Forecasting Hospital Bed AvailabilityUsing Simulation and Neural Networks

摘要

The availability of beds is a critical factor for decision-making in hospitals. Bed availability (or alternatively the bedrnoccupancy level) affects required staffing levels, scheduling of surgeries, patient acceptance decisions in emergencyrndepartments, and many other important hospital decisions. To better enable a hospital to make these decisions, thisrnpaper presents a forecasting methodology for predicting hospital bed availability based on the current status of thernhospital. A two-phased approach is used. The first phase involves building a computer simulation model of arnhospital, to determine the significance of various factors on bed availability. In the second phase, an artificial neuralrnnetwork (NN) is developed, using input variables and training data from the validated simulation, to create bedrnavailability forecasts. The NN features the use of time as an explicit input, to model the significant changes in thernsystem over time such as time dependant arrival and departure rates. An associated tolerance interval for thernforecasts is calculated to establish a measure of the quality of forecasted value. This method is applied to scenariosrnof varying forecasting time horizons illustrating the general effectiveness of the method for generating forecasts ofrnhospital bed availability.