Using smartphone-based assessment of social experience to predict relapse in first-episode psychosis.

Social-affective changes are early indicators of psychosis relapse, yet their dynamic and subjective nature makes them difficult to capture between routine clinical assessments. Smartphone-based ecological assessments offer a method for capturing real-time social experiences in daily life. To address this gap, we examined whether social contact and social experience, measured through smartphone-based assessments, were associated with relapse following a First Episode of Psychosis (FEP), and evaluated whether smartphone data could predict relapse using machine learning. This longitudinal study included 256 individuals with FEP who completed at least one smartphone assessment, recruited from 10 Early Intervention Services across England and Wales between 2020-2023. Participants completed baseline clinical assessments and up to 30 days of daily smartphone surveys assessing social contact (alone vs with others) and social experience (a composite measure of emotional response to social contact). Relapse was assessed via structured interviews and electronic health records at 4, 8, and 12 months. Of 256 participants analysed (mean age 25.7 years [SD 5.3], 44.5% female), 8.2%, 13.5%, and 20.1% relapsed within 4, 8, and 12 months. Higher social experience consistently predicted reduced relapse risk across follow-ups, with similar effects for daily and momentary measures, while social contact showed inconsistent associations. Machine learning models using five smartphone assessments within 30 days achieved a validation accuracy of 0.790 (AUC 0.829) for predicting 12-month relapse, while data from a single smartphone assessment within 7 days achieved an accuracy of 0.715. These findings suggest that brief smartphone-based self-reports of social experience provide a low-burden, scalable tool for detecting early relapse risk and supporting more timely, personalised interventions in psychosis care.