Developing Medicaid Core Set measures for States

META KREINER: Hello, welcome, everyone. My name is Meta Kreiner. I'm a population health specialist with Optum. And I'm going to let Yamini introduce herself.

YAMINI RAJU: My name is Yamini Raju, and I'm with the Commonwealth of Virginia as a data analytics manager. I started my career in Medicaid, in Missouri Medicaid, in 2000. And I've been with Medicaid ever since, multiple states, and for the past 10 years with Virginia. I'm going to give it to Kasia.

KASIA GRUSZKA: Good afternoon, everyone. My name is Kasia Gruszka. I work for the Michigan Department of Health and Human Services. It's always a mouthful to me. I am the Quality Improvement and Program Development Manager within the Managed Care Plan division, which basically means that I lead the quality improvement efforts for the 2.5 million Michiganders covered under Medicaid, no pressure whatsoever. I have been with the State of Michigan for seven years now. The first six years, I spent under the Financial Alignment Initiative, which was the CMS’s demo for the duly eligible end release. And after six years of trying to combine Medicare and Medicaid and all of the LTSS services, I moved to the Medicaid side, the Medicaid Managed Care side, thinking how complicated can that be? It's just Medicaid. It is not. So I have been leading the quality for the Managed Care Medicaid for the last year, as of two weeks ago. So I just had my anniversary, and I officially cannot be fired as a new employee.

META KREINER: So this afternoon, what we want to talk about is our journey with the CMS Core Sets. And we'll start with the background, just to level set with everyone on what the CMS Core Sets are. Then we're going to focus a little bit on the timelines for Virginia and Michigan. And then we'll really dig into actually calculating the measures, what that has looked like for these two state experiences, and then how calculating those measures and getting those results can drive quality improvement.

So for those who may be less familiar, the CMS Adult and Child Core Sets come from sections 1139A and 1139B of the Social Security Act, which require a core set of health care quality measures for Medicaid-enrolled adults -- that's the adult core set -- and children enrolled in Medicaid and the Children's Health Insurance Program, or CHIP -- that's the child core set. The adult and child core sets support the CMS move towards a national system of quality measurement, reporting, and improvement. And the first voluntary core sets were released in 2011 for children and 2012 for adults. So they've been around for a while. They are updated annually, so the measures themselves, the list changes every year, and the specifications change every year. So you have to be responsive to that. And the big milestone we just passed is mandatory reporting. The first year that was required was 2024. So that was the activity of last fall.

This just gives a little bit more information. When reporting became mandatory in 2024, that included all of the child core set measures and the behavioral health core set measures in the adult core set. Both core sets include a number of health domains, including not just behavioral health, but also chronic conditions, dental measures, maternal health, preventive services, experience of care, and long-term services and supports.

So how does this work exactly? States submit state-level data annually. CMS then publishes that data when at least 25 states have submitted data that met quality standards. And another milestone last year was that there was the development of a CMS core set dashboard. That includes data up to 2022. So that's still when the data was voluntarily reported by states. And the dashboard includes state-specific performance and national medians, with a stated goal being to develop actionable analytics that support efforts to advance health care quality and improve health outcomes for Medicaid and CHIP beneficiaries. So if you haven't seen it yet, there's a link there on the slides. You can go to the CMS core set data dashboard. For those of us who like quality measures, it's pretty cool. You can look up states. You can look up measures. And it basically provides the numerator, denominator, and rate, and then a median that you can compare to.

[00:05:00]

So what we want to talk about today is this looks pretty simple, it looks pretty straightforward, we just want to unpack it a little bit. There's actually a tremendous amount of work over many, many years to get to the point of just being able to put together what might seem like a straightforward dashboard like this for each state. So we've got representatives from Michigan and Virginia here to talk about their timelines. This is just a super high-level representation of that effort, which is just intended to show -- goes back to both states started reporting in 2020-12, the child core set reporting started first. Then you can see the dotted lines is the adult core set reporting. And those numbers are kind of slightly going up, slightly going up, as we get to 2023, which is the last year of voluntary reporting before mandatory reporting started. And here, just to focus in on the Virginia timeline for a little bit, I'll pass it over to Yamini.

