A0990
Title: Opportunities for causal inference and reinforcement learning in real-world interventional mobile health studies
Authors: Zhenke Wu - University of Michigan at Ann Arbor (United States) [presenting]
Abstract: Twin revolutions in wearable technologies and smartphone-delivered digital health interventions have significantly expanded the accessibility and uptake of personalized interventions in multiple domains of health sciences. For example, push notifications to promote healthy behaviors can be sent via mobile devices that are adapted to continuously collect information on an individual's current context. These time-varying adaptive interventions are hypothesized to lead to meaningful short- and long-term behaviour change. Key scientific questions in statistical terms will be formulated. However, standard assumptions such as non-interference and stationarity might be violated in real-world mobile health studies due to peer influence and long monitoring periods. We will present two methodological solutions, the first for estimating a new type of peer effects and the second for optimal policy learning under possibly non-stationary environments. We will use a multi-institution cohort of first-year medical interns in the United States to illustrate the ideas. We will also highlight that teams of engineers, and clinical and data scientists can collaborate to build statistical models that extract scientific insights from noisy and longitudinal interventional mobile health data.
