Hopkins PSMG Meeting Randomized Designs for Person, Place, and Time For Effectiveness Trials and Beyond C Hendricks Brown University of South Florida Peter.

Hopkins PSMG Meeting

Randomized Designs for Person, Place, and Time

For Effectiveness Trials and Beyond

C Hendricks BrownUniversity of South Florida

Peter Wyman U RochesterJing Guo USF Juan Peña U Rochester


Support

R01-MH40859 (Brown) Methodology for Mental Health / Drug Abuse Prevention & Early InterventionDesigns & Analyses for Mental Health Preventive Trials

NIMH and NIDA

3R01MH040859-15S1 (Brown)Methodology for Population-Based Approaches to the Prevention of Suicide

NIMH Office of Rare Diseases CDC Injury Prevention Center

R34MH071189-01 (Wyman) RCT of Gatekeeper Training for Suicide Prevention

NIMH

P20MH071897-01 (Caine) Developing Center On Public Health and Population Interventions For The Prevention Of Suicide

NIMH


Brown C.H., Wyman P. A., Guo J, and Peña J. (2005). Dynamic wait-listed designs for randomized trials: New designs for prevention of youth suicide. Submitted for publication.


Outline

A. Types of Scientific Questions

B. Typology of Trials Matching these Questions

C. Randomization involving Person, Place, and Time

D. Dynamic Wait-Listed Design for suicide prevention

E. Randomizations of Person/Place/Time


Efficacy

Effectiveness

Sustainability

Going to Scale

Sustaining Systemwide

Efficacy

Effectiveness

Sustainability

Going-to-Scale

Disseminating & Sustaining Systemwide

A. Types of Questions: Phases of prevention


Do we need random assignment as we move to effectiveness and beyond?

Nonrandomized designs Blueprints Project U Colorado Denver

Alternative Study Designs for Evidence-based Practice: Harnessing Natural Variation for Effectiveness Research

Agency for Healthcare Research and Quality (AHRQ),


Prevention Field Trials Needed to Determine Precise Information

about

What WorksFor Whom

Under What Conditions

Because …


There are very few Broad Street Pump Handles Left to Remove

John Snow’s Map of London


To Obtain Precise Answers

• Need to rely on Very Carefully Designed Intervention Studies

• Random Assignment needed,

• In novel ways ……


B. Typology of Research Questions/Designs

1) Efficacy – Impact in optimally controlled settings2) Effectiveness – Impact in real world settings3) Implementability – What level of intervention produces

what level of impact; what does it take to implement?4) Adaptability – How does variation in intervention

delivered affect impact?5) Extensibility – What impact is achieved when delivered to

different persons/settings?6) Sustainability – Does impact continue after training ?7) Scalability– What impact is achieved when the

intervention is expanded system wide, or in larger contexts?

8) Disseminability/ Adoptability – What interventions are effective in having new communities adopt such a program?


Extensibility Trials

• Selection Bias -- ASU Braver

• Participation Bias – UM, ASU, GW

Price/Vinokur/Sandler/Howe

-- U Miami Szapocznik


Three Stages of a Trial Design

Target Population

Located Contacted Eligible Consented Assessed at Baseline

Randomized and Intervention Begins

Intervention Ends

Selection Bias

Participation Bias

Attrition

Participation Adherence

Followed Up

Adapted from Brown & Liao, 1999

Pre-Intervention Design

Intervention Design

Post-Intervention Design


Extensibility Trials

Participation Trials

Can participation be increased and if so, how do these people benefit compared to those who would normally have participated?


Extensibility Trial with Two Randomizations

Trial Inside a Trial U Michigan, GWSample

Intervention Control

Routine Intensity

High Intensity

Randomize

Randomize to Invitation

Measure Participation and

Measure Outcomes

Model Participation and Measure Outcomes

Can participation be increased, and if so, how do these people benefit compared to those who would normally have participated?


Newer Examples of Experimental Manipulations

Extensibility Trial –

Intervening to Increase Participation

Effectiveness Trial --

Combinations of Intervention Components

Sustainability Trial –

Schedule for Supervision of a defined intervention

Scalability Trial –

Systems level intervention


C. A General Framework for Randomizing Using Person Place,

and Time

Person

PlaceTime


Three Descriptive Epidemiologic Concepts Viewed Three Different Ways

Concepts Epidemiology Analytical Methods

Randomized Designs

Person Person Multivariate or Person-level Analyses

RCT: Individual Level Randomized Trials

Environment Place Multi-level Analyses

Group/Cluster/Place Based Randomized Trials

Developmental Course

Time Longitudinal Analyses

Wait-Listed Designs, Randomizations at Different Times


Units that can be Randomly Assigned in a Trial

• Persons – Inclusion/Exclusion Criteria • Places/Groups – Random assignment at

the group level Group/Cluster Randomized Trials- Murray

Place-Based Randomized Trials -- Boruch

• Time -- when to intervene


Person Level Assignments

Persons Target Population

Inclusion / Exclusion Criteria Universal, Selective, Indicated, Treated

Extensibility Trials: Does intervention work effectively for Prevention

Lower risk subjects?Less likely to be identified or participate?

TreatmentPlacebo Nonresponders?Genetic Risk?


