Evolution of a Large Language Model for Preoperative Assessment Based on the Japanese Circulation Society 2022 Guideline on Perioperative Cardiovascular Assessment and Management for Non-Cardiac Surgery

Takahiro Kamihara; Masanori Tabuchi; Takuya Omura; Yumi Suzuki; Tsukasa Aritake; Akihiro Hirashiki; Manabu Kokubo; Atsuya Shimizu

doi:10.1253/circrep.CR-24-0019

Evolution of a Large Language Model for Preoperative Assessment Based on the Japanese Circulation Society 2022 Guideline on Perioperative Cardiovascular Assessment and Management for Non-Cardiac Surgery

Circ Rep. 2024 Mar 15;6(4):142-148. doi: 10.1253/circrep.CR-24-0019. eCollection 2024 Apr 10.

Authors

Takahiro Kamihara¹, Masanori Tabuchi², Takuya Omura³, Yumi Suzuki⁴, Tsukasa Aritake⁴, Akihiro Hirashiki¹, Manabu Kokubo¹, Atsuya Shimizu¹

Affiliations

¹ Department of Cardiology, National Center for Geriatrics and Gerontology Obu Japan.
² Department of Nursing, National Center for Geriatrics and Gerontology Obu Japan.
³ Department of Metabolism, National Center for Geriatrics and Gerontology Obu Japan.
⁴ Department of Surgery, National Center for Geriatrics and Gerontology Obu Japan.

Abstract

Background: The Japanese Circulation Society 2022 Guideline on Perioperative Cardiovascular Assessment and Management for Non-Cardiac Surgery standardizes preoperative cardiovascular assessments. The present study investigated the efficacy of a large language model (LLM) in providing accurate responses meeting the JCS 2022 Guideline. Methods and Results: Data on consultation requests, physicians' cardiovascular records, and patients' response content were analyzed. Virtual scenarios were created using real-world clinical data, and a LLM was then consulted for such scenarios. Conclusions: Google BARD could accurately provide responses in accordance with the JCS 2022 Guideline in low-risk cases. Google Gemini has significantly improved its accuracy in intermediate- and high-risk cases.

Keywords: Guidelines; Large language model; Virtual scenarios.