How AI can eliminate bias and inefficiency in health care billing [PODCAST]

Listen and followApple PodcastsSpotifyYouTubeMore apps · Browse all episodes

Physician executive Tim Wetherill discusses his article, “Why AI is the perfect neutral arbiter for health care claims.” Tim explores how deeply embedded incentives, inefficiencies, and subjectivity in the current claims adjudication process create opportunities for manipulation and inefficiency. He explains how AI, when trained on clear, bias-free rules, can standardize decision-making and eliminate ambiguity, reducing the need for multiple vendors and decreasing billing errors. Tim argues that AI is uniquely positioned to serve as an unbiased third party—improving outcomes for health plans, providers, and ultimately, patients. His insights point to a future where fairness and efficiency drive health care claims resolution.

Our presenting sponsor is Microsoft Dragon Copilot.

Want to streamline your clinical documentation and take advantage of customizations that put you in control? What about the ability to surface information right at the point of care or automate tasks with just a click? Now, you can.

Microsoft Dragon Copilot, your AI assistant for clinical workflow, is transforming how clinicians work. Offering an extensible AI workspace and a single, integrated platform, Dragon Copilot can help you unlock new levels of efficiency. Plus, it’s backed by a proven track record and decades of clinical expertise and it’s part of Microsoft Cloud for Healthcare–and it’s built on a foundation of trust.

Ease your administrative burdens and stay focused on what matters most with Dragon Copilot, your AI assistant for clinical workflow.

VISIT SPONSOR → https://aka.ms/kevinmd

SUBSCRIBE TO THE PODCAST → https://www.kevinmd.com/podcast

RECOMMENDED BY KEVINMD → https://www.kevinmd.com/recommended

Transcript

Kevin Pho: Hi, and welcome to the show. Subscribe at KevinMD.com/podcast. Today we welcome Tim Wetherill. He’s a physician executive. Today’s KevinMD article is “Why AI is the perfect neutral arbiter for health care claims.” Tim, welcome to the show.

Tim Wetherill: Hey, thanks, Kevin. Excited to be here.

Kevin Pho: All right, so briefly tell us a little bit about your story and then the article itself for those who didn’t get a chance to read it.

Tim Wetherill: Sure. I’m a general surgeon by training. I grew up in the Philly area, then trained at the University of Kansas and did everything from trauma to robotics to bariatrics in private practice. Then I moved to the VA, where I was chief of the VA for Montana for a bit. Then, not by any plan, I just gravitated to the payer world one day and worked for some Blues plans for about ten years, and then transitioned to an AI startup, where I’m the chief clinical officer today, at Machinify.

Kevin Pho: All right. Excellent. So, for those who didn’t read your article, “Why AI is the perfect neutral arbiter for health care claims,” tell us what this article is about.

Tim Wetherill: Yeah, it’s really about using AI to read through medical records and claims and find the truth. We all know that with medical records, especially when you’re dealing with long inpatient stays, the average length is like 800 pages now. We’ve even seen one that was 100,000 pages. That actually happened.

You’d think that was for somebody who was in the hospital for six months. I think it was for a bill that was under $5,000. So it was for something ridiculous, but trying to filter through all that is really difficult. And we all know that humans are humans, and we have error rates, and we have different ways to interpret. I mean, go ask two different orthopods about how to treat a fracture. It’s OK to get different opinions sometimes. We all know how it is. It’s part art, part science, and that comes through, unfortunately, in reviewing claims and medical records. It’s not completely a hard science, and there’s bias.

There’s also pressure from above. In some organizations, the pressure might be for denials. I never saw that, but I think some organizations might have that. But the AI isn’t about that. It just spits out the facts and the truth. What you do with the truth is a different story. That’s up to you. But our job is to give the reviewer, whether it’s a nurse or a physician, enough information to decide what actually happened, whether the code is valid, and how the determination should be rendered.

Kevin Pho: So give us some insight in terms of the payer side. Before the advent of AI, how were claims typically evaluated on a routine basis?

Tim Wetherill: Yeah. So, you know, claims are the bills that come in, and they’re pretty superficial. They’ve got ICDs and CPTs or HCPCS codes and patient identification. So there’s not much there, and a lot of that is automated. If they do get flagged, it’s usually flagged for reasons that are not great, like there’s a certain dollar threshold, or an unlisted code came in, or, I don’t know, something fishy is there, like a bill for eighteen wheelchairs in a day, something like that.

Then when it gets handed off to a reviewer, it can be anybody from a coding specialist to a clinician, who usually looks at the medical record to decide, “OK, what actually happened here? Does this claim match up to the record?” Now, the payers, still to this day, have a tremendous manual process. It’s not super sophisticated. They have a tremendous amount of policies, guidelines, and rules that need to be applied. Some are payer-specific; some come from CMS and the feds. But when you have so many rules, and these rules can be thirty-page documents, and you’re asking a team of, you know, fifty clinicians to apply them, I mean, you can do the math and figure out what happens.

