AI tool simplifies cardiology reports making them ‘more understandable’ for patients
A pioneering study has revealed that an artificial intelligence (AI) tool can simplify heart test results into explanations that are accurate, relevant, and comprehensible to patients.
The focal point of the research was the echocardiogram—a diagnostic tool that employs sound waves to visualise the flow of blood through the heart’s chambers and valves. Traditional echocardiogram reports are replete with technical jargon and numerical data relating to heart function, dimensions, vessel pressure, and tissue thickness, all indicative of potential cardiac ailments. However, such reports are notoriously difficult for patients to decipher, often causing undue anxiety, according to the researchers.
To mitigate this issue, NYU Langone Health explored the potential of a specific AI technology capable of generating contextually appropriate language predictions. This AI, developed by OpenAI and known as GPT-4, was accessed in March 2023 when NYU Langone secured one of the earliest “private instances” of the tool, allowing them to safely experiment with real patient data under stringent privacy regulations.
The study, which was published online on July 31 in the Journal of the American College of Cardiology (JACC) Cardiovascular Imaging, evaluated 100 cardiologist-authored echocardiogram reports. GPT-4 was tasked with reformulating these into patient-friendly summaries. These AI-generated explanations were then assessed by five board-certified echocardiographers using a five-point scale to rate accuracy, relevance, and understandability. Impressively, 73 percent of these summaries were deemed accurate enough to be shared with patients without any modifications.
The evaluations showed that 84 percent of AI-generated summaries were completely accurate, and 76 percent included all critical information. The significance of these findings is underscored by none of the AI-generated explanations being rated as “potentially dangerous” due to missing information.
“Our study is the first of its kind to test GPT-4 for this purpose, and our findings suggest that generative AI can significantly aid clinicians in communicating complex echocardiogram results to patients,” explained Dr. Lior Jankelson, MD, PhD, the corresponding study author and an associate professor at NYU Grossman School of Medicine. He noted that such rapid, precise explanations could alleviate patient anxiety and reduce the overwhelming number of inquiries clinicians receive.
The need for such advancements is partially driven by the 21st Century Cures Act of 2016, which mandates the swift release of test results directly to patients, leading to a spike in patient inquiries due to misunderstandings about their health data.
Dr. Jacob Martin, MD, the study’s lead author, emphasised the ongoing need for AI tools to deliver results explanations promptly as they become available, thus potentially easing the burden on healthcare providers who must manually input vast amounts of data into electronic health records.
However, the study also revealed that 16 percent of AI explanations contained some inaccuracies. For instance, one AI summary incorrectly described the size of a pleural effusion—a mistake attributed to an ‘AI hallucination’, a known issue where AI tools generate erroneous or fabricated data. This underscores the essential role of human oversight in verifying and refining AI-generated drafts before they are finalised for patient review.
Additionally, the study involved a survey of non-medical participants to gauge the clarity of AI-generated explanations compared to traditional reports. A resounding 97 percent found the AI versions easier to understand, which often reduced their anxiety.
“This analysis confirms the potential of AI to enhance patient understanding and reduce anxiety,” said Dr. Martin. He revealed that the next steps would involve integrating these AI enhancements into regular clinical practice to improve patient care and reduce the workload on healthcare providers.
The research was a collaborative effort involving multiple specialists from NYU Langone’s Leon H. Charney Division of Cardiology and other departments, highlighting the interdisciplinary approach necessary for such innovative healthcare solutions.