Artificial Intelligence as a Triage Tool During the Perioperative Period: Pilot Study of Accuracy and Accessibility for Clinical Application

Abstract

BACKGROUND: Artificial intelligence (AI) has been rapidly evolving and is awaiting integration into healthcare. With ChatGPT continuing to captivate public attention, greater consideration of its potential impact on plastic surgery is warranted. As a natural language processing software, ChatGPT is inherently dialogistic. Given this property, our aim was to determine if this AI function can be used as a patient directed self-service tool whereby patients can have their clinical questions directly answered by AI. The goal of this research is to determine the functionality of incorporating AI as a clinical tool to help manage patient questions and clinical concerns in the perioperative period. While such tools introduce complex medico-legal questions, which are beyond the scope of this initial pilot study, our objective was to assess the content, accuracy, and accessibility of AI generated content regarding common perioperative questions and complications for reduction mammaplasty. METHODS: AI interface ChatGPT (OpenAI, February Version, San Francisco, CA) which is publicly accessible was utilized to query 20 common patient questions or complications that arise in the perioperative period of a reduction mammaplasty. Searches were performed in duplicate, where a query was performed for a general term (“breast reduction bleeding”) and repeated with a specific clinical question (“I had a breast reduction yesterday and now I have bleeding. What should I do?”). Query outputs were analyzed both objectively for metrics including output length, sentence structure, and readability scores and subjectively for tone, content, and accuracy. Microsoft Excel (Version 7, Seattle, WA) was used for performing descriptive statistics, t-tests, and chi-square tests where appropriate with a predetermined level of significance of p<0.05. RESULTS: A total of 40 AI generated outputs were analyzed. Mean word length was 191.8 words with 998.9 characters. Mean Fresh-Kincaid Grade Level was the 13th Grade and Mean Flesh Reading Ease was 39.7 (1-100 where 100 is most readable). Regarding content, out of all query outputs 97.5% were on the appropriate topic. Medical advice was deemed to be reasonable in 100% of cases. General queries more frequently (16/20) reported overarching background information whereas specific queries more frequently reported prescriptive information (18/20) (p<0.0001). Specific queries recommended discussion with the surgeon in 100% of cases, while general queries recommended the same in 95% of cases. AI outputs specifically recommended following surgeon provided postoperative instructions in 86.8% of instances. Notable interesting responses included instances of a congratulations and an apology. CONCLUSIONS: Currently available AI tools, in their nascent form, are capable of providing recommendations for common perioperative questions and concerns for reduction mammaplasty. While the reading level of these outputs are higher than ideal, this represents a first step in developing a plastic surgery specific AI application that can serve as the first resource for patients undergoing surgery. Limitations include a potential delay in patients seeking urgently needed medical care. With further calibration, AI interfaces may serve as a tool for fielding patient queries in the future, however patients must always retain the ability to bypass technology and be able to contact their surgeon.

Ähnliche Arbeiten