Kevin Pho: Yeah. So give us a sense of your particular AI system. How would it work in practice?

Tim Wetherill: Yeah. So sepsis is probably the most commonly abused diagnosis submitted for an inpatient stay because it pays the most. If you are a facility that uses DRG, a DRG is a grouping. It basically tries to say, “Well, not all conditions are the same, so let’s pay the hospital based off of how much resources were used.” So sepsis is highly abused because it pays a lot. Sepsis criteria are a little nebulous and difficult to tabulate, but we work with a payer per their policy and their guidelines. And, you know, we don’t make up the science; neither does the payer. We use clinical literature, whether it’s Sepsis-3 criteria or what have you, based off of sound science, and we use those rules, and we train the AI to look through the record and find the specific details.

So sepsis is about: Do you have an infection? Is there a dysregulated immunity response? Is there organ dysfunction? We are categorizing that and giving the user an output. We are not necessarily making a determination. We’re not in that business. We still believe that this is not self-driving cars just yet. So the nurse or the physician who’s reviewing the output of the AI system will then also have the record there to validate and confirm that the AI was correct, accurate, and consistent.

What we have seen is that the clinicians are now reviewing records much more quickly. That means that providers are getting paid sooner. The accuracy is improved, and the consistency is absolutely improved, which is good for everybody.

Kevin Pho: Do you find that there’s any discrepancy between the human reviewer after the first pass with the AI?

Tim Wetherill: Yeah, you know, there’s always room for error. Some of it is just the OCR, the optical character recognition, just doesn’t work. It fails because the PDF is really dirty or messy, and so the numbers might be misinterpreted. The other thing is, there are those corner cases. We all know it: the zebras. We train the system to actually call that out and say, “I’m not sure,” or, “This needs further investigation.”

Again, we respect, and I’m very proud to work for a company that hired a clinician to say medicine is not black and white. It’s binary. So you are going to need that expertise to interpret and make that judgment call. And absolutely, what we really focus on is, let’s get the easy cases out of the way so that the clinicians can use their brain power and their expertise to do the job that they were hired to do. So it’s really about less paperwork and more skill set.

Kevin Pho: Now, is there any feedback mechanism in terms of the AI evaluation getting back to the clinician and the local hospital system, in terms of how they could improve their documentation to better fit the coding that they’re submitting?

Tim Wetherill: Yeah, there’s been feedback from payers to providers, to hospital systems, and to doctors all the time. What’s crazy is that I used to get those letters. I never read them. I tore them up and would throw them away. So I’m a bad example of learning and adapting, but what payers know is that most providers actually don’t change their behavior, even though that behavior is very simple to change.

With the AI system, there is absolute feedback. It’s both automated and manual. We learn from the users plugging in information or adjusting the output, and then we also have feedback sessions constantly with the teams to say, “What did you like? What did you not like? What was good? What was bad?” So there’s lots. Especially at this stage of the game, there’s a tremendous amount of feedback going on.

Kevin Pho: So how common is AI penetration in terms of the payer evaluating billing claims? Is it fairly common that this is happening? Because it seems like an intuitive next step.

Tim Wetherill: Yeah, you know, payers, from a tech standpoint, are not the most sophisticated industry. They still use faxes and DOS-based systems that aren’t even supported by the companies that made them twenty years ago. And I’m not kidding. There are still systems that were made in the ’90s that are core functionalities of payers. So they are trying. Sometimes the motivation just isn’t there. Yeah, the feds pass rules and everybody’s complaining, but till there’s something pretty harsh, like a new standard or a new requirement from CMS, it’s not going to happen overnight.

I think there is the will and intention, but this industry moves, you know, so damn slow. And unfortunately, this is one area where it’s not that difficult to put some of this in play, especially as you have software systems that are cloud-based, where you really don’t need to integrate much with your legacy systems. You can adapt, but it’s more of a culture problem than anything.

Kevin Pho: So currently, as we’re speaking in April 2025, would you say just a minority of payers are using AI tools to adjudicate billing claims?

Tim Wetherill: Yeah, I bet it’s the minority of the claims, actually. I mean, look, there’s probably some, and I’m using air quotes here, “AI” that touches the majority of claims, but it’s pretty unsophisticated stuff. It’s like matching an ICD to a CPT code. You can call that AI, but it’s really just a rules engine. But true benefit of improving quality output with AI, I think, is absolutely a minority right now.

Kevin Pho: So I can see some usefulness, certainly in a primary care setting, because, as you know, I’m a primary care physician. A lot of doctors, we just simply put our own code in after writing the notes. So I can see a scenario where AI can just evaluate whatever we’ve written, evaluate the chart, and just kind of give a suggested code based on our A, and have us not worry about it.

Tim Wetherill: Yeah, I think that’s absolutely an opportunity. I think the only concern I have with solutions like that, and you can flip that to the payer side as well, is that I personally don’t want AI to have a bias or intention underneath that says, “I want Kevin to get maximum profit out of this visit.” In my mind, that should not be the rule. It should be about, “Kevin, you documented this. Here’s what the CPT code should be for that visit, that problem, that procedure, what have you,” and not try and nudge you.

