ChatGPT and Large Language Models in Healthcare: Opportunities and Risks

Abstract

ChatGPT, a pre-Trained large language model (LLM), has the potential to transform healthcare by providing valid clinical insights and reducing doctors' workload. There are already signs that such tools can be useful for automating the generation of patient discharge reports, clinical vignettes, and radiology reports. Such tools can also capture the vast medical knowledge base as demonstrated by ChatGPT clearing the United States Medical Licensing Examination (USMLE). Such tools promise to make healthcare more accessible, scalable, and efficient, leading to better patient outcomes. However, such tools are far from perfect and well-known to be susceptible to error, misinformation, and bias. In this paper, we review the potential applications of ChatGPT in healthcare and also identify potentials risks that must be addressed before ChatGPT and other LLM tools can be safely adopted in healthcare. First, we offer case studies on using ChatGPT for passing USMLE, identifying prevention methods for cardiovascular disease, generating patient discharge reports, generating clinical vignettes, and generating radiology reports. Second, we present the opportunities that ChatGPT offers in healthcare. By leveraging its language generation and processing capabilities, ChatGPT can streamline and improve a range of healthcare tasks, from digitizing clinical notes and improving the accuracy of diagnosis to revolutionizing medical education and empowering patients with personalized healthcare information. Finally, we reflect on the associated risks and conclude that caution is advised in interpreting the results of ChatGPT as these studies are preliminary and not entirely error-free.