Evaluating ChatGPT-4o and DeepSeek-R1 for Patient Education in Nasopharyngeal Carcinoma Radiotherapy: a Comparative...

Nasopharyngeal carcinoma (NPC) is one of the most common head and neck tumors and is particularly prevalent in certain geographical regions, especially Southeast Asia. Radiotherapy remains the cornerstone of treatment; however, patients often exhibit misconceptions due to limited health literacy, which may compromise treatment adherence and outcomes. Large language models (LLMs) provide a novel approach to patient education, yet their reliability and readability in the context of radiotherapy for NPC have not been systematically evaluated. In July 2025, we conducted a comparative evaluation of ChatGPT-4o and DeepSeek-R1 in addressing educational questions related to NPC radiotherapy. The DISCERN instrument was used to assess response quality and reliability, while text readability was measured using the Flesch-Kincaid Reading Ease Score (FRES), Flesch-Kincaid Grade Level (FKGL), and Coleman-Liau Index (CLI). Statistical analyses were performed using RStudio (v4.2.2). Both models achieved overall DISCERN scores of 51-62, indicating a "good" quality rating, with strengths in relevance and neutrality. However, deficiencies were noted in the areas of evidence currency, guideline references, and long-term side effects. DeepSeek-R1 demonstrated significantly higher readability compared with ChatGPT-4o, with a 25.6% reduction in FKGL and a 27% decrease in mean sentence length, making it more accessible for populations with limited health literacy. LLMs show substantial potential in supporting patient education for NPC radiotherapy, particularly by enhancing readability. Nonetheless, current models remain limited in terms of source transparency and completeness of clinical details. Future development should incorporate multimodal educational formats, real-time guideline integration, and structured output templates to further improve information reliability and patient support.