Comparing AI-Driven and Heart Team Decision-Making in Multivessel Coronary Artery Disease

Abstract

<b>Background/Objectives</b>: Multivessel coronary artery disease (CAD) remains a challenging condition requiring multidisciplinary decision-making, particularly when determining between percutaneous coronary intervention (PCI) and coronary artery bypass grafting (CABG). Recent advancements in artificial intelligence (AI), particularly generative language models like ChatGPT, present an opportunity to assist in the decision-making process. However, their ability to replicate human clinical judgment in complex scenarios, such as multivessel CAD, remains untested. <b>Methods</b>: The aim of this study was to evaluate the concordance between recommendations from AI (ChatGPT) and those from heart team (HT) in the management of multivessel CAD, with a focus on comparing treatment strategies such as PCI and CABG. A retrospective observational study was conducted on 137 patients with multivessel CAD, discussed at multidisciplinary HT meetings in 2024. Standardized clinical vignettes, including clinical and anatomical data, were presented to ChatGPT for treatment recommendations. The AI's responses were compared with the HT's decisions regarding PCI or CABG. Statistical analysis was performed to assess the level of agreement and predictive value of ChatGPT's recommendations. <b>Results</b>: ChatGPT achieved an overall accuracy of 65% in its recommendations. The agreement rate was higher for CABG (82.4%) than for PCI (44.4%). Discordance was identified in 48 patients, with a notable bias towards recommending CABG. Factors such as age, diabetes, and chronic kidney disease were predictors of discordance, although no significant factors emerged for the PCI or CABG subgroups. <b>Conclusions</b>: AI, particularly ChatGPT, demonstrated modest concordance with HT decisions in the management of multivessel CAD, especially favoring CABG. While AI offers potential as a decision-support tool, its current limitations highlight the continued need for human clinical judgment in complex cases. Further research is required to optimize AI integration into clinical decision-making frameworks.

Ähnliche Arbeiten