Early warning signals observed in motor activity preceding mood state change in bipolar disorder

• Published in: Bipolar disorders Vol. 26; no. 5; pp. 468 - 478
• Main Authors: Jakobsen, Petter, Côté‐Allard, Ulysse, Riegler, Michael Alexander, Stabell, Lena Antonsen, Stautland, Andrea, Nordgreen, Tine, Torresen, Jim, Fasmer, Ole Bernt, Oedegaard, Ketil Joachim
• Format: Journal Article
• Language: English
• Published: Denmark Wiley Subscription Services, Inc 01.08.2024
• Subjects: Affective disorders; Algorithms; Bipolar disorder; Homeostasis; Mood; mood disorders; Motor activity; nonlinear dynamics; recurrence; Statistical analysis; Statistics; systems analysis; Time series; unsupervised machine learning; recurrence; bipolar disorder; unsupervised machine learning; mood disorders; systems analysis; nonlinear dynamics; motor activity
• ISSN: 1398-5647; 1399-5618; 1399-5618
• DOI: 10.1111/bdi.13430

Introduction Alterations in motor activity are well‐established symptoms of bipolar disorder, and time series of motor activity can be considered complex dynamical systems. In such systems, early warning signals (EWS) occur in a critical transition period preceding a sudden shift (tipping point) in the system. EWS are statistical observations occurring due to a system's declining ability to maintain homeostasis when approaching a tipping point. The aim was to identify critical transition periods preceding bipolar mood state changes. Methods Participants with a validated bipolar diagnosis were included to a one‐year follow‐up study, with repeated assessments of the participants' mood. Motor activity was recorded continuously by a wrist‐worn actigraph. Participants assessed to have relapsed during follow‐up were analyzed. Recognized EWS features were extracted from the motor activity data and analyzed by an unsupervised change point detection algorithm, capable of processing multi‐dimensional data and developed to identify when the statistical property of a time series changes. Results Of 49 participants, four depressive and four hypomanic/manic relapses among six individuals occurred, recording actigraphy for 23.8 ± 0.2 h/day, for 39.8 ± 4.6 days. The algorithm detected change points in the time series and identified critical transition periods spanning 13.5 ± 7.2 days. For depressions 11.4 ± 1.8, and hypomania/mania 15.6 ± 10.2 days. Conclusion The change point detection algorithm seems capable of recognizing impending mood episodes in continuous flowing data streams. Hence, we present an innovative method for forecasting approaching relapses to improve the clinical management of bipolar disorder.