A data-driven digital twin of emergency department flow to quantify congestion dynamics and test operational resilience.

Emergency department (ED) overcrowding is associated with delayed care, adverse outcomes, and reduced system efficiency. Conventional models often fail to capture the nonlinear and state-dependent dynamics of real-world ED operations. Digital twins, computational replicas that evolve with operational data, offer a promising framework for evaluating system resilience and testing organizational policies. To develop and validate a data-driven digital twin of ED upstream flow capable of reproducing congestion dynamics and quantifying the impact of operational interventions on performance and extreme delays. We used routinely collected data from 41,818 ED visits to build a closed-loop digital twin of patient flow. Process times were modeled using regularized regression and stochastic simulation with dynamic updates of system state. The model was calibrated on January-September data and evaluated out-of-sample on October-November visits. Counterfactual operational scenarios were then simulated. In the test period (7596 visits), median ED length of stay was 266 min, with 20.1% exceeding 8 h. The digital twin reproduced observed performance (simulated median 276 min). A 20% increase in arrivals produced nonlinear deterioration, with more than 90% of visits exceeding 8 h. Reducing post-triage waiting time produced the largest isolated improvement, lowering prolonged stays >8 h to 19.3%, while the combined intervention reduced them to 17.9%. A closed-loop digital twin can reproduce ED congestion dynamics and identify high-impact operational levers to support resilience planning and decision-making.