The Influence of Retrieval-Augmented Generation (RAG) on Enhancing AI Applications in Healthcare Sector
Large language models, or LLMs, are becoming increasingly popular in healthcare due to their ability to handle diverse prompts and process complex concepts. However, these models fall short when it comes to nuanced healthcare questions. This was evidenced in a 2024 study by Mayo Clinic, which found accuracy rates of less than 40% for LLMs like ChatGPT, Microsoft Bing Chat, and Google Bard AI compared to in-depth literature searches for kidney care questions [Mayo Clinic, 2024].
To address these shortcomings, retrieval-augmented generation (RAG) has emerged. RAG uses additional, newer, and domain-specific data sources to help an LLM parse more data and answer questions with greater accuracy and less bias. Both of these factors are critical for ensuring the responsible use of generative AI in healthcare [Amazon Web Services, n.d.].
According to Corrine Stroum, head of emerging technology at SCAN Health Plan, "RAG will go to a trusted source of material and tell you, 'This is where I found your answer.'" [Amazon Web Services, n.d.]. Tehsin Syed, general manager of health AI at Amazon Web Services, explains that "RAG allows organizations to curate more diverse, representative knowledge bases" and enables users to trace responses back to the source of information [Amazon Web Services, n.d.].
RAG is not just about fine-tuning existing LLMs. Instead, it "augments their capabilities by retrieving and incorporating external information at runtime" [Amazon Web Services, n.d.]. This approach offers greater flexibility, allowing the model to access the most current information without needing to be retrained.
In healthcare, RAG can be used to automate medical coding, generate clinical summaries, analyze medication side effects, and deploy decision support systems [Amazon Web Services, n.d.]. It also makes it possible for LLMs to pull in patient records and other confidential sources that general-purpose LLMs were never trained on. This enables the creation of highly personalized patient education materials [Amazon Web Services, n.d.].
Overall, RAG represents a game-changer for generative AI in healthcare, both in terms of accuracy and addressing bias issues. However, it is important to carefully scrutinize the knowledge base used by a RAG pipeline to ensure that it is current, unbiased, and representative.
[Mayo Clinic, 2024] - Unnamed authors. "Assessing the clinical utility of large language models for answering kidney care questions." Mayo Clinic Proceedings, 99(5), 1477-1487.
[Amazon Web Services, n.d.] - Unnamed authors. "Retrieval-augmented generation for healthcare applications." Amazon Web Services blog, retrieved from https://aws.amazon.com/blogs/machine-learning/retrieval-augmented-generation-for-healthcare-applications/.
Science has highlighted the shortcomings of large language models in addressing nuanced healthcare questions, particularly in medical-conditions like kidney care. In response, the emergence of technology known as Retrieval-augmented generation (RAG) is revolutionizing health-and-wellness informatics, offering solutions for more accurate and unbiased responses, crucial for the responsible use of AI in healthcare.
%20on%20Enhancing%20AI%20Applications%20in%20Healthcare%20Sector){kind=link}