YAMINI RAJU: So as Meta said, in 2011 and '12, it was voluntary. So it was not mandatory. So at that time, the Commonwealth of Virginia, we got the data directly from MCOs. So we call it the Interactive Data Submission System, IDSS. So ours is a huge managed care state like Michigan. We have about close to 2 million members. And 90% of our members are managed by managed care. We have five managed care organizations. So they used to submit the data, and the state used to upload it to CMS. But then when we came to know that it's going to be mandatory in 2024, we started the process in 2019, if you can see, to transition to custom development, so in-house development. So I work for the health care analytics. So we have built all these measures using the data from the data warehouse. And the first time custom built measures were reported was in 2023, for the period 2022, it was like a mock run. But the actual mandatory reporting was for 2024, for the data of 2023. But if you can see, there's a dip in the measures, because we decided to focus only on the required measures, just while we are custom coding. So that's our timeline.

KASIA GRUSZKA: So I'm going to focus on something a little bit different. We started at the same time. Michigan chose to voluntarily report both adult and child core sets from the very, very beginning. We started developing our child core set. That was the first one introduced. So you can see this is the first one. And that was kind of state-driven, how we were approaching that. And then CMS introduced or offered to the states the CMS adult core set grant, in which we participated. And so once we started with the child, then we

participated in the grant, which was very specifically leading us how to approach that. And the interesting part to that, at least for me when I started with the program a year ago, was that we are approaching the adult core set very differently than the child core set. And this is -- I finally understand why -- we had very different processes of how we were approaching the reporting for both of them. And the point I'm making here is that there are different ways to slice the same pie, even within the same state, right? We know that all of the states -- and we will talk about this a little later, too, in the further slides -- but Medicaid in every state is so different. So the approach to how you develop the processes to report those can be very different. But they can be very different even within the state, depending which core set you are looking at.

The other point I wanted to make here is that we really leveraged all of the support that was available to develop and integrate into the reporting of the core set measures. So we used the adult core set grant through CMS. We also participated in the contraception measure grant that was offered by CMS. But we were also looking at different affinity groups. We participated in the HIV viral load measure, and the maternity and infant health. We also did a dental measure affinity group. So as we were developing that, we were trying to do as much work as we could on our own, but also leveraging what other states were doing, what other organizations that had a stake in those measures, in the stakeholders, what they were doing. And I am actually happy to say that now, for a number of measures, we participate in those affinity groups. And we are the state that actually shares their experience, right? So we are trying to support other states that might not be -- I don't want to say as far as we are -- but just still need a little extra to put those processes in place, so they don't have to reinvent the wheel, but they can get tricks or the best practices from us.

[00:10:12]

That said, from the very beginning, we were committed to reporting all of the measures, so as they were added, we were adding them. So we are right now reporting all of the adult and all of the child. You can see it…so it's a little different than how Virginia is approaching that. Even though the child is fully mandatory, what is optional to report, we have been reporting that. And the way I think about that is, that means that maybe that did not put us much ahead, because as Meta mentioned, the CMS continues to add measures. Every year there are new measures. So it continues to be a process for us to add to those processes. But at least we are not falling really behind, right? We are keeping up with what's happening. And my big suggestion or encouragement to anyone, especially from this perspective, is to look for all of those groups that are doing that. And there is no need to re-invent that. A lot of the states are already doing this. There are best practices. And you obviously cannot use fully the same approach, because every system is different, but pull the little pieces and learn from our mistakes and our growth.

META KREINER: Thank you. So now we're going to dig into actually calculating the measures -- what goes into those lines going up and up and up. And we wanted to talk about defining populations, and then stratifications, establishing secure data linkage, developing the analytics themselves, and then validating the results. So first, we start with defining the population. And this seems like something that's really straightforward,

right? The specifications just say, for all measures, denominators must include all Medicaid and CHIP beneficiaries who satisfy all specified criteria. But for those of us who work in Medicaid, we know that Medicaid isn't really one program, right? It's many programs all together. And they all have slightly different benefits. And that nice, neat circle really looks something a little bit more like this. So if you think of that big circle sort of as your full-benefit Medicaid, a lot of states have Medicaid managed care covering most of that circle. But then there's some fee-for-service outside of that that maybe has a slightly different benefit or covers some portion of their services. You've got CHIP. CHIP can be managed care, maybe not, children, adults. Medicare/Medicaid duals, that's a very important population. But there's the Medicare component, and every state will also have state-specific programs as well with different benefits. So what does it mean to include these populations? And I'll pass it on to Yamini and Kasia to just talk a little bit about their decisions.

