Things are moving forward on a number of fronts at Mercy Health System, a five-hospital, 80-clinic integrated health system based in Janesville, Wisconsin. In January 2015, the former Rockford (Illinois) Health System merged with Mercy to create the combined, community hospital-based system.
Significantly, both the old Mercy Health and the old Rockford Health were participants in the Medicare Shared Savings Program; the old Mercy had joined the MSSP program on Jan. 1, 2014, while the old Rockford had joined on Jan. 1, 2015. Senior leaders at the merged health system are currently preparing to bring the two ACOs together as a single ACO in 2017. That ACO will coordinate the care of 20,500 Medicare beneficiaries—11,000 from the Mercy ACO and 9,500 from the Rockford ACO.
Given the extensive involvