AI-Generated Heart Test Reports Found Accurate and Patient-Friendly in Study
An artificial intelligence program developed to generate patient-friendly explanations of heart test results has shown promising results in a recent study. Researchers at NYU Langone Health tested the AI's ability to produce accurate and easy-to-understand summaries of echocardiograms. Traditionally, the complex language and numerical data in these reports have caused confusion and anxiety among patients.
The study, published in JACC Cardiovascular Imaging, evaluated the AI tool's performance in creating explanations that patients could comprehend. Specifically, the AI, based on OpenAI's GPT-4 model, was tasked with generating simple summaries of echocardiogram results. The summaries were then reviewed by five board-certified echocardiographers who rated them on accuracy, relevance, and clarity.
73 percent of the AI-generated explanations were deemed suitable for patient communication without needing any changes. The AI was highly accurate, with 84 percent of its explanations rated as entirely true and 76 percent containing all the important information.
However, the study also highlighted some limitations, as 16 percent of the AI-generated explanations contained inaccuracies. One notable error involved the AI incorrectly describing the size of a pleural effusion, which underscores the necessity of human oversight in reviewing AI-generated content before it reaches patients.
NYU Langone Health's initiative is part of a broader effort to leverage AI to reduce the burden on clinicians and enhance patient care. With the federal mandate for the immediate release of test results under the 21st Century Cures Act, clinicians have faced an overwhelming volume of patient inquiries. The use of AI could streamline the process by providing patients with immediate, understandable explanations of their test results, potentially reducing anxiety and the number of follow-up questions to clinicians.
The research team also included feedback from participants without clinical backgrounds to assess the clarity of the AI-generated explanations from a layperson's perspective. The results were encouraging, with 97 percent of these participants finding the AI explanations more understandable than the original reports. Many reported that the AI-generated summaries alleviated their concerns about their health.
Looking ahead, the NYU Langone team plans to refine these AI tools further and integrate them into clinical practice. By doing so, they aim to improve patient understanding, reduce clinician workload, and ultimately enhance the overall patient experience.
Reference:
Study evaluates AI that creates cardiology reports for patients. Medical Xpress - medical research advances and health news. July 31, 2024. Accessed August 23, 2024. https://medicalxpress.com/news/2024-07-ai-cardiology-patients.html.