AI and Heart Teams Disagreed on Treatment Decisions
A study found AI models often mismatched human heart team care plans for patients with complex coronary artery disease.
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?
Between 2019 and 2024, researchers assessed how well AI models aligned with multidisciplinary heart team revascularization recommendations for 546 patients. The study revealed that agreement rates varied significantly depending on whether the clinical information was provided as unstructured text or a structured pro forma.
Why it matters
Understanding how AI performs as a decision-support tool is critical, as the study linked disagreements between the AI and human teams to a 1.67 higher odds ratio for major adverse cardiovascular events (MACE).
This retrospective study of 546 patients, with a mean age of 66, analyzed AI performance against historical multidisciplinary heart team decisions. Researchers found that providing AI with unstructured clinical narratives resulted in 75% alignment, compared to 35% for structured case forms.
The players
ChatGPT-4o
A large language model developed by OpenAI used here to simulate clinical decision-making.
Gemini 2.0
An artificial intelligence model developed by Google evaluated for its potential role in medical decision support.
JSCAI
The Journal of the Society for Cardiovascular Angiography & Interventions, which published the study findings.
The details
Researchers tested ChatGPT-4o and Gemini 2.0 by inputting clinical case data to generate revascularization recommendations. The models struggled with consistency when using structured case pro formas, and even including formal revascularization guidelines in the prompts failed to improve performance. The study suggests that how information is presented significantly influences the AI's ability to mirror the nuanced clinical judgment of human heart teams.
Timeline
Heart team discussions for the 546 study patients occurred between 2019 and 2024.
The study findings were published in the journal JSCAI in September 2026.
Health Landscape
The study sits at the intersection of evolving artificial intelligence integration in cardiology and the established standard of care involving multidisciplinary teams. These findings underscore the current limitations of generative models compared to the human expertise required for complex heart disease.
If you are managing complex coronary artery disease, these findings reinforce that your treatment path should be determined by a human multidisciplinary heart team rather than automated tools. You should always discuss your specific care options and any concerns about your treatment plan with your cardiologist.
The takeaway
AI models currently lack the clinical consistency of human heart teams when recommending treatments for complex heart disease. Patients should prioritize decisions made in consultation with a team of human specialists and discuss any questions regarding their revascularization plan with their doctor.
Further reading
For more on managing advanced cardiac conditions, visit our 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?
