Novel Methods for Evaluation and Implementation of Behavioral Intervention Technologies for Depression



 DESCRIPTION (provided by applicant): Major depressive disorder (MDD) is projected to be a leading cause of burden of disease globally and in the United States. In the United States, in 2012 alone, an estimated 16 million adults aged 18 or older (7% of all adults) had at least one major depressive episode. While psychological treatments are effective at treating depression, the high prevalence of MDD makes it impossible to meet the needs in the population with standard one-to-one intensive psychological treatments. Behavioral intervention technologies (BITs) use technologies such as mobile phones to support behavior change to improve mental health, and have been shown to have similar effects to psychotherapy and pharmacotherapy. With the growing number of mobile phone users, BIT is a viable and promising option for delivering psychotherapy. On the other hand, the current evaluation framework of new interventions is not adequate for evaluating and implementing BITs, because of the rapidly evolving BIT landscape and the complexity of the interventions. This research aims to develop and validate novel concepts and evaluation framework to address these two challenges in the dissemination and implementation of BITs in MDD patients in pragmatic settings. We plan to achieve this research goal in four steps. First, we will develop a new statistical design, called open-ended adaptive randomization (OAR) procedure, which will enable us to continuously evaluate BITs that enter and leave a care delivery system. The OAR also aims to improve quality of care given to the participating patients, by sequentially allocating patients away from inferior BITs based on the interim evidence during deployment. Second, we will develop a data analytical technique, called regularized Q-learning, which will enable us to perform variable selection in high-dimensional settings and retain only the important predictors of health outcomes in the learning model. While the original Q-learning is a cutting-edge technique originating from the computer science literature, the research will extend its capability to handle high-dimensional data and enrich the learning model by incorporating regularized regression. Third, we will prepare for the next implementation phase of the proposed methods, by calibrating the methods with computer simulations, creating an initial knowledge base by analyzing data from current randomized clinical trials, identifying partnerships with healthcare providers and app curation plaftorms. Fourth, we will advocate for the general implementation of the proposed methods by producing publications, building cognitive computing systems, and tracking the source of citation and adoption of the published results by the broader health research community. Our long-term goal is to enhance our capability of deploying complex interventions such as BITs to depressed patients in a personalized and evidence-based manner throughout the healthcare system.