|
Abstract:
|
When using smartphone sensor data for healthcare research, large amounts of data need to be converted to clinically meaningful outcomes, exposures, and behaviors. However, smartphone data collection poses three major challenges. First, large amounts of data are missing, potentially hampering measurement of outcomes, exposures and behaviors. Most sensors are battery-intensive, rendering continuous collection of data from smartphones infeasible. Second, different participants may use different phone types (e.g. Android or iPhone). Third, participants may use their phones in different ways (e.g., worn on-body, left in a bag or on a table). It is important that data processing algorithms provide reliable and valid results for different phone types and in different situations. In this session, Dr. Anna Beukenhorst will showcase statistical methods that address these three problems. She will showcase an open source software repository that contains methods for converting passively-measured smartphone data (e.g. GPS, accelerometry, communication logs) into clinically-meaningful exposures and outcomes.
|
