Large language models— generative artificial intelligence tools that process and generate text — are proving popular in healthcare. Because LLMs can respond to diverse prompts and process complex concepts, they show promise for augmenting medical research, patient education and clinical documentation.
That said, models trained on general data often fall short when it comes to nuanced healthcare questions. A 2024 study from Mayo Clinic found accuracy rates of less than 40% for prompts on ChatGPT, Microsoft Bing Chat and Google Bard AI compared with in-depth literature searches for questions on kidney care. “In critical areas like healthcare decision making, the impact of such inaccuracies is considerably heightened,” the authors write, “highlighting the need for models that are more reliable and precise.”
This is where retrieval-augmented generation comes into play. RAG draws on additional, newer and domain-specific data sources. This lets an LLM parse more data than it was initially trained on and answer questions with greater accuracy and less bias — both of which are critical for ensuring the responsible use of generative AI in healthcare.
“The world we live in has its own set of canonical literature, whether it’s medical policies or claims processing manuals or technical literature,” says 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.’”
Click the banner below to achieve meaningful AI transformation in healthcare with expert guidance.