This AI model can predict mood episodes using wearable devices

A new model has emerged from a collaborative research effort, promising to transform the way mood episodes in mood disorder patients are predicted. Led by Chief Investigator KIM Jae Kyoung of the IBS Biomedical Mathematics Group and Professor LEE Heon-Jeong of Korea University College of Medicine, the team developed a method that relies solely on sleep and circadian rhythm data collected through wearable devices. This innovation reduces the cost and complexity of mood prediction while enhancing its feasibility for real-world application.

How does it work?

The model harnesses the data captured by wearable devices like smartwatches, focusing exclusively on sleep-wake patterns and circadian rhythms. Over a span of 429 days, data from 168 mood disorder patients were analyzed to extract 36 unique features related to sleep and circadian behavior. These features were then applied to machine learning algorithms, achieving exceptional predictive accuracy for various mood episodes:

• Depressive episodes: AUC 0.80
• Manic episodes: AUC 0.98
• Hypomanic episodes: AUC 0.95

Daily variations in circadian rhythm emerged as key predictors. Delayed circadian rhythms were linked to a higher risk of depressive episodes, while advanced rhythms increased the likelihood of manic episodes. This focus on specific sleep-wake dynamics simplifies data collection without compromising accuracy.

Why does it matter?

Mood disorders are often tied to disruptions in sleep and circadian rhythms. While previous models required diverse and often expensive data inputs, this new model uses a single, accessible source — wearable devices.

As Professor LEE Heon-Jeong noted, "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."

The implications extend beyond prediction. With personalized sleep recommendations delivered via smartphone apps, patients can proactively manage their condition, potentially preventing disruptive mood episodes.

Chief Investigator KIM Jae Kyoung emphasized, "By reducing the cost of data collection, we have significantly improved clinical applicability."

The context

The popularity of wearable devices has grown exponentially, offering an unprecedented opportunity to collect health data seamlessly in everyday life. Yet, until now, the full potential of this data remained untapped in psychiatry. This study, published in *npj Digital Medicine* on November 18, sets a new standard for how wearable technology can integrate with medical care.

The research is a testament to interdisciplinary collaboration, combining expertise in mathematical modeling, machine learning, and psychiatry. Conducted by teams from KAIST, the Biomedical Mathematics Group at IBS, and Korea University College of Medicine, it exemplifies how cutting-edge technology and healthcare can converge to improve lives.

In a world where mental health challenges are on the rise, innovations like this represent a beacon of hope, enabling cost-effective and accessible interventions for mood disorder patients.