Beau Bailey, M.D., medical director of appeals and denials for Utah-based Intermountain Health, recently spoke with Healthcare Innovation about the role AI-based tools are playing in clinical documentation integrity and in making health system denial appeals processes more efficient. 

Healthcare Innovation: Dr. Bailey, could you talk about your role as medical director of appeals and denials? And does Intermountain have a centralized appeals and denials unit within it?

Bailey: I am the medical director of of appeals and denials for what we call Canyons and Deserts — that’s what we would say is the old version of Intermountain prior to the merger. So I cover Utah, Nevada, and Idaho, and I have a counterpart who does the same thing over the Colorado and Montana market. But it is centralized in the sense that we do everything pretty much together — written appeals and peer to peer. There is a little bit of a regional basis to that, just because we like to maintain some relationships with the medical directors for the payers. But yes it's all centralized and integrated. 

HCI: With an organization that's as large and complex as Intermountain, it seems that some kind of standardization would be important.

Bailey: Absolutely. We do that in a number of ways. We have a tool we call the FIT tool — financial impact tool — that streamlines the process. It has accounts that are workable at every level simultaneously, so case management can receive the denial and put it into the tool. It automatically routs it to whatever step it needs to go to next, whether there's a peer-to-peer opportunity or a written appeal opportunity. That's where our physicians document their work and outcomes. The nice thing is that it gives us some objectivity to our outcomes. 

We use quite a bit of AI already to standardize our approach. We have an agent that helps standardize our written appeals and another one that helps with our peer-to-peer preparation. There's no question that in appeal work, there's a ton of variation that's stylistic based on who's doing it, but we're trying to eliminate that variation and define best practices because honestly, a weird thing about medicine is that everything we do is as evidence-based as possible, but there's just no evidence to dictate what we do from an appeals denial standpoint really. So we kind of invent it as we go.

HCI: I read that you have something called Intermountain Physician Advisor Services that integrates utilization management, documentation improvement, and denial prevention. Could you talk about that program?

Bailey: We have a very robust physician advisor group that has several prongs. We have a utilization review team, and that is led by a physician medical director and an operational dyad partner. We have another branch that is appeals and denials. I lead that with an operational dyad, and she is exceptional. She is our statistician guru, too. Then we have an education branch, and that has a medical director who has regionally appointed education specialists who do education with providers by service line and hospital systems. Clinical documentation is integrated into the whole process. We have a very robust physician advisor team that we call the IPAS or Intermountain Physician Advisor Services, and we work really closely with nursing case management. It’s just a big integrated team. 

HCI: You mentioned an AI element to some of the tools. What other kinds of AI clinical documentation improvement tools has Intermountain implemented and what kinds of outcomes have you seen?

Bailey: For years we've used Solventum products that are nudges for real-time documentation improvement, and we track natural compliance with that, and then basically nudge compliance. How often do providers actually respond to those nudges?

[Intermountain has used Solventum CDI Engage One and 360 Encompass CDI. CDI Engage One uses natural language understanding to analyze documentation in real time and provide automated prompts to physicians and advanced practice providers when additional specificity or clarification may be needed, while 360 Encompass provides CDI specialists with access to supporting evidence and the documentation workflow.]

I don't know if you'd call it AI, but certainly we use technology for query escalation, and then we track that by response rate. We use some self-developed agents to standardize our letter format and we also use an agent that helps with peer-to-peer preparation. 

We're in a number of conversations with companies that provide a lot of these administrative services. We don't have contracts with any of them yet. But what we have found is that based on our in-house development from a physician advisor standpoint, we really haven't found anything yet that augments what we already do with our own FIT platform. We have had a lot of talks with people, but we haven’t ended up moving forward yet on a lot of it because we've already developed much of it ourselves. But it is the future. 

I would say, from a physician advisor standpoint, the biggest outcome is by far efficiency and just the amount of manpower work that goes into the workload without sacrificing quality. 

HCI: Do you track metrics about improvement in either number of denials overturned or the speed at which all of this is happening?

