Uncertainty-Informed Decision Making for Just-in-Time Adaptive Interventions (JITAIs) - PROJECT SUMMARY/ABSTRACT The long-term objective of this K99/R00 application is to facilitate Dr. Asim Gazi’s development into a scientific leader in engineering and data science methods that enable personalized mobile health (mHealth) support during everyday life. The K99 phase of the project supports Dr. Gazi’s development in three key areas to facilitate his transition to scientific independence. First, Dr. Gazi will develop expertise in core topics relevant to the proposed K99/R00 research, including statistical machine learning methods for uncertainty quantification (UQ) and the behavioral science underlying smoking cessation and suicide prevention. Second, he will train in the research methods necessary to advance uncertainty quantification for machine learning in mHealth and reinforcement learning for just-in-time adaptive interventions (JITAIs), mHealth systems that intelligently intervene by adapting interventions to a patient’s state. Finally, his third training objective will be to develop professionally by growing his collaborative network and gaining experience in grant writing and academic leadership. These career development goals will be achieved while forming the foundation for a sustained line of research on uncertainty- informed decision making for JITAIs. Uncertainty-informed decision making for JITAIs requires (1) UQ algorithms to assess how confident a machine learning model is in its predictions of a patient states; and (2) uncertainty- informed reinforcement learning algorithms to incorporate these measures of confidence into a JITAI’s decision making (e.g., how should a JITAI intervene differently if suicidal risk is predicted with 51% confidence rather than 99% confidence?). The proposed research is divided into two aims accordingly. Aim 1 is to design, evaluate, and deploy UQ methods for prediction models that leverage passive biosensor data as input. Aim 2 is to design uncertainty-informed reinforcement learning algorithms that improve a JITAI’s efficacy by accounting for uncertainty in predictions of a patient’s state when intervening or interacting with the patient. The outcome of this research will be a set of algorithms that enable JITAIs to make uncertainty-informed decisions when adapting interventions and interactions with a patient to their predicted state. These algorithms will remove a significant obstacle in leveraging JITAIs to extend health care support outside the clinic in settings that are high risk or settings that would benefit from machine learning predictions of state. These settings include suicide and addiction, two behavioral health applications that will be investigated as part of this research. This project thus aligns with NIBIB’s mission to transform, through technology development, the ability to prevent and treat disease. The proposed research also fits within NIBIB’s Digital Health Program’s priorities and areas of interest in mHealth. Dr. Gazi will pursue these research and career development goals as a mentee of Dr. Susan Murphy and with the institutional commitment of the School of Engineering and Applied Sciences at Harvard University. This provides an ideal environment of resources and support to help him achieve his training and research goals.
