A carregar...
Publicação
Breast cancer patients’ questions about radiotherapy-induced skin toxicity: a comparative analysis of ChatGPT-4.0 and Gemini 2.5 flash
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 817.94 KB | Adobe PDF |
Orientador(es)
Resumo(s)
Introduction: Artificial Intelligence (AI) is increasingly being deployed in health communication, a trend particularly visible in specialised domains, such as breast cancer radiotherapy. The subsequent clinical and educational value is fundamentally determined by the reliability and clarity of the responses generated. However, the comparative performance of different AI models in addressing patient concerns about radiotherapy side effects remains unclear, creating uncertainty regarding the optimal tool selection for patient education and support. This study aimed to compare the performance of ChatGPT 4.0 and Gemini 2.5 FLASH using a mixed-methods analytical-descriptive approach. Methods: Twenty-three unique questions, derived from a literature review and rephrased to simulate patient enquiries about the skin effects of radiotherapy, were submitted to both models. Sixteen expert Radiation Therapists (RTTs) independently assessed the responses using a seven-point Likert scale. The analyses included semantic cosine similarity and linguistic readability (Flesch Reading Ease/Flesch-Kincaid Grade Level), with statistical comparisons performed using Mann-Whitney tests. Results: Gemini 2.5 FLASH achieved higher median scores (6/7) than ChatGPT 4.0 (5/7), demonstrating particular strengths in clinical detail and empathy. Conversely, ChatGPT 4.0 produced more direct and structured answers, although it occasionally simplified complex concepts. Models showed low semantic similarity (median 0.78). Readability analysis revealed that ChatGPT aligned with an 8th-grade level, whereas Gemini operated at an 11th-grade level. Expert agreement was robust, with Gemini achieving greater consistency (α = 0.78; κ = 0.70) than ChatGPT (α = 0.72; κ = 0.65). Conclusion: Gemini was more effective for complex psychosocial issues, whereas ChatGPT provided more accessible guidance, necessitating ongoing professional validation for reliable integration into patient education workflows. Implications for practice: The use of AI in counselling patients undergoing breast radiotherapy enhances accessibility but requires ongoing professional validation to ensure clinical reliability.
Descrição
Palavras-chave
Artificial Intelligence AI Breast cancer Large language models Patient education Radiotherapy Skin toxicity ChatGPT Gemini
Contexto Educativo
Citação
Caetano M, Vicente D, Caetano L, Carolino E, Grilo A. Breast cancer patients' questions about radiotherapy-induced skin toxicity: a comparative analysis of ChatGPT-4.0 and Gemini 2.5 flash. Radiography (Lond). 2026;32(5):103469.
Editora
Elsevier BV