Enabling trustworthy healthcare decisions with retrieval

Photo: Dejan Marjanovic/Getty Images

The use of generative artificial intelligence (AI), particularly large language models (LLMs), is growing exponentially in the healthcare sector.1 LLMs can support greater efficiencies in operations and care delivery, by allowing users to search for answers buried deep within medical, claims or analytics data.2 As a result, said Kayt Leonard, Director, Marketing for SAS’ Global Health and Life Sciences practice, this evolving technology is slowly but surely changing the face of healthcare.

“Generative AI is enabling a stronger approach to healthcare,” she said. “It allows physicians to move more quickly with the research they conduct to get answers, test theories and get clinical decision support. It helps validate data more quickly for claims and operational needs. And it also supports public health entities’ effectiveness as they track diseases and respond to outbreaks.”

Yet not all generative AI is created equal. Algorithms are only as good as the data they rely upon, and many models can return “hallucinations,” or inaccurate or inconsistent responses, which have the power to negatively affect critical decisions. That’s why, Leonard said, AI will only be a trusted and reliable decision tool if models draw from approved, preselected sources, which is what retrieval-augmented generation (RAG) enables.

“When you are relying on generative AI in a healthcare setting, you need to be sure the model is always working with the right type of data,” she noted. “And RAG can deliver more accurate, source-grounded responses that users can rely upon to make decisions.”

Fast, more accurate insight generation

RAG combines two important AI capabilities, retrieval and generation, to enhance the quality of AI outputs. By pairing semantic search with LLMs, RAG can retrieve relevant data from the right sources, even those, like patient records, that contain unstructured data. By drawing from approved, pre-selected sources, often internal and proprietary, RAG delivers citation-backed, source-grounded responses, enabling clinicians and staff members to feel confident about using the technology, according to Leonard. And RAG makes it easy for providers to delve into the data without knowledge of code or analytics.

“RAG allows for real democratization of AI across the healthcare system,” she said. “You don’t have to work with a data scientist or analyst to get answers. Instead, RAG unlocks the ability of physicians, nurse practitioners and even practice administrators to get the information they need to make decisions in a fast and accurate manner.”

A trusted AI approach for the healthcare arenaThere are a wide range of real-world scenarios where RAG can benefit healthcare stakeholders, Leonard said. It can assist with clinical policy medical review, claims and payment integrity, disease outbreak management and more. These more accurate AI models boast strong data governance to support a safe, compliant framework for use.

“Physicians don’t want to be down in the weeds all the time trying to find the data they need. They want to be able to easily access the right information to make the right decision for the patient,” Leonard pointed out. “RAG can empower faster decision making at scale, and its use has the power to supercharge the way that healthcare is delivered in the future.”