The role of mobile sensing in behaviour change Q Sense; a ... · The role of mobile sensing in behaviour change – Q Sense; a context aware smoking cessation app Felix Naughton Behavioural

The role of mobile sensing in behaviour change – Q Sense; a context aware smoking

cessation app

Felix Naughton Behavioural Science Group

University of Cambridge

[email protected] @FelixNaughton

Collaborators Neal Lathia Sarah Hopewell Chloë Brown Jo Emery Rik Schalbroeck Cecilia Mascolo Andy McEwen Stephen Sutton

What is tailoring?

What is tailoring?

• Tailoring: support customised to individual using information about them

What is tailoring?


• Targeting: support customised to group based on shared characteristics

What is tailoring?


• Targeting: support customised to group based on shared characteristics

• Generic: one-size-fits-all

Generic leaflet (n=105)

Tailored leaflet (n=102)

Does tailoring increase effectiveness?

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Reported receiving leaflet Read at least once

Generic

Tailored

* p<0.05 ** p<0.01

*

**

Naughton et al, 2012 Nicotine & Tobacco Research

Does tailoring increase effectiveness?

0

1

2

3

4

Leaflet provided new info Written for me

Generic

Tailored

** **

* p<0.05 ** p<0.01

5-point rating scale

Naughton et al, 2012 Nicotine & Tobacco Research

Existing cessation apps


97.3%

2.7%

No tailoring

Tailoring

Hoeppner et al, 2015 Nicotine & Tobacco Research

% of cessation apps (N=225) that 'remembers' user input to tailor interactions





Examples of tailoring variables

• Demographics/social characteristics

• Clinical characteristics/symptoms

• Past behaviour

• Beliefs




• Past behaviour

• Beliefs

Relatively stable over short periods




• Past behaviour

• Beliefs

• What they are currently doing

• What is in their current environment

• How they are currently feeling





• Past behaviour

• Beliefs




Transitory and ever-changing





• Past behaviour

• Beliefs




characteristic tailoring

context tailoring

Impact of early lapse on relapse

Lapse in first week of quit

attempt (FU 6 months)




x5

Ashare et al, 2013, Journal of Addiction Medicine




x5


Induced lapse in first week

of quit attempt (FU 14 days)




x5


Induced lapse in first week

of quit attempt (FU 14 days)

x2

Shadel et al, 2011, Health Psychology

Episodic craving (cue induced craving)

Major cause of lapse

Shiffman et al, 1996 Journal of Consulting & Clinical Psychology; Ferguson & Shiffman , 2009, J Subst Abuse Treat





Implicated in 44% of lapses

Shiffman et al (1996)




Implicated in 44% of lapses

Shiffman et al (1996)


Half of lapses occur within 11 minutes of

episodic craving


Sense

Sense

SET QUIT

DATE

Sense

SET QUIT

DATE

Sense

SET QUIT

DATE

IF REPORTS > THRESHOLD THEN ACTIVE GEOFENCE

CREATED

Sense

SET QUIT

DATE

IF REPORTS > THRESHOLD THEN ACTIVE GEOFENCE

CREATED

Sense

AFTER QUIT

DATE

Sense

AFTER QUIT

DATE

MRC

framework

phase

Phase 0 Phase 1 Phase 2 Phase 3 Phase 4

Acceptability study

Intervention refinement

Theory and evidence generation

Intervention targets, modelling & barriers

Feasibility, acceptability & trial parameters

Definitive RCT Examine implementation in practice

Pilot/exploratory RCT

Q Sense development

Intervention development

Feasibility study

Effectiveness RCT

Theory

&

literature

MRC

framework

phase


Acceptability study







Q Sense development


Feasibility study

Effectiveness RCT

Theory

&

literature

Feasibility study

• Median time to report smoking 13 secs

• Underreporting on around half of days

• Reporting barriers – Forgetting

– Not wanting to appear rude

– Driving

– Relapse

Naughton et al, in press, JMIR mHealth uHealth

MRC

framework

phase


Acceptability study







Q Sense development


Feasibility study

Effectiveness RCT

Theory

&

literature

Acceptability study

• Objectives 1. Assess acceptability of Q Sense among target population 2. Estimate speed of engagement with geofence support 3. Estimate disengagement from app

• Design & methods • Mixed methods design (app data, follow-up survey & 1-to-1 interviews) • Smokers, receiving/motivated to receive cessation support (N=42)

– 55% female – 50% were over 35 years old – 74% smoked first cigarette after waking within 30 minutes

