.png){kind=icon}
.png){kind=icon}
Return
An LLM-based framework for event-driven operating room scheduling
.png){kind=icon}
PRE
AI
DOI：10.1080/00207543.2026.2728019.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Emergency operating room scheduling constitutes a complex online decision problem under uncertainty, characterised by stochastic arrivals and strict deadlines. While mixed-integer programming (MIP) offers optimality guarantees, it lacks real-time adaptability to exogenous disruptions. This study proposes a hybrid framework integrating long short-term memory (LSTM)-based forecasting, large language model (LLM) constraint synthesis, and exact MIP optimisation within a verifiable pipeline. Specifically, an LSTM model predicts patient arrivals, while a retrieval-augmented LLM translates unstructured operational updates into formal constraints validated via an intermediate representation. We evaluate the approach on an open dataset from a public emergency department in Iran. The adaptive use of short-horizon forecasts reduces median scheduling costs by 20.4% under high congestion and by 12.6% overall. In addition, a retrospective replay evaluation using ten historical emergency-surgery episodes from Ningbo First Hospital in China suggests that the framework can reduce the model-defined scheduling objective primarily by reducing latest-start violations while maintaining comparable waiting times. This work develops a framework for embedding adaptive prediction and verifiable LLM-based synthesis into exact online optimisation, supporting data-informed decision support for emergency operating-room scheduling.
Keywords:
Emergency department
online scheduling
long short-term memory prediction
large language model
mixed-integer programming
Applications in healthcare systems
optimisation
scheduling
Journal
IF:
7.3
Papers:
1.1W
Citations:
3.7W