Crowd-Sourced Annotation of Longitudinal Sensor Data to Enhance Data-Driven Precision Medicine for Behavioral Health



 DESCRIPTION (provided by applicant): Longitudinal sensor data collected passively from mobile phones and other wearable sensors will transform behavioral science by allowing researchers to use "big data," but at the person-level, to understand how behavior and related environmental exposures impact health outcomes. Computers will analyze individual-level data streams to permit unprecedented, individual-level precision in research and intervention. This type of precision medicine enables targeting of science and medicine to a particular individual's genetic makeup, past and current situation, and behavioral health exposures. Mobile phones, smartwatches, and common fitness devices are already capable of generating rich data on behavior, but developing algorithms to interpret that raw data using the latest machine learning algorithms requires practical strategies to annotate large datasets. We propose to develop and test the feasibility and usability of a mobile and online crowdsource-based system for cleaning and annotating behavioral data collected from motion sensors, mobile phones, and other mobile devices. Our goal is to demonstrate how individuals playing mobile and online games - the "crowd" - can collectively, affordably, and incrementally clean and add important metadata to raw sensor data that has been passively collected from individuals, similar to that from population-scale surveillance studies (e.g., the National Health and Nutrition Examination Survey (NHANES) and UK Biobank) and those planned for studies such as the White House's Precision Medicine Initiative. The game-playing crowd will thereby dramatically improve the utility of the datasets collected for a variety of scientific studies. We will validate our prototye system on datasets collected from motion monitors used to study physical activity, sedentary behavior, and sleep, but we will demonstrate how the system could be extended for use on the increasingly rich datasets that are being collected with mobile devices and that include not only motion data, but also sensor data on location, light, audio, and person-to-person proximity. We will then refine the system, foster a community of crowd game players interested in citizen science, and release the source code to the system as an open source project so that other researchers can adapt the technique for their own work.