Minimally Verbal Children with AutismPI: Kasari(UCLA), Kaiser(Vanderbilt), Smith(Rochester), Lord(Cornell), Almirall(Mich)

Non‐Responders (Parent training no feasible)

JASC (joint attention and

play) Re‐engage /continue JASC

JASC + Parent Training R

DTT (Discrete training)

Re‐engage /continue DTT

DTT + Parent Training

Responders(Blended txt unnecessary)

R

Non‐Responders(Parent training not feasible)

Responders(Blended txt unnecessary)

R

JASC+DTT

Re‐engage / continue JASC

R

JASC+DTT

Re‐engage /continue DTT

R



Discussion

Thank you! Questions?

Find papers on SMART:I http://www.lsa.stat.umich.edu/∼samurphy/ (Susan Murphy)I http://methcenter.psu.edu (Linda Collins)

These slides will be posted on my website:I http://www-personal.umich.edu/∼dalmiral/

Email me with questions about this presentation:I Daniel Almirall: [email protected]


Adaptive Treatment for Children with ADHDPI: Pelham, Florida International University

Continue Medication Responders

Medication Increase Medication Dose

Add Behavioral Intervention

R Continue

Behavioral Intervention Behavioral

Intervention Increase Behavioral

Intervention

Add Medication

Non-Responders R

Responders

Non-Responders R

Treatment for Alcohol DependencePI: Oslin, University of Pennsylvania

Early Trigger for NR: 2+ HDD CBI

CBI + Naltrexone

R

Late Trigger for NR: 5+ HDD

CBI

CBI + Naltrexone

Non-Response R

Non-Response R

Naltrexone

TDM + Naltrexone

8 Week Response R

Naltrexone

TDM + Naltrexone

8 Week Response R

Extra Slides


Hypothesis-generating Observational StudiesPost-hoc Analyses Useful for Building Adaptive Interventions

I Give examples of different observational study questionsthey can examine using data from a previous 2-arm RCT

I Standard observational study caveats apply:I No manipulation usually means lack of heterogeneity in txt

options (beyond what is controlled by experimentation inoriginal RCT)

I Some RCTs use samples that are too homogeneousI Confounding by observed baseline and time-varying factorsI Unobserved, unknown, unmeasured confounding by

baseline and time-varying factors


Hypothesis-generating Observational StudiesPost-hoc Analyses Useful for Building Adaptive Interventions

I There exists a literature for examining the impact oftime-varying treatments in observational studies

I Marginal Structural Models (Robins, 1999; Bray, Almirall, etal., 2006) to examine the marginal impact of observedtime-varying sequences of treatment

I Structural Nested Mean Models (Robins, 1994; Almirall, etal., 2010, 2011) to examine time-varying moderators ofobserved time-varying sequences of treatment

I Marginal Mean Models (Murphy, et al., 2001): to examinethe impact of observed adaptive interventions


Early precursors to SMART

I CATIE (2001) Treatment of Psychosis in Patients withAlzheimer’s

I CATIE (2001) Treatment of Psychosis in Patients withSchizophrenia

I STAR*D (2003) Treatment of Depression


Other Alternatives

I Piecing Together Results from Multiple TrialsI Choose best first-line treatment on the basis of a two-arm

RCT; then choose best second-line treatment on the basisof another separate, two-arm RCT

I Concerns: delayed therapeutic effects, and cohort effects

I Observational (Non-experimental) Comparisons of AIsI Using data from longitudinal randomized trialsI May yield results that inform a SMART proposalI Understand current treatment sequencing practicesI Typical problems associated with observational studies

I Expert Opinion


Why Not Use Multiple Trials to Construct an AIThree Concerns about Using Multiple Trials as an Alternative to a SMART

1. Concern 1: Delayed Therapeutic Effect

2. Concern 2: Diagnostic Effects

3. Concern 3: Cohort Effects

All three concerns emanate from the basic idea thatconstructing an adaptive intervention based on a myopic, local,study-to-study point of view may not be optimal.


Why Not Use Multiple Trials to Construct an AIConcern 1: Delayed Therapeutic Effects, or Sequential Treatment Interactions

Positive Synergy Btwn First- and Second-line Treatments

Tapering off medication after 12 weeks of use may not appearbest initially, but may have enhanced long term effectivenesswhen followed by a particular augmentation, switch, ormaintenance strategy.

Tapering off medication after 12 weeks may set the child up forbetter success with any one of the second-line treatments.


Why Not Use Multiple Trials to Construct an AIConcern 1: Delayed Therapeutic Effects, or Sequential Treatment Interactions

Negative Synergy Btwn First- and Second-line Treatments

Keeping the child on medication an additional 12 weeks mayproduce a higher proportion of responders at first, but may alsoresult in side effects that reduce the variety of subsequenttreatments available if s/he relapses.

The burden associated with continuing medication an additional12 weeks may be so high that non-responders will not adhereto second-line treatments.


Why Not Use Multiple Trials to Construct an AIConcern 2: Diagnostic Effects

Tapering off medication after 12 weeks initial use may notproduce a higher proportion of responders at first, but may elicitsymptoms that allow you to better match subsequent treatmentto the child.

The improved matching (personalizing) on subsequenttreatments may result in a better response overall as comparedto any sequence of treatments that offered an additional 12weeks of medication after the initial 12 weeks.


Why Not Use Multiple Trials to Construct an AIConcern 3: Cohort Effects

I Children enrolled in the initial and secondary trials may bedifferent.

I Children who remain in the trial(s) may be different.I Characteristics of adherent children may differ from study

to study.I Children that know they are undergoing adaptive

interventions may have different adherence patterns.

Bottom line: The population of children we are makinginferences about may simply be different from study-to-study.


Documents

Getting SMART about Developing Individualized Sequences of ...dalmiral/slides/IMPACT_NC_2nov2012.pdf · Getting SMART about Developing Individualized Sequences of Health Interventions