Detecting transitions in bipolar disorder (BD) is essential for implementing early interventions. Our aim was to identify the earliest indicator(s) of the onset of a hypomanic episode in BD. We hypothesized that objective changes in sleep would be the earliest indicator of a new hypomanic or manic episode. In this prospective, observational, contactless study, participants used wearable technology continuously to monitor their daily activity and sleep parameters. They also completed weekly self-ratings using the Altman Self-Rating Mania Scale (ASRM). Using time-frequency spectral derivative spike detection, we assessed the sensitivity, specificity, and balanced accuracy of wearable data to identify a hypomanic episode, defined as at least one or more weeks with consecutive ASRM scores ≥10. Of 164 participants followed for a median (IQR) of 495.0 (410.0) days, 50 experienced one or more hypomanic episodes. Within-night variability in sleep stages was the earliest indicator identifying the onset of a hypomanic episode (mean ± SD): sensitivity: 0.94 ± 0.19
specificity: 0.80 ± 0.19
balanced accuracy: 0.87 ± 0.13
followed by within-day variability in activity levels: sensitivity: 0.93 ± 0.18
specificity: 0.84 ± 0.13
balanced accuracy: 0.89 ± 0.11. Limitations of our study includes a small sample size. Strengths include the use of densely sampled data in a well-characterized cohort followed for over a year, as well as the use of a novel approach using time-frequency analysis to dynamically assess behavioral features at a granular level. Detecting and predicting the onset of hypomanic (or manic) episodes in BD is paramount to implement individualized early interventions.