Enhancing Coronary Revascularization Decisions: The Promising Role of Large Language Models as a Decision-Support...

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Clinical Decision Support Systems

Background:

• Clinical guidelines recommend multidisciplinary heart team (MDHT) discussions for coronary revascularization.
• Variability in MDHT implementation necessitates innovative decision-support tools.
• Language learning models (LLMs) like ChatGPT may bridge implementation gaps.

Purpose of the Study:

• To evaluate the concordance between MDHT recommendations and LLM-generated recommendations for coronary revascularization.
• To assess the performance of ChatGPT-3.5 and ChatGPT-4 in this decision-making support role.

Main Methods:

• Analysis of 86 coronary angiography cases referred for revascularization.
• Comparison of MDHT decisions with recommendations from ChatGPT-3.5 and ChatGPT-4.
• Evaluation of model performance using accuracy, sensitivity, specificity, and kappa statistics.

Main Results:

• ChatGPT-4 achieved high concordance with MDHT decisions (accuracy 0.82, kappa 0.59).
• ChatGPT-3.5 showed significantly lower concordance (accuracy 0.67, kappa 0.12).
• Detailed case presentation enhanced ChatGPT-4's accuracy, particularly for left main disease, 3-vessel disease, and diabetic patients.

Conclusions:

• Advanced LLMs like ChatGPT-4 show potential as supportive tools in coronary artery disease revascularization decision-making.
• High accuracy in specific patient subgroups suggests tailored applications for AI in cardiology.
• LLMs may help standardize and improve the consistency of revascularization recommendations.