Mood Prediction of Patients With Mood Disorders by Machine Learning Using Passive Digital Phenotypes Based on the...

Area of Science:

• Chronobiology
• Computational Psychiatry
• Digital Health

Background:

• Circadian rhythms are fundamental to human physiology and closely linked to mood disorders.
• Disturbances in circadian rhythms are strongly associated with mood disorders.
• Digital technologies offer new avenues for collecting data to understand and predict mood states.

Purpose of the Study:

• To evaluate mood states, activity, sleep, light exposure, and heart rate over two years using digital log data.
• To develop and investigate a machine learning-based mood prediction algorithm utilizing passive data phenotypes linked to circadian rhythms.

Main Methods:

• A prospective observational cohort study involving 55 patients with major depressive disorder (MDD) and bipolar disorder (BD I, BD II) over two years.
• Collected data included daily mood scores, light exposure, activity, sleep, and heart rate via smartphone apps and wearable devices.
• Developed a mood prediction algorithm using random forest, processing 130 features derived from circadian rhythm-based digital phenotypes.

Main Results:

• The mood prediction algorithm demonstrated accuracies of 65% for predicting mood states in all patients over the next three days.
• Accuracies for predicting specific episodes (no episode, depressive, manic, hypomanic) were high, reaching up to 94% for manic episodes.
• Bipolar II patients showed distinctively balanced and high prediction accuracies for no episode, depressive, and hypomanic episodes.

Conclusions:

• This study presents a robust model for future research by developing a mood prediction algorithm based on chronobiology principles and machine learning.
• The algorithm effectively processes and reclassifies digital log data, offering significant academic and practical value.
• The findings are expected to aid in improving the prognosis of mood disorder patients through practical clinical applications, leveraging the growth of digital technology.