This study presents a programmed discrete event simulation model of an emergency surgery department that utilizes patient data to predict schedule outcomes. Patient data, including parameters such as name, operation code, start time, ASA-class, BMI, and age, are used to generate process times and worker demands for each patient.
The study investigates and discusses the limitations of using patient data and regression algorithms for time prediction and by extension deterministic simulation outcomes, highlighting the low correlation between process times and patient parameters that according to work experience has the largest impact. Despite this limitation, a simple neural network is constructed and found to perform comparably to the current method of using a moving average for time predictions. However, significant performance improvements are needed for reliable and useful predictions.
To address the limitations of regression algorithms and deterministic simulation outcomes, the study suggests employing Monte Carlo simulations to generate outcome distributions to investigate the uncertainty of schedules. By analyzing variations in input and output distributions, valuable information about schedule uncertainty can potentially be obtained. The study emphasizes the importance of high-quality data, particularly regarding resource requirements in terms of worker hours and specific tasks performed on each patient, to improve the model's practical applicability.
David, Petr and Martin