Turning AI Into Performance in Value-Based Care Contracts

Artificial intelligence (AI) is rapidly reshaping healthcare, yet many provider organizations still struggle to move from interest to practical application particularly when it comes to value-based care contracts. While AI promises efficiency and insight, turning it into measurable performance and financial advantage remains a challenge. 

In this episode of Value-Based Care Insights, host Daniel Marino explores how AI can be used to address one of healthcare’s most complex challenges: evaluating and managing value-based contracts. Drawing on real-world experience, Daniel explores how AI can be used to build performance models that connect contract design with operational outcomes across clinically integrated networks, ACOs, and other value-based arrangements.

Joined by Eddie Diaz, a data scientist with more than 15 years of experience in value-based care analytics, the discussion highlights how AI can bring clarity to attribution, risk, quality, and financial performance to help CFOs, CMOs, and managed care leaders better understand contract opportunities, risks, and results. Together, they explore how AI can move beyond hype to deliver real, actionable value in value-based care contracts.  

LISTEN TO THE EPISODE:

 

 

Daniel Marino:

Welcome to Value-Based Care Insights. I am your host, Daniel Marino. Like the rest of the world, all of us are trying to figure out what artificial intelligence means. And, for many of our listeners, if you've been tuning in, we've had a number of episodes, a number of conversations with guests talking about, the impact of artificial intelligence, AI, on healthcare. Everything from business operations to the impact on clinical diagnoses, how we can incorporate that in clinical workflow, and so forth. So, within our consulting firm, within Lumina Health Partners, we as well have been trying to figure out how to incorporate artificial intelligence and I've thought for a while that as we start to incorporate this into a lot of the work that we're doing, it could certainly create a lot of efficiencies, but like with everybody else, we were trying to figure out where. And for years, with a lot of the work that we've been doing, one of the areas that I've constantly been challenged in is coming up with a contract management tool, contract evaluation tool, that is really geared towards value-based contracts, And a tool that could really help CFOs, chief medical officers, managed care officers, a tool that could help them evaluate value-based contract. And how to really see the opportunities and the performance outcomes, that would come out of a contract. And a value-based contract is challenging, right? Because you have a lot of other variables that aren't in a fee-for-service contract. You have attribution, you have, you know, different definitions of risk. You know, you're thinking about the different programs, and oh, by the way, quality is also an important component. So it's been very difficult to create a contract management tool, and frankly, I've worked with a number of analytic companies to try to create one.

Well, I'm really excited to share with you today the work that we are doing around artificial intelligence, and how we've created our own performance model, that is driving, or let's say, correlating a value-based contract with actual operational performance outcomes within a clinically integrated network, an ACO or some type of operational value-based performance. It's quite exciting. And here to help me talk through this is my colleague, Eddie Diaz, and Eddie is phenomenal. He's a data scientist by nature. He's got 15, over 15 years of experience within value-based care, and has worked with numerous organizations on creating different analytic products, and is just really a super, super knowledgeable in artificial intelligence capabilities. Eddie, welcome to the program!

 

Eddie Diaz:

Thanks, Dan. Appreciate it. Excited to talk to you.

 

Daniel Marino:

So, Eddie, let's talk a little bit about the performance model. You know, we've worked on this with a few clients and have been refining this. For our listeners, talk a little bit about the robustness of the model, and maybe you can start a little bit with the data ingestion.

 

Eddie Diaz:

Sure, sure. So, the model fundamentally looks at a variety of different data formats, right? So, if we work with healthcare data, we have a lot of data in image files, or native versus you know, image PDFs. There's also a lot of documents that exist in terms of, PowerPoints. And then there's a lot of flat file data, like Excel-based data. All that kind of is disjointed. And the beauty of what we've done and what we've accomplished is that we've been able to extract all of that with extremely high fidelity, unify it to make it understandable, so from the contract, whether it's a PDF, from the payer reports, whether they're in PDF or Excel-based versions, we can unify all that and give it a semantically representational layer that can start to put intelligence in connecting all the dots.

 

Daniel Marino:

Well, and I'll tell you, one of the things that impressed me about what we've been able to do is that for one particular client, for one particular health system that we worked with, we ingested a tremendous amount of data, right? If I… if memory serves, it was something like, you know.

 

Eddie Diaz:

I think it was about 100 billion, right? So 100 gigabytes of data.

 

Daniel Marino:

Yeah, claims data, plus the documents that came back from the payers, plus the contracts, right? And we did this, you did this, in a relatively short amount of time. I think it was 7 or 8 weeks, which was unheard of before AI. I could recall a couple years ago working with a large analytics vendor, one of the top analytics vendors, and, you know, I think we've ingested probably a quarter of this data, and it took, at the minimum, 3 months. And to semantic align this data still was a bit of a challenge. I mean, it was really impressive what you all were able to do.

