AI Performance Compared to Heart Teams in Study
Researchers evaluated how LLMs matched multidisciplinary heart team revascularization decisions for 546 patients.
Updated on Sept. 22, 2026 in Heart Disease
Live Poll
Would you trust an AI model to help decide your treatment for a complex medical condition?
A study published in JSCAI evaluated the ability of AI models like ChatGPT-4o and Gemini 2.0 to mirror human heart team recommendations for complex coronary artery disease. Analysis of data from 546 patients treated between 2019 and 2024 revealed varying agreement rates based on input formats.
Why it matters
The research aimed to determine the viability of using LLMs as decision-support tools for complex cardiovascular cases. Findings suggested that clinical input structure significantly influences whether AI models align with expert human consensus.
The study included 546 patients with a mean age of 66 years, 75% of whom had triple-vessel disease and 12% severe left main disease. Researchers reported an odds ratio of 1.67 for major adverse cardiovascular events (MACE) in cases where AI and heart team recommendations were discordant.
The players
JSCAI
This peer-reviewed medical journal focuses on research regarding cardiovascular interventions.
ChatGPT-4o
This large language model developed by OpenAI was one of two AI systems used to generate revascularization recommendations.
Gemini 2.0
This AI model created by Google was utilized alongside other systems to compare medical decision-making capabilities.
The details
Researchers input clinical data from patient cases into AI models using both unstructured narratives and structured case pro formas. Results indicated that while AI reached high alignment with human teams when processing narrative text, performance dropped significantly when using rigid, structured input formats, even when augmented with formal revascularization guidelines.
Timeline
Patient heart team discussions occurred between 2019 and 2024.
Study findings were published in JSCAI in September 2026.
The Big Picture
This research follows a pattern set by ongoing efforts to integrate generative AI into clinical decision-support systems for coronary artery disease management. By testing these models against multidisciplinary teams, it highlights the current performance ceiling for AI in complex cardiovascular triage.
The findings suggest that current AI tools for heart health still require human oversight due to potential discrepancies in complex diagnosis. Patients should continue to rely on traditional multidisciplinary heart teams for critical revascularization decisions rather than autonomous AI assessments.
The takeaway
This study underscores that the structure of medical data input is vital for the accuracy of AI decision support. Physicians and patients should remain cautious of AI recommendations in cardiovascular care until model performance is more consistent across varied clinical formats.
Further reading
For more on the latest research in this field, visit our /health/diseases/heart-disease/ section.
Source note: This article includes information reported by Tctmd.
Live Poll
Would you trust an AI model to help decide your treatment for a complex medical condition?
