Advanced forecasting of emergency surgical case arrivals: Enhancing operating room performance

Background and Objectives: This study, conducted at a major regional hospital in Australia, aims to enhance operating theatre performance by developing a two-step forecasting method for emergency case arrivals. By analysing data from 2018 to 2022, the study seeks to improve operating room efficiency and reduce cancellations through accurate predictions of emergency surgery demands. Methods: In the first step, several forecasting models, including Prophet, ARIMA, SARIMAX, LSTM, and AgentBased Simulation, were evaluated for their effectiveness in predicting daily emergency case arrivals. Each model was trained on 80 % and tested on 20 % of data to replicate real-world forecasting conditions. Performance was assessed using error metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE), along with the model's ability to capture monthly seasonality, general trends, and day-of-week patterns. The second step involved using a non-homogeneous Poisson process to provide more precise hourly forecasts for each day. Results: The SARIMAX model emerged as the most accurate, with the lowest error metrics (MAE: 1.01, MSE: 2.21, RMSE: 1.48), excelling in capturing seasonality, trends, and weekly patterns. It also demonstrated high robustness and scalability, making it the most reliable model. The non-homogeneous Poisson process provided precise hourly forecasts, further improving resource allocation and operating room scheduling. Conclusions: The two-step forecasting approach, particularly the use of SARIMAX and the non-homogeneous Poisson process, has the potential to significantly enhance operating room performance by reducing cancellations and improving efficiency. This research lays the groundwork for future advancements in operating theatre emergency management through data-driven decision-making.