Symposium 99 - Using Machine Learning in Scalable, Adaptive Digital Mental Health Interventions

Saturday, November 18, 2023

4:15 PM – 5:45 PM PST

Location: Chiliwack (402), Level 4

CE: 1.5

Keywords: Technology / Mobile Health, Treatment/ Program Design, Research Methods
Level of Familiarity: Basic to Moderate
Recommended Readings: Nahum-Shani, I., Smith, S. N., Spring, B. J., Collins, L. M., Witkiewitz, K., Tewari, A., & Murphy, S. A. (2018). Just-in-Time Adaptive Interventions (JITAIs) in Mobile Health: Key Components and Design Principles for Ongoing Health Behavior Support. Annals of behavioral medicine : a publication of the Society of Behavioral Medicine, 52(6), 446–462. https://doi.org/10.1007/s12160-016-9830-8, Figueroa, C. A., Aguilera, A., Chakraborty, B., Modiri, A., Aggarwal, J., Deliu, N., Sarkar, U., Jay Williams, J., & Lyles, C. R. (2021). Adaptive learning algorithms to optimize mobile applications for behavioral health: guidelines for design decisions. Journal of the American Medical Informatics Association : JAMIA, 28(6), 1225–1234. https://doi.org/10.1093/jamia/ocab001, Trella, A. L., Zhang, K. W., Nahum-Shani, I., Shetty, V., Doshi-Velez, F., & Murphy, S. A. (2022). Designing Reinforcement Learning Algorithms for Digital Interventions: Pre-Implementation Guidelines. Algorithms, 15(8), 255. https://doi.org/10.3390/a15080255, ,

Chair(s)

Miranda Beltzer, Ph.D. (she/her/hers)

Postdoctoral Research Fellow
Northwestern University Feinberg School of Medicine
New York, New York

Discussant(s)

AA

Adrian Aguilera, Ph.D.

Associate Professor
University of California, Berkeley
Berkeley, California

Presenter(s)

KM

Kevin Matos

Northwestern University, Feinberg School of Medicine
SM

Susan Murphy, Ph.D. (she/her/hers)

Professor
Harvard University
Boston, Massachusetts
Miranda Beltzer, Ph.D. (she/her/hers)

Postdoctoral Research Fellow
Northwestern University Feinberg School of Medicine
New York, New York
Jeremy Eberle, M.A. (he/him/his)

Graduate Student
University of Virginia
Charlottesville, Virginia

Although about 1 in 5 American adults experience symptoms of a diagnosable mental health condition each year, less than half of them receive treatment, and of those, few receive evidence-based care. Digital mental health interventions aim to improve population health and wellbeing by making evidence-based treatment more accessible, by addressing many of the barriers to care, including cost, stigma, geography, and provider time. However, building digital mental health interventions involves designing around many challenges, not least of which is tailoring the intervention to an individual and their circumstances. Because digital interventions are not bound to one session a week, they have the opportunity to reach people in their daily lives with an intervention tailored to their in-the-moment needs, and they have the ability to collect a variety of contextual data to inform what that intervention should optimally be. However, designing and implementing such just-in-time adaptive interventions (JITAIs) is challenging, and researchers use a variety of methods to make decisions about what to present to which users and when to present it. Here, we present four digital mental health interventions that use various forms of machine learning to make decisions to adapt to the user, with the goals of sharing methods for design and analysis, decision-making processes, lessons learned, and results with those building digital mental health tools and using similar methods in their work.

First, we will explore a human-centered design process that is laying the groundwork for the first (to our knowledge) dyadic JITAI that will use machine learning in delivering tailored chemotherapy adherence support to adolescents/young adults with cancer and their caregivers. The next two presentations will describe the development processes and early results for online reinforcement learning algorithms used to decide the framing, content, and timing of prompts to send to users in two different digital mental health interventions: one focused on reducing problematic cannabis use and one focused on depression and anxiety self-management skills. The last presentation describes the development and performance of a machine learning algorithm to predict adherence to a web-based cognitive bias modification intervention for anxiety, which is used to determine which users are assigned to supplemental coaching to improve engagement and adherence. Together, these presentations will take us from questions to consider early in the design process, through decisions made in selecting variables and defining algorithms, and finally through results on model performance. Our discussant, a leader in the field who works to make digital mental health interventions more equitable for diverse populations and who has extensive experience developing adaptive interventions, will synthesize presentations and provide commentary on their broader implications.

Learning Objectives:

At the end of the session, the learner will be able to:

Describe 3 advantages of adaptive digital mental health interventions.
Discuss 3 challenges in implementing machine learning algorithms in digital interventions.
Identify appropriate variables to include in machine learning algorithms.
Describe collaborative design processes for adaptive digital interventions.
Describe how adaptive digital interventions are being developed to address a variety of mental health conditions.