Computational Methods for Personalizing Mobile Health Interventions

The increasing use of smart devices such as smartphones and wearables has enabled new opportunities for health research that leverage rich multimodal, multiscale data streamed from embedded sensors to monitor and deliver timely interventions to patients when and where they need them most using recent computational advances in machine learning. This new approach to intervention provides more accessible, scalable, and cost-effective options to reach individuals. This relatively new intervention framework called just-in-time adaptive intervention (JITAI) aims to provide the right type/amount of support, at the right time, by adapting to an individual’s dynamically changing internal and contextual state. The success of JITAIs depends on accurate models for recognition of internal states such as an individual's emotional state and other contextual states relevant to health. This data can in turn be used to design an intervention policy that leads to improved user engagement, lower attrition rates, and lower symptom burden. In this dissertation, we propose multiple computational techniques that move us towards more personalized JITAIs for mental health.

To demonstrate the efficacy of our proposed computational approaches, we leverage real-world data from multiple mobile sensing studies from a population of college students to (1) personalize affect recognition for subgroups of individuals, (2) learn context-aware intervention policies for emotion regulation (ER), and finally culminating by combining approaches 1 and 2 into a (3) subgroup-based, context-aware intervention policies for emotion regulation. These methodologies contribute to a growing body of approaches that moves us closer to the realization of just-in-time interventions in mobile health.