ICU Delirium as a Failure of Predictive Synchronization: A Two-Agent Active Inference Model

Area of Science:

• Computational Psychiatry
• Artificial Intelligence
• Neuroscience

Background:

• Delirium is a common and serious complication in Intensive Care Units (ICUs).
• Current understanding often views delirium as solely a patient-specific condition, overlooking environmental interactions.
• The active inference framework offers a novel lens to model complex brain-environment dynamics.

Purpose of the Study:

• To present a computational model of delirium in the ICU based on the active inference framework.
• To conceptualize delirium as a relational phenomenon emerging from predictive failure between patient and environment.
• To investigate the role of environmental inference in the development and manifestation of delirium.

Main Methods:

• Developed a two-agent Partially Observable Markov Decision Process (POMDP) model.
• Agents represent the patient and the ICU room environment, each with generative models.
• Operationalized synchronization via directional surprisal metrics (SR→P and SP→R).

Main Results:

• Environmental (room) inference is critical for the synchronization-delirium relationship; its absence leads to high delirium rates.
• A non-inferring room environment results in a weak association between synchronization and delirium.
• Patient and environmental learning shape the phenotypic gradient of delirium, reducing ceiling effects.

Conclusions:

• Delirium is a relational phenomenon, not solely a patient-state, emerging from persistent environmental predictive failure.
• Environmental inference capabilities are essential for maintaining patient-environment synchrony and preventing delirium.
• The model provides a computational hypothesis for ICU delirium, suggesting implications for environmental design and monitoring systems.