AAO 2024: Applying large language models in revenue cycle management

This transcript has been lightly edited for clarity.

T.Y. Alvin Liu, MD: Hello everyone. My name is Alvin Lu. I'm a retina specialist at the Wilmer Eye Institute, and I'm also on the AI Operations team at Johns Hopkins Medicine, and this newly established team with a purview over all things AI related in the clinical, operational and imaging domains. And later today, I have the honor of giving a talk that focuses on using large language models (LLMs) for revenue cycle management.

As you know, the vast majority of medical transactions in the US are carried through medical insurance, and therefore revenue cycle and management is extremely important. That's exactly how health systems and providers get paid by insurance payers. RCM management, in itself, is a big business. It's estimated that each year, about $150 billion are involved in RCM management. However, this process is also very tedious, with a lot of pain points and inflections. Therefore, one of the most promising use cases of large language models to use LLMs for revenue cycle management such as prior authorization.

We at Johns Hopkins Medicine recently performed a pilot for this exact purpose, and so far, we have been able to demonstrate that we improved productivity by about 25% and in over 70% of all the prior authorizations, all the required documents to be uploaded were, in fact, recommended by this LLM. This is certainly a very exciting area of development for healthcare AI overall, and certainly for ophthalmology, because, as you know, obtaining the correct prior authorizations is important for ophthalmology, especially in the context of intravitreal injections in retina.