3 - (SYM 44) Using Passive Sensor Data to Enhance Mental Health Risk Assessment in Community-dwelling Adults

Mental health treatment has improved substantially in the last 50 years; however, best estimates suggest that only 40-50% of people who need mental health care are able to access it, and, for those who do get treatment, less than half receive an adequate dose (Reinert et al., 2021). Novel research methodologies such as digital phenotyping may help address these limitations by allowing us to identify individuals in need of mental health services before the severity of their symptoms impedes their access to treatment. Digital phenotyping refers to the application of computational techniques to passive sensor data (e.g., GPS, accelerometry, call/text logs) collected from personal digital devices (e.g., Smartphones, Smartwatches) in order to identify “digital biomarkers” that are relevant to mental health (Torous et al., 2016). Thus far, digital phenotyping studies have demonstrated that it is possible to identify and predict mental health events (e.g., manic episodes) in psychiatric samples (e.g., Gershon et al., 2016) and that passive sensor data can improve prediction accuracy beyond what is achievable with self-report measures (Auerbach et al., 2022), while incurring little additional burden to the individual. Therefore, as an extension of this existing work, we are investigating whether passive sensor data may also be used to improve the identification of community-dwelling individuals at risk for a future mental health event. In this presentation, we will describe an ongoing study that includes collecting medical records and self-reported psychosocial functioning over a 12-month period, as well as 7 days of daily diary reports, audio recordings, and continuous Smartphone passive sensor data from a sample of 250 community dwelling adults. We will particularly highlight innovative aspects of our study design and data sources that would likely be valuable for a wide range of applications. Some of the aspects we will highlight include using a stratified random sampling design grounded in a transdiagnostic operationalization of mental health risk, and deriving novel, sensor-based measurements of behavioral and contextual features such as time spent in different social and environmental contexts (e.g., bars, parks, houses of worship, etc.) by combining GPS data with place labels from opensource geographic information systems (e.g., OpenStreetMap).