Using Machine Learning to Optimize User Engagement and Clinical Response to Digital Mental Health Interventions

Summary

Digital mental health interventions are a cost-effective and efficient approach to expanding the accessibility and impact of psychological treatments; however, little guidance exists for selecting the most effective program for a given individual. In the proposed study, decision rules will develop for selecting the digital program that is most likely to be the optimal intervention for each user. These treatment recommendations can be implemented in the context of large healthcare delivery systems to improve the delivery of digital mental health interventions at scale. The overarching aim of the current study is to better understand for whom and how leading digital interventions work in a large healthcare setting. The study builds on the existing literature and follows expert recommendations by using machine learning (ML) methods to develop precision treatment rules (PTRs) for three leading digital interventions for emotional disorders (e.g., anxiety, depression, and related mental health disorders). Specifically, ML methods will be used to develop PTRs to optimize clinical outcomes and associated intervention engagement. This study will leverage a unique partnership between Boston University (BU), SilverCloud Health (SC)--a leading provider of digital mental health care--and Kaiser Permanente (KP)--one of America's leading health care providers. A clinical trial (RCT) will be conducted to evaluate the relative effectiveness of three distinct empirically supported digital mental health interventions (from SC's existing library of programs) in a sample recruited from KP primary care and other clinical settings. Data from this trial will be used to develop theoretically and empirically informed, reliable selection algorithms for managing treatment delivery decisions. Algorithms will be validated in a separate "holdout" dataset by examining whether allocation to predicted optimal treatment is associated with superior outcomes compared to allocation to a non-optimal treatment. The role of user engagement will be determined, and other mechanisms in treatment outcome.

Description