(PS8-77) Investigating the Link Between Sleep Variability and Affect Fluctuations: Computational Insights from Digital Phenotyping

Clinical assessment of mood and anxiety changes often relies on lengthy self-report scales, which can be burdensome and infrequent. Smartphones enable continuous collection of active and passive data streams, including daily mood fluctuations, sleep variability, home time, and screen duration. Integrating these digital markers with clinical symptom scores offers a scalable and potentially more valid approach to monitoring patient state changes with greater granularity than standard weekly assessments.

This study evaluates the feasibility and validity of smartphone-based digital phenotyping, focusing on the role of sleep variability in predicting fluctuations in affect over time. Using machine learning and anomaly detection algorithms, we assess the predictive power of passive sensing data—particularly by implementing a novel passive sleep algorithm—on mood and anxiety fluctuations. Across multiple international datasets, our models predict daily and weekly symptom changes with AUC scores between 0.65 and 0.8.

Results demonstrate that passive measures, particularly sleep variability, are critical predictors of symptom change. Findings are replicated across different patient groups, countries, and clinical settings, highlighting the generalizability of this approach. This work underscores the potential of computational methods combined with digital phenotyping to enhance early detection, monitoring, and personalized interventions for mood and anxiety disorders.