Staging the Stagers: Clinicians vs. Large Language Models in Head and Neck Cancer Staging

Background:Accurate TNM staging is fundamental to treatment selection, prognosis, and multidisciplinary management of head and neck squamous cell carcinoma (HNSCC).However, staging requires integration of clinical examination, endoscopy, pathology, and imaging, creating potential for inter-rater variability.Large language models (LLMs) may offer a reproducible method of synthesizing these data, while retrieval-augmented generation (RAG) may further improve performance by providing models with relevant staging guidelines.We evaluated agreement in HNSCC staging among clinicians and LLMs and assessed the impact of RAG on LLM staging decisions. Methods:Fifty consecutive presurgical patients with HNSCC were independently staged according to the AJCC 8th edition by 13 raters: five clinicians of varying experience and eight LLM configurations.Models included Gemini, GPT-5.5, and Claude with and without RAG, as well