Conrad Coopersmith, general manager of coding automation at AGS Health, recently joined Trending NOW, a Healthcare NOW Radio production hosted by Shahid N. Shah, to discuss the intersection of artificial intelligence (AI), automation, analytics, and professional services in medical coding.
If this industry is increasingly going to be dominated by AI and automation, what is it about human labor that remains indispensable? There’s no way to get rid of human labor. What aspects of coding will AI never be able to replicate in technology?
Coopersmith: As I reflect on your questions, I would say they are linked. And what I mean by that is I don't think about skilled labor versus automation and AI. I think about augmenting skilled human labor with automation and AI. I think there are very specific use cases that actually fit both.
Let me give you a couple as I expand on it.
When we think about what I would call simpler high-volume visits for things like diagnostic imaging or emergency medicine, as examples, there are a lot of things that a machine can do faster and arguably more precise to code than a person can do because the nature of those visits is largely consistent and knowable, which presents different opportunities to save time and money, not to mention to save precious resources, which are skilled coders.
If you contrast that in more complex areas of medicine like neurology, cardiology, and oncology, to name a few, we're going to need a combination of coding automation and people in the loop to be able to render decisions because there's more information to account for. Not only is there more information to account for, but the rate of change with that information is greater, and that information has to be contextualized to arrive at final codes in order to bill. And that's taking into account everything from the history of the patient to clinical documentation from EMR, notes from the physician, orders, labs, diagnostic data from imaging, and so on. In short, the process itself is far more nuanced, requiring the expertise of the skilled coder in conjunction with automation.
So, whether it's a relatively simple visit or a more complex one, there's an opportunity with both to enhance coding outcomes when you marry a human's judgment with technology. The irony of that is that it ultimately accelerates the capacity for the technology to learn. I think both people and different aspects of automation are going to exist in tandem for quite some time.
Coders with the right technology can become even better because the technology gives them additional clinical and coding connections. What can generative AI and modern AI systems provide for human coders that we couldn't even dream of five or ten years ago?
Coopersmith: First, I'd like to talk a little bit about the biggest myth that I've encountered to date, which is that AI is going to magically and autonomously replace the profession of medical coding by moving literally everything directly to billing with no human intervention at all in the coding process regardless of the visit type or the complexity. I'd like to publicly reject that as a false premise because I think all it's really done is create a lot of noise in this industry.
So we start with the premise that medical coding by people is not going away and there are additional opportunities to gain efficiency through automation. In some cases, the specialty to autonomously code charts means they are directly billed without human intervention. We should be focusing on pragmatism with all of those scenarios in mind.
This is as much about a healthcare organization’s mindset and the division of labor as it is about technology itself.
So, when I think about the adoption of medical coding automation, it starts with essentially assessing what the jobs to be done are today and what opportunities can create greater productivity, which will affect the bottom line. And once that has been called out, the work's been segmented, and the opportunity has actually been quantified in terms of what people do, what automation does, and what autonomous technologies can do, there need to be time frames established to solve for that and there needs to be a representation from key stakeholders in solving for it. It needs to include all of the key stakeholders, including the physician community, coding, IT, and revenue cycle, and there has to be support from the C-suite and the willingness to adapt operations in the future with that change in mind. I think the willingness of healthcare leaders to fully reimagine those processes and then to be able to demonstrate patience to achieve an end state and, more importantly, to support the team that's doing it, is as important as the outcomes that can result in the form of savings when it comes to time, money, and resources.
Describe your term achievable automation and how senior leaders can use it to help quantify the art of the possible and achievable.
Coopersmith: It starts with defining the actual use case. I think there are several where people will fit with using some form of automation. Automation may exist solely based on the complexity and the type of the visit. It starts with defining that use case to quantify whether the problem's even worth solving. A lot of times, people take a ready-fire-aim approach to problems, so it's easy to get ahead of ourselves, and that can present challenges.
Does AGS Health have ways to quantify, categorize, and classify to assist in this process?
Coopersmith: Sure. Part of our core business is delivering medical coding services, which is something that we're steeped in. It's in the DNA of the company. So, it's not just about technology. It's more about understanding the operational impact of what you're trying to solve for.
What I would also share is that we've all been a part of automation projects that have not gone well. And there are pretty simple reasons for it. When it doesn't go well, people usually don't think through the desired outcomes in advance. They don't involve the appropriate stakeholders integral to the process. There are dependencies on those stakeholders when it comes to data and none of that gets memorialized at the initiation of the project by the executive sponsor and revisited on a regular basis to see if they're actually getting the outcomes.
Drawing on several years of learning, I would categorize two factors that lead to challenges with implementing automation. It’s as simple as people and data.
When we think about people, there's often a lack of awareness as to who should be involved with the project, the role they're going play and, in the use case of coding automation, it should be coders themselves, representation from physicians and clinicians, IT, and revenue cycle and all those parties play a role.
If you peel it back another layer, physicians and clinicians are on the front lines of care delivery, which manifests itself in services rendered. IT is the group that enables the delivery of data for coding and it's coming from a myriad of sources, including EMRs and other systems housing documents and diagnostic data. Coders have the role of bringing all that information together, putting relevancy to it, and making sure that in this context, patients don’t overpay for care and providers aren’t underpaid for the care. Finally, the revenue cycle is responsible for dropping the bill and making sure it's paid. So that's the people side of it.
