Artificial Intelligence

Leveraging artificial intelligence to advance organizational strategy

May 31, 2022 5:41 pm

When bringing artificial intelligence into their business models, health systems have the opportunity to improve revenue cycle management by maximizing efficiencies. With the potential to improve workflow and consolidate resources, automation and machine learning require organizations to examine their infrastructure, from staffing to datasets.

Investing in automated processes to reduce variation and empower staff to focus on higher-level tasks, organizations are also exploring more complex machine learning to work across provider and payer systems. In this roundtable, sponsored by Optum, revenue cycle leaders discuss challenges and opportunities that accompany implementation of automated processes and artificial intelligence.

CHUCK ALSDURF: What are you doing to move beyond just automation to leverage artificial intelligence (AI)?

MOTTI EDELSTEIN: We have tools where we’re learning when to contact patients, as we recently signed a deal with Cedar to text and email patient statements. Based on patient behavior, it’s AI-driven, focused on taking the manual part out of the process for cost savings and better patient experience.

BARRETT BLANK: We’re early in the process, trying to build out our unstructured database, leveraging the appropriate schemas to be able to leverage AI.

MELANIE WILSON: A couple years into our journey, we’ve learned that everybody defines AI, automation, RPA, a little bit differently. When I think about true AI, I think about digital workers, who can take multiple workstream flows and string them together in an automated fashion. 

EDELSTEIN: Bots, in our language, is removal of the manual process, and AI is the thinking that goes along with it. 

BLANK: Have your algorithms improved as related unstructured content, such as being able to take actual text, chat — those kinds of things — to different decision processes that it can learn from?

EDELSTEIN: Eligibility, authorization, patient billing, tracking some of the denials; we’re learning down to the benefit level, which patients have what type of benefit structures and then what the outcome’s going to be through the cycle. 

KURT HOPFENSPERGER: How have you all overcome resistance to AI?

WILSON: We started computer-assisted coding about 10 years ago. Our teams thrive on automation because it takes those monotonous tasks away, especially on the back end. 

MARY BETH REMORENKO: We’ve been developing a tool for the last 10 years that leverages NLP and machine learning to replace coding for diagnostic services. We started with radiology services, leveraging different components of the medical record. The tool can take all the data and figure out the coding with a level of accuracy that is on par with a real coder. What started us on our journey was physician burden.

AARON BOUW: How do your coders feel about AI?

REMORENKO: So far, our strategy hasn’t been to completely replace all coders. It’s more of an augmentation strategy, reducing future cost of staffing. For our teams, we’ve been able to explain our strategy, and they appreciate being able to work at the top of their skill set and license since the simpler coding can be handled by AI. 

ALAN MENDELOFF: AKASA published a white paper with HFMA in which they drew a distinction between automation and AI — that automation is task completion, and AI is about taking large amounts of data and being able to identify trends and probabilities to help guide decision-making. Being able to guide that discussion for greater consistency is where the state-of-the-art is right now.

HOPFENSPERGER: If you look at computer-assisted physician documentation, that’s sort of the end game. There are some adjustments that physicians can do that AI can pick up on. If the assessment plan lists the key concerns of the physician and [the patient’s] expected length of stay, whether it’s a commercial or Medicare case, the AI is pretty accurate. 

WILSON: The payers want more information, more documentation, more data. I think the payers have to get to a place where they would be willing to accept that level of information before we could ever trust it to actually make it a reality.

ALSDURF: Are there any other areas that you’re looking to invest in AI or something like that?

REMORENKO: We’re thinking about the whole area of decision support, which really kind of opens up a whole other realm.

WILSON: We’re using it for our empanelment work and driving that to get patients scheduled, so more proactive outreach.

BLANK: Our business tends to be more on the plan enrollment side of the equation. There’s a lot of very difficult populations to get a hold of and bring into care, and that repeatable outreach is similar, but the enrollment side is the equation that we’re automating.

WILSON: For the automation piece, we’re using an external vendor solution to help us identify the patients and then drive the next steps back into the medical record. We use MyChart for most of our messaging, and the vendor tells us who needs the message.

ALSDURF: What’s your end goal with some of these investments that you’re making?

BLANK: We’re a fairly high-growth-path organization, and scalability is critical for us. We’re mostly focused on throughput so we can go after a different market segment and upscale.

REMORENKO: With the pandemic and impact of labor shortages, it’s now part of our business continuity strategy. That’s not originally what we intended, but it’s assisting us with stabilization.

EDELSTEIN: Most revenue cycle shops are not set up for spikes, and the pandemic gave us lots of spikes for testing, for vaccines. We manually couldn’t keep up with it. If it can be automated and technology driven, we can be ready for future spikes.

BLANK: We’re looking at different business models, finding a way to leverage technology to become more of a managed service, moving towards a SaaS model. 

ALSDURF: The financial investment seems like a no-brainer ROI. What about the IT lift? How does that impact your adoption and how are you overcoming some of those challenges?

WILSON: Where IT gets bogged down in our organization is when we come to them with a half-baked idea, and then they’re involved in the total build out of that idea. So we’re trying to vet out the workflows as much as we can before getting IT involved.

REMORENKO: We’ve worked closely with our IT colleagues to advise us on how to approach building out our tool. It minimizes a lot of concerns about how the data is handled and meeting all the requirements for SOC compliance and HIPAA.

JEANNETTE WOOD: We use Athena,  which has a two-step process for physicians to create a claim. You have to review, and mark saved, then go back to create the claim; a lot of physicians forget to hit that button, and so we created a bot to do that for them. We also use bots to apply prepayment monies to outstanding patient balances. Before that, I had staff who would have to manually apply those prepayments to the claims once the claims were adjudicated. Now these staff are working on denials instead of spending their time applying money.

