Artificial intelligence-generated patient information leaflets: a comparison of contents according to British Association of Dermatologists standards

Abstract

Abstract Background Patient information leaflets (PILs) can supplement a clinical consultation and provide additional information for a patient to read in their own time. A wide range of PILs are available for distribution by the British Association of Dermatologists (BAD) and undergo rigorous review ahead of publication. In the UK, 7.1 million adults are estimated to have the reading age of a 9-year-old child and 43% are unable to comprehend written health information. Objectives To determine whether artificial intelligence (AI) can produce PILs that include a similar degree of content to current BAD PILs. Methods Using the AI tool ChatGPT, 10 PILs were generated, and their contents compared with those of existing BAD PILs using an author-generated list of commonly included themes. Omissions were noted and a repeat series of PILs generated using targeted request phrasing. The readability of AI-generated PILs was also analysed. Results AI-generated PILs were found to include similar factual content to BAD PILs but excluded information that was felt to be more pertinent to patient concerns such as curability and heritability. Targeted request phrasing saw AI generate PILs including this content. The readability of AI-generated PILs was beyond that of a large number of UK adults. Conclusions Where a condition-specific PIL is not readily available, an AI-generated PIL can provide relevant information of lesser quality than existing BAD PILs, which may be inaccessible to some patients. Specific caution is advised regarding AI-generated medication-specific PILs.