A key advantage of mobile health (mHealth) is its capacity for in-situ data collection and intervention, which overcomes several problems of in-clinic assessment and treatment (e.g., recall bias, not in-the-moment) [8]. Mobile phones are also pervasive, especially among youth, as the majority of teens (73%) and emerging adults (85%) own a smartphone [30]. However, the promise of mHealth greatly depends on the quality of the collected data. Although sensor technologies have made some health data collection nearly effortless [10], many important health-related data can be only manually captured (e.g., food intake, stress appraisal, mood, substance use) [11]. Unfortunately, sustaining user engagement with manual data collection is a challenge because of high burden. Indeed, previous work has demonstrated that the 30-day retention rate for using health and fitness apps is only 47% where the usage is 2.7 times a week on average. [31]. Financial incentives are one way to engage participants; however, it may not be cost-effective to provide financial incentives for population-scale deployments. New research is required to find scalable solutions that can engage participants with little or no dependence on financial incentives.

Fortunately, engagement is closely related to motivation and persuasion [1]. Skinner’s pioneering work on operant conditioning [2] and schedules of reinforcement [3] demonstrate how rewards, and their frequency, can shape behaviors. Cialdini [4] showed how humans are influenced in a social setting. Recent work by BJ Fogg suggests that persuasion involves not only desirability of the incentives, but also the ease and accessibility of the task [5]. Although these theories give general guidelines to engage users, they do not directly translate to the mHealth setting. Furthermore, it is also not clear how these incentives interplay with each other over time and what adjustments are necessary for specific sub-populations in the mHealth setting.

To this end, we introduce SARA, a novel application that integrates various kinds of rewards to incentivize data collection. SARA stands for “Substance Abuse Research Assistant”. This first version of SARA is developed for adolescents and emerging adults (14–24 years) who report past-month binge drinking and/or marijuana use. This is a challenging population to engage because the transition from adolescence to adulthood is a developmental period characterized by increased novelty seeking [6], which is reflected in national data showing escalating rates of binge drinking and other drug use, which peak during emerging adulthood [30]. Furthermore, substance use frequently co-occurs with depression and other mental illnesses, which may further reduce motivation to engage [26].

In SARA, participants are asked to complete, every day between 6PM and 12AM, a survey with 7 questions about their emotions (e.g., stress level, mood, loneliness), hopefulness, and daily reflections (e.g., amount of free time, excitement of their day), as well as 2 active tasks [7] which measure their spatial memory and reaction time. SARA incentivizes its users in several different ways. First, SARA has an aquarium theme, and new fishes are awarded as participants complete surveys or active tasks (Figure 1). Sometimes SARA provides rewards in the form of memes and gifs. After participants log their data for a number of days, SARA gives visualizations of past data. SARA also provides some financial incentives, but the total financial incentive is minimal (nine dollars for a 30 day study). The key underlying idea to provide the multitude of different rewards is to provide enough novelty so that participants bear the burden of data collection.

In this paper, we give a detailed description of the SARA application. We then briefly discuss our future plans to conduct user studies with SARA and how the data from these user studies can inform the development of just in time adaptive interventions for engagement [9].

SARA is an Android application developed for youth and emerging adults at risk for substance use. The goal is to capture substance use data for extended periods of time in order to inform substance use and addiction science and treatment. The key innovation of SARA is the multitude of different rewards it provides to increase novelty and keep youth engaged in data collection [18].

These rewards are both culturally relevant and theoretically grounded. Also, to the best of our knowledge, SARA is one of the few systematic investigations of different forms of rewards to incentivize mHealth data collection. Therefore the lessons learned can provide guidance for increasing engagement in mHealth in general.

SARA is comprised of two modules: (1) a data collection module (2) an engagement module. In the following, we first briefly describe the data collection module. Then we discuss the engagement module, which contains different strategies to reward data collection.

SARA is implanted on the Android platform. The aquarium animations, badges, and other non-notification related reinforcement are implemented with a combination of Ionic and Phaser frameworks [32,33]. The location and activity sensing are implemented in native android codes. Finally, all the data stored in Amazon S3. The communication between Amazon S3 is encrypted with RSA 2048 and AES-256 in order to be HIPAA compliant.

We have already implemented an initial prototype of SARA and will be conducting a focus group and a field test of SARA in the summer of 2017. The goals of these studies are several folds:

The focus group study will be conducted before the field trial. The purpose of this focus group is to investigate the acceptance of SARA and gather feedback to improve SARA before the field trial. We will recruit undergraduate students between the ages of 18–24 old. 32% of the undergraduate students use alcohol or marijuana according to our data from past studies. We will recruit 25 students and divide the students up into 4 separate focus groups. Focus groups will be audiotaped, transcribed and analyzed for emerging themes by trained undergraduate RAs.

Following the focus group, we will improve SARA and then deploy SARA on a cohort of 60 adolescents and emerging adults. Participants will be recruited from the University of Michigan Health System’s Emergency Department and enrolled in the study if they meet the following criteria: are 14–24 years old, report past-month binge drinking (4/5 drinks in a day, tailored by gender) or recreational cannabis use, can download the app, and provide written consent/assent. Enrolled participants will use SARA for one month and complete a follow-up interview to provide user feedback.

This field trial data will provide early evidence of the efficacy of SARA. In addition, several engagement strategies (i.e., memes/gifs, life-insights, notifications) in SARA are randomized and we can thus analyze these data to measure their causal effects on increasing engagement. In other words, the analysis can inform the development of science for just-in-time engagement interventions. Also, the size of the subject pool is sized to detect small-standardized intervention effects (<0.2). We refer readers to the work on “micro-randomization” for details on this trial design and the analysis methods [21, 22].

Furthermore, the collected data can be used to train algorithms that provide engagement when they are based on the data. In other words, the dataset can be used to train policies. Policies are data-based decision rules that can dictate which engagement strategy to use in each context (such as stress level, mood etc.). Policy learning is a topic from the reinforcement learning literature and a full discussion is outside the scope of the paper. But, we refer readers to [23, 24, 25] for preliminary work on the application of reinforcement learning algorithms to adapt interventions in mHealth.