Novel wearable sensor calibration and validation for automated measurement of screen time in children - Abstract Over the past decade, the use of digital media and electronic screens has grown substantially. Existing methods for studying effects of digital media on child health and wellbeing are insufficient to assess the intermittent, on- demand, and interactive forms of media (e.g., tablets, smartphones) that are intermeshed within families’ daily activities. Calls for higher quality research have been made to better understand and evaluate the effect of electronic screens on children’s health outcomes. While scientific progress has been made with advanced analytics and data processing techniques using wearable devices for other health behaviors including physical activity, sedentary time, and sleep, insufficient research has been conducted on the development and calibration of wearable sensor measurement of electronic screen use. Wearable devices that include a color light sensor combined with advanced machine learning methods is an emerging and promising measure of electronic screen exposure in adults. However, there is a scientific need to extend this approach to free-living calibrations with natural observation and validation in children. The overarching aim of this project is to develop and validate a device-based measure of electronic screen use for children. The specific aims are to: 1) evaluate estimation accuracy of machine learning algorithms developed under controlled and free-play screen and non-screen activities using features extracted from a wearable multi-sensor, 2) compare accuracy of electronic screen use estimation across activities (e.g., watching tv, reading a book), screen type (e.g., TV, smartphone, tablet), body position (e.g., sitting, lying, standing), and ambient light level (e.g., florescent room light, natural light), and 3) assess wear method placement, day to day variability, and compare estimates with the established Comprehensive Assessment of Family Media Exposure (CAFE) tool obtained during 7 days of free-living wear. Our highly qualified research team will address these aims by using hybrid-structured and semi-structured activity observation to train and refine sensor-derived, machine-learned algorithms for assessing screen time as compared to direct observation. Additionally, a free-living calibration protocol will then be used to evaluate wearable sensor algorithms for the estimation of screen time in naturalistic settings. The results of this work will- -for the first time--provide an innovative and translatable approach to assess free-living electronic screen use in children.
