A data-driven multi-fidelity simulation optimization for medical staff configuration at an emergency department in Hong Kong

Overcrowding at emergency departments in Hong Kong has been a critical issue for hospital managers recently. In this study, we focus on optimizing the medical staff configuration to alleviate overcrowding. According to the service requirements proposed by the Hong Kong government, 90% of urgent patients should receive treatment within 30 min. However, this condition is rarely satisfied in the practical situation. Therefore, we formulate the problem as minimizing the proportion of urgent patients that violate the service requirements while satisfying the service requirements of the other categories and cost constraints, thereby resulting in an optimization problem with a stochastic objective and several stochastic constraints. To solve this problem efficiently, we proposed a multi-fidelity simulation optimization framework containing a low- and a high-fidelity process. We utilize an evolutionary algorithm with violation-constrained handling assisted by a surrogate model as a low-fidelity process to shrink the solution space and generate an elite population. In the high-fidelity process, we exploit the optimal computing budget allocation method to identify the best solution in the elite population based on a data-driven simulation model. A case study is also discussed, and the results demonstrated that with a limited labor cost, there is a 52.05% reduction on average in the waiting time of urgent patients. Meanwhile, our proposed multi-fidelity simulation optimization framework proves to save 98.4% of the simulation time.