Abstract: Operating rooms (ORs) serve as core resources in hospitals, with their utilization efficiency directly impacting patient waiting times and hospital operational benefits. This paper investigates the single-day joint scheduling problem for emergency and elective patients, considering the uncertainties in emergency arrivals and surgery durations. By constructing a joint ambiguity set incorporating patient feature and environmental variables based on machine learning, a distributionally robust optimization model is developed to minimize OR operational costs. The model aims to ensure the completion of elective surgeries while reserving reasonable OR resources for same-day emergency patients. The distributionally robust optimization model proposed in this paper can be equivalently transformed into a mixed-integer linear programming model with the same complexity as the deterministic model. Computational experiments based on the real operational data of hospitals show that, compared with traditional stochastic programming methods, the distributionally robust optimization method improves the solution efficiency. Although the average operational cost and overtime duration increase by 0.28% and 3.29% respectively, the performance indicators under the worst-case scenario improve by 0.94% and 18.13%, providing a more robust surgical resource allocation scheme for addressing uncertainties in healthcare service systems.

Key words: surgery scheduling, emergency arrivals, distributionally robust optimization, machine learning, healthcare operations management