Designing healthcare strategies requires anticipating clinical and operational impacts before rollout, a task that usually relies on health‑system simulation. Such simulation demands a model of how patient states evolve and generate care demand. Conventional approaches tailor compact state transitions to a single policy question and ignore history‑dependent dynamics, forcing a redefinition of states and transitions whenever a new question calls for different clinical information. We propose to learn a patient world model from longitudinal electronic health records using a neural network that provides a shared clinical representation of patient trajectories, applicable across simulation scenarios. Using ICU operations as an example, we train the model on MIMIC‑IV, freeze it, and connect its time‑queryable predictions to shared ICU calendars, resources, and queues via scenario‑specific mappings, so a single model can generate admission demand, care tasks, and discharge proposals for simulations of bed capacity, care dispatch, and post‑ICU transfer capacity.
Experiments show that the model matches Markov and semi‑Markov baselines in aggregate occupancy and directional responses to capacity and dispatch policies, and it reproduces policy gains observed in replayed trajectories. For patients entering the same support state from different prior trajectories, the learned representation captures a 24‑hour recovery difference (‑6.4% predicted vs. ‑6.7% observed), a contrast missed by compact state models, which propagates into distinct simulated care demand, queue burden, and bed usage. These results indicate that a learned world model of patient dynamics can replace question‑specific state models, allowing new care policies to be evaluated by reconfiguring the scenario without rebuilding the model.
Review: The paper convincingly demonstrates the feasibility and benefits of a deep‑learning‑based patient world model for ICU simulation, highlighting the critical role of history‑aware representations for accurate prediction and policy evaluation.