YAMINI RAJU: So for Virginia, we have included the duals, duals meaning a person who has both Medicare and Medicaid. So if you see the CMS measures, they don't dictate what we should include or exclude. But whatever we do, we should be uniform for all the measures. So we at Virginia, we have included all the dual population in all our measures. But then if you see the Plan first, that's a limited benefit member. So we have excluded that for other measures. But only for contraceptive care, postpartum measures, we have included the Plan First members. I'm going to pass it on to Kasia.

KASIA GRUSZKA: So I am really, really jealous that Virginia was able to include the duals because we are way, way behind on this one. And I speak as someone who worked for the duals division for six years. And on my to-do list when I left, that was at the very top: You guys need to figure out how to get Medicare claims, because it's coming. Don't know where they are. But the idea in Michigan is that, and potentially for a lot of states, which I think CMS is recognizing with the duals population, a lot of states actually do not have access to the Medicare claims. We have Medicare claims in our data warehouse, but we do not validate them. So this is not really any reliable source of calculating the rates for quality measures. CMS is recognizing that. Right now, the dual eligibles are excluded. You can report them, but you don't have to ask for an exception. It's a given exception for the duals.

[00:15:00]

We know that this is going to be a big focus because the duals population, which is the Medicare over 65, is really increasing. There is going to be more and more. And then especially with the focus on the high DSNPs that CMS has been introducing in the last few years. So this is like something to really start asking questions. How are we going to approach this? Are we going to work with CMS and have the agreement for the data exchange and get those Medicare claims from them, knowing that there is going to be a delay, right? Because they will have to go through their process of validation, and we are going to get them much later. So that's when you start getting delays in your rates because you don't have the data. We already have six months delay with Medicaid data before we get it from the health cares, and we validate all of that. So then your Medicare data becomes even older. Or do we collect all of this Medicare data, and we find

processes to internally validate them, just how we do this for Medicaid claims. Which begs the question of capacity, having the staffing, having the expertise in Medicare, so we know what we are actually doing and if all of the information is correct. So out of all of that, this is like one circle. Like how many questions do we have around this one circle of getting the data? That's why I'm very jealous. And when we were prepping for this, I was like, how did you do this? Like, you just did it. And I've been trying to get this done for six years, and I left. I gave up on this one. It is on their to-do list.

But the other thing I wanted to talk about is what Meta kind of introduced, that we think of Medicaid -- when I switch on like Medicaid, how complicated is it, right? It's Medicaid. And it's very complicated. And not only from the perspective of that every state has it set up very differently, but even the Medicaid benefit in itself, you have all of those partial coverage programs under Medicaid. So Yamini already mentioned the Plan First. And Medicaid CMS requires us to, if someone is receiving the benefit, we have to include them in the measure. So now we are kind of looking through all of the benefits that we have, like what do they receive? Do they meet the criteria for the measure? Because also the measure has the enrollment criteria. How long someone has been involved? Do they receive this benefit that falls into that measure? So we have some groups that, out of all of the core set measures, they only have one. But we have to figure out how to incorporate them into our rate -- so a lot of different questions. Like knowing what enrollment benefits you have, what they are covered, what measures do they fall in. We actually had to ask -- even though we have been doing this for so long -- but when CMS started providing more guidance, we were like, well, we don't include this. We didn't know that was included. This is just a partial coverage. Why is this included? And we were told that if there is any measure that they receive the service that that measure measures, that needs to be included. So we have been working through all those little circles to get more accurate data. My example is, we have the Plan First, but the other example is Mom's Program, which is under CHIP. And it covers all of the maternity benefits, right? So we have to include all of the maternity measures, which is a big chunk of the core set. Just for this separate population, we needed to build it into the current reporting that we had.

META KREINER: So once you've worked out the population and who you're going to include, then the next step is making sure you have all the data. And what we're trying to capture in this slide is there's actually a wide range of data sources that are necessary to report the core set. The big blue circle in the middle is typically maybe seen as one of the easier circles for state government, for Medicaid claims and encounters. But even those may be coming in through multiple systems. You've got pharmacy, behavioral health might be a slightly different system. Dental might be a different system. You need your eligibility data, because you need to know how many months someone's been enrolled. That may be in a different system. Same with demographics. And then what we're also trying to capture here is that some of the measures require linkage with public health data, vital records, immunization registries. And that requires both the technological infrastructure to securely link that data, as well as the compliance and the agreements to share that data. Other sources can be your managed care organizations and their reports, survey data. And then really the gold standard often for quality measures is considered the medical record, the EHRs. That can be challenging to access.

