Doctors sometimes use a procedure called percutaneous coronary intervention (PCI) to open blocked heart arteries. Deciding which patients need it can be complex. Researchers wanted to see if artificial intelligence (AI) language models could help with these decisions. They tested 15 different AI models using past medical records from 93 patients at one hospital in China.
The study found that different AI models gave different recommendations. Some were more aggressive, suggesting the procedure more often, while others were more conservative. When the researchers combined several models in a smart way, this 'ensemble' performed better than any single model. However, the AI's performance was clearly affected by the patient's age, with specific age points (like 73, 75, and 76) showing gaps in how well it worked.
It's important to understand what this study did and did not show. The research only looked at how well the AI models agreed with each other on classifying cases; it did not test whether following the AI's advice would lead to better patient health. No safety issues were reported because the AI was not actually used on patients. The main reason for caution is that this was a small, single-center study using old data. The results are a technical first step, not proof that AI is ready to help make these medical decisions. Readers should see this as an early look at a possible future tool that requires much more testing in real-world, diverse hospital settings before it could ever be considered for use.