Can ChatGPT help patients understand radiopharmaceutical extravasations?

• Published in: Frontiers in Nuclear Medicine
• Main Author: Madeleine Alvarez
• Format: Article
• Language: English
• Published: Frontiers Media S.A. 2024-11-01
• Subjects: radiopharmaceutical extravasation; SNMMI; diagnostic imaging; medical event reporting; AI in healthcare; nuclear medicine procedures
• Online Access: https://www.frontiersin.org/articles/10.3389/fnume.2024.1469487/full

A previously published paper in the official journal of the Society of Nuclear Medicine and Molecular Imaging (SNMMI) concluded that the artificial intelligence chatbot ChatGPT may offer an adequate substitute for nuclear medicine staff informational counseling to patients in an investigated setting of 18F-FDG PET/CT. To ensure consistency with the previous paper, the author and a team of experts followed a similar methodology and evaluated whether ChatGPT could adequately offer a substitute for nuclear medicine staff informational counseling to patients regarding radiopharmaceutical extravasations. We asked ChatGPT fifteen questions regarding radiopharmaceutical extravasations. Each question or prompt was queried three times. Using the same evaluation criteria as the previously published paper, the ChatGPT responses were evaluated by two nuclear medicine trained physicians and one nuclear medicine physicist for appropriateness and helpfulness. These evaluators found ChatGPT responses to be either highly appropriate or quite appropriate in 100% of questions and very helpful or quite helpful in 93% of questions. The interobserver agreement among the evaluators, assessed using the Intraclass Correlation Coefficient (ICC), was found to be 0.72, indicating good overall agreement. The evaluators also rated the inconsistency across the three ChatGPT responses for each question and found irrelevant or minor inconsistencies in 87% of questions and some differences relevant to main content in the other 13% of the questions. One physician evaluated the quality of the references listed by ChatGPT as the source material it used in generating its responses. The reference check revealed no AI hallucinations. The evaluator concluded that ChatGPT used fully validated references (appropriate, identifiable, and accessible) to generate responses for eleven of the fifteen questions and used generally available medical and ethical guidelines to generate responses for four questions. Based on these results we concluded that ChatGPT may be a reliable resource for patients interested in radiopharmaceutical extravasations. However, these validated and verified ChatGPT responses differed significantly from official positions and public comments regarding radiopharmaceutical extravasations made by the SNMMI and nuclear medicine staff. Since patients are increasingly relying on the internet for information about their medical procedures, the differences need to be addressed.