[00:20:02]

Health Information Exchange can be a source for that. Health Plan Partners can be a source for that. So lots of different pieces to get all of the necessary data, maybe just for one or two measures. And I'll let Yamini talk a little bit about this.

YAMINI RAJU: I want to focus on the center circle, the claims. It looks like it's easy, right? Medicaid is claims. But even then, as Meta was saying, it comes from different sources. For us, the people service claims comes from a vendor. The encounters, they come from MCOs and it's processed in-house by us. And then the pharmacy is from another vendor. So we have to get that data. And then dental is from another vendor. So just the claims alone, it comes from different data sources. Storing them in the data warehouse, making it the same format, making sure all the fields are correct, even before we build the measures. And then, as Meta was talking about, even building two measures. So we want to build the two measures. One is for immunization for adults and immunization for children. So we had to sign an agreement. Though all state, but different agencies. So we have to sign our interagency agreement with our Virginia Department of Health to get the immunization records. And then, I mean, you know how it is. The language. So we just can't say, I need the data. I need the data for what purpose. And this is how I need the data. So to build those two measures, even to get the data, it took so long. So though it looks like, OK, it's simple immunization. But what we are trying to prove here is that to get the data from outside what you have into the warehouse before you build it. And make sure the data is clean, acceptable. We are not using surveys or other things, but we are using public data for the immunization. And of course, member and demographics, we do have that. So we use that. So I'm going to give it to Kasia.

KASIA GRUSZKA: So I have two examples for this one. In this one, actually, claims in Michigan are a little simpler. But we don't have Medicare. And I will be upset forever about this. But our claims were a little bit more straightforward. So the way I've been looking at this is that claims are like your basic information, like your base. And then, to get full and complete and accurate information, we started working on data sharing and developing different use agreements with different entities. And oftentimes, with entities, as Yamini was saying, within our own kind of agency, but we did not really have access to the data. So the two examples that I have is the bottom yellow circle, the public health registries. We worked with them for probably a year, with the public health side of it and with our lovely data partners, who are translating all of those pieces for us, for about a year to get the data for HVL-AD, which is the HIV viral load suppression measure, which is thinking about just one measure out of, I don't know, 45 or 50 for the core set. It took us a year to be able to get data for just one of those measures. And those are folks that work with us. We are the same agency. We're MDHHS. But we did not have access to that data. And HIV is especially sensitive for privacy and security. So we have to go through all kinds of other hoops with our compliance department, who's also part of our organization. And you would think they would be supporting us in this endeavor. But now we know that we are all compliant and secure. But we had to update existing DOAs with the plans so they can access the data. Because as we will talk a little later, we are not

just trying to get this data for ourselves so we can ship it to CMS and be checked off, done and done. We are really using this data to drive quality. The other one that was interesting to me -- and I was quite surprised, and I also count as the big success of my first year, and that's why I should not get fired -- is the Health Information Exchange, which the plans have been using to translate the medical records data for a really, really long time. We all know about that. But us as a state, who is actually the big partner of MiHIN, which is the Health Information Exchange in Michigan, we didn't have access to the data as a state. And we worked for a year to be able to get the agreement, to get the all plan file -- I'm not sure, I'm making this up, OK? I'm quality, I'm not the data -- to get the file to ingest it into our data warehouse, to know where it is, to know what security levels it has to have. And we just completed getting this data.

