LA JOLLA, CA— Artificial intelligence (AI) can now integrate multiple data types: electrocardiograms that track your heart, fundus imaging data that capture the inner parts of your eyes, and electronic health records that consolidate your patient information. In combining all of this data, AI technologies are transforming how we detect, predict and ideally treat cardiovascular diseases.

As the potential for AI to reshape cardiology becomes a reality, a new review article describes the importance of properly integrating these technologies into cardiovascular medicine, including how to approach AI with both optimism and caution.

In the Lancet Series article published August 29 in The Lancet Digital Health, co-authors Giorgio Quer, PhD, director of artificial intelligence at Scripps Research Translational Institute, and Eric Topol, MD, executive vice president of Scripps Research and the director and founder of the Scripps Research Translational Institute, call for regulatory pathways that can vet the accuracy of medical devices that apply AI technology, as well as their ability to safekeep patient data.

They specifically explore how AI tools trained to understand and generate human-friendly text (large language models like chatbots) will intersect with technologies that collect and analyze a range of data types. These include unimodal AI that detects one type of data (such as biosensors), and also multimodal AI that analyzes and interprets a range of medical information (combining smart device records, medical imagery, genomics, and/or electrocardiogram data). While the goal of merging these technologies is to better personalize patient care, the appropriate human oversight of their medical applications needs careful consideration.

“One of the issues is that AI tools, particularly large language models, are continuously receiving new input and are learning at the same time as they are acting—making them difficult to validate,” says Quer, who is also an assistant professor of digital medicine at Scripps Research. “This means we need to be careful and not use new technologies without proper verification, which involves running clinical trials to prove these tools properly work for a specific task, publishing the results and alerting the community of new developments.”

For example, while AI programs could assist with early detection of cardiovascular conditions or predict someone’s risk for heart problems, if the results generate false positives or leave personal information vulnerable, then it will be challenging to convince patients they are safe to use. Another common issue with AI models is that the data on which they are trained may exclude historically under-represented communities, so the technology could be accurate for some individuals but not others.

On the clinical side, large language models are already beginning to enhance cardiologist-patient relations. These programs are summarizing text, extracting and organizing patient information, and translating complex medical information into easily understandable text or multiple languages. While AI models could help with even more advanced tasks, such as diagnosing diseases, human supervision remains important as these tools are known to fill in knowledge gaps with made-up information.

“We’re seeing remarkable advancements in AI’s ability to assist with diagnosis, assess cardiovascular risk and even help with patient communication,” says Topol, a leader in individualized medicine who was recently featured in the inaugural 2024 TIME100 Health list of most influential people working in global health. “However, we must balance our enthusiasm with a rigorous approach to validation and implementation.” The researchers advocate for both regulatory oversight and discussions between AI experts and clinicians, ensuring there’s a robust framework for evaluating these tools and their specific applications. “AI should augment and enhance the doctor-patient relationship, not replace it,” says Quer. “Our ultimate goal is to harness these technologies to improve cardiovascular outcomes and quality of life for patients worldwide.”

Quer and Topol are the authors of the Lancet Series article, “The potential for language models to transform cardiovascular medicine.”

This work was supported by funding from the National Center for Advancing Translational Sciences/National Institutes of Health grant UM1TR004407. Topol is a member of Abridge AI Scientific Advisory Board.