Physical analytics to model health behaviors

Abstract

Mobile phones are a pervasive platform for opportunistic sensing of social and health related behaviors. In this talk, I discuss how sensor data from mobile phones can be used to model and predict health outcomes. The talk starts with a review of research at the MIT Media Lab, and then transitions into how Ginger.io has built a commercial platform to collect, annotate, analyze and drive healthcare interventions at scale, deployed with major US hospital systems and healthcare providers. The Ginger.io three-part platform -- patient app, behavioral analytics engine, and provider dashboard -- applies this technology to give care providers a window into their patients' health between office visits. Our mobile app uses smartphone sensors to passively collect information about a patient's daily patterns. Using this data, our machine learning models are able to detect at-risk patients significantly better than the standard of care. Any concerning changes in behavior are communicated to the provider through our simple, action-oriented web dashboard. Ginger.io is part of the care solutions at institutions such as Kaiser Permanente, Novant Health, UCSF, Duke Medical and Cincinnati Children's.