[00:25:00]

Initially, it was driven by using it for, again, just one measure. We were unable to get the diabetes care for people with serious mental illness. Initially, we did this crazy workaround -- we pulled our enrollment data, so we knew who was qualifying for it, but we didn't have access to the medical records and data for the A1C levels. So we sent those lists securely to the plans, and we asked them to give it to us. This was our way of doing this. And in the meantime, we were trying to figure out a little bit more streamlined way, and more accurate, I suppose, and worked with our health information exchange to have the agreement so we can actually use their data. So that's another piece. If we don't have it, we find workarounds, creative workarounds. We all work for Medicaid. We all are good at creative workarounds. But then trying to find a more sustainable and long-term solution to that. And when we were discussing this slide and thinking about that, it's like, how much data do we actually have as a state, but how much data do we as Medicaid have access to? This is like two very separate questions. So as you look at all of the measures, it's a good way to first figure out the population, and then look at, those are the resources or the sources of data, do I have access to it? If I don't, what do I need to do? And as silly as it is, you go one measure by one measure. But knowing that CMS is expanding -- because we also have to justify it, we have to say, “This is what we need it for, this is the measure.” But now that we have access to this data, we can see what else is there, and we can leverage it for -- I think we have another four or five measures planned where we can supplement. That's the word -- supplement the data that we get from the encounters, which is from the Medicaid health plans. But we can supplement through the medical records to get a more accurate picture of that.

META KREINER: OK, so we've got the population. We've got the data. And now it's time to calculate the measures themselves. And Michigan and Virginia have both taken slightly different approaches. In Michigan, there is a license for a large measures engine software that includes hundreds of measures that can all be run at the same time, and that includes core set measures. One of the benefits there is that the software is updated annually, and so that keeps the measures up to date with the specifications. Additional measures that need to be custom coded, that's also calculated in that process. And there's now an automated process that kind of weaves both of those processes together, runs quarterly, and then those results can both be made available at the same time in the same format. So basically, the measures engine is run quarterly, and the custom queries are run

at the same time. There's a data validation built into that automated process. Once they're reviewed and approved, the data is loaded. And it's stored in two different ways. One is the summary results. So that's just your numerator, denominator, and rate, which is great for ease of use. But also at the detailed beneficiary level, and that allows for stratification and support of more like care coordination applications and things like that where you'd want that data at the member level. And Yamini, I'll let you talk about how it's done in Virginia.

YAMINI RAJU: I mean, I'm passionate about this slide, but because of lack of time, I'll just keep it to two sentences or so. So this, it looks really complicated, but these are the design documents that we developed before we even start coding the measure. So that's what Meta was saying, that we did it a little different. So we optimize our data warehouse vendor, and they've not only supported us with our data, but people like Meta, who have worked in this industry for a long time, they help us to understand what the measure says. Reading the CMS measure line by line, this is what it is. And then we work with the developer, and then the developer and the business people, we all talk before we even begin to code. And convert…it's like English to English. So this design document talks about the requirement on the left side. If you see on the right side, there are pieces of code, but it talks about what column we are going to use, what table we are going to use. And this template, though it took so much time to develop that and this approval process for this, but in the long run, it served as the purpose. Like if a person went and a new person joined the team. It was so easy. OK, this is the design document. So they didn't have to start from the scratch. They don't have to debug somebody's code.

[00:30:00]

So the code was there, and some of the pieces of code could be reused. How to look up a value set. Where is the reference table? Which value set to use? So this was useful. So we have this template, and our email address is at the back, the last slide. So if you need this, we can share it with you. You can go to the next slide.

So this is, as Meta said, like for Michigan, we have the same thing. We developed something called as a Datamart. So that has proved us well. We developed this for the 1115 waiver for the arts program, and we replicated the same for the adult and child core measure sets, too. So we have a table for the summary level, which just says the numerator and the denominator. But then we have detailed tables, and then we can pull like at the member level. It has everything. More than what we report to CMS, it has additional information, which have served us well for many reports. Like if somebody comes and asks me, “What's the vaccination rate in this particular region in Virginia?” So I can go to this table, and then filter by the FIPS code. OK, this region, this is how the vaccination rate is. So this has served us well to compare one region in Virginia or one locality with other localities. So this has proved us well. Thank you.

META KREINER: And then we wanted to just talk about, OK, so now you've calculated the measure. You have a result. Is it accurate? And what are the steps that you go through once you have results to be sure about that? And there are many different options. Some of them that have been done are listed here, including peer code review,

internal benchmarks from previous measure analyses. There's often other sections within departments that are also calculating quality measures. You can compare results to data validation meetings with internal and external partners. There's often many quality experts within managed care organizations and other partners who you can also draw on to look at your results. Comparison to national benchmarks, that's where things like the CMS dashboard come in. But also there are academic studies and other national research being done that you can look at results for similar measures, similar populations. And of course, you also have the resource of your external quality review organization. So if there's anyone else in the audience who has ever calculated a result for a Medicaid program, you're very excited and you share it. And almost always, the first question you get back is, well, what about for this population? What about for this region? Can you stratify this for me by age? And so one of the key elements, I think, of this is that ability to stratify the results when you have it at the beneficiary level. And what you have here are the lists of many of the stratifications that have been developed by Michigan and Virginia. Some of these are going to be mandatory for CMS reporting in the upcoming years. But these lists go far beyond that, as I think both states have found that there's many additional stratifications that are useful to help them drive quality and see where there are gaps in care, not only in terms of demographics, programmatic stratifications. And then also you can see some social determinants of health and health status stratifications as well.