On the data side of it, it’s key to make sure that all systems are accounting for where the information's coming from, that interfaces are mapped, and that technical dependencies are clarified because when the flow of data for coding and the disposition of that data for billing is not understood it tends to lead to downstream workarounds, and usually what that means is time, money, and resources are worse and revenue loss, which defeats the purpose of putting in automation in the first place.
Tell us about a good way to use this AI push from executives to get our data house in order.
Coopersmith: It starts with an assessment of where the data is coming from and what it looks like. And as you know, there are lots of different types of data. There is structured data and unstructured data, all of which flow into a model, whether it's coding automation, using some natural language processing and computational linguistics, and the other aspects that go into building AI models for full-blown artificial intelligence. It starts by creating an inventory of processes and systems, and then making sure that data can be delivered in a way that's repeatable and consistent.
How much does AGS help in creating an inventory and other processes?
Coopersmith: We're part of it on the front end and during the delivery phase, helping our provider partners think through where it should be coming from and what it's going to look like based on what they're trying to solve for.
What are the key qualities that healthcare leaders should be looking for in their technology partners?
Coopersmith: In deference to technology in general, like anything else, I would begin with the end in mind by defining what the organization's trying to accomplish, fix, or avoid by going through a process to begin with. Once there's an actual tangible use case identified, I would focus on involving the right people and convening a small team to conduct research on new technology. As an organization starts to diligence partners, and we help organizations with this quite a bit, what I would pay attention to is the kind of technology partners in the market, what they believe they are really good at in terms of core competencies, their track record of delivering those things, and a specific knowledge of the business problem, in this case coding, that the organization is trying to solve for. And I would be looking for actual outcomes from current customers.
The other thing that I would say is I would test the partner's knowledge of coding to understand if they understand what has to happen operationally to take advantage of the tech because there's a lot of companies that look at problems and opportunities through a technology only lens and that often does not comport to reality. If they speak in absolute statements, if they use three-letter acronyms and paragraphs, if they can't provide proof of results, you have good reasons to be skeptical, and I would push on that.
What do you see in the marketplace as it relates to potentially replacing some percentage of billing and medical coding professionals?
Coopersmith: Based on real-world experience, I haven't met a provider customer in the last five to 10 years who's looking for less people. They're actually looking for more people and looking to do more with less because that's the reality of the situation. We're not creating more medical coders. In fact, they're retiring and leaving the profession, so they need to figure out how to do more with less. It's more about the division of labor, segmenting the work, and redefining what their jobs are today. Through automation, we're seeing that traditional coders focus on items like audit and review, and those jobs are changing over time. We're also seeing healthcare organizations invest in educational tracks that are more related to programming, math, and statistics in the discipline of medical coding because they're able to better use coding automation and AI, and they appreciate the data that results from that. So, the jobs are changing. I don't necessarily think that fewer people are needed; they just need to figure out how to channel them in a different way.
What are your recommendations for medical coders and billers so that they can remain at the forefront of their profession and grow their skill sets as it relates to auditing, improving their reviews, and knowledge of the clinical side while leveraging technology, automation, and AI?
Coopersmith: As we discussed, medical coding is going to require a person in the loop for the foreseeable future, especially with more complex specialties. As things become more automated, the role of the traditional coder is changing to more of an auditor, providing final review and sign-off prior to chart adjudication and billing. Knowing that that shift is happening now, I would encourage investment in career paths that incorporate base understandings of AI and automation so they can better appreciate the output of it.
Pragmatically speaking, knowing what to do is different from knowing how to do it. In my experience, the best way to prepare a team to handle complex and different tasks when jobs inevitably change over time is to pick an actual use case implemented in your organization and iterate on it. In this context, it could mean cross-training existing coders on audit functions and then using automation to help facilitate that.
What would you recommend leaders do in a specific use case example?
Coopersmith: I would recommend visiting a noncompetitive environment and looking at their existing service line from soup to nuts in terms of a person being admitted, a person receiving those services, coding for those services, and billing for those services. I would also recommend that the executive team actually go through those functions to internalize what that might mean to their own organization and use it as a live lab.
Thinking more visionary, what does the future look like? Where do you see the technology going that we couldn’t even imagine a few years ago?
Coopersmith: We've been talking about the Affordable Health Care Act since 2013, which was over 11 years ago. The natural or the underlying premise of it, in addition to making health care more affordable for Americans, was to set the stage for value-based contracting. And as value-based contracting continues that evolution towards reimbursement for outcomes and quality and computing accelerates, we're going see additional opportunities for coding automation in more complex medical specialties and we're also going see it in nontraditional care sites. Core to that acceleration and the expansion of technology will be access to data from nontraditional data sources, such as telehealth and ambient voice recognition. There are going to be very few companies that are able to harness data for new sources, code that data, tie it to billing operations, and maintain flexibility to accommodate a division of labor, which is going to continue to require people to perform medical coding, use of coding automation and true autonomous technology. I think those new tranches of data are going be a unique opportunity, and there's going be very few companies that are going be able to take advantage of that. As a result, I think we're going to see tremendous consolidation across revenue cycle coding and the whole continuum of the process itself.
That sounds like a great visionary view. Conrad, thank you so much for your insightful perspectives on this very evolving role of automation and medical coding and balancing technology with skilled labor.
Author
