Abstract No. 202 Application of GPT-4 in Creating Lay Person Summaries of Interventional Radiology Procedural Reports in Multiple Languages

To explore the potential application of GPT-4 in the summarization of interventional radiology (IR) procedural reports in multiple languages to increase patient comprehension of their care and experience with IR. 200 reports dated 1/1/2022 - 1/1/2023 were sorted into 8 intervention categories and simplified using ChatGPT (GPT4 September 2023 version) in English using a specific prompt. These simplified English summaries were then translated into Spanish, Korean, Chinese, and Swahili using ChatGPT, after which all languages were evaluated by bilingual IR evaluators (n=8 total, 2 of each language) for usefulness, accuracy, misinformation, and critically wrong/missing information. For the 16 evaluated reports, mean original report Flesch-Kincaid reading level was 11.0 ± 1.2 which decreased to 7.1 ± 1.2 (p < 0.01) for the simplified reports in English. 95.3% of the simplified reports were strongly considered as "useful patient-centered representation" of the original report. No significant difference was observed in accuracy or critically incorrect information after translation. However, the number of misinformation instances per summary was slightly increased post-translation, predominantly due to a rise in misinformation in the Korean translations compared to the simplified English reports (Table 202.1). Our study demonstrates that GPT-4 successfully simplifies IR reports to below the 8th grade reading level of the average US adult and are patient-centered. These reports were successfully simplified and translated into multiple languages with few misinformation/critically wrong information. This shows potential in enhancing understanding and engagement of a diverse IR patient population.