UCLA

Consumer wearable devices and smartphones are ubiquitous and generate valuable health-related data that remain under explored. These data have the potential to enhance our understanding of depression by bridging gaps left by traditional methods that often rely on retrospective self-reports. By leveraging machine learning, we identified relationships between specific passively measured behaviors and retrospective self-reports related to depression severity, reward functioning, and sleep quality. Focusing on sleep quality, our findings indicate that self-reported and physiologically measured sleep quality assess different constructs and offer distinct insights into depression. Anomaly detection (AD) methods were examined and aimed at identifying correlations between deviations from typical behavior, as recorded by mobile health (mHealth) devices, and changes in depression severity and symptoms. Although no significant relationship was found, the AD methods effectively detected multivariate anomalies, indicating potential applications beyond depression. Additionally, real-time data from wearable devices proved effective in detecting momentary reward functioning and affect, with models performing above random chance and performance varying across demographic and clinical groups. This dissertation highlights the importance of nuanced approaches in using consumer device-generated data to passively detect depression symptomology.