Summary
Applying AI to Support Operational Decision-Making
The Emergency Department Prediction project utilizes patient arrival data from the University of Kentucky’s Chandler and Good Samaritan hospitals. Through this data, CAAI can determine an accurate count of how many walk-ins the emergency departments (EDs) at each location see each day. The goal of the project is to accurately forecast the number of walk-in patients that will arrive at each ED in a given day, up to a week in advance. With these predictions, informed decisions could be made by hospital staff to determine, for example, how many transfer patients to accept on that day to avoid overwhelming the hospital or the staffing needs in the ED that day.
The forecasting is performed using a Temporal Fusion Transformer model architecture, which is a deep learning method specifically built for time series forecasting. This model can learn from previous trends in the data to create forecasts for future arrivals. It can use temporal relationships between the data to identify trends and patterns. For example, over the course of an average week, the number of ED arrivals will follow a regular pattern over each day of the week that can be modeled and forecasted into the future. Additional variables, called covariates, can also be used to improve the prediction accuracy. For example, we use local weather data, such as temperature and precipitation, as these factors have also been found to be associated with changes in ED arrivals. By leveraging these forecasting capabilities, the Emergency Department Prediction Project supports proactive resource planning in emergency healthcare settings.