Matching patients to the treatment most effective for them can accelerate recovery and meaningfully reduce the growing burden of mental health conditions. Key barriers to tailoring care are the lack of objective data that can predict treatment response and effective approaches to translate data to improved clinical outcomes. As a result, many patients experience multiple treatment trials before recovery and a substantial proportion do not recover. The combination of mobile behavioral tracking and machine learning holds promise to overcome this barrier. Smartphones and wearable sensors can collect passive, continuous and objective measures and can be used to administer scalable, active behavioral tasks that capture constructs central to mental health. These highly dense data can be combined with genomics and clinical records, and machine learning holds promise to extract meaningful signals from these rich, multidimensional streams of information and facilitate the development of accurate predictive models. Our long-term goal is to increase the effectiveness of mental health treatments and the capacity of our mental health care system. Our objective in this application is to identify factors that can be used to effectively match patients to treatments. We will recruit 4,400 patients initiating outpatient mental health care in a network of primary and specialty clinics into the COMPASS Study (Comprehensive Mobile Precision Approach for Scalable Solutions in Mental Health Treatment) as part of the IMPACT-MH program. Subjects will be tracked through a wearable device and smartphone and complete active behavioral tasks. Because evidence-based digital interventions are increasingly widespread, patients will first be followed as they are randomized to receive one of two evidence-based digital interventions: cognitive behavioral therapy (CBT); or 2) mindfulness training. Subsequently, patients will be followed as they receive the array of treatments selected by their clinical teams. Our overarching hypothesis is that, through the use of mobile technology and machine learning in a large cohort before and during mental health care, we can develop individualized prediction models that will optimize mental health treatments. Our study is designed to test this hypothesis with the following specific aims: (1) Develop predictive models for personalized digital intervention treatment; (2) Develop predictive models for personalized, clinic-based mental health treatment; (3) Assess patient and clinician preferences for and perceptions of, the use of predictive modeling and behavioral tracking in mental health care; and (4) Actively participate in cross-IMPACT-MH project activities. Our approach is innovative because it applies scalable technology and analytic tools to a large and diverse sample of subjects receiving treatment under real-world conditions. Further, the project is designed to lead directly to an organization-level intervention that matches patients to treatments. Finally, this project is significant because it has the potential to greatly accelerate recovery by identifying the treatments from which each person is likely to derive the most benefit.

PROJECT NARRATIVE Efforts to address the high burden of mental health conditions have been hindered by the lack of data to match patients to treatments, resulting in delayed or insufficient care and a growing public health burden. The proposed research study will establish baseline factors and track behavior and clinical markers among individuals as they enter and progress through mental health treatment. Ultimately, this study seeks to develop scalable treatment decision tools that can be readily adopted and hasten recovery for the millions of Americans suffering from mental health conditions.