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

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Digital interventions offer a highly scalable and relatively cost- and time-efficient approach to the delivery of accessible mental health services. However, evidence for efficacy comes from nomothetic group averages, overlooking the fact that a treatment that is effective for one patient may be less effective or even harmful for another. Further, guidance on matching individuals to their optimal intervention is lacking. These decisions are primarily based on clinical judgment or “trial and error,” which results in many patients receiving ineffective treatment or requiring multiple courses of treatment before achieving remission. Machine learning (ML) algorithms offer an alternative to conventional clinical decision-making by generating empirically derived precision treatment rules (PTRs) for selecting an optimal treatment. To date, research on the development of PTRs has been hindered by major design and statistical issues, including sample size limitations and lack of random assignment. The primary objective of the proposed study is to develop and test PTRs, using ML, for three evidence- based digital mental health interventions, within an existing digital healthcare system, SilverCloud Health (SC). A secondary objective is to better understand user-engagement as a mechanism of treatment response. In partnership with primary care physicians at Kaiser Permanente (KP), we will conduct a large (N = 1,800) randomized clinical trial where participants will be randomly assigned to one of three digital interventions in SC’s suite: Unified Protocol, Space from Depression, and Space for Resilience. Aim 1 will evaluate the overall effects and engagement patterns of the three digital interventions. Aim 2 will use ML to develop treatment- matching algorithms and determine the extent these precision treatment rules lead to improvements in clinical outcomes and engagement. Aim 3 will determine if user engagement and other common and specific factors (e.g., working alliance, negative thinking) are mechanisms of treatment response. The results of this study will provide a definitive answer regarding the relative effectiveness of three leading digital interventions, determine the value of developing PTRs for CBT interventions with different purported mechanisms of action, and further the understanding of common and treatment-specific mechanisms of change.