(P-IT-8) Minutes, Not Hours: Saving Time on Non-Clinical Activities with Artificial Intelligence

Artificial Intelligence (AI) has the potential to revolutionize the practice of medicine.  Patient care always takes priority, and some tasks are delayed in order to provide uninterrupted care or mitigate risks from multitasking.  Reliable ways to complete tasks more quickly while not sacrificing quality are welcome. While most individuals are aware of the potential for AI, encountering utilization of AI is rare in the course of clinical work, education, administration, or leadership activities. Presented here are some of the ways AI can be used in timesaving and quality-preserving ways, as well as caution against over-reliance on the veracity of AI output.

Study

Design/Methods:

Typically, numerous hours are required to create a high quality de novo document. The workflow starts with a specific request to the AI software, followed by review of the AI response, with follow-up requests for additional detail, examples, resources, and references. The final AI response is reviewed by a senior transfusion medicine physician. 

Results/Findings: New documents were created in minutes, not hours: new guidelines, educational documents, SBAR communication documents for resource requests, and goals and objectives for trainees. Creation of three separate drafts of Transfusion Medicine Rotation goals and objectives for three types of trainees (Anesthesiology, Pathology, and Transfusion Medicine) was accomplished within five minutes total, less than two minutes per initial request to generate a response and be ready for review and editing.
The review and editing stage revealed limitations in AI performance. For example, when asked for irradiation guidelines for transfused products, all physician reviewers disagreed with the recommendations to irradiate any blood product intended for patients receiving prednisone. Queries to discover the source of the recommendation yielded a disclaimer on the general use of irradiated blood in immunosuppressed patients.

Conclusions:

While the integration of AI offers numerous benefits in theory, caution must be exercised in its adoption.  Understanding the current limitations on the accuracy and ability of AI output must be carefully considered to ensure the ethical and safe implementation of AI technologies. Furthermore, the potential for overreliance on AI systems and the loss of human judgment in critical decision-making processes pose significant challenges that must be addressed prior to utilization.  Implementing the aforementioned irradiation recommendations without revision by an experienced and knowledgeable reviewer would lead to significantly increased number of irradiated products, which may have negative clinical consequence as well as workflow challenges. A balanced approach is needed, combining the strengths of AI with the expertise of experienced blood bankers to optimize patient care while upholding ethical standards, resource stewardship, and patient safety.