Prediction of impending mood episode recurrence using real-time digital phenotypes in major depression and bipolar disorders in South Korea: a prospective nationwide cohort study

• Published in: Psychological medicine Vol. 53; no. 12; pp. 5636 - 5644
• Main Authors: Lee, Heon-Jeong, Cho, Chul-Hyun, Lee, Taek, Jeong, Jaegwon, Yeom, Ji Won, Kim, Sojeong, Jeon, Sehyun, Seo, Ju Yeon, Moon, Eunsoo, Baek, Ji Hyun, Park, Dong Yeon, Kim, Se Joo, Ha, Tae Hyon, Cha, Boseok, Kang, Hee-Ju, Ahn, Yong-Min, Lee, Yujin, Lee, Jung-Been, Kim, Leen
• Format: Journal Article
• Language: English
• Published: Cambridge, UK Cambridge University Press 01.09.2023
• Subjects: Algorithms; Bipolar disorder; Circadian rhythm; Circadian rhythms; Clinical interviews; Cohort analysis; Consortia; Depressive personality disorders; Disruption; Ecological momentary assessment; Emotional disorders; Heart rate; Hospitals; Impending; Mental depression; Mental disorders; Mood; Mood disorders; Original Article; Patients; Phenotypes; Predictions; Recurrence; Rhythm; Sleep; Smartphones; Symptom management; Wearable computers; South Korea; Circadian rhythms; prediction; wearable devices; machine learning; mood disorders
• ISSN: 0033-2917; 1469-8978; 1469-8978
• DOI: 10.1017/S0033291722002847

BackgroundMood disorders require consistent management of symptoms to prevent recurrences of mood episodes. Circadian rhythm (CR) disruption is a key symptom of mood disorders to be proactively managed to prevent mood episode recurrences. This study aims to predict impending mood episodes recurrences using digital phenotypes related to CR obtained from wearable devices and smartphones.MethodsThe study is a multicenter, nationwide, prospective, observational study with major depressive disorder, bipolar disorder I, and bipolar II disorder. A total of 495 patients were recruited from eight hospitals in South Korea. Patients were followed up for an average of 279.7 days (a total sample of 75 506 days) with wearable devices and smartphones and with clinical interviews conducted every 3 months. Algorithms predicting impending mood episodes were developed with machine learning. Algorithm-predicted mood episodes were then compared to those identified through face-to-face clinical interviews incorporating ecological momentary assessments of daily mood and energy.ResultsTwo hundred seventy mood episodes recurred in 135 subjects during the follow-up period. The prediction accuracies for impending major depressive episodes, manic episodes, and hypomanic episodes for the next 3 days were 90.1, 92.6, and 93.0%, with the area under the curve values of 0.937, 0.957, and 0.963, respectively.ConclusionsWe predicted the onset of mood episode recurrences exclusively using digital phenotypes. Specifically, phenotypes indicating CR misalignment contributed the most to the prediction of episodes recurrences. Our findings suggest that monitoring of CR using digital devices can be useful in preventing and treating mood disorders.