Capturing crisis dynamics: a novel personalized approach using multilevel hidden Markov modeling.

Area of Science:

• Psychology
• Psychiatry
• Data Science

Background:

• Crisis is a complex, multidimensional phenomenon with poorly understood temporal dynamics.
• Effective prevention requires a deeper appreciation of how crises evolve over time.
• Existing models lack granularity in capturing the interplay of factors during a crisis.

Purpose of the Study:

• To clarify crisis dynamics by clustering fluctuations in cognitive, affective, and behavioral (CAB) factors.
• To identify latent states representing different crisis trajectories within individuals.
• To inform the development of personalized crisis prevention strategies.

Main Methods:

• Utilized ecological momentary assessment (EMA) with frequent self-report questionnaires (3x daily).
• Collected data on five CAB symptoms from 26 patients over a prolonged period (60 measurements per patient).
• Applied multilevel hidden Markov models (HMM) to isolate crisis states and dynamics.

Main Results:

• Identified four distinct, ascending CAB-based crisis states.
• Found that remaining in the current state was most likely, decreasing with higher states.
• Observed significant patient heterogeneity in state transitions and persistence, with difficulty returning to lower states.

Conclusions:

• Multilevel HMM effectively quantifies and visualizes individual crisis trajectories.
• Patient heterogeneity highlights the need for personalized crisis prevention.
• Future statistical models should facilitate individualized approaches to mental health crises.