And now we'll talk about what you can kind of start to do with this data. Now that you have these measure results, you're running them regularly and you're able to stratify them. And I'll let Yamini talk about this slide.

YAMINI RAJU: So here we have given one measure: How do the members follow up after a mental illness? They've gone to the emergency department. How do they follow up? Right within 30 days. So here we have stratified based on the region. Virginia has divided into six regions. So we have stratified according to the region. But if you see the graph here, the Tidewater region is below the average for Virginia. So this is about comparing region to region within Virginia. So this gives us a better gauge to focus on which region needs improvement. So this opened our eyes to see not only to work with the MCOs in that region, but also for us to develop our own network adequacy dashboards. Do we have enough providers in this region? Why are people not going back for follow up checkup? Why are they going for emergency and not following up? Because there's no provider there. Maybe there's a provider there and he or she did not get an appointment for the next 30 days. So this has helped us so much. Developing these measures is not just for CMS purpose, but to improve the quality for Virginians. It has helped us a lot.

[00:35:00]

And the next slide, if you can see, is the same measure, but this is stratified based on the eligibility category. So this is a higher level granular. We have more stricter level granularity. But here, if you see, the first one is the limited benefits. So they don't have the follow up checkup. They don't have the benefit. They have the benefit to go to emergency, but they don't have the benefit to go back and checkup. So that level is less.

But others, you can see they're consistently constant, like equal to the average. So this helps us to see based on the category of members.

META KREINER: And here are some examples from Michigan. The first one there is a measure of diabetes short-term complications by a managed care organization over multiple years. The next one there is the developmental screening measured by region. You can see geographic variations, like Yamini was talking about in Virginia. We've got childhood-led screening by geographic designation. So you can see big differences there between rural and urban. And the last example is Well Child Visits. The measure itself stratifies by age, but when you stratify that as well by gender, you can see that it's not really a factor under 18, but once you get over 18, you start to see some disparities there. And I will pass it on to Kasia to talk about this slide.

KASIA GRUSZKA: My favorite slide. A little busy, but yes. I think this is absolutely my favorite part. I am the quality person. So I have data. I need it, and I need to ask my questions, and I want the data to answer those questions, or at least lead me in the direction of an answer. So I love the previous slide for stratifying by age, but then by gender. This is like this amazing thing about having measures set up, and then you have a rate, and congratulations, that's exciting. But this is how you drive the continuous quality improvement. You stratify it by all kinds of different things to see where you have a gap. If you can close the gap, you improve your rate. But you can also look at it from a perspective of evaluating potential impact of either policies or additional benefits, or even like a medication that is added to a formulary on outcomes, and what impact do they have on the members. So here, you can see that we are looking at doula services that we in Michigan added a year, year and a half ago. It's a pretty new service that the Medicaid is covering. And we are starting, we are building a baseline for trying to see if there is a potential impact of doula utilization. So we look at the members who have encounters of doula, and then we look at the outcomes on different measures. This one is for the low birth weight. So the fact that we have all of those measures built into the system, because it was required for us to have for the core measure reporting, we are able to leverage that to look into how some of the benefits and policies are potentially, because we know there is not like a direct correlation. It just leads us into an area to investigate. We are actually, Michigan just introduced in lieu of services for food support. And that's kind of the same idea once we get it off the ground. We're still working on it. It takes some time. That we are going to create a very similar dashboard to see if the food services, the ILOS, is impacting the folks that qualify for it through whatever different measures that we have. It's kind of the same idea. But because we have all of those measures already there, we can kind of tie both elements and make informed decisions about whether the money that we are spending at Medicaid on those services is actually impactful, and get additional funding, push it through the legislation. And we always have to prove that what we do is actually doing what we say it's doing.

