Evaluating a large language model for radiotherapy patient scheduling under realistic clinical and operational constraints: A feasibility study

BACKGROUND: Radiation Therapy (RT) scheduling requires coordination of clinical constraints and operational objectives. Large language models (LLMs) may provide a natural-language interface for generating candidate schedules. PURPOSE: To evaluate the feasibility of using an existing LLM to generate RT patient schedules under clinical and operational constraints through simulation and retrospective clinical comparison. METHODS: A one-year, three-linear-accelerator (LINAC) simulation included 1400 new patients across 12 treatment categories modeled after clinical practice. Structured prompts supplied ChatGPT with patient data, the current schedule state, clinical rules, and scenario-specific objectives. Performance was evaluated across scenarios addressing weekly treatment-time consistency, LINAC continuity, gap-constrained temporal relaxation, and infeasible request handling. LLM-generated schedules were benchmarked against heuristic and CP-SAT schedules under two objective settings. A six-week retrospective comparison evaluated 2043 appointments against a clinical schedule. RESULTS: 18.4 min while maintaining low overtime. In infeasible request tests, the framework identified predefined constraint violations and returned interpretable corrective suggestions. The optimization benchmarks also generated feasible schedules, but their results showed strong objective-weight trade-offs. In the retrospective comparison, all computational approaches scheduled the workload without overtime; the LLM strategies reduced mean gap by 50.8%-77.1% and increased time consistency from 86.7% to 99.5%-99.9%. CONCLUSION: Structured LLM-assisted scheduling was feasible in the simulation and retrospective clinical comparison, producing candidate schedules under predefined constraints. The findings support further investigation of LLMs as flexible, human-in-the-loop decision-support tools for translating clinical priorities into feasible candidate schedules.

Authors

Institutions

Publication Details

Journal
Medical Physics
Published
2026-11-18
DOI
https://doi.org/10.1002/mp.70767
Citations
1
Primary Topic
Healthcare Operations and Scheduling Optimization
Type
article
Field-Weighted Citation Impact
2.30
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Evaluating a large language model for radiotherapy patient scheduling under realistic clinical and operational constraints: A feasibility study

Michael B. Roumeliotis, Youfang Lai, Annette N. Souranis, Eric Zhang et al.
1 citations
Medical Physics
Healthcare Operations and Scheduling Optimization
2.30
article

Evaluating a large language model for radiotherapy patient scheduling under realistic clinical and operational constraints: A feasibility study

Michael B. Roumeliotis, Youfang Lai, Annette N. Souranis, Eric Zhang, Georgia Paparoidamis, Wen Li, Xun Jia
article en
1 citations

Abstract

BACKGROUND: Radiation Therapy (RT) scheduling requires coordination of clinical constraints and operational objectives. Large language models (LLMs) may provide a natural-language interface for generating candidate schedules. PURPOSE: To evaluate the feasibility of using an existing LLM to generate RT patient schedules under clinical and operational constraints through simulation and retrospective clinical comparison. METHODS: A one-year, three-linear-accelerator (LINAC) simulation included 1400 new patients across 12 treatment categories modeled after clinical practice. Structured prompts supplied ChatGPT with patient data, the current schedule state, clinical rules, and scenario-specific objectives. Performance was evaluated across scenarios addressing weekly treatment-time consistency, LINAC continuity, gap-constrained temporal relaxation, and infeasible request handling. LLM-generated schedules were benchmarked against heuristic and CP-SAT schedules under two objective settings. A six-week retrospective comparison evaluated 2043 appointments against a clinical schedule. RESULTS: 18.4 min while maintaining low overtime. In infeasible request tests, the framework identified predefined constraint violations and returned interpretable corrective suggestions. The optimization benchmarks also generated feasible schedules, but their results showed strong objective-weight trade-offs. In the retrospective comparison, all computational approaches scheduled the workload without overtime; the LLM strategies reduced mean gap by 50.8%-77.1% and increased time consistency from 86.7% to 99.5%-99.9%. CONCLUSION: Structured LLM-assisted scheduling was feasible in the simulation and retrospective clinical comparison, producing candidate schedules under predefined constraints. The findings support further investigation of LLMs as flexible, human-in-the-loop decision-support tools for translating clinical priorities into feasible candidate schedules.

Medical PhysicsVol. 53(12)
Johns Hopkins University (US), Instrumental Polymer Technologies (United States) (US)
Good health and well-being
Openalex Percentile: Top 18%
Healthcare Operations and Scheduling Optimization
2.30
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon