Summary: Researchers have developed a machine learning-based algorithm that uses sleep-wake data from smartphones and smart bands to predict mood episodes in individuals with depression and bipolar disorder, enabling early intervention and potential prevention of recurrences through lifestyle adjustments.
Key Takeaways:
- Irregular sleep patterns and lifestyle habits can contribute to mood episodes, but individuals may not easily recognize disruptions in their circadian rhythm.
- Unlike existing machine learning models that require multiple data types, this novel approach relies solely on sleep-wake history and prior mood episodes, making it more practical for real-world applications.
By leveraging sleep-wake data collected using a smartphone and smart band, researchers at the Korea University College of Medicine have developed a novel prediction algorithm that forecasts future mood episodes in people with depression and bipolar disorder. This could help patients with mood disorders prevent such recurrences by adjusting their lifestyle habits.
Irregular sleep and lifestyle habits can contribute to depressive and manic episodes, though individuals may not easily discern disruptions in their circadian rhythm. This makes predicting mood episodes a significant challenge.
Wearable devices offer a promising solution by facilitating the passive collection of circadian rhythm indicators, including sleep, heart rate, and step-count data. This data can be leveraged for mood episode prediction in patients with mood disorders. However, current machine learning-based prediction models often necessitate various data types, hindering their practical application in real-world settings.
Researchers, led by Heon-Jeong Lee, a professor of psychiatry at Korea University College of Medicine and the university’s vice president of research affairs, demonstrated in a new prospective observational cohort study that simple measurements of circadian rhythms using a smartphone and a smart band can predict the recurrence of mood episodes. This work was done in collaboration with the team of Jae Kyung Kim, PhD, at KAIST, Korea. Their work was recently published in npj Digital Medicine.
They propose a novel algorithm that predicts future depressive and manic episodes using only sleep-wake data collected easily using smartphones and wearables. Notably, it is also trained using the person’s sleep-wake history and prior mood episodes. The team collected longitudinal data from 168 patients and identified 36 sleep and circadian rhythm features through mathematical modeling.
Among them, daily circadian phase shift was found to be the most significant predictor, with delays associated with depressive episodes and advances to manic episodes. “We found that the features helped effectively predict depressive, manic, and hypomanic episodes for the next day, with the respective area under the curve values of the XGBoost machine learning model being 0.80, 0.98, and 0.95,” says Lee in a release.
This suggests that patients with mood disorders can receive early insights into their condition, enabling them to manage their habits and potentially prevent the recurrence of depressive and manic episodes.
“Our findings could lead to the development of digital therapeutics that help individuals become aware of changes in their circadian rhythm, enabling them to adjust their routines and mitigate mood episodes,” Lee says.
