The research team led by Chief Investigator KIM Jae Kyoung (IBS Biomedical Mathematics Group, Professor at KAIST) and Professor LEE Heon-Jeong (Korea University College of Medicine) has developed a novel model that can predict mood episodes in mood disorder patients using only sleep and circadian rhythm data collected from wearable devices.

Mood disorders are closely associated with irregularities in sleep and circadian rhythms. With the growing popularity of wearable devices like smartwatches, it is now easier than ever to collect health data in everyday life, highlighting the importance of analyzing sleep-wake patterns for predicting mood episodes. However, existing models require diverse data types, making data collection costly and limiting practical application.

To overcome these limitations, the research team developed a model that predicts mood episodes using only sleep-wake pattern data. By analyzing 429 days of data from 168 mood disorder patients, the team extracted 36 sleep and circadian rhythm features. Applying these features to machine learning algorithms, they achieved highly accurate predictions for depressive, manic, and hypomanic episodes (AUCs: 0.80, 0.98, and 0.95, respectively).

The study found that daily changes in circadian rhythm are a key predictor of mood episodes. Specifically, delayed circadian rhythms increase the risk of depressive episodes, while advanced circadian rhythms increase the risk of manic episodes. This discovery opens new possibilities for tracking individual circadian rhythm changes to predict future mood episodes.

Professor LEE Heon-Jeong commented, “This study demonstrates the potential of using only sleep-wake data from wearable devices to predict mood episodes, increasing the feasibility of real-world applications. We envision a future where mood disorder patients can receive personalized sleep pattern recommendations through a smartphone app to prevent mood episodes.”

Chief Investigator KIM Jae Kyoung added, “By developing a model that predicts mood episodes based solely on sleep-wake pattern data, we have reduced the cost of data collection and significantly improved clinical applicability. This study offers new possibilities for cost-effective diagnosis and treatment of mood disorder patients.”

The results of this study are published online in npj Digital Medicine on November 18, presenting a new paradigm in the prediction of mood episodes.

Accurately Predicting Mood Episodes in Mood Disorder Patients Using Wearable Sleep and Circadian Rhythm Features, npj Digital Medicine (2024)

About the Research Team

The study was conducted by a collaborative research team from the Department of Mathematical Sciences at KAIST, the Biomedical Mathematics Group at IBS, and Korea University College of Medicine. The team has extensive expertise in mathematical modeling, machine learning and mood disorders.

Figure 1. Development of a mood episode prediction model using only sleep-wake data
The joint research team from IBS, KAIST, and Korea University has developed a new prediction model that overcomes the limitations of existing mood episode prediction models. While previous models required various types of data, the new model relies solely on sleep-wake wearable data.

Figure 2. Results of predicting mood episodes in mood disorder patients using sleep-wake data
The results of predicting mood episodes in mood disorder patients using sleep-wake wearable data. The model predicted depressive (left), manic (center), and hypomanic (right) episodes with AUCs of 0.80, 0.98, and 0.95, respectively.

Figure 3. Delayed and advanced daily circadian phases are linked with depressive and manic episodes, respectively
(Left) The likelihood of depression (red dots) increases as the circadian rhythm is delayed. (Right) The likelihood of mania (blue dots) increases as the circadian rhythm advances.

Figure 4. Researchers involved in this study
(From left) LIM Dongju (IBS/KAIST, co-first author), JUNG Jaegwon (Korea University, co-first author), LEE Professor Heon-Jeong (Korea University, co-corresponding author), CI KIM Jae Kyoung (IBS/KAIST, co-corresponding author).