researchsquare.comSource publication:
Loading content…
Full Capacity Protocols in emergency departments usually start only after crowding thresholds are crossed, yet hallway beds, temporary nurses and transfer restrictions take time to arrange. A discrete-event simulation evaluation embedded a machine learning model forecasting waiting-room counts six hours ahead into the emergency flow and used those forecasts to trigger the protocol early. Over a simulated year, early activation significantly shortened time above the crowding threshold and average protocol duration at all three threshold settings, while changes in average emergency length of stay and wait time were not statistically significant.
The timing of Full Capacity Protocol activation moves ahead of crowding rather than reacting after waiting counts cross a threshold. Conventional practice triggers the protocol from near real-time patient-flow counts, so interventions often begin only after crowding has developed. Across a 365-day simulation with 10 independent replications, early activation significantly reduced the percentage of time above the crowding threshold and average protocol duration at all three threshold settings.
Early activation is driven by a short-horizon forecast of waiting-room counts that feeds directly into the simulation loop and decides when the protocol opens. Most prior prediction studies evaluated forecast accuracy without testing whether acting on forecasts changes downstream system behavior. The model predicts waiting counts six hours ahead from the most recent 48 hours of hourly counts; support vector regression with a radial basis function kernel reached an average mean absolute error of 2.46 patients and an average R² of 83.3% in five-fold cross validation.
Early activation shifts interventions from few and long to more frequent and shorter, with about ten times as many activations but each roughly 125 minutes shorter at the moderate threshold. The reactive strategy relies on fewer but longer activations because the protocol opens only after crowding has crossed the threshold. At the μ+2σ threshold, early activation cut time above threshold by about 5%; at the more sensitive μ+1σ threshold the reduction was about 21%, while differences between strategies narrowed at the restrictive μ+3σ setting.
The evaluation covers a simulation of one academic medical center emergency department in the southeastern United States, calibrated from 308,196 visits between January 2019 and July 2023 with the COVID-19 period excluded. Simulated and observed mean emergency length of stay differed by 0.21% (330.0 versus 329.3 minutes), with a two-sample t-test p value of 0.4848. The intervention bundle was held fixed at 10 hallway beds, five temporary nurses and cancellation of 3% of inbound transfers, with only the activation logic changed. The authors frame this as a preliminary step before real deployment, with a controlled pilot at the partner hospital as the next phase.
The results are presented as evidence of operational feasibility and potential impact, not as definitive real-world effectiveness. The decision logic assumes the protocol is implemented fully and immediately once activated, and communication, coordination and behavioral delays are not explicitly modeled. Sensitivity analysis shows the duration advantage of early activation is not uniform when activation delay grows, and high-delay conditions need more resources to keep the benefit. A single prediction model, single-institution parameters and a fixed intervention bundle limit extrapolation.