There was a use case where the DOJ nailed an EHR company. This was a few years ago; I think it was 2019. There was a kickback scheme with an opioid manufacturer, and so docs were getting prompted to prescribe more opioids. Well, we all know how that works out. That kind of nefarious business practice, we need to all be aware of as physicians, and know, “OK, if you’re going to give me an AI tool, that’s fine, but let me understand what the intent is and how it was designed.” That is a step that I think we are missing.

I know that a lot of companies do have physicians, a lot of companies don’t have physicians, and there is essentially zero federal oversight over this stuff. There’s probably oversight in the clinical stuff, but on the business side, there’s very little oversight, and that’s what makes me nervous: that this could just accelerate the current problems we have.

Kevin Pho: So right now you’re saying there’s just not a lot of safeguards, because the AI perhaps can be manipulated with a prompt saying, “What more can the doctor add to maximize revenue?” Or even for the payer side, “How can the AI deny certain procedures or medications?” So right now we’re kind of in the Wild West. There’s not a lot of safeguards both ways.

Tim Wetherill: Yeah. I mean, the only safeguard that we have is the oath that you and I and others took, and that we have a role in this. That might sound scary to some people, but even though I’m not practicing anymore, I still live by that oath. Ultimately, I respect that anything I do, even though we’re so far removed, there’s a very powerful human-to-human interaction that happened, and I am not going to make that worse. If anybody starts pushing me to do that, I’m gone.

Kevin Pho: So on this podcast and on KevinMD, we always talk about the different ways AI can intersect with the various aspects of health care. Specifically as it relates to billing and coding, and you’re an expert in that, where do you see the next steps being? Because so much has changed over the last several years or so. So what do you see in the foreseeable future, specifically with that intersection between AI and billing and coding?

Tim Wetherill: Yeah, I think there’s a huge opportunity, again, to have this neutral arbitrator that both sides, payer and provider, can rely on and trust, and that can do two things. It can stop the fighting and improve the timeline, so everybody can get paid accurately and quickly, and there’s no guessing about what’s going to happen. If we can get out of the way, there’s greed on both sides. No one’s innocent here. Let’s not play games.

If we can get beyond that and realize that this arms race that we’ve been in for decades about coding and billing, if we just agree that this is essentially a mutually assured destruction scenario, and just say, “Hey, let’s just agree on using this one platform or system that can arbitrate neutrally,” I think that’s the way to go. And I think that is possible. But again, the tech’s the easy part. It’s the culture part that is difficult.

Kevin Pho: So, a tangential issue: You could easily see a scenario where AI can take over all of the coding-related issues when it comes to a chart. What about the profession of coding itself? Do you worry about that whole profession of coding just going by the wayside, being replaced by AI?

Tim Wetherill: No, not necessarily. In fact, they’re really important to help. AI is based off of rules, and coding is really complex. I remember just trying to learn it, and I still get taught by my coders every day about these little nuances. There are so many scenarios, and ultimately it’s based on the language we use, and it gets a little bit legalese in how you interpret records. But you need those experts to help train the models, to teach it and learn it.

I think where we really want coders to evolve is, we keep getting more and more complex and confusing with coding, and maybe I’m naive, but I’d love to see us go the other way and say, “Let’s try and simplify this.” I mean, every time we try. Like DRGs: They were started in the ’80s, and they were a way to simplify and try and control costs by putting different services in different buckets. To my understanding from looking at the literature, there’s never been an indication that it has controlled cost at all. In fact, all it did was create loopholes that people were able to take advantage of. So I wish there was a way. Maybe, again, I’m just being naive to think that we could do that.

Kevin Pho: We’re talking to Tim Wetherill, who’s a physician executive. Today’s KevinMD article is “Why AI is the perfect neutral arbiter for health care claims.” Tim, let’s end with some take-home messages that you would like to leave with the KevinMD audience.

Tim Wetherill: Yeah, sure. I think one of the biggest lessons that I learned: I had been in pure private practice. I was employed. I worked at the VA. I did some prior reviews for Blue Cross plans, and I thought I knew a lot. Then once I got in the payer space, I didn’t know squat. It is incredible how complex and confusing this world can be. And there are a lot of things that payers do that are good.

I think that a lot of the physician voices are very loud and uninformed, and they’re working on denial letters and their opinion about care. The problem is that because you don’t understand the other side, it creates more chaos, especially when physicians are helping design legislation. Some of the prior auth legislation looks good on paper. It’s actually increasing denial rates. So we need to have people who are curious and willing to learn about both sides, and then hopefully design solutions based off of that.

Kevin Pho: Tim, thank you so much for sharing your perspective and insight, and thanks again for coming on the show.

Tim Wetherill: Thanks, Kevin. Really appreciate it.

Listen and followApple PodcastsSpotifyYouTubeMore apps · Browse all episodes