META KREINER: Keep the microphone there.

KASIA GRUSZKA: That's right. I'm next. So now that we talked about the different certifications and what the data does, this is what do we do with the data. And in Michigan, we do two things. First of all, we don't want to just have the rate, right? Even have these fancy stratifications for ourselves so we know what's happening. That's one side. But we really want to empower the health plans and feed them all of this information so they can start. Because at the end of the day, that's what managed care is supposed to do. The managed care plans are supposed to be managing the care and improving the outcomes. And us as a state are monitoring that and incentivizing.

[00:40:00]

So the first level, or the first step, is to provide all of this information to the health plans so they do receive a quarterly report with their performance towards the set benchmarks that we have. We have minimum standards that the plans are expected to meet on a number of measures. And we have a whole section that is dedicated to the CMS's core set measures, which again, like Meta was saying earlier, include a variety of different categories. We have the behavioral health, even though that's a little tricky in Michigan, because we have a bifurcated delivery system. Behavioral health is separate from the physical health there. But we are still working very hard to integrate it despite the challenges that the law imposes on us. But we also have dental care. Michigan Medicaid covers dental benefits, so all of our members receive dental. So we are really monitoring a lot of that, chronic conditions, and maternal health.

So the plans receive statewide rates for all of the different kind of Medicaid -- different division of Medicaid, whether it's the fee-for-service, whether it's the entire Medicaid. Again, as you guys saw, Medicaid is much more than just the managed care. They receive rates, the denominator, numerator, and the rate for all of the health plans. We have nine health plans in Michigan. But also, they are able to export a beneficially level data, which at that point, that's when they can reach the members at the member level. We are no longer just looking at the rate and being like, OK, this is the rate. But they can see who is receiving the services, who is not receiving services. They can identify all of those gaps. They know exactly how we are measuring the rates, so they know how it's going to look like. And I know we say that it allows. We hope that the plans are using this to improve the quality and reach their annual targets. And they receive this data quarterly, which I think is also very important, because that's an annual rate. This is much harder to kind of impact. But when you receive this data quarterly, you really can see which way you're trending. You can catch it much faster. And because this is all based on encounters, plus all the additional data that we are trying to get and incorporate it into our data set, this is about six months old, right? So this is as real-time data as we can get.

The other part -- so we talked about giving all of this data to the health plans, right? With the hope and trust that they are using it for what they are supposed to be using it. But on top of that, knowing that we have all of those measures, that we can stratify them in all kinds of different ways, and we can identify where there are gaps. We have two quality improvement incentive programs based on the data that we created for the core set reporting. So the one on the left is our annual quality withhold program. So that's when we withhold a percentage of the capitation payment that the plans receive based on the performance. And that includes a number of core set measures. We have, I think, like six different sections there that focus on different areas. Core set measures are one of them. And the big work that we are trying to do here, and that's what's pictured on the screen, is

that we stratify it by race and ethnicity. And the plans are incentivized to not just reach a certain performance benchmark, but also improve existing disparities. And the second piece, where we incentivize the performance, that is, again, tied to money -- I mean, that's the language that we speak -- is the other assignment algorithm. And that's what determines number of members that are all assigned to each plan. Which, again, this is their capitation payment, right? So they all want more members. And this is all based on the plan's performance. And because, again, we have all of this data, we share it quarterly. The measures that we use change quarterly, which allows us to incentivize more measures, right? So they have to work on more measures because we switch it every quarter. And we also stratify it by region. So they get the assignments based on the performance in the region. So if you have a plan that -- we have 10 regions in Michigan and nine plans, so it's a complicated kind of calculation, but I have a brilliant person who does that. But if a plan is performing well in one region, they are going to get more members. But if they are not performing so well in another region, they are going to get less members.

[00:45:00]

Because we want to make sure that where we send the members, the performance, the access to care, the care outcomes are higher. So that's how we incentivize the plans that perform better. And it allows us to have it weighted by region. So they are not based on the size. They cannot kind of swing it. They cannot swing it based on this region always performs better. So we are able to kind of shift it to another region and pretend like that's what we do there. So it allows us to really dig deep and send members where we are a little bit more confident that they are receiving the care and for the plans to start improving the care because they are competing for members.

META KREINER: And do you want to pass to Yamini?