Bailey: We track the amount of time that we commit — that’s the best way to say that. We used to be somewhere between an hour and a half to two hours per written appeal. We've got that down to about 33 minutes by our best estimation. We look at overturn rates for our physician advisors from an appeal standpoint, but one of the issues that complicates overturn rates is when the work becomes more efficient, you also do more work, so we find that we argue cases that we may not have in the past. 

HCI: I saw you quoted in another publication as saying that AI helps clinicians, coders, and administrative teams speak a more common language. Could you explain what you meant by that and how it helps?

Bailey: I would say that the biggest problem with burnout in healthcare is the burden of administrative work that clinicians really don't understand or have the capacity to do.

When I talk to providers who are frustrated, it's because of all the additional administrative tasks. One, they don't understand why they're necessary, which is what our education role is, and two, the way that we speak about the business of medicine is so divergent from the way that we speak about medicine clinically. 

As a brief example, when I write a note about a patient that I think is quite stable, and I'm communicating to the care team about the clinical scenario, I might focus on how much clinical improvement there is — look at how stable this patient is. They're responding to therapy. But when a payer sees that, they see trivialization of an already sick patient. This is all relative to the acuity in which you find the patients. So it's pretty easy for clinical documentation, just by nature, to clash with medical necessity. AI can give us the opportunity to say, "Hey, you're saying this heart failure patient looks relatively stable. Are there ways to augment your documentation with more specificity to demonstrate the severity of the symptoms that you're managing, the severity of the diagnoses that you're managing?” But there's no point in any training where an internal medicine resident gets told when you document atrial fibrillation, there is a coding difference between permanent and persistent. Those are just not common nomenclatures we use in clinical medicine. There also are a lot of assumptions in clinical medicine in term of the way that we see things, that we talk about things, and those assumptions don't exist in the coding world.

HCI: There’s been a lot of talk about a recent Blue Cross Blue Shield Association analysis that just came out that purported to show a connection between health system AI deployments and increased billing. Is there an increasing tension between providers and payers on the impact the tools are having, and an escalation in terms of whose AI tools are better?

Bailey: You mean an arms race? Yes. I mean this is the unfortunate nature of the beast. I would say the incentives for the payers vs. the providers just feel divergent. The first conversations about AI were about weaponizing it for profit margins as opposed to improving clinical or compensatory outcomes, and really that's because the motivations are different, right? We'd like to say that everybody's trying to be good stewards of the dollar, and we're trying to reduce fraud, and we're trying to be compliant, and we're trying to appropriately resource healthcare. And healthcare is too expensive. There's no question that that's true, but it's also a gigantically profitable business that is not patient-centric. I think that AI has the potential to add to the problem even more so than to resolve any of the issues that have been developing over the last 15 to 20 years.

HCI: Has Intermountain developed partnerships with any payers to improve data sharing or use these AI tools to make the flow of data between the two smoother and cut down on denials that way?

Bailey: We haven't yet, from a proactive denial prevention standpoint. We do a ton of collaborative work on the back end, but I would say that that's still more relationship-based work. We are active in all sorts of conversations about how we could improve clinical data access to the payers in a way that doesn't end up being harmful or detrimental as well because there are proprietary things, there are patient care issue things. There's so much that goes into access to the EMR, and I would say that so far Intermountain has been very, very cautious with that. I do not oversee that. I have no idea of specifics about that, but it's a very hot topic.

HCI: Does the prior authorization workflow also fall under your purview, or is that a separate thing?

Bailey: That's separate. Certainly we educate around denials related to prior authorization. If there's a payer workflow, a prior auth checklist that needs to be completed, we delve into that every once in a while to identify the concerns, but we don't have ownership or accountability over any of those prior auth processes. We certainly do a ton of the appeal work on the back end, then validate that the process had been followed, but our service lines take accountability over more of those prior auth issues.

HCI: Are there other areas where AI or agentic tools might be able to help streamline even more?

Bailey: I don’t oversee clinical decision support tools, but there's no question that AI is going to help improve clinical thought processes, especially when it involves rare and infrequently seen problems. The other thing I think it might be really useful for is in identifying the next step of care workflows, like early identification of patients who may need SNF or in supporting SNFs in weekend authorization management because that's a huge problem that's delaying discharges. So there are a lot lot of areas we could really use it to improve status determination, diagnosis, decision support, and transitions of care.