Data-driven queueing modelling: a simulation case study of emergency department crowding

Comprehensive Summary

This study aimed to develop and validate a novel data-driven queueing model to analyze emergency department (ED) crowding and evaluate the impact of Unscheduled Care Services (UCS). The authors used real-world data from 34,358 ED visits at the Troyes ED in eastern France between October 2018 and May 2019. The queueing network model was based on statistical modeling of patient arrivals, care pathways, and station-specific departure rates, incorporating non-homogeneous Poisson processes for arrivals, logistic regression for patient routing, and machine learning-based Poisson regression for departure probabilities. Model validation demonstrated low bias and high accuracy in reproducing observed crowding measures, including occupation level, waiting time, and congestion, with mean absolute relative errors generally below 5% for aggregated measures. Simulations incorporating UCS demonstrated approximately a 5% reduction in overall ED occupation, along with measurable improvements in waiting times and staffing requirements across multiple ED stations.

Outcomes and Implications

Emergency department crowding is a complex phenomenon that cannot be adequately captured by single indicators such as occupancy alone and requires models that account for arrival patterns, patient flow, and resource allocation. This study demonstrates that a data-driven queueing methodology can quantify crowding dynamics and evaluate the impact of system-level interventions, such as Unscheduled Care Services, using routinely collected operational data. The model provides unbiased and accurate estimation of multiple crowding measures simultaneously, including occupation levels, waiting times, and congestion across different ED stations. The authors conclude that this approach offers healthcare regulators an alternative retrospective evaluation framework for assessing the effects of new healthcare service structures on emergency department crowding when controlled or randomized study designs are impractical.