YAMINI RAJU: And we are running out of time. And we do need time for questions. I'm going to keep mine super quiet. OK. For us, there are four bullets. But this is basically two. So based on the input, we have developed public dashboards on our public website for the MCOs. So we compare each of the five MCOs, how they are performing when compared with the national median. Like Meta said, on the CMS side, we went with the national median. So we see that uses us to encourage our MCOs to do better and talk to them. If somebody is constantly not performing or under in some measures, we have divided into five domains. And we focus most on the behavioral health and other things. And then the next thing is that, as I said, based on the data, we have used it for ad hoc reporting. Our governor started an initiative for one of the regions, which was not performing. The members, especially the postpartum care, it was so below the average of the rest of the Virginia. Because the members have to drive 30 minutes to go to a postpartum care. So those variables, we were able to find through these measures, through the Datamart. And then this story has a good ending. So we measured it after a year. So the whole state was focused on that particular region. So more providers were sent there, more gynecologists. And after the next year, when we did that, there was an increase in all the rates for that particular region, not only for postpartum, but for dental,

for well-child visits, and everything. So the measures that we developed have been used in other areas, too.

META KREINER: Thank you. And so just to conclude, we hope that we've communicated that really the scope of activities required to develop the CMS Medicaid core set measures were more extensive than we initially anticipated. The complexity of the populations, the data linkage required, all of the calculations, there were many, many steps. And the approaches taken by Michigan and Virginia reflect state-specific priorities, program features, and their health IT environment. So you can see that they really took slightly different routes to get to the same end place that I think really produced enhanced analytical capacity and a range of tools to support improving population health for both state Medicaid programs. And of course, the work is never done. Next steps include development of new measures. CMS has already released additional measures for next year. New data linkage, dashboards to provide greater capacity to trend and to stratify results. And I think we did leave -- yes, we successfully left some time for questions, if there are any in the room.

KASIA GRUSZKA: We did so well. We explained it so well, there are absolutely no questions.

JAMES: To what extent do those states’ measures include the clinical data, like lab data and things like that, and how do you get the data?

KASIA GRUSZKA: So we just finished our agreement with the MIHiN, which is the Health Information Exchange. Here I'm looking at my experts to give me the words. So we now have access to lab data. That was the one that we were looking for diabetes, that we were asking the plans to give it to us kind of backwards. And we got direct access to the data that they were using. So we just started. We were initially only using the encounters. Now we have data, and we're also investigating what else is there, what else we can pull from this data. And we know that it's growing, because all of the SDOH, HRSN, the Health Related Social Needs screenings now are reported as lab data based on long codes. So there is going to be having access to that, and knowing that the providers and the health plans have to start using it for other measures, other quality measures that we will be able to leverage it as it grows and as it develops.

[00:50:09]

So we started. We are hoping that -- we are actually right now, we got it, we are sorting through it in our data warehouse, and we are going to start comparing what did we have, and how having this data improves the rates. What is the difference?

META KREINER: And just to add to that, I think another source for the clinical or lab data is public health data sets. They often collect that data for specific conditions for the entire state. And it's different data agreements, data sharing agreements that are required to access that data by Medicaid for the purposes of reporting. But that can be another option for state Medicaid programs.

KASIA GRUSZKA: Which we do one measure at a time. I think we are getting bonus points for the vaccinations and immunizations, because we just have one, and that keeps growing. We are adding the adolescents. The adult immunizations measure is a new measure. It used to be collected through CAHPS, and now it's going to be actually a HEDIS measure. And it's going to be moved away. The CAHPS is going to move away from just self-reported. So I think we like scored on this one. We developed this, and we are now benefiting, because this is growing. So that's kind of the hope, right? That once we get this agreement on something, that we'll be able to use our wonderful data partners to mine the data and see what else we can get to, again, improve the rate. Not improve the rate, to improve the accuracy. Accuracy and completeness of the rates, because that's what we want. That's the only way to improve is when we actually know what's happening.

YAMINI RAJU: We don't use the clinical data in Virginia. We've not even thought about it, but we do have the data. So it just explains how states are different. We do have the clinical data. We do have the long data. But we have not thought about using it. We always say, “Oh, it’s sensitive data, put it aside, put it aside.” So we have to use it. But we have not used it, James, in Virginia.

META KREINER: Any other questions? Well, thank you very much for coming today. And we appreciate talking about this topic.