Why the distribution matters: using discrete event simulation to demonstrate the impact of the distribution of procedure times on hospital operating room utilisation and average procedure cost

Background: Economic evaluations include estimates of surgical procedures costs that have usually been derived by allocating operating room (OR) costs in proportion to the average duration of different procedure types. However, ORs run with average utilisation below 100%, due to idle time between procedures and at the end of the day. Longer and less predictable procedures generate greater OR idle time, for a given tolerance of schedule over-runs. Estimates of surgical procedure costs that are based on average procedure duration alone as a measure of OR resource consumption will not capture the impact of the length and variability of procedure duration on OR idle time and capacity utilisation. Objective: To demonstrate how real-world OR scheduling practices lead to different levels of resource consumption than predicted by simple micro-costing approaches based on average procedure duration, and how those differences can vary between procedures with significantly different distributions of duration. Methods: We use a discrete event simulation model, calibrated with real-world data from a single surgical centre in Belgium, to compare simulated resource consumption, including idle time, for two alternative surgical procedures for ablation for atrial fibrillation. Results: We demonstrate that simple micro-costing approaches can under-estimate effective resource consumption between 31% and 48% for a procedure with long and unpredictable duration. For a shorter and more predictable procedure the under-estimate is only 15%. Conclusion: Simple approaches to estimating procedure costs may under-estimate resource consumption and do so in a way that is biased against technologies with shorter and more predictable procedure duration. For health technology decisions where a substantial part of costs are OR resources, a more sophisticated approach, taking account of the real-world implications of the distribution of procedure durations, should be used to avoid potential bias