Amy Raymond, senior vice president of revenue cycle and deployment at AKASA, led a discussion with 18 hospital and health system executives at the Becker's 14th Annual Meeting in Chicago in April 2024 to learn about their pursuits, opportunities and adoptions of generative AI and large language models in the revenue cycle.
The conversation touched on the potential applications of generative AI and LLMs in areas such as automation, healthcare revenue cycle management, and patient access. Participants also discussed the challenges and opportunities associated with implementing this technology, including the need for strong leadership and controls on potential impacts to patient care.
Here are four central takeaways and main points made throughout the discussion:
1. AI tools have respective roles and capabilities. At the top of the discussion, Ms. Raymond distinguished large language models, or LLMs, and generative AI.
LLMs are a type of AI model that can understand complex language and reasoning at human levels based on patterns they have learned from a large amount of data. This technology holds promising transformative potential within the healthcare sector, as Ms. Raymond noted.
Generative AI is powered by LLMs and uses these models to “generate” content, such as text, images, music, etc. It works by filling in the blank — predicting words or parts of words, given what it has seen. "Generative AI allows the technology to predict the next thing," Ms. Raymond said. "So if I am writing a sentence and leave a word out, generative AI can figure out what that word is."
And this technology is making an impact on healthcare. More than 70% of healthcare leaders are actively considering the use of generative AI, with 60% eyeing it for the revenue cycle, according to an AKASA survey.
2. The capacity of Generative AI and LLMs to learn, evolve, and autonomously update represents a paradigm shift for healthcare technology systems. This is an important shift for organizations that have historically depended on laborious manual updates and single-stage revisions.
"One of the things I hear from many of my colleagues is the issue of, 'Yeah, we asked for those changes a million years ago, it hasn't happened,'" said Ms. Raymond. "Or, 'We have a whole position dedicated to making sure we're updating this information and keeping it updated,' or 'We changed our own coding guidelines and it's not reflected in here, and now people just aren't paying attention to it. They're deleting everything and starting over.'"
Tools that slow down professionals or make their work cumbersome may deter their ongoing use. However, the integration of advanced AI facilitates seamless integration and continuous improvement, ensuring users can access the latest features and enhancements without disruptions or delays. This enhances the overall user experience by streamlining workflows and reinforcing continued utilization of the technology.
"A tool like this learns from the actions taken by the end user and continuously updates," said Ms. Raymond. "Users don't have to put in a ticket to get the tool updated when it's not able to do what they need it to."
3. The technology involves a nuanced balance of personalization and generalization. One participant asked whether the self-teaching capabilities of generative AI and LLMs result in intelligence that is progressively more specific to individual organizations. Ms. Raymond's answer for AKASA's products? A little bit of both.
"We have what's called a foundation model that incorporates data across the board, but then for each individual health system, we're able to tailor it to varying levels for that health system," she said. Contractual clauses outline how one's information can be utilized within a company's cohort, particularly in the context of technologies that rely on vast amounts of data, Ms. Raymond mentioned.
4. Trust remains a significant concern for end users. Revenue cycle professionals have faced ongoing challenges with platforms handling basic functions such as billing, scheduling and denials management.
The integration of another tool has the potential to impact employee morale and productivity. Therefore, leaders must assess the existing level of trust in current platforms before introducing additional technology. As noted by one health system leader, there is a prevailing sense of low trust in the platforms that currently support them. This only underscores the importance of good leadership in navigating the introduction of AI, as it has the potential to either alleviate or exacerbate existing mistrust.
Ms. Raymond stressed the need for active and vocal leadership that is closely attuned to the integration process. "Change management is a huge piece of implementing any technology — not just internally understanding the impacts, process and flow. But making sure that's called out early on and mitigated for any patient impact," she said.