Background/Case Studies: The application of artificial intelligence (AI) in the medical and scientific community has been increasing. However, there may be unfamiliarity and skepticism towards the use of AI, including highly specialized medical specialties including Transfusion Medicine, raising concerns about its potential and limitations. Appropriate uses of AI and knowledge of its limitations remain uncertain and are worth exploring.
Study
Design/Methods: Using the ChatGPT-4o model, we provided input for several example prompts. We assessed the ChatGPT responses for relevance, accuracy, and thoroughness, and to determine the value added for clinical practice and teaching. The specific queries were considered relevant for a typical day at an academic institution, ranging from immunohematology to therapeutic apheresis and transplantation.
Results/Findings: Summarized in Table 1, a total of seven prompts were inputted into ChatGPT with a broad range of clinical applications in Transfusion Medicine, including topics on blood donor screening infectious disease testing, basic blood banking, blood product transfusions, therapeutic apheresis, stem cell transplantation, and medications in coagulation. Regarding accuracy of responses to various transfusion medicine prompts, ChatGPT produced results with moderate accuracy with intermittent minor errors or omissions, including lack of source verification, failure to adjust for clinical nuances, and missing occasional critical details. Conclusions: Our analysis demonstrates areas of utility of AI in some areas of Transfusion Medicine while also identifying some limitations. More so, this investigates the realm of specialty-specific knowledge recall, calculative abilities, and reasoning. We observed decreased accuracy and thoroughness as query complexity increases. In conclusion, we support the thoughtful use of AI in Transfusion Medicine, while advocating for enhancing the accuracy and capability of the tool, including consideration that the initial search may yield incomplete results that require attention to detail and refinement to obtain the best possible responses.
