jitai development inpeople.seas.harvard.edu/~samurphy/seminars/obbi 05.19.16session… · 05.19.16...
TRANSCRIPT
![Page 1: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/1.jpg)
JITAI Development
in
Mobile Health
Susan Murphy, Walter Dempsey, Peng
Liao & Tim NeCamp
05.19.16
HeartSteps
JITAI
JITAIs
JITAIs
JITAIs
![Page 2: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/2.jpg)
2
Overview
• Session 1: Introduction to JITAIs: Just-in-
Time Adaptive Interventions
• Session 2: Micro-Randomized Trials for
Developing mHealth JITAIs
• Session 3: Data Analytics for Developing
JITAIs
![Page 3: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/3.jpg)
3
Session 2: Micro-Randomized Trials
• HeartSteps
• Micro-Randomized Trial
• Sample Size Considerations
![Page 4: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/4.jpg)
HeartSteps
HeartSteps Activity Coach
o Wearable band measures activity, phone
sensors measure busyness of calendar,
location, weather, ..…
o In which contexts should the smartphone
ping and deliver activity
recommendations?
4
![Page 5: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/5.jpg)
5
HeartSteps
• Goal: Develop a Just-in-Time Adaptive
Intervention for Encouraging and Maintaining
Physical Activity
+
![Page 6: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/6.jpg)
6
Distal Outcome:
Activity over the 42 day study.
Proximal Response:
Proximal activity (step count) over next 30
minutes.
HeartSteps
![Page 7: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/7.jpg)
7
Treatments:
Whether to provide Tailored Activity
Recommendation? Yes/No
Decision points:
Approximately every 2-2.5 hours
HeartSteps
![Page 8: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/8.jpg)
8
Tailored Activity
Recommendation
No Message or
![Page 9: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/9.jpg)
9
Potential Tailoring Variables:
Sensor data: activity recognition (walking,
driving, standing/sitting), weather, location,
busyness of calendar, adherence, step count
Self-report: usefulness, burden
HeartSteps
![Page 10: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/10.jpg)
10
Session 2: Micro-Randomized Trials
• HeartSteps
• Micro-Randomized Trial
• Sample Size Considerations
![Page 11: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/11.jpg)
11
Micro-Randomized Trial
Randomize each participant between
treatments at each decision point
�Each person may be randomized 100’s or
1000’s of times.
These are sequential, “full factorial,” designs.
Extension of A/B testing & Single Case Designs
![Page 12: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/12.jpg)
12
Examples
• HeartSteps: 5 times/day*42 days=210 possible
randomizations
• Sense2Stop: 60 times/hour*10 hours/day*10
days=6000 possible randomizations
![Page 13: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/13.jpg)
Micro-Randomized Trial Elements
1. Record outcomes
– Distal (scientific/clinical goal) & Proximal
Response
2. Record potential tailoring variables
3. Randomize among treatments at decision
points
4. At end of trial use resulting data to assess
effects, moderation, construct decision rules
13
![Page 14: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/14.jpg)
14
Why
Micro-Randomization?
• Randomization (+ representative sample) is a
gold standard in providing data to assess the
causal effect of a treatment.
• Sequential randomizations will enhance
replicability of data analyses (moderation,
decision rule development) .
![Page 15: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/15.jpg)
15
HeartSteps
• Whether to provide a tailored activity
recommendation (at the decision points).
• Aim for an average of 2 recommendations/day.
• 210 decision points for the tailored activity
recommendations.
Yes ���
No ���
Tailored Activity
Recommendation?
Randomization Probability
![Page 16: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/16.jpg)
16
Sense2Stop
• Whether to provide a reminder to practice
stress-regulation exercises (at the decision
points).
• Aim for an average of 3 reminders/day over 10
days post-quit.
• Randomization probabilities depend on current
stress and are determined by an algorithm.
![Page 17: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/17.jpg)
17
Micro-Randomized Trial
These sequential factorial trials are used to build
JITAIs…
First Question to Address: Do the treatments
differentially impact the proximal response?
AKA: is there a signal here?!
--Test for main effects of the treatments on the
proximal response
![Page 18: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/18.jpg)
18
Time-varying Main Effects
Time varying potentially intensive/intrusive
treatments � potential for accumulating
habituation and burden
→
Allow main effect of the treatments on
proximal response to vary with time
![Page 19: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/19.jpg)
19
Sample Size
Determine sample size to detect a time-varying
main effect of the treatments on the proximal
response
• HeartSteps: Aim to detect time-varying effect of a
tailored activity recommendation on step count over
the 30 minutes following each decision point.
• Sense2Stop: Aim to detect time-varying effect of
reminder to use stress regulation exercise on % time
stressed over the hour following each decision point.
![Page 20: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/20.jpg)
20
Availability & the Main Effect
• Treatments can only be delivered at a decision
point if an individual is available.
