Ecological Momentary Assessment in Substance Abuse Research

1
Ecological Momentary Assessment in Substance
Abuse Research
James W. Hopperjhopper_at_mclean.harvard.edu
Zhaohui Su, Alison R. Looby, Elizabeth T.
Ryan,David M. Penetar, Christopher M. Palmer,
Scott E. Lukas
Behavioral Psychopharmacology Research
LaboratoryMcLean Hospital and Harvard Medical
School Statistical and Data Analysis
CenterHarvard School of Public Health
2
Overview

• Ecological Momentary Assessment (EMA)
• General EMA research issues
• Our minimalist EMA study
• Minimalist EMA strengths and limitations

3
Ecological Momentary Assessment

• Collection of near-real-time data, in real-world
  environments of daily life
• Standard methods
• Participants carry handheld computers running
  electronic diary software
• Computer-initiated random assessments
• User-initiated event recordings (e.g., smoking a
  cigarette) with assessments

4
Standard EMA Methodology

• Relatively extensive assessments of
• Setting Location, alone or with others, target
  behavior allowed in setting, etc.
• Activity Interacting, working, leisure, phone
  recent food and substance consumption etc.
• Mood, emotions, urges, craving, etc.

See Shiffman et al., 1996, J Consult Clin
Psychol, 64, 366
5
Insights Yielded by EMA Example

• Nicotine patch, smoking lapse and relapse
• After initial lapse, staying on patch associated
  with decreased likelihood of full relapse
• Largest effect in study
• Complete opposite of current instructions to
  patch users, i.e., stop patch if smoke
• Progression from lapse to relapse not amenable to
  retrospective assessment, never previously
  assessed in outcome studies

Shiffman et al., 2006, J Consult Clin Psychol,
64, 276
6
General EMA Research Issues

• Training participants
• Piloting methods
• Monitoring data quality and completeness
• Managing and reducing extremely large data sets
• Appropriate descriptive and inferential analyses
• Reporting findings effectively in tables and
  figures
• Best methodology papers
• Stone Shiffman, 1994, Ann Behav Med, 16, 199.
• Schwartz Stone, 1998, Health Psychol, 17, 6.
• Stone Shiffman, 2002, Ann Behav Med, 24, 236.

7
My Experience with EMA

• Study designed by Scott Lukas
• My role Oversee final stages of data collection
  and cleaning work with statistician write paper

8
EMA Study of Polysubstance Use by Regular Ecstasy
Users

• Background
• Retrospective studies have found high rates of
  polysubstance use among ecstasy users
• Popular, media-promoted belief Ecstasy use is
  highly associated with, even causes, use of other
  illicit drugs
• Antidote to limited methods and speculation EMA,
  especially in party and rave situations

9
EMA Study of Polysubstance Use by Regular Ecstasy
Users

• Research Goals / Questions
• Assess patterns of ecstasy use and its
  relationship to other drug use in daily lives of
  regular ecstasy users
• Compare illicit drug use on nights when ecstasy
  was used vs. similar nights on which ecstasy was
  not used
• Assess patterns of craving for ecstasy
• Over hours preceding and following its use
• Over days of weeks ecstasy was used vs. not used

10
Minimalist EMA Approach
ActiWatch-Score Device
11
Minimalist EMA Approach

• Wrist actigraphy device, ActiWatch-Score
• User-initiated data entry (drug use events)
• Device-prompted data entry (craving)
• Simpler and less obtrusive than standard EMA
• Intent More acceptable to polysubstance users,
  especially in drug use settings
• Major limitation No data on situations,
  activities, mood, etc.

12
Sample

• Convenience, via net and newspaper ads
• 22 (of 34 enrolled) participants completed
  protocol and used ecstasy at least once during
  study
• Age 19 to 38, mean 22.8 (?13.5) 13 males
• Heavy recreational drug users
• 55 current regular smokers

13
EMA Training for Participants

• Demonstrations, verbal instructions, practice in
  lab, and written instructions
• Enter numeric code each time used a drug (given
  small laminated card with codes)
• One or more ecstasy pills
• Individual drink or cigarette
• Marijuana use session, cocaine line or rock
• Single dose of other drugs (e.g., sedative,
  opiate)
• Respond (within 10s) to audible prompt, with 0-9
  rating of current craving
• Write down mistakes (e.g., entered 4 when meant
  5)

14
EMA Procedures

• Wear device 24/7 for 6 weeks (except in water)
• Enter every drug use event, respond to every
  audible prompt with rating of current craving
• Daily diaries of drug use, wake and sleep times
• Phone check-ins over first 2-3 days, to ensure
  compliance with protocol and answer questions
• Return to lab every 7 days to provide ActiWatch
  and daily diary data

15
Pilot Work

• Key issue Hear and respond to audible prompts
• Pilot participants wore device 1-3 weeks
• Calibration of prompt volume Loudest possible
  without awakening participants or sleep partners
• 85 response rate during waking hours, i.e., good
  compliance rate found in other EMA studies

16
Monitoring EMA Data Qualityand Completeness

• Question participant
• Evidence of ActiWatch malfunction?
• Responding to auditory prompts when awake?
• Any incorrect ActiWatch entries?
• Inspect ActiWatch data files immediately after
  downloading (every 7 days)

17
Data Quality and Completeness

• Across all participants, on 90 of study days
  data was entered into device, it was worn the
  entire day and did not malfunction
• Number of days participants entered data
• 10 participants, exactly 42 days/6 weeks
• 7 participants, gt 42 days (max 47)
• 4 participants, 31-41 days
• 1 participant, 17 days

