For executives who have lived through the marketing of such technological developments as big data, the cloud and the Internet of Things, the prospect of taking on the latest hot concept, AI, might not be appealing.
But ignoring AI and all its variations may be a mistake. Although there are different technology solutions being sold under the name AI — robotic process automation (RPA) and machine learning (ML) being two — the important things to know are the problem needing to be solved and whether the technology solves or mitigates the problem. These more established techniques also may assist with the adoption of a form of AI that is attracting so much attention, the large language model (LLM), of which one prominent example has the commercial name ChatGPT.
Some hospitals already have begun using LLMs. Scripps Health, San Diego, has taken its first steps in implementing the approach. In August, the health system started testing a tool that drafts a response to patient inquiries. And about a year and a half ago, Scripps began using another large language tool that turns a physician-patient conversation into a medical progress note.
Both have produced promising results.
“We’re continuing to learn through the implementation how to help improve the experience overall, most importantly for the patient and the members of the care team,” said Shane Thielman, corporate senior vice president and CIO for Scripps. “If we can take some of the administrative burden off of the physician and the members of the care team, we can speed up the cycle time [to when] a patient gets a response.”
Scripps executives are looking ahead at ways the technology can be applied in both clinical and nonclinical ways, and like many, they believe it will be transformational.
Read case examples of other healthcare organizations that are applying AI to address revenue cycle management challenges in a sidebar to this article, “Applying AI to Revenue cycle management.”
Changes in store
“When I sat down and used ChatGPT for the first time, I got a sense that I hadn’t gotten since I [first] played with the internet way back in the ’90s,” said David Wetherhold, MD, chief medical information officer for ambulatory systems at Scripps Health, and an internal medicine physician at Scripps Clinic Anderson Medical Pavilion in La Jolla.
“This is a game changer,” said Wetherhold.
There are a number of ways that those changes will affect the revenue cycle.
“[Among the] potential future applications that we see for revenue cycle specifically is the opportunity to summarize a lot of the data and information that’s necessary to generate charges and to support accurate coding moving forward,” Thielman said.
Using an LLM that can comb a patient record and sense keywords through prompts that a business leader would be able to identify, the provider can drive the coders to the information that’s going to be most useful to them to generate the codes and get the claims out the door, Thielman said.
Industrywide AI status
Not all health systems are in a position to adopt ChatGPT-like applications, and there’s considerable confusion about when and how to use AI in nonclinical areas, and what kind of ROI CFOs can expect.
“Over the next decade, organizations will need an AI strategy in the revenue cycle, and the sooner you start with very mature use cases like RPA, the better,” said Bret Anderson, principal at Chartis. “It’s good to develop organizational muscle memory so you can be prepared to evaluate options in the future.”
Anderson’s colleague, Prashant Karamchandani, director and revenue cycle transformation practice co-leader at Chartis, said the question should not be, “How can I use AI?” CFOs must ask: “What problem am I trying to solve?”
“A macro-level answer is not what you need,” Karamchandani said. “It’s about going one or two layers below that. You don’t want to end up buying something and then realize you had a different problem and don’t end up getting the ROI you thought you’d get.”
Consider a hospital that can’t recruit and retain enough staff to perform claims follow-up. This organization may benefit from RPA that checks payer portals for claim payment status, said Karamchandani. It may not need ML or NLP, which can be more expensive and complicated to deploy.
However, now consider a hospital that struggles with staffing shortages across the board and decides to hire novice coders and train them internally. In this case, an NLP or ML solution may be worthwhile if the existing RCM team includes more experienced coders who can train the technology, said Michelle Wierczorek, senior vice president, coding and clinical integrity, at CSI Companies.
“The ML and NLP can be really helpful in terms of accelerating accuracy for new coders,” she added.
Setting the AI record straight
Defining AI is difficult. Part of the confusion stems from the fact that AI is often used as an umbrella term for several more specific technologies, said Bret Anderson, principal at Chartis Group. While many RCM vendors say they provide AI-powered solutions, this characterization doesn’t necessarily tell CFOs what technology works.
The big three established technologies associated with revenue cycle AI are robotic process automation (RPA), machine learning (ML), natural language processing (NLP) or some combination of all three. While RPA typically uses structured data to replace manual functions, ML and NLP take it one step further to ingest unstructured data and generate insights, Anderson said. RPA typically requires little or no human intervention while ML and NLP typically augment (not replace) human work. Executives at Scripps Health note that some don’t consider those tools to be true AI.
“Over the next decade, organizations will need an AI strategy in the revenue cycle, and the sooner you start with very mature use cases like RPA, the better,” said Anderson. “It’s good to develop organizational muscle memory so you can be prepared to evaluate options in the future.”
6 tips for deploying AI in RCM
Experts provide tips to help CFOs develop an AI strategy in revenue cycle management (RCM).
- Define your ‘why.’ What problem are you trying to solve?
- Look at your native system. Does functionality already exist within the electronic health record (EHR) to solve the problem? Some EHRs may include forms of AI.
- Evaluate vendors. If the vendor refers to the solution as AI, what does that actually mean? For example, is it RPA, ML or NLP? And will that technology solve the problem at hand? If so, how?
- Gauge how much human input is required to promote a positive ROI Does the organization have the right type of internal expertise to train the solution? Also, when software updates occur, does the vendor ensure all historical ML is maintained? If so, how?
- Ask for references. What other organizations have used the technology and how? What ROI have they seen and why?
- Focus on change management. Help RCM staff understand how technology can elevate — not necessarily replace — them.