Generative Pre-Trained Transformers (GPT) Artificial intelligence – Assessing the Accuracy of ChatGPT as an Adjunct for Peri-operative Care

Abstract

PURPOSE: As artificial intelligence (AI) innovation blossoms, minimal advancements have occurred in its integration within plastic surgery; this is especially disadvantageous given AI’s potential for improved patient outcomes and experience. Most recently, a novel machine learning (ML) model using Generative Pre-Trained Transformers (GPT) has been making headlines for its ability to pass the United States Medical Licensing Exam (USMLE) and its capability to converse with the public on a wide array of topics. Chat GPT3 developed by OpenAI and released in late 2022 utilizes a deep learning model through neural networks to recognize data patterns, and through supervised and reinforced human learning is able to answer a broad range of questions. With the US healthcare system facing a physician shortage, increasingly shorter clinical visits, and a substantial administrative burden, we investigated Chat GPT’s capability and accuracy in addressing common peri-operative questions by plastic surgery patients, with the ultimate goal of utilizing a GPT model as an adjunct to assist surgeons in peri-operative care. METHODS: Misconceptions on various common plastic surgery procedures and plastic surgery as a field among the public were identified using a literature search. Surveys on breast reconstruction, silicone implants, bariatric surgery, and preconceptions of cosmetic versus plastic surgery were adapted into questions for the Chat GPT platform (1-4). Chat GPT answers were then assessed for accuracy and compared to published literature addressing these misconceptions. RESULTS: In addressing questions on common misconceptions regarding various plastic surgery procedures including risks and complications, Chat GPT answered 100% of the questions correctly. However, when answering questions on PRS procedure costs Chat GPT accuracy dropped to 30% and was at the lower range of price estimated when compared to ASPS. Lastly, in addressing differences between plastic and cosmetic surgery, it answered 62.8% of questions correctly, and frequently confused the term plastic and cosmetic surgeons, which may lead to further public confusion. The model did have a preference towards plastic surgeons relative to other providers and surgical sub-specialties when asked to decide between subspecialties for common plastic surgery procedures such as breast implants and rhinoplasty. CONCLUSION: ChatGPT’s ability to answer common perioperative questions and misconceptions with 100% accuracy illustrates its broad medical knowledge. While the model was able to answer most questions accurately, its answers were often basic and not nuanced. Regardless, a GPT AI model has significant potential to be a clinical adjunct to aid patients in answering peri-operative questions to allow for more enhanced and efficient patient care. References: 1. Gusenoff JA Pennino RP, Messing S et al. Post-Bariatric Surgery Reconstruction: Patient Myths, Perceptions, Cost, and Attainability Strategies. Plastic and Reconstructive Surgery. 2008;122(1) 2. Schneider LF, Mehrara BJ. De-Mythifying Breast Reconstruction: A Review of Common Misconceptions about Breast Reconstruction. Journal of the American College of Surgeons. 2015;220(3) 3. Shah A, Patel A, Smetona J, Rohrich RJ. Public Perception of Cosmetic Surgeons versus Plastic Surgeons: Increasing Transparency to Educate Patients. Plastic and Reconstructive Surgery. 2017;139(2) 4. Rohrich RJ, Kaplan J, Dayan E. Silicone Implant Illness: Science versus Myth? Plastic and Reconstructive Surgery. 2019;144(1)

Ähnliche Arbeiten