Dynamics of data-driven microstates in bipolar disorder.

Area of Science:

• Psychiatry
• Computational Neuroscience
• Data Science

Background:

• Current bipolar disorder mood models are often discrete or continuous, relying on aggregate scores.
• Existing models have limitations in capturing the complex dynamics of mood states.

Purpose of the Study:

• To propose a novel framework combining discrete and continuous mood models for bipolar disorder.
• To utilize machine learning to detect subtle, individual mood patterns.
• To characterize the dynamic nature of mood through microstate transitions.

Main Methods:

• Developed a machine learning pipeline analyzing item-level assessment data.
• Constructed latent factors and clustered them into 'microstates'.
• Employed a discrete-time Markov chain to model transitions between microstates.

Main Results:

• Identified a key factor associated with irritability and aggression.
• Discovered hierarchical patterns within depressive and manic mood microstates.
• Validated findings using an independent dataset from a separate cohort.

Conclusions:

• The novel framework effectively captures the dynamic nature of bipolar disorder mood.
• Item-level analysis and microstate modeling offer deeper insights than aggregate scores.
• Results demonstrate the generalizability and validity of the proposed approach.