Large Language Models in Cardiology: Systematic Review.
Signal observed that large language models demonstrate variable performance across cardiology applications, with ChatGPT-4 achieving 66% accuracy on physician education tasks and 91% accuracy interpreting electrocardiogram vignettes. Evidence suggests meaningful potential for educational applications and ECG interpretation, though limitations exist in emergency guidance readability and comprehensive clinical task performance. Worth noting the systematic evaluation identified performance gaps in acute care scenarios, with key myocardial infarction first aid steps omitted in 25-45% of cases.
Limitations in emergency guidance and readability, as well as small in silico study designs, highlight the need for multimodal models and prospective validation.
Relevance: Systematic review of large language models in cardiology applications including heart failure, acute events, and diagnostic testing. Broad cardiology scope relevant to heart failure (research interest) and general cardiology education/management across diverse tasks.