Generative AI for patient education in cancer care: A scoping review of evaluation practices and emerging trends

Generative AI (GenAI) tools, particularly Large Language Models (LLMs), are increasingly used across clinical contexts; including to support patient information needs. As these technologies become more prevalent, understanding their utilisation and evaluation in practice is critical. This scoping review aimed to map existing literature on GenAI applications in education for patients with cancer and identify trends in evaluation practices.

A scoping review was conducted following PRISMA-ScR guidelines. PubMed and Medline databases were searched for studies published between January 2019 and November 2024. Fifty-four eligible articles were analysed for GenAI models used, treatment modalities, education contexts, prompt sources, and evaluation domains and metrics.

Most studies (81.5 %) were published in 2024, with over half (55.6 %) originating from the USA. ChatGPT-3.5 and ChatGPT-4 were the most frequently used models. Decision-making and general disease information were the predominant education contexts. Evaluation of GenAI outputs was reported in 96 % of studies, with accuracy (61.1%), readability (42.6 %), and quality (29.6 %) as the most common domains. More than half (50.8 %) of evaluation metrics were custom scales, indicating limited use of standardised tools. Patient-centred frameworks were rarely applied.

GenAI shows promise in enhancing patient education for cancer care, but evaluation practices lack standardisation and cultural responsiveness. Future research should prioritise validated frameworks, patient-centred metrics, and prompt engineering strategies to ensure safe, equitable and effective integration of GenAI in clinical care.