TLDR: A new study explores how passively sensed smartphone data and self-reported contexts influence user receptivity to adaptive mental health interventions. Researchers used a custom Android app, LogMe, with 70 students to deliver Just-in-Time Adaptive Interventions (JITAIs). Findings reveal that factors like time of day (evening best), battery status, screen interaction, physical activity (walking, eating), and location (academic buildings, sports regions) significantly impact whether users accept and can act upon mental health prompts. The study highlights the potential of using real-world contextual data to optimize the delivery of personalized digital mental health support.
The landscape of mental health support is rapidly evolving with the advent of mobile health (mHealth) technologies. These innovations allow for real-time monitoring and intervention, leveraging data from our smartphones. A particularly promising area is Just-in-Time Adaptive Interventions (JITAIs), which aim to provide personalized support precisely when and where it’s most needed, adapting to a user’s changing environment and emotional state.
While previous research has looked at how context influences responses to general notifications, less attention has been paid to how it affects engagement with actual mental health interventions. Crucially, understanding ‘receptivity’ – a user’s willingness and ability to engage with and act upon an intervention – has been a missing piece. This study delves into this concept, defining receptivity through two components: ‘acceptance’ (engaging with a prompt) and ‘feasibility’ (ability to act given the situation).
Researchers conducted a two-week study involving 70 university students using a custom-built Android application called LogMe. This app collected passive data from smartphone sensors, such as screen interactions, battery status, physical activity, app usage, call activity, and GPS location. Additionally, participants actively reported their current activity and social context. The adaptive intervention module within LogMe utilized a reinforcement learning algorithm called Thompson Sampling to optimize when and what interventions to suggest.
Key Insights from Passive Smartphone Data
The study explored how various passively sensed smartphone features influenced whether students accepted intervention notifications and how quickly they responded. It found that the time of day significantly impacted engagement. Completion rates (the percentage of interventions responded to) were higher, and response times were quicker in the afternoon and evening compared to the morning. This suggests that evenings might be a more opportune time for delivering mental health support.
Interestingly, while weekends might seem like a good time for interventions due to more free time, the study found lower completion rates during weekends. This could be attributed to a less structured routine, leading to missed notifications or delayed responses. Phone battery status also played a role: higher battery levels generally correlated with higher completion rates, and responsiveness decreased when battery levels were critical (below 10%) or low (10-20%). When phones were charging or full, completion rates dropped, possibly because users were away from their devices.
Device interaction patterns provided further clues. When the phone screen was on or the device was unlocked, participants were more likely to engage with notifications and respond faster. This highlights the importance of real-time device interaction as a signal for intervention timing. Physical activity, as detected by Google’s Activity Recognition API, also influenced receptivity. Walking was associated with the highest notification acceptance and quickest response times, while being ‘still’ (stationary) was linked to lower engagement, perhaps because users were focused on other tasks or had set their phones aside.
Analyzing app usage revealed that participants responded more quickly when using communication or social media apps (like WhatsApp or Instagram), but were more likely to ignore notifications when using productivity tools (like Google Docs). Surprisingly, response rates increased when participants were on a phone call, possibly because they were already actively engaged with their devices. Location data, derived from GPS, also offered valuable insights. Engagement was higher when participants were near academic buildings, aligning with the study’s design of sending notifications at the 55th minute of the hour, a common break time between classes. Conversely, engagement was lowest near sports facilities, likely due to physical activity and phones being put away.
Feasibility of Adaptive Interventions
Beyond just acceptance, the study also investigated the ‘feasibility’ of interventions, measured by an ‘average reward’ metric that combined acceptance and the ability to perform the suggested task. The findings showed that interventions delivered in the evening received higher average rewards, indicating greater willingness and feasibility at that time. This might be linked to mood fluctuations, where individuals might be more open to mental well-being support as the day progresses.
When participants actively reported their current activity, significant patterns emerged. The average reward was highest during exercise, followed by eating and walking, suggesting that interventions are more feasible during these activities. For example, a simple breathing exercise might be more readily accepted while walking than during a lecture. However, being in a ‘relaxing’ state showed the lowest feasibility. Interestingly, the social context (whether a participant was alone or with others) did not significantly influence their willingness to perform interventions.
The study also found that the highest average reward was observed when participants were in sports regions, possibly because they felt physically exhausted and found the intervention helpful. Academic locations also showed high feasibility, suggesting that the designed interventions, which were simple and quick, were acceptable in structured educational settings.
Also Read:
- Advancing Depression Assessment: A New Dataset and AI Reasoning Approach
- PrivCLIP: A New Approach to User-Controlled Privacy in Sensor Data
Designing Future Interventions
This research provides crucial insights for designing more effective and context-aware adaptive mental health interventions. By leveraging passively sensed smartphone data – including time of day, battery status, screen interaction, physical activity, app usage, call state, and location – developers can create intelligent models that predict optimal moments for delivering support. For instance, an intervention could be timed for the evening, when a user is walking, and their phone screen is unlocked, to maximize engagement and the likelihood of the intervention being both accepted and feasible. These findings pave the way for real-world, personalized mental health support that is not only timely but also actionable, ultimately contributing to improved well-being for students and potentially broader populations. For more details, you can refer to the full research paper here.
