Can a Smartphone App That Includes a Chatbot-based Coaching and Incentives Increase Physical Activity in Healthy Adults?

Investigating Different Intervention Components of a Smartphone App to Promote Physical Activity: The ALLY Micro-Randomized Trial

The investigators conduct a micro-randomized trial to test main effects and moderators of three different intervention components of Ally, a mHealth intervention to promote physical activity that is offered to customers of a large Swiss health insurance. Interventions include the use of different incentive strategies, a weekly planning intervention and daily message prompts to support self-regulation. The Health Action Process Approach (HAPA) as well as principles from behavioral economics were used to guide the development of interventions. Further, sensor data is collected in order to enable prediction of latent contextual variables. These data can be used to build prediction models for the user's state of receptivity, i.e. points in time where the user is able and/or willing to receive, process and utilize the support provided. The results of this study enable the evidence-based development of a just-in-time adaptive intervention for physical activity.

Just-in-time adaptive interventions (JITAIs) have recently been proposed as framework for health interventions that exploit the potential of mobile health information and sensing technologies. By obtaining contextual information for example from smartphone sensors (e.g. location, time of day), a JITAI adapts the provision of interventions over time with the goal to deliver support when the person needs it most (state of vulnerability) and is most likely to be receptive (state of receptivity).

To facilitate the development of a JITAI for physical activity, the present study has the following objectives:

1. To quantify main effects and interactions of three intervention components of Ally, a mHealth intervention for physical activity.
2. To identify moderators for these intervention components to formulate evidence-based decision rules.
3. To train machine learning models that predict the user's state of receptivity

A micro-randomized trial design is used to meet the objectives of the study. Customers of a large Swiss health insurance company will use Ally over a 10-day baseline and a 6-week study period. During the baseline period, participants only have access to the dashboard of the app and no interventions are administered. During the intervention period, Ally provides daily personalized step goals and different interventions via an interactive chatbot interface based on the MobileCoach system (www.mobile-coach.eu). We investigate the following intervention components as between-subject or within-subject experimental factors during the intervention period: daily self-regulation coaching (two levels, within-subjects), a weekly planning intervention (3 levels, within-subjects) and different incentive strategies (3 levels, between-subjects).

Primary outcome will be the difference in achievement of the daily personalized step goal between intervention and control conditions for all intervention components. We expect all intervention components to increase the probability of goal achievement. Sensitivity analyses will be conducted for per protocol analysis and adjustment for covariates. Moderators of intervention components will be investigated exploratively.

To reach objective 3, we will collect a wide range of smartphone sensor data as well as usage logs of the Ally app throughout the study.

Inclusion Criteria:

• Possession of iPhone (5s or newer) or Android smartphone (Android 4.0 or higher)

Exclusion Criteria:

• not enrolled in a complementary health insurance plan
• actively using an activity tracker or comparable smartphone app
• working night shifts
• presence of medical condition(s) that prohibit increased levels of physical activity