ALSDURF: On the security side, are you getting any kind of pushback on the process with these technologies? What’s the damage control and risk mitigation?

REMORENKO: We just had a bot break the other day, and it locked the medical record in Epic for a patient who was in the ER. That was an emergency phone call to get that bot turned off. We have spent a fair amount of time working on scenarios where we have some controls in place. On the AI side, we actually have it as a separate tool that’s outside of our medical record system. We have some internal firewalls, and our information security team has been involved on how to architect all of that so it’s compliant and secure.

EDELSTEIN: We build infrastructure to keep an eye on technology so that when something does break, we know about it as soon as possible. We normally use that type of technology when we do any type of EMR or other system upgrades.

REMORENKO: We had some good advice from our IT colleagues early on who said, ‘Don’t just think about our own system. We have to also think about external systems such as payers that we sometimes interface with.’ 

BLANK: We’re building data warehouses to aggregate that data from multiple systems. And from that system is where the AI basically applies.

WILSON: When we started our journey, we found that we were better served going externally and buying the service that would do the monitoring for us because of the constantly changing environment.

REMORENKO: We have a separate company that we started called CodaMetrix. We realized that we don’t want machine learning off of bad data, and so we worked with data scientists to make sure that we have the right level of quality for AI.

ALSDURF: Are most of you using some kind of data warehousing strategy or using the proprietary databases from your applications?

BLANK: It’s dependent upon your framework. If you’re dealing predominantly with an open framework, you have a better opportunity to directly leverage those applications. If you’re not using an open-source concept, then it’s proprietary and you’ve got to take it outside of that system to get it integrated. 

EDELSTEIN: Because not everything’s done in Epic, we’ve got other information such as lab, retail pharmacy, EMS and other information in the data warehouse, and that becomes the source data. 

ALSDURF: From a talent perspective, are you having a hard time finding people that have the skill set or filling the gaps to manage this technology and understand it?

EDELSTEIN: I have no control over who gets hired related to data and IT, but there are people who have been assigned from IT that we consider part of my team. 

REMORENKO: It’s a competitive environment for hiring data scientists and having a tie to the mission of healthcare is important.

ALSDURF: How are you training teams to manage these technologies?

REMORENKO: We set up a center of excellence that manages our RPA internally and the support of the bots as well. I have four dedicated FTEs for the revenue cycle for both hospital and physician side.

HOPFENSPERGER: When we put AI in, we added a training program that case managers and physician advisers have to go through. 

WILSON: We don’t have any dedicated people who do training or monitor the bots. It’s kind of baked into our leaders over each of their areas, monitoring their own metrics. Because we do AI as a service, we have a service company that we partner with; we get reports from them that share information that we monitor.

ALSDURF: Where do you think you need to go next in terms of AI within the revenue cycle? What do you think could be really impactful for your organization?

REMORENKO: For us, patient experience is an area we’d look at tying into our strategy. One of the challenges is that we really need to understand the data, and it takes time, resources and investment. 

WILSON: Making that transition within revenue cycle from fee-for-service to value-based medicine. How do we mine that data to feed our outpatient world so that we’re connecting the right patients to the right preventative medicine?

BLANK: Thinking from the gross business model perspective, we’re hoping to shift from payer-driven to focus more on provider-driven data to give us a better understanding of how contracting should be done. 

REMORENKO: In the area of prior authorization, the sweet spot might be an area where both providers and payers can achieve cost savings and value. With COVID, we’ve been hearing that payers have been hit pretty hard with staffing issues, impacting their ability to answer calls in a timely fashion, make corrections to claims — all these workflows. 

WOOD: One of our top denials is for prior authorizations, and getting the right procedure authorized since most of the payers only allow you to authorize three CPT codes at a time. So if you get a prior authorization for a laparoscopic procedure, and the procedure turns into an open procedure, you only have 48 hours to go back to the payer to change the authorization to the new CPT codes.

HOPFENSPERGER: Nurses and physicians are expensive, and utilization review takes time away from higher-level clinical activities. Finding words or phrases in a chart, or matching those words or phrases with guidelines is an already-solved problem with AI in 2021. I think it’s puzzling that providers and payers have large numbers of people doing this when those resources could be put to better use. 

ALSDURF: With the technology in place, we need to ask how we can increase the value to ourselves, our payers, payer partners and solve problems of cost. 

HFMA Roundtable




















About Optum

Optum is a leading health services innovation company dedicated to creating simple, effective and comprehensive health solutions for organizations and consumers across the whole health care system. Optum is one of two distinct businesses within UnitedHealth Group (NYSE:UNH). 

The published piece is provided solely for informational purposes.  HFMA does not endorse the published material or warrant or guarantee its accuracy.  The statements and opinions by participants are those of the participants and not those of HFMA.  References to commercial manufacturers, vendors, products or services that may appear do not constitute endorsements by HFMA.


googletag.cmd.push( function () { googletag.display( 'hfma-gpt-text1' ); } );
googletag.cmd.push( function () { googletag.display( 'hfma-gpt-text2' ); } );
googletag.cmd.push( function () { googletag.display( 'hfma-gpt-text3' ); } );
googletag.cmd.push( function () { googletag.display( 'hfma-gpt-text4' ); } );
googletag.cmd.push( function () { googletag.display( 'hfma-gpt-text5' ); } );
googletag.cmd.push( function () { googletag.display( 'hfma-gpt-text6' ); } );
googletag.cmd.push( function () { googletag.display( 'hfma-gpt-text7' ); } );
googletag.cmd.push( function () { googletag.display( 'hfma-gpt-leaderboard' ); } );