Emergency Room overcrowding is a pervasive problem worldwide, which impacts on both performance and safety. Staff are required to react and adapt to changes in demand in real-time, while continuing to treat patients. The multifactorial nature of the problem does not suggest a single solution. Previous studies found contributing factors included increases in elderly presentations, complex presentations, lowacuity presentations, lack of staff, reduced access to alternative services, and declining bed base. These factors point toward a local solution, which includes demand prediction, demand management and capacity management.
Demand management can take two forms. Firstly, patients can be provided with additional information that can support decisions about the most appropriate place to attend, based on their beliefs regarding the acuity of their condition, and knowledge of current wait times. Secondly, as per the aim of this paper, demand management can take the form of redirecting appropriate patients to alternative services as queues become unmanageable.
As one of the emergency room overcrowding solutions, this paper employs a case study to propose a hybrid application of discrete-event simulation (DES) and time-series forecasting across multiple centers in an urgent care network.
To support an integrated hybrid model, the DES model, which runs in minutes, has been built in AnyLogic, as the download and parser programs are written in Java. Due to the availability of time-series forecasting libraries in Python, the forecasting models use a Java/Python interface. The development of the integration framework, which brings together the near real-time data, forecasting models and real-time simulation, is a work in progress.
Patients can enter the model via the walk-in route or ambulance. A one-week hourly schedule of arrival rates (from 2018 data) defines entry into the model, and patients are allocated a severity level (triage category) on arrival, according to historical probability. Data for 2016/17/18 is stable for this distribution, and each triage category conforms with the overall daily arrival pattern. Patients are allocated a probability of X-Ray according to their triage category. Patients can be discharged home via any component part and the performance monitoring ‘clock’ stops for discharge home, or admission to the EAU, CDU or inpatient wards. We are working on the mechanisms to automate the emergency room model execution process such that as soon as new data is downloaded, it is parsed, the model variables are assigned relevant data items, and the execution of the emergency room overcrowding solution model starts.