Making Clinical Inroads with AI via RCM Use-Cases with Alvin Liu of John Hopkins Medicine
This is part of our weekly executive insights series where Elion CEO Bobby Guelich speaks with healthcare leaders about their tech priorities and learnings. For more, become a member and sign up for our email here.
Role:
Endowed Professor and Inaugural Director, James P. Gills Jr. MD and Heather Gills Artificial Intelligence Innovation Center
Co-chair of Imaging AI Subcouncil, Artificial Intelligence and Data Trust Council
Member, AI Operations Team
Organization: Johns Hopkins Medicine
What’s your role at Johns Hopkins Medicine with respect to technology evaluation and selection?
My work at Johns Hopkins involves four main areas:
Research in AI for ophthalmology: I’m a vitreoretinal surgeon and faculty member at both the School of Medicine and School of Engineering. My research focuses on applying artificial intelligence (AI) in ophthalmology. I’ve been working on how AI, particularly deep learning, can assist in analyzing retinal images to provide new insights.
Real-world AI implementation and scaling: I’ve led the implementation of autonomous AI for diabetic retinopathy screening (the first FDA-approved fully autonomous medical AI device in any medical field) across the health system since 2020. Additionally, I’m involved in a pilot using large language models (LLMs) for revenue cycle management, focusing on making administrative processes like prior authorization more efficient.
AI leadership: I was named the inaugural director of a new AI center at the Wilmer Eye Institute. It is the first endowed AI center at the Johns Hopkins University School of Medicine. In that role, I will push for strategic collaborations between academia and non-traditional partners like big tech, startups, and venture capital to maximize the benefits of AI in healthcare.
AI governance: JHM recently established the Artificial Intelligence and Data Trust Council, within which I will serve as co-chair of the imaging AI subcouncil. I will also serve on the AI Operations Team, which will oversee the work of all three AI subcouncils (imaging, clinical and operational).
What AI applications are you most excited about right now?
There are two main areas I’m really excited about:
AI in research: In the past few years, AI in the form of deep learning has made incredible advances, particularly in medical imaging and diagnostics. For example, AI can now analyze pathology images and predict the genetic makeup of tumors—tasks that are impossible for humans. It can also look at a retina photo and predict someone’s age, blood pressure, smoking habits, cardiovascular risk profile, etc. These superhuman capabilities in specific contexts are extremely exciting for me as a scientist and physician. However, clinical tools are very difficult to implement and scale in a sustainable manner in the US, because one has to think about the regulations, FDA clearance, and who’s going to pay for it. So that’s the big caveat.
AI in RCM: Specifically because of that caveat in clinical AI tools, I’m really excited about the application of LLMs in RCM. I personally believe that the application of LLMs in revenue cycle management will be one of the fastest, most scalable applications of this technology. RCM processes are ideal for demonstrating a clear return on investment (ROI), which will make the technology sustainable. Once we establish ROI in areas like prior authorization, we can start exploring how to implement more clinical applications of AI.
I’d like to talk about both of those, but let’s start with RCM. Are you currently piloting AI for revenue cycle management, and what areas are you focusing on first?
We’re starting with prior authorization, which is a big pain point for health systems. It’s a complex process where payers often require multiple documents for approval, and it just takes a human a long time to get all the different sources for those documents. This is where large language models really shine as a “co-pilot” to automatically gather and present the necessary documents to the human user. The human can then review and submit them, significantly reducing the time and effort involved. While prior authorization is the starting point, there are many other areas within RCM—like coding optimization and appeals management—where LLMs could also be applied.
Let’s talk about the clinical side. Given your experience, how do you get clinicians more comfortable with AI in clinical settings?
It’s important to engage clinicians—doctors, nurses, and other end users—early in the product development process. This ensures the technology is designed with their needs in mind, making it user-friendly and more likely to be adopted. On a broader scale, large AI initiatives need strong support from the health system’s leadership. I think, ideally, there should be a good amount of technical knowledge—specifically when it comes to AI—in the C-suite to really understand the technology and essentially be an advocate for these new tools. This kind of high-level buy-in is crucial to scaling AI solutions across an organization.