Artificial intelligence (AI) may soon play a significant role in decision-making within cardiology, specifically in choosing between surgery and percutaneous interventions. A recent proof-of-concept study suggests that AI, particularly large language models (LLMs), could become valuable tools in this area.
The Role of AI in Cardiology
AI platforms, including well-known models like ChatGPT, have already made strides in the medical field. These systems can generate echocardiography reports, pass clinical exams, and even author scientific papers. Despite some skepticism about the extent of AI’s role in medicine, many professionals now anticipate its integration into clinical practice, though challenges remain.
Dr. Edward Itelman, lead author of the study from Rabin Medical Center in Petah-Tikva, Israel, emphasized that AI will not replace physicians but will enhance their capabilities. “These tools will never replace doctors,” Dr. Itelman stated. “They will augment the role of the physician, particularly in decision-making support systems, but this must be done through a rigorous regulatory process.”
AI in Practice: Current Findings
In their study, published as a research letter in JACC: Cardiovascular Interventions, Dr. Itelman and his team evaluated three commercially available LLMs—ChatGPT v3.5, ChatGPT v4.0, and the now-discontinued Google Bard. The researchers asked these models to recommend treatments for 20 structural and 20 coronary cases, based on current medical guidelines from the European Society of Cardiology. The cases were classified for either transcatheter interventions (TAVI or PCI) or traditional surgery (SAVR or CABG), with input from expert interventional cardiologists.
The results showed:
- Structural Cases: Both ChatGPT models correctly recommended TAVI or SAVR in all cases. However, Google Bard was less accurate, correctly recommending only 70% of the time.
- Coronary Cases: The task was more challenging, with ChatGPT v3.5 and Google Bard correctly assigning only 70% of the cases to PCI or CABG. In contrast, ChatGPT v4.0 successfully selected the appropriate procedure for all cases.
Limitations and Future Directions
The AI models often provided logical reasoning aligned with current guidelines but sometimes included outdated references. Dr. Itelman noted that improvements in citation capabilities and access to recent knowledge are expected with future models. Additionally, the study highlighted limitations in how AI models handle incomplete information. While LLMs can process data such as age and sex, they lack the ability to review imaging and conduct patient interviews comprehensively.
Addressing these issues is crucial as AI technology evolves. Dr. Itelman pointed out that the study’s goal is not to compare the LLMs’ current capabilities but to emphasize the need for validated models across medical topics. “We want a large language model that is validated on most medical topics and can be used safely and securely by doctors and patients alike,” he explained.
Moving Forward
As AI continues to advance, its integration into clinical practice is likely to increase. However, it is essential to establish a clear path for how AI can best support decision-making in cardiology. With ongoing improvements and regulatory oversight, AI could soon become an integral part of the clinical toolkit, enhancing decision-making and patient care in cardiology.