• Used app prequit (~7 days) and postquit up to 28 days • Follow up survey at 28 days post quit date (n=30 out of 42; 71%) • Purposive sample invited to interview (n=9)

% receiving geofence-triggered support

• 70% of those eligible (16/23)

1. Acceptability

17%

23%

76%

0% 20% 40% 60% 80% 100%

Negative reminder*

Privacy concerns

Use app again

Agree

Neutral

Disagree

* Subsample followed up who received geofence triggered support

1. Acceptability

“When [the messages] actually came through it was as if

the programme was written for me. Seriously that is what I did feel...because it was coming through at the times when I felt that I would have smoked and that’s when the support was there.” (ppt 1)

1. Acceptability

“When [the messages] actually came through it was as if

the programme was written for me. Seriously that is what I did feel...because it was coming through at the times when I felt that I would have smoked and that’s when the support was there.” (ppt 1)

“Some of [the messages] were useful and some of them seemed very daft. Yes. Some were very irrelevant to me personally, I thought.” (ppt 21)

Self-monitoring

Self-monitoring

“…inputting it I’d think, “Am I really that stressed? Am I

really that anxious?” (ppt 42)

Self-monitoring

“…inputting it I’d think, “Am I really that stressed? Am I

really that anxious?” (ppt 42)

“And that was the most important thing to start off with, is realising where in your day the pinch points were going to be and to sort of see a pattern of how much you were smoking and when you were smoking.” (ppt 24)

(more) self-monitoring

“Have a button, ‘I’m not smoking’…” (ppt 7)

“…it almost like solidifies your decision to not smoke whereas you might, if you didn’t have a button as it currently is, five minutes later you think, “Oh I still do want one” (ppt 24)

2. Engagement

• 2,879 interaction episodes (> 1 minute apart)

– Mean of 70 (SD 75) per participant

• Of 3,090 notifications, 1,483 (48%) engaged with

• Of 769 GF notifications, 432 (56%) engaged with

2. (Speed of) engagement


geofence

Geofence messages

Median time to response after geofence message notification (n=15) = 4.5 mins

79% viewed within 30 minutes

Median = 4.5 minutes


daily support geofence

Geofence messages Daily support messages

Median time to response after geofence message notification (n=15) = 4.5 mins

79% viewed within 30 minutes 54% viewed within 30 minutes

p<0.001

Median = 4.5 minutes Median = 24.2 minutes


MLM - Response time (DV), n=15 Estimate F p value

Time 0.02 4.83 0.029

AR1 rho (serial correlation) 0.217 Wald Z = 1.84

0.066

Fixed effect: GF situation (home vs. work) -0.92 0.02 0.882

Fixed effect: GF event (entry vs. dwell) -2.61 0.38 0.536

Fixed effect: Situational craving (low vs. high) 0.25 0.002 0.968

Fixed effect: Time of day -9.51 0.50 0.481

Response time Mean (SD) = 22.4 (37.3) mins Median = 4.9 mins

Response time Mean (SD) = 17.7 (33.7) mins Median = 3.8 mins

Home Work

3. Disengagement

• Last completion of an app survey or rating a message

3. Disengagement

Approx. end of automated support (38 days)

Median 25 days (IQR 7-41)


3. Disengagement


Shiffman et al, 2007 Drug & Alcohol Dependence

Summary

Summary

¾ would use Q Sense again

Summary


Over half of geofence messages engaged with, most

viewed within 5 mins

Summary


Over half of geofence messages engaged with, most

viewed within 5 mins

Desire for more self-monitoring

Summary 2

~ ½ of interaction episodes driven by

notifications

But only 10% of smoking cessation apps include any type of proactive notifications Hoeppner et al (2016) Nic Tob Res

Future steps

Future steps

“You can vote the messages can’t you? So that’s great as well because it means the very best and most popular ones come up first and foremost.” (study 2, ppt 24)

Future steps


Future steps


Future steps

AIRO wristband

Scholl & Van Laerhoven, 2012

Source: Qualcomm

http://somatixinc.com/smokebeat/

Thank you

Felix Naughton

Behavioural Science Group

University of Cambridge

[email protected] @FelixNaughton

Collaborators Neal Lathia Sarah Hopewell Chloë Brown Jo Emery Rik Schalbroeck Cecilia Mascolo Andy McEwen Stephen Sutton

Documents

The role of mobile sensing in behaviour change Q Sense; a ... · The role of mobile sensing in behaviour change – Q Sense; a context aware smoking cessation app Felix Naughton Behavioural