18
Data Quality and Completeness

• Mean rates of responding to craving prompt
• 58 ecstasy use nights
• 67 ecstasy non-use nights
• Groupings of response rates
• 21 participants with gt20
• 18 with gt40 rates
• 10 with gt66

19
Managing and Reducing the Data

• EMA data sets are huge
• Data reduction methods depend on
• Research questions and study design
• What is being described, compared
• Sampling frequencies (e.g., of prompts)
• Temporal aspects of phenomena studied (e.g.,
  pharmacokinetics of substance)
• Statistics used for data analyses

20
Reducing the Data

• Research questions and study design
• Describe patterns of ecstasy use hour and day
  bins
• Compare ecstasy use vs. ecstasy non-use nights
• Needed to define night. Chose 5pm to 9am based
  on inspection of EMA data
• Needed non-use nights valid for comparison. Used
  Fridays and Saturdays, i.e., typical party
  nights
• Craving analyses hour and day bins
• Dependent on sampling rate of prompts
• 3-hour bins, entire days

21
Reducing the Data

• Temporal aspects of phenomena studied
• Pre Before using ecstasy
• During When ecstasy taken to intoxication
  plateau
• After Coming down phase
• Multiple ecstasy uses after indexed to last
  use
• Requirements of statistics
• Need equal lengths for pre, during, and after
  periods
• Choice 4-hour duration, based on likely (a)
  delay to ecstasy intoxication onset (40-90 mins)
  and (b) duration of intoxication plateau (3 hrs
  after onset)

22
Statistical Analyses

• Even descriptive statistics may not be simple, as
  they are inseparable from data reduction choices
• Inferential statistics require high level
  expertise with esoteric statistical methods
• Preliminary analyses are essential, to determine
  distributions, covariance structures, etc.
• Key complex data feature Nesting of variables
  within temporal periods within participants

23
Descriptive Findings
24
Descriptive Findings
25
Descriptive Findings
26
Statistical Analyses Data Features

• Difficult characteristic of non-ecstasy drug use
  data
• High between-subjects variability in number of
  times ecstasy was used during study
• Requires nesting drug use events not only within
  ecstasy use night, but within participant so that
  heavy users arent driving results
• But this solution means very esoteric stats

27
Esoteric Stats
Binary (yes/no) drug use outcomes were modeled
with generalized estimating equations
(GEE). Logit link function and binomial error
distribution of multivariate responses (i.e.,
before, during, and after periods) were
specified. (For these and all GEE analyses
described below, multiple covariance structures,
including unstructured, autoregressive and
exchangeable, yielded identical findings.)
28
Effectively Reporting Findings

• EMA provides real-time, real-life data so ones
  presentation of findings should reflect that
• Related to events and rhythms of daily life
• Easily and intuitively grasped
• However, depending on the research questions and
  statistics, presentations of findings may
• Be complex and detailed
• Demand close attention of readers

29
Inferential Findings

• No drug was significantly more likely to be used
  on nights that ecstasy was used than on
  comparison Friday and Saturday nights.
• After correction for multiple tests, a trend for
  cocaine use more likely on ecstasy use than
  ecstasy non-use nights.

30
(No Transcript)
31
Interpretive Summary of Prior Table
For nights involving ecstasy use, use of ecstasy
and other drugs appeared to follow a natural
history that typically began with alcohol,
progressed to a period involving use of a highly
intoxicating drug, in this case ecstasy, which
was followed by significantly decreased
likelihood of using any intoxicating substance.
32
Findings unchanged by when analysis limited to 18
with response rates to craving prompt gt40 or 10
with rate gt66
33
Our Minimalist EMA ApproachLimitations and
Challenges

• That drug-use event record isnt present may not
  mean that drug-use event didnt occur
• Inherent limitation of all real-time EMA event
  recording methodologies
• Could result in incorrect classification of drug
  use nights as non-use nights, thus bias results
• No data on whether compliance varies with time
  relative to ecstasy use or time of night
• Trend for lower rate of responding to craving
  prompt on ecstasy use nights does suggest some
  bias

34
Our Minimalist EMA ApproachLimitations and
Challenges

• Low response rates to audible prompts
• Much lower than pilot testing rates 85
• Loud environments associated with drug use?
• When debriefed participants said volume was
  issue, but may not have acknowledged ignoring
  prompts
• Difficulty responding in 10s window allowed by
  device? Especially when active and/or
  intoxicated?
• Possible solutions Vibration prompt (battery
  life) compensate participants based on response
  rate
• Issue not addressed here Concordance between EMA
  and daily diary data

35
Conclusions I

• EMA is a powerful research methodology
• Data from real-life situations and activities
• Little to no retrospective self-report bias
• Challenges common to all EMA research
• Effective training of participants
• Monitoring data quality and completeness
• Appropriately reducing huge databases
• Complex and esoteric statistical issues

36
Conclusions II

• Benefits of minimalist EMA
• Minimal demands on participants time
• Easy, unobtrusive wrist watch-like form factor
• Limitations and challenges of minimalist EMA
• No EMA data on situations, activities, mood, etc.
• Ensuring prompts are noticed and responded to
• Future directions
• Greater data capacity per event record/
  assessment
• Option to postpone responses to prompts

37
Ecological Momentary Assessment in Substance
Abuse Research
James W. Hopperjhopper_at_mclean.harvard.edu
Zhaohui Su, Alison R. Looby, Elizabeth T.
Ryan,David M. Penetar, Christopher M. Palmer,
Scott E. Lukas
Behavioral Psychopharmacology Research
LaboratoryMcLean Hospital and Harvard Medical
School Statistical and Data Analysis
CenterHarvard School of Public Health