Place-Based Randomized Trials

Places/Groups – Classroom, Schools, Child Welfare Settings, MH/Drug Abuse Clinics

Schools -- Sloboda 88 HS’s New DARE 7th & 9th

Classrooms within Schools – Balt Prev PgmBrown & Liao (1999) Am J Commun Psychol

Rural Communities – PROGRESA, Mexico


Time Assignments

Time -- when to intervene

Wait-listed design: Half initially half later

Crossover Design A-B B-A

Multiple Baseline Designs


Time (Persons)Randomized Clinical Trial with Blocking on Time

A B B A A B B A A B A B


Person-Place Randomizations

Interventions at Two Levels

Example: Linking Classroom Prevention and an individually based Service Intervention within Schools


Person/Place

Unit Combinations of Interventions

School Assign Available Services

No Available Services

Classrm Universal No Universal

Universal No Universal

Child Moderate

High

Moderate

High

Moderate

High

Moderate

High


Place and Time

Group Based, Wait-Listed Design

Group Based, Multiple Baseline Design


Randomization in Place and Time:

D. Dynamic Wait-Listed Designs

• Divide a set of schools into replicates

• Randomly assign WHEN a replicate converts from wait-listed to active intervention

Brown et al. (under review)

Wyman, Brown Ga Gatekeeper Trial: NIMH


Randomization in Place and Time:School-Based Gatekeeper Training

for Suicide Prevention QPR: (Quinnett, 1995): Question a person (showing warning signs) about

suicide Persuade the person to get help Refer the person to the appropriate resource

Identify recognizable behaviors (“threats of suicide”)

Develop knowledge and skills to take action


Primary Endpoint

Rate of referral of youth by school staff for suicidality

QPR should

• Increase rate of referral in those schools that have been trained

• Maintain this increase rate over the study


Prevalence of Suicidal Behavior in Middle and High School

SmartTrack Survey Results(2003-2004 / 3291 Surveys)

10%

7%

6%

10%

6%

3%

0% 2% 4% 6% 8% 10% 12%

Ever tried to kill self

Tried to kill self in last year

Tried to kill self in last fourweeks

Eighth Grade Tenth Grade


Proportion of Suicidal Attempters Identified by School System

6%


Probability that a Single Staff Member Would be able to Identify

and Refer a Suicidal Child

0.03%


Probability of Referral as Function of Proportion Trained and Training

EffectivenessProbability of Someone Referring Suicidal Child

as Function of Proportion of Staff Trained and Training Effectiveness

Proportion Trained

Pro

babi

lity

of R

efer

ral

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.1

0.2

0.3

0.4

0.5

1

1.52

3

4

5

10


Comparison of Classic versus Dynamic Wait-Listed Designs

Standard Wait-List Design

Half are assigned immediately, half later

Dynamic Wait-Listed Design

Divide into replicates, then assign intervention or wait-listed control within each replicate

Train in 1st replicate early intervention schools, then in 2nd replicate, etc.


Classic and Dynamic Wait-Listed DesignYear Time Block Wait-Listed Design Dynamic Wait-Listed

Design Time Intervention Wait-Listed Intervention Wait-Listed

1 4 28 2 8 24 3 12 20

1

4

16 16

16 16 5 20 12 6 24 8 7 28 4

2

8

32 0

32 0


Comparing Power of Classic vs Dynamic Wait-Listed Design

Assumptions• Poisson Rates of Referral that vary randomly

over time but are proportional to time interval• Intervention Effect is Linear on the rate• General least squares analysis

• Efficiency in Asymptotic Variance fctn of Intervals 2 / (Intervals 2 – 1) 3 18% 4 25%Limit 33%


More Realistic Setting

Poisson counts for each interval with a varying multiplicative rate parameter that has a Gamma distribution

Log-Linear model for intervention impact

Maximum likelihood analysis


Efficiency Increases with Number of Time Blocks and Low Variability

in Rates over Time


Recommendation for the Current Trial

Currently completed first half of trial with half receiving intervention, half wait-listed.

Switch to a dynamic wait-listed design in the second phase, 4 groups of 4 schools


Added Power from Dynamic Wait Listed Design

Effect = Intervention Rate / Control Rate

Po

we

r

1.0 1.1 1.2 1.3 1.4 1.5

0.2

0.4

0.6

0.8

1.0

32 Time Blocks16 Time Blocks8 Time Blocks6 Time Blocks4 Time BlocksStandard Wait List


E. Randomizations involving Person Place and Time

Timing of Multi Level Interventions

Testing of Intervention in the Presence of Gene by Environment Interactions


Timing of Universal Classroom and Individual Level Interventions

Universal Classroom First Universal Classroom Intervention Intervention Conditions Yes No

Yes A Child Intervention No B C Child Intervention First

Universal Classroom Intervention Intervention Conditions Yes No

Yes D F Child Intervention No E


Serotonin Transporter Gene SLC6A4 Long and Short Forms

Short form associated with increased depression, bipolar disorder, violent suicide

Low CSF serotonin occurs among short allele monkeys who are raised in deprived environments, but not in mother raised environments


Comparison of Interventions for Those with Genetic Risk

(Short Alleles)Early Intervention

Reduce Environmental Risk Intervention Conditions Yes No

Yes A * B Child Intervention No C * D Late Intervention

Reduce Environmental Risk Intervention Condition Yes No

Yes E ** F Child Intervention No G ** H


Summary

For Effectiveness Trials and BeyondOften Dictated by Community Needs

“Everyone Gets This by a Certain Date”

Randomization involving TIMEDynamic Wait-Listed Designs are Efficient

Newer Designs often require randomization involving 2 or 3 ofPERSON, PLACE, and TIME

Hopkins PSMG Meeting Randomized Designs for Person, Place, and Time For Effectiveness Trials and Beyond C Hendricks Brown University of South Florida Peter.

Documents

hopkins psmg meeting

hopkins psmg meeting

hopkins psmg meeting

hopkins psmg meeting

hopkins psmg meeting

mexico slide

phases of prevention

time person place time