Predicting Emotional Exhaustion with Multimodal Sensor Data During Return-to-Work Trajectories: A 6-Month Longitudinal Study

PURPOSE: Emotional exhaustion is a core component of burnout and impacts return-to-work trajectories following sick leave due to burnout or stress. Previous research identified links between emotional exhaustion and sleep, physical activity, mobility, and smartphone usage, yet little is known about how these associations vary within and between individuals over time. This study examined how emotional exhaustion can be monitored and predicted using multimodal data from smartphones and smart rings during return-to-work trajectories. METHODS: Eighteen employees on sick leave in the Netherlands due to stress or burnout symptoms were recruited via occupational physicians. For six months, participants completed daily Ecological Momentary Assessments (EMAs) of emotional exhaustion and provided multimodal sensor data on sleep and physical activity (Oura ring), mobility, and phone usage patterns (Avicenna app). Between- and within-person associations were examined using correlational analyses and linear mixed models. Subject-dependent Random Forest (RF) models were trained to assess the predictive performance of multimodal features. RESULTS: Subject-dependent RF models achieved an average Spearman's ρ̄ of 0.38 (range: 0.10-0.72). Sleep and physical activity features showed more consistent associations with emotional exhaustion than mobility and smartphone usage patterns, which were more heterogeneous. The strength and direction of associations, as well as the most predictive features, varied substantially between participants. CONCLUSIONS: Multimodal device data can modestly predict emotional exhaustion, with performance varying by individual. Results indicate that emotional exhaustion patterns are highly variable among employees, necessitating individualized approaches to return-to-work counselling. Future research should incorporate longer monitoring periods and examine inter- and intra-individual variability.