• The main effect of a treatment at a decision point is
the difference in proximal response between
available individuals assigned an treatment and
available individuals who are not assigned a
treatment.
• Availability is not the same as adherence!
![Page 21: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/21.jpg)
Main effect of activity recommendation on
proximal response is likely time-varying
β(t), t=1,…,T
• What does
this main effect
mean? 21
Main Effect
(Example: HeartSteps)
Main Effect
β(t)
![Page 22: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/22.jpg)
Micro-Randomized Trial Elements
1. Record outcomes
– Distal (scientific/clinical goal) & Proximal
Response
2. Record potential tailoring variables
3. Randomize among Treatments at decision
points
4. At End of Trial use Resulting Data to
assess effects, moderation, construct decision
rules 22
![Page 23: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/23.jpg)
23
Outline
• Adaptive Interventions and Just-in-Time
Adaptive Interventions
• HeartSteps
• Micro-Randomized Trial
• Sample Size Considerations
![Page 24: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/24.jpg)
24
Sample Size Calculation
• We calculate the number of participants to
test H0 : no effect of the treatment, i.e.,
• Size to detect a low dimensional, smooth
alternate H1.
– Example: H1: β(t) quadratic with intercept, β0,
linear term, β1, and quadratic term β2 and test
H0 : β(t) = 0, t = 1, 2, ....T
![Page 25: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/25.jpg)
25
Sample Size Calculation
• Our test statistic uses estimators from a
“generalization” of linear regression.
• The test statistic is quadratic in the estimators of
the β terms.
• To calculate a sample size we need to specify a
clinically/scientifically important effect size to
detect.
![Page 26: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/26.jpg)
26
Sample Size Calculation
Alternative hypothesis is low dimensional
→ assessment of the effect of the activity
recommendation uses contrasts of between
participant responses + contrasts of within
participant responses.
--The required number of participants will be
small.
![Page 27: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/27.jpg)
27
Specify Alternative for Sample Size
Calculation
SPECIFY:
• Standardized main effects:
– main effect on first day,
– average main effect over trial duration
• Day of maximal main effect.
• Average availability
![Page 28: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/28.jpg)
Standardized effects:
– Initial effect: 0
– average standardized main effect over trial
duration: ?
– day of maximal effect: 28
– average availability: ?
28
HeartSteps (42 day study)
![Page 29: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/29.jpg)
29
Standardized Average Main Effect over 42 Days
Sample Size For
70% availability or 50%
availability
0.06 81 or 112
0.08 48 or 65
0.10 33 or 43
HeartSteps Sample Sizes
Power=.8, α=.05
![Page 30: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/30.jpg)
30
Sample Size Calculations
Beta version of the sample size calculator:
https://pengliao.shinyapps.io/mrt-calculator/
Try calculator for Sense2Stop study:
• 10 days, 600 decision points per day
• Availability of 25% so essentially 600*.25=150
available points per day
• For an average of 3 interventions per day 3/150=.02
randomization probability
![Page 31: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/31.jpg)
31
Micro-Randomized Trial
1) Be conservative in planning the trial!
1) Under-estimate the amount of time
participants are available for treatment.
2) Under-estimate the average proximal
effect
![Page 32: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/32.jpg)
32
Micro-Randomized Trial
2) Power to detect effect of treatment on
the proximal response is robust to
interactions and to delayed effects (e.g.,
burden)
3) Secondary data analyses concern time
varying effect moderation and data
analyses to construct data-driven
decision rules for the JITAI
![Page 33: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/33.jpg)
33
Micro-Randomized Trials: When are
they (not) useful?
• NOT USEFUL: When malleable circumstances are
rare: Want to learn the best type of alert to prevent
suicide attempt
• USEFUL: When malleable circumstances occur
frequently: Stress, urges to smoke, adherence,
physical activity, eating
• NOT USEFUL: Proximal response cannot be
feasibly assessed.
• USEFUL: Proximal response can be unobtrusively
sensed or unobtrusively self-reported.
![Page 34: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/34.jpg)
Practice designing a
micro-randomized trial!
Some Questions:
1. Which treatments do you want to (micro-) randomize?
2. How long should your study last?
3. How many decision points are there per day?
4. On average, how many treatments do you want per day?
5. What will determine availability in your study?
6. What level of availability might you expect?
7. What should your randomization probabilities be?
8. What might the time-varying main effect look like?
34
![Page 35: JITAI Development inpeople.seas.harvard.edu/~samurphy/seminars/OBBI 05.19.16Session… · 05.19.16 HeartSteps JITAI JITAIs JITAIs JITAIs. 2 Overview • Session 1: Introduction to](https://reader034.vdocuments.mx/reader034/viewer/2022050117/5f4e0237558aa1349f08f141/html5/thumbnails/35.jpg)
35
Collaborators