Classifying and clustering mood disorder patients using smartphone data from a feasibility study.

Area of Science:

• Digital phenotyping
• Computational psychiatry
• Mobile health

Background:

• Distinguishing bipolar disorder from major depressive disorder presents clinical challenges.
• Novel methods for monitoring disorder phenotypes are needed.
• Smartphone-derived behavioral metrics offer potential diagnostic insights.

Purpose of the Study:

• To assess the feasibility of using smartphone data for phenotyping bipolar disorder (BPI, BPII) and major depressive disorder (MDD).
• To evaluate the ability of digital phenotyping to differentiate between patient groups and controls.

Main Methods:

• A 12-week observational study involving patients with BPI, BPII, MDD, and healthy controls.
• Utilized the mindLAMP app on smartphones for collecting geolocation, accelerometer, and screen-state data.
• Applied machine learning models (random forest, logistic regression, k-means clustering) for data analysis.

Main Results:

• A random forest model achieved an AUC of 0.91 for classifying controls versus non-controls.
• A logistic regression model showed an AUC of 0.62 for differentiating MDD from bipolar disorder.
• K-means clustering yielded a silhouette score of 0.46 and ARI of 0.27.

Conclusions:

• Digital phenotyping demonstrates potential for clustering individuals with depression, bipolar disorder, and healthy controls.
• Current accuracy limitations necessitate additional data streams for clinical implementation.
• Further research is required to refine smartphone-based diagnostic tools.