Improving Engagement and Efficacy of mHealth Micro-Interventions for Stress Coping: an In-The-Wild Study

TL;DR

This study tackles the persistence-accuracy trade-off in real-world mHealth stress interventions by proposing the Personalized, Context-Aware Recommender (PCAR), which leverages context, personality, and Last Switch Dependent dynamics to optimize when and what intervention to deliver. Implemented as a four-week, in-the-wild experiment with 29 parents of young children via a WhatsApp chatbot, PCAR is compared against random delivery and a no-intervention control, showing improved engagement and sustained short-term stress reduction, with one-minute interventions yielding statistically significant relief ($p = 0.001$). The work identifies transitional periods, such as shifts between daily activities, as optimal timing for interventions, and demonstrates long-term benefits via reduced Perceived Stress Scale (PSS) scores from $18.3$ to $16.0$, highlighting the practical potential for scalable, personalized stress coping support. Overall, the findings advance JITAI literature by providing an economical, context-aware scheduling approach and insights into how to maximize real-world adherence and efficacy of micro-interventions in parent populations.

Abstract

Sustaining long-term user engagement with mobile health (mHealth) interventions while preserving their high efficacy remains an ongoing challenge in real-world well-being applications. To address this issue, we introduce a new algorithm, the Personalized, Context-Aware Recommender (PCAR), for intervention selection and evaluate its performance in a field experiment. In a four-week, in-the-wild experiment involving 29 parents of young children, we delivered personalized stress-reducing micro-interventions through a mobile chatbot. We assessed their impact on stress reduction using momentary stress level ecological momentary assessments (EMAs) before and after each intervention. Our findings demonstrate the superiority of PCAR intervention selection in enhancing the engagement and efficacy of mHealth micro-interventions to stress coping compared to random intervention selection and a control group that did not receive any intervention. Furthermore, we show that even brief, one-minute interventions can significantly reduce perceived stress levels (p=0.001). We observe that individuals are most receptive to one-minute interventions during transitional periods between activities, such as transitioning from afternoon activities to bedtime routines. Our study contributes to the literature by introducing a personalized context-aware intervention selection algorithm that improves engagement and efficacy of mHealth interventions, identifying key timing for stress interventions, and offering insights into mechanisms to improve stress coping.

Figures (14)

• Figure 1: The framework developed to improve engagement with the stress-reducing micro-interventions: Mobile sensor data from the user is gathered via a specialized application and uploaded to a centralized database. A dedicated process analyzes this data to infer the most opportune moment for intervention, aiming to maximize user engagement. When the timing is deemed appropriate, a specific intervention is chosen, triggering a chatbot interaction with the user. User feedback is subsequently stored in the cloud for future analysis and refinement.
• Figure 2: Just-A-Minute Component: Timing Inference Pipeline. The pipeline is divided into two main processes: the Offline Process, which involves dual jobs for model retraining using aggregated user data, and the Real-Time Process, which ingests and encodes data. A decision to initiate a stress intervention is made based on the calculated likelihood of user acceptance and current budget constraints.
• Figure 3: Phase-wise User Cluster Differentiation: Initial allocation included 7 users in the control group and 21 in the random intervention group (Phase 1). After two weeks, 10 users remained in the random intervention group, while 16 were transitioned to PCAR intervention (Phase 2). 1 user didn't complete any phase and 2 users completed the first phase, totalling to 3 dropouts in the study.
• Figure 4: Stress Reduction Micro-Intervention Example Dialog: (a) Our system inferred an ideal time for a stress reduction micro-intervention and initiated a WhatsApp conversation. (b) If the user decides to accept the intervention and engage with our chatbot, he is asked whether it's a good time to intervene. (c) Once the user replies "Yes", he is asked for the current stress level. (d) Our chatbot suggests a stress-coping activity. (e) 10 minutes later, our chatbot asks again for the current stress level.
• Figure 5: Average Weekly Stress Reduction Across Phases and Groups: (a) This figure is divided into two sections, each corresponding to a different study phase as outlined in Section \ref{['method:Study_Phases']}. The y-axis shows the "reward", calculated as the difference between initial and subsequent stress levels on a 7-point Likert scale. The x-axis is divided into weeks. Error bars denote 95% confidence intervals. We see that for PCAR, the efficacy is maintained week over week while for the other settings, there is a decrease, especially in phase 2. We also observe a significant stress reduction within intervention groups. (b) This figure shows the change in the average reward of each group over the corresponding weeks within the phase.