Use of Generative Artificial Intelligence for Development of Plain Language Summaries: A Blinded Assessment of Education Preferences of the UK Sickle Cell Disease Community

Objective: Generative artificial intelligence (GenAI) supports medical content development, including plain language summaries (PLS), thereby supporting accessibility of education. We examined preferences of PLS format created from a study assessing UK sickle cell disease (SCD) community social media conversations. Methods: Three written PLS were developed, human-written, AI-generated, and hybrid AI-human (person with SCD edited AI version for understandability). Videos were developed of each written PLS using the AI tool Synthesia. Adults (n = 93) with SCD and carers completed a survey assessing understandability of each written PLS and preference for written versus video. Participants were blinded to PLS source. They were presented at random with a PLS, rating understandability on a 5-point scale. They then answered multiple-choice questions to assess understanding, rated the other two written PLS and ranked each in order of understandability. Finally, participants watched the video of their top-ranked written PLS and stated their preferred format. Results: There was no significant difference (p > 0.05) in participant-rated understandability for the first PLS viewed: mean (standard deviation) 3.9 ± 1.1 (human-written), 4.1 ± 0.9 (AI), and 3.9 ± 0.8 (AI-human). Answers were similar for the multiple-choice questions regardless of which PLS participants saw first. Forty-one participants (44%) ranked the human PLS in first place for understandability, 31 (33%) the AI, and 21 (23%) the AI-human. Fifty-four participants (58%) preferred the video PLS, 27 (29%) preferred the written PLS and 12 (13%) had no preference. Conclusion: GenAI generates PLS as informative as human-written PLS, achieving similar readability scores for people living with SCD.