Solving operating room scheduling problems using large language model based multi-agent architecture

Efficient operating room scheduling is a complex task due to the diversity of surgical types, high patient demand, and limited hospital resources. Since OR operations account for a substantial share of hospital costs, effective scheduling solutions are critical. Large language model (LLM)-based multi-agent systems offer new opportunities to support optimization for this problem. This study evaluates an LLM-based multi-agent framework for generating and refining optimization code for OR scheduling. Leveraging iterative collaboration among specialized agents, the proposed iterative multi-agent (IMA) framework consistently outperformed both single-agent and traditional multi-agent baselines. Across 10 LLMs and 35 scenarios, IMA achieved an average feasible schedule rate of 82.91%, compared with 65.59% and 52.53% for the baselines. Experiments on 1050 synthetic cases and three real OR datasets further showed that IMA-generated code produced solutions close to those of expert-written mixed-integer programming models, while a case study of hospital as-is schedules demonstrated improvements in OR utilization and overtime reduction. These results suggest that LLM-based multi-agent systems can provide a practical and effective approach for OR scheduling optimization.