"Now You're Talking My Language" - Improving Health Literacy and Patient-Directed Knowledge of Scientific Abstracts Through Provision of Plain Language Summaries Created by Artificial Intelligence: A Cross Sectional Infodemiology Study.

Background:Due to free and digital availability of scientific abstracts in medical journals, as well as search engines including PubMed, many patients are increasingly looking to these as reliable and trusted sources of information, amidst an information ecosystem of potential mis- and disinformation. However, such scientific abstracts are difficult-to-read by the lay community, as they are not written purposefully for a lay audience. The Plain Language Summary now offers such readers a new medium to engage with, thereby helping with their health literacy and understanding of the research findings being described. The aims and objectives of the present study were to: calculate the readability of all scientific abstracts published in the Ulster Medical Journal over the five year period 2020 - 2024 (n=48), (ii) using artificial intelligence, prepare a plain language summary of each scientific abstract (n=48) with (a) minimal prompts and (b) with extensive prompts and (iii) calculate the readability of AI-generated plain language summaries. Methods:Readability was calculated using Readable software, defined by the (i) Flesch Reading Ease (FRE), (ii) Flesch-Kincaid Grade Level (FKGL), (iii) Gunning Fog Index and (iv) SMOG Index and four text metrics [word count, sentence count, words/sentence, syllables/word] on abstracts from all original clinical papers (n=48) published in the Ulster Medical Journal in the last five years (2020-2024). Plain language summaries were created from the existing scientific abstract using artificial intelligence with (a) minimal prompts and (b) extensive prompts. The readability of all AI-created plain language summaries was further determined. Results:Scientific abstracts had a mean FRE and FKGL score of 24.2±14.1 (standard deviation) and 14.4±2.8, respectively (Reference target values of ≥60 and ≤8, respectively). AI created plain language summaries with improved readability scores of 59.8±7.4 and 8.9±1.6, respectively for summaries with minimal prompts, thereby almost meeting reference readability targets. AI-created summaries with extensive prompts had mean readability scores of 71.3±6.1 and 6.3±0.9, respectively, with 46/48 (96%) of scientific abstracts now reaching reference readability target values. Scientific abstracts and Plain Language Summaries were statistically different (p<0.0001) in terms of both FRE and FKGL scores. Inputting the necessary and appropriate prompts to the AI-tool is critical to attaining the desired readability values. Conclusions:Medical journals may reach out to lay readers, including service users, patients, family and friends, through new innovation with the inclusion of a Plain Language Summary. Scientific abstracts are written at a level which is beyond the average reading age of 11 years old in the UK. Computational creativity through the employment of AI platforms can successfully generate narrative text for specific reading ages, with optimal readability. Effective communication of medical research findings from medical and scientific papers is vital for service users to enhance their health literacy, thereby helping promote better clinical outcomes, as well as promoting inclusivity for lay readers. With thorough checks and controls by the authors of clinical papers, AI-created plain language summaries may provide a new medium for medical journals to communicate with patients and service users, the results of clinical and original studies. The ability to create fit-for-purpose and easy-to-read Plain Language Summaries allows the lay public and service users to now become included in the family of readers of the journal and further supports the health literacy of patients and service users.