Patient understanding of AI-simplified oncology imaging reports requires further validation

Ribeiro and colleagues offer timely evidence that large language models can make oncology imaging reports more accessible to radiologists and patient and public involvement representatives. Their findings also highlight an important distinction between a report that is easier to read and one that patients understand accurately. As the authors acknowledge, the patient-facing assessment included three representatives with previous experience in oncology imaging research and was restricted to reports that had received high factual-correctness scores from radiologists. This approach was suitable for comparing presentation preferences, although it provides limited evidence about comprehension in routine practice, particularly for lower-scoring or clinically ambiguous outputs. Further studies could examine unselected reports in patients with varied health and digital literacy, using outcomes such as comprehension, recognition of uncertainty, emotional response, and intended action. In future adequately powered studies, accounting for ratings clustered by reader and report may provide more precise estimates. Such evidence would help clarify the role of AI-generated explanations in patient-facing oncology care.