Harnessing smartphones for real-time detection of affective disturbance and future depression risk in adolescents - Project Summary/Abstract Negative affective (NA) states (e.g., high sadness, anger, and anxiety) increase substantially during adolescence, which may heighten risk for the onset of affective disorders, in particular major depressive disorder (MDD), which surges during the adolescent years. Over the past decade, affective disturbances and MDD have been rising in adolescents, and the COVID-19 pandemic has only exacerbated this alarming trend. As a result, the Surgeon General and national pediatric organizations (American Academy of Pediatrics, American Academy of Child and Adolescent Psychiatry, and Children’s Hospital Association) recently declared a national state of emergency for youth mental health. Accordingly, there is an acute need to develop personalized data-driven approaches to predict and ultimately interrupt states of markedly high NA as they occur in the daily lives of teens. In addition to the immediate benefits of alleviating acute states of affective distress, reducing the frequency and duration of episodes of high NA may serve to reduce the risk of depression onset in youth. Relevant in this context, the majority (88%) of U.S. teens own a smartphone, which can continuously and unobtrusively measure behaviors predictive of affective disturbance, including activity levels, location, phone use, sleep, and proxies of social interaction. In addition, smartphone data may also predict risk of MDD onset. The ability to prospectively predict MDD prior to its onset would have important clinical implications for the early identification of – and targeted deployment of interventions for – at-risk youth, which is strongly aligned with the NIMH Strategic Plan. To address these gaps, adolescents ages 12-16 (the age range corresponding to the largest developmental increase in depression) will complete repeated ecological momentary assessment (EMA) surveys of NA (i.e., assessing different negative emotional states) over 30 days. During this period, smartphone sensors and a wrist-worn actigraphy band will collect data on activity levels, location, phone/screen use, calls/texts and estimates of relevant sleep variables (e.g., sleep onset, offset, and duration). The project has two aims. First, a personalized machine learning approach recently developed by the study team will test the accuracy of predicting states of high NA in the daily lives of teens from these passively derived features (Aim 1). The ability to accurately predict states of high NA at the individual level could ultimately inform the development of highly scalable and personalized smartphone-delivered interventions matched to the current affective state (e.g., high sadness vs. anger) of a given teen. Second, during a follow-up phase, participants will be contacted every 6 months to assess changes in symptoms, along with bursts of passive sensor data collection and EMA. Machine learning analyses will test whether passive data, in combination with affect dynamics, predict subject-specific risk of future depression onset with sufficiently high sensitivity and specificity to be clinically useful (Aim 2). To the extent that a data-driven approach could be developed to predict individual risk of future depression onset, it could ultimately inform the development and delivery of individualized, targeted, and timely prevention efforts.
