Evaluation of large language models as decision support tools for head and neck cancer management: A blinded multidisciplinary simulation study

Abstract

• LLMs generated guideline-concordant management recommendations for complex head and neck cancer scenarios. • Retrieval augmentation was associated with higher ratings for appropriateness and feasibility in this blinded simulation. • Eight blinded surgeons rated outputs across appropriateness, clarity, specificity, and feasibility domains. • RAG-enabled models demonstrated consistent performance across repeated generations. • Findings support LLMs as adjunctive decision-support and educational tools requiring expert oversight. The management of head and neck cancer relies on multidisciplinary expertise; however, access to tumor boards remains variable. Large language models (LLMs) may support guideline-based decision-making, although performance in complex oncologic scenarios is not well defined. Fourteen synthetic cases based on real tumor board encounters were evaluated. Five blinded comparator arms produced recommendations: a human expert, Non-RAG-GPT-4, Non-RAG-GPT-5, RAG-GPT-4, and RAG-GPT-5. Eight head and neck oncologic surgeons scored each recommendation for appropriateness, clarity, specificity, and feasibility using 5-point Likert scales. Paired permutation testing and inter-rater reliability were assessed. LLM outputs showed close alignment with expert recommendations. RAG-based models achieved the highest mean scores across domains, with some statistically significant differences versus the expert comparator in appropriateness and clarity; however, absolute differences were modest. Inter-rater reliability was strong (ICC 0.73–0.87). Advanced LLMs can generate guideline-concordant management recommendations in simulated head and neck cancer cases, supporting potential utility for decision support and education; prospective validation and expert oversight remain essential.

Ähnliche Arbeiten