DIGITAL PHENOTYPING TO UNDERSTAND HEALTH-RELATED OUTCOMES: A SCOPING REVIEW

Abstract

Abstract Modeling passive data collected by smartphones, or digital phenotyping, may help detect the early signs of declines in daily health and lead to steps to prevent declines. Health-related outcomes have been typically acquired through self-report measures, however this broad approach has several limitations, such as recall and social desirability bias. A potential solution is digital phenotyping. Guided by the Arksey and O’Malley methodological framework, the aim of this scoping review was to identify studies that modeled passive smartphone-sensor data into behavioral markers that correlate or predict health-related outcomes. A literature search of peer-reviewed or conference studies in PubMed, Scopus, Compendex, and HTA yielded 3,170 articles. 40 studies met the inclusion criteria. Studies were organized and assessed for their: 1) data collection approaches (e.g., smartphone passive sensor data and questionnaire), 2) feature extraction (e.g., converting bedtime or wake time into sleep), 3) data analytics (e.g., identifying accelerometer data into sleep patterns using machine learning), 4) behavioral markers, which is observing the activity pattern for an extended period of time and then modeling the activities into appropriate behaviors, and 5) health-related outcomes. The findings of this study demonstrate how to transform passive smartphone-sensor data into behavioral markers that correlate or predict health-related outcomes, which could provide information for more grounded predictive use and successful implementation of digital phenotyping in future research.