Using Wearable Device and Machine Learning to Predict Mood Symptoms in Bipolar Disorder: Development and Usability...

Area of Science:

• Digital health
• Machine learning in psychiatry
• Biomarker discovery

Background:

• Bipolar disorder (BD) is characterized by recurrent mood episodes.
• Early detection and intervention are crucial for improving patient prognosis.
• Preventing mood symptom recurrence is a key clinical goal.

Purpose of the Study:

• To develop machine learning models for predicting bipolar disorder symptoms.
• To utilize digital biomarkers from wearable devices for symptom prediction.

Main Methods:

• Recruited 24 participants with BD.
• Collected digital biomarker data from wearable devices.
• Employed six machine learning algorithms to build predictive models.

Main Results:

• Depressive symptom prediction model: 83% accuracy, 0.89 AUROC, 0.65 F1-score.
• Manic symptom prediction model: 91% accuracy, 0.88 AUROC, 0.25 F1-score.
• Interpretable AI identified high resting heart rate, low activity, and poor sleep as potential predictors of depressive symptoms.

Conclusions:

• Digital biomarkers show promise in predicting manic and depressive symptoms in BD.
• This predictive capability can aid in early symptom detection and timely treatment.
• The models may help prevent mood symptom recurrence in bipolar disorder.