 

Eddie Diaz:

Yeah, one of the things we can highlight is how even just a year ago, with the advancements of these models, they're becoming more and more robust with increasing efficiency and productivity, that a year ago, we would have needed probably 2 or 3 other resources. And today, and maybe 6 months of time, whereas today we have weeks, and we need less resources. So it's an incredible achievement, and it's incredible how we could pass that on to the clients.

 

Daniel Marino:

So, a lot of times in these contracts, where I see the breakdown occurring is you've got the managed care team negotiating what they feel is a pretty strong value-based contract with the payers. And, you know, again, the value-based contract could include an MLR target, or a total cost of care target, maybe a RAF score, some quality components, and that sort of thing. And, you know, so that exists in the contract. And where I see the real gap occurring is the alignment with how that information then is either supported by clinical operations, or how clinical operations could support the performance of the contract, right? So one of the things that, you know, I'd love for you to talk a little bit about is how the performance model helped to close that gap. What did you see? And let's start with maybe MLR, because I think that's… this is a big area of focus. What did you see as an output of the model that helped to close the gap around the medical loss ratio, in terms of what's in the contract and the performance. By the, you know, by clinical operations.

 

Eddie Diaz:

Yeah, so one of the things that we were… that we were seeing is, because we can extract the key terms and the key drivers of the contract performance directly from the contract. Measure that up against the financial and quality outcomes. We can have a very clear line of sight on terms… in terms of what are the performance levers that these individual providers need to achieve. So, we could discover that if there's a gap, like we've seen in certain clients, of 5-10% between their MR and their actual performance, there are certain key utilization areas that we can highlight where they're significantly underperforming. Right? As well as quality gaps. So, in terms of utilization, we could identify, you know, high specialty to, to, high specialty per 1,000 performance, right? Right. Or, you know, or conversely, low PCP access, you know, as measured by PCP visits per 1,000. So those are some of the key levers that we were highlighting to say these are where your areas are. If you start to kind of change that juxtaposition, we can improve that baseline and get your MLR to where it needs to be.

 

Daniel Marino:

Well, and I'll tell you, that's a big challenge for a lot of organizations, because especially when, you know, you're thinking about the performance of these Medicare Advantage contracts, a lot of times, and what we've been seeing is that some of the… some of these… the healthcare organizations, you know, they'll have an MRR… MLR of 88%, 90%, 95%, and yet the target, you know, has to be 85, right? I mean, they've got to get it down, and frankly, if you're going to make money, you actually need to get it close to 80%. But what they struggle with is where to focus the attention on bringing down the medical loss ratio. And what's that blueprint? So, how did the model help create that path? That knowledge base in terms of where to put resources, how to really be able to manage the care. How did the model support that?

 

Eddie Diaz:

So, one of the things that I did, Dan, that we worked on was essentially helping the clients understand the relationship between the MLR and all of the downstream metrics, when we talk about inpatient utilization. ER utilization, physician services, like we were talking about before, pharmacy utilization. So, there is… there is a very clear line of sight in terms of the impact that each one of these specific measures has on the MLR. And based on their operational, competencies, right, and their most effective use of operations. We might be able to help them pinpoint, if you focus on these key areas, on avoidable ED visits per 1,000, on your physician PMPM rates, on your generic, you know, pharmacy prescription rates. They need to have the blueprint that works the best, that's complementary to their existing operations, in addition to understanding the roadmap to enhance their future operational performance.

 

Daniel Marino:

Yeah, you're absolutely right. And, you know, when you think about utilization, obviously utilization does impact the MLR, right? Because it's all driving around cost, but there's certain utilization categories that have a greater cost impact than others, right? Particularly, like, inpatient utilization. If you have a strong transitional care program that helps. But it's not just having the transitional care program. It's focusing on the right cohort of the patients, right? The high-risk patients, or those, let's say, high… those DRGs that are creating an additional, say length of stay, an additional day in the hospital that you can focus on to really bring down some of the costs. And the exciting thing about the model is you were able to identify that. I mean, almost down to that specific profile of a cohort of a patient that we needed to focus on.

 

Eddie Diaz:

Yes, that's the… that's the beauty of the overlay and the integration between the claims data, because we are allowed to double-click into different cohorts by disease care management programs, or by disease category and disease state. That there will be a high impact. In the entire population, right? And like we're talking about with risk, there's… there's always the… the fact that most of these organizations, especially, primarily fee-for-service health systems that… that are… that are not intentionally, but they're overlooking their rising risk populations. Sometimes accidentally. And that is an area that we've identified in particular that can really start driving some really good performance down the line by giving them those tactical insights.

 

Daniel Marino:

If you're just tuning in, I'm Daniel Marino, and you're listening to Value-Based Care Insights. I'm talking with my colleague, Eddie Diaz, and we're discussing the performance model that we've used to help organizations close the gap between their value-based care contract with payers and their operational performance. Eddie, I'm gonna turn the attention, or the discussion a little bit around, the risk adjustment factor, the RAF indicator, that is oftentimes used within the contracts to identify risk, and certainly within Medicare Advantage, it's an important component. And I'll tell you, in my experience. It's often difficult for maybe the clinical leaders or physicians or, you know, even some of the finance folks to understand what really influences the RAF, what the impact is on the RAF related to the contract, and what we need to do in order to properly have the RAF reflect our population. And just, you know, by way of explaining that, the risk adjustment factor is a way that you're able to quantify what the, what the risk is, or how sick your population is that you are managing. Most of the time, the RAF score is underrepresented of the… let's say, the sickness of the population, or the acuity of the population. So, Eddie, talk a little bit about how the model identified a gap with what was identified as the RAF score, but then also the acuity of the population.

 

Eddie Diaz:

Absolutely. So, we could think of RAF, as being calculated in multiple different, segments or entities that are essentially stakeholders of that same patient. You have the government, which is CMS, you have the provider, and you have the payer, right? So each one of them are kind of responsible, the provider being responsible for the activities that are necessary to enable a proper risk adjustment factor, and then that recognition when, through billing processes, gets recognized by the payer, and ultimately the government. But there's often wide gaps, and what those gaps… what we can do with the claims data, as well as with the benchmarking data that is provided by the payer, is see what that gap is by using claims data. So I could create, essentially a RAF score based on the population, and find a delta of where that underfunding potential might be.

 

Daniel Marino:

And the government, if I could just jump in for a second, you know, this past year, the government changed a lot of the regulations moving into V28, which changed a lot of the HCC coding, which obviously influenced the RAF. So, you know, I would assume as you're looking at the model, the model took all that into consideration.

 

Eddie Diaz:

100%, because over the last few years that there was a blended model between V24 and V28, and now we're going into the full-blown V28. And we're seeing, you've seen this, like I have across all clients across the country, is a downward pressure. Which makes it very challenging. It means you have to be increasingly efficient. And it's becoming very challenging to be… to perform well. Our… the model is allowing our clients to be able to understand where those gaps are, whether it's at a cohort level, or it's an operational level, or it's a documentation level, right?

 

Daniel Marino:

Yeah, absolutely. And, you know, one of the things that I was impressed with, that I think, you know, our client was particularly impressed with, was the model not only showed, how the coding needed to be improved and optimized within HCC, and where the gaps were. But it was done in a compliance format, because, you know, that's a concern, right? You don't want to overcode, and I think V28 is helping with that. But it also showed what the financial impact would be. And I think that's the other thing that's an important takeaway for our audience listening to this, is, you know, again. Anybody can point to a number, but it all is dependent upon what the acceptance is and how well, say, your physicians and the coders, actually activate improved coding, right? So, you end up having a low probability, moderate probability, high probability with different dollar figures in there, and the model almost moves it from a retrospective review to really a prospective analysis of working through what that coding is and plotting the path and measuring the path in order to achieve those values.

 

Eddie Diaz:

100%, Dan, and to add on to that, what the model can do is move beyond just traditional natural language processing, or NLP for short, which creates a lot of false positives, right? Which ends up becoming a manually intensive validation process. And we really want to move away from that because it allows these providers to not have to redo work. They can quickly get to where they need to be, and move on to the next patient, right?

 

Daniel Marino:

Yeah, absolutely, absolutely. The last area that I want to talk a little bit about is quality. You know, as we know, I mean, if you know, you can do as well as you want on, cost and managing the population, but if you don't hit the quality targets and the quality thresholds, then, you know, then you're going to be challenged in really maximizing your potential under these contracts. So the quality piece becomes really important. I touched a little bit of this on a couple of my comments, but how were you able to use the performance model to help the organizations think about looking at quality and sort of care management from a retrospective perspective, if you will, to more of a prospective way of thinking through working with these patients, getting out in front of these patients. How did the model help in that regard?

 

Eddie Diaz:

So, one of the things I wanted to highlight first, before we jump into that topic, is how… what really stood out to me, once I overlaid all this information is some of these targets that they were setting up, or had negotiated for, were actually unachievable, right? And then there's… they're also the inverse, where if they're outperforming a low target, they should raise that target, because they can make more of a bonus, right? So I thought that that was really interesting, and giving that insight to the client, because oftentimes, the quality data is sitting in one kind of department, and it's never being married back to what kind of contract terms that we negotiate, right? So, but in terms of the actual operational implementation of quality programs, I think what's… what is quite, what stands out, especially if we… if we look at not only just a single client, but them in relation to other clients that are performing, that are working with similar contracts, is what targets are they working on that are the most competitively advantageous based on their operational footprint? Similar to what we discussed earlier, right?

 

Daniel Marino:

Yeah, I mean, it's that contract harmonization, right? I think that's a critical point that you just brought up, because a lot of times, you either have something in the contract that isn't aligned with operations, or vice versa. They may be doing really well, and yet it may not be reflected in the contract. So that contract harmonization is important, but I think you bring up another good point, too, because the robustness of the model not only took into consideration what the current state was, but also looked at what the market activity was.

 

Eddie Diaz:

Correct. Correct. And seeing what their competitors, so to speak, were doing in terms of their target benchmarks, but also which measures they were using to measure themselves, in terms of their targets. So that combination of targets that they had set up, whether it be on specific HEDIS outcomes or a more truncated version, was, I think, eye-opening to a client. Whereas they could see that they can either improve or get rid of certain measures to be more optimal that's consistent with their operational expertise.

 

Daniel Marino:

Right, so we were able to actually get some recommendations that would go to the managed care folks in terms of how to incorporate in their next round of negotiations, which,

 

Eddie Diaz:

Correct.

 

Daniel Marino:

Some of these contracts are negotiated annually, so that's an important point. So, with the quality piece, though, let's talk a little bit about that, because, you know, again, I think, obviously, quality is critical. You know, when we think about quality, you know, some of the quality outcomes, some of these are more, let's say, operational performance-led versus clinically led. How did the model help to kind of close that gap? You know, when we think about, let's say, measurement of the chronic disease, or you mentioned some of the HEDIS measures, how did the model align that with some of the performance and, let's say, the performance dollars?

 

Eddie Diaz:

So, so yeah, absolutely, right? When we think about the HCCs, or the chronic conditions, there's a kind of a dual aspect of that. If I'm capturing appropriate coding based on the appropriate risk of my patient population. I'm also seeing them, so I'm also meeting certain HEDIS criteria, right? I'm doing those FOBT tests, I'm doing breast cancer screenings. because I'm engaging with them, right? So there's this kind of dual… Process of access constraints. And limited access creates a deleterious effect across a myriad of outcomes, not only just MLR, but also quality. But inversely, if you focus on quality, because you're getting those folks into being seen, it starts to have positive impacts from a quality perspective, but also on the MLR perspective, as well as the RAF lift. And this kind of symbiotic nature is very interesting to our business, right? And it's interesting for our clients, and it's what they have to focus on, right?

 

Daniel Marino:

Absolutely right, and I'll tell you, just to kind of bring this together, one of the things that was exciting to me was when we were working through this, was identifying access, right? And the question really became, this particular client knew that they did not have enough PCP access, so the question really was, well, what's the right level of access, and what should it be when you think about the number of visits per 1,000, and the model was able to create that, is what as well as then balance that with what the specialty access visits were. Really powerful, and I think, again, it plotted the course, which I think was something that we… I personally, having been doing this for contracting and working with organizations for over 20 years in value-based care, I've never seen this before. It was quite exciting.

 

 

Eddie Diaz:

Being able to tell them, if I point… if I… if I show you what your PCP visits per 1,000 could be, and what that downstream impact on everything else, it's really hard to bring that to life, and that's what we've achieved, which I think is extremely helpful for our clients.

 

Daniel Marino:

Absolutely. Well, Eddie, you know, we just barely scratched the surface on this. I'm excited about what this direction is. You know, again, I think this is the first time that I've actually seen and realized the powerful nature of artificial intelligence. We've been able to, again, do this in, I think, 7, 8 weeks' time compared to months and months and months prior to this, and the costs were considerably less. Certainly an incredibly realized value coming out of artificial intelligence. Eddie, if any of our listeners today are interested in learning a little bit more, or maybe connecting with you, can you share your contact information?

 

Eddie Diaz:

Yeah, absolutely, Dan. I'm on LinkedIn, and that is, you can look me up on… on… as… as Eduardo Diaz, or healthcare, or Eduardo Diaz Healthcare, and then you can also reach out to me by email, ediaz@LuminaHP.com.

 

Daniel Marino:

Well, thanks, Eddie. I really appreciate it. Nice job with all of your work. You and the team have done a great job with this. Really appreciate it. If any of our listeners are interested in learning a little bit more about this, we're going to be doing quite a bit of write-up, and we have white paper and some other things. It's quite exciting. If you are interested in learning about this, please feel free to contact Eddie or reach out to me at dmarino@luminaHP.com. You could also visit luminaHP.com. for, additional information. But want to thank everyone for tuning in. Until our next Insight, I am Daniel Marino, bringing you 30 minutes of value to your day. Take care.

About Value-Based Care Insights Podcast

Value-Based Care Insights is a podcast that explores how to optimize the performance of programs to meet the demands of an increasingly value-based care payment environment. Hosted by Daniel J. Marino, the VBCI podcast highlights recognized experts in the field and within Lumina Health Partners