Smokers’ engagement with a context-tailored cessation smartphone app: real-time engagement, context effects and disengagement profile

Abstract

Background: Sensors on smartphones enable real-time behavioural support to be triggered and tailored to changes in a user’s situational context, representing a new type of online computer-tailoring. However, we do not know if users engage with such support in a suitably rapid timeframe or how this might be influenced by situational context. Methods: A single arm study of a smoking cessation app that delivers context tailored lapse and relapse prevention support (Q Sense). Smokers were invited to use the app before and 4-weeks after a quit attempt (N=42). Data from all app-generated notifications (notification type, location, context features, speed of viewing), including context-triggered support messages, were analysed using multi-level modelling (n=3,090). The disengagement profile was generated using the last formal engagement episode. Findings: Three participants withdrew. Of 3,090 notifications, 1,483 (48%) were engaged with (mean 38 per participant). For those context-triggered support message notifications engaged with (56%), the median speed of viewing was 4.5 minutes. The equivalent speed of viewing for schedule-triggered support messages was 24.2 minutes (p<0.001). Controlling for time and serial correlation, context features (home vs. work, situation entry vs. exit, low vs. high situational craving, time of day) were not associated with speed of viewing notifications. Median time to disengagement of using app was 25 days (IQR 7-41). Discussion: While not all context-triggered support messages were engaged with, most that are engaged with are viewed within 5 minutes. Context features do not appear to influence the speed to engagement, supporting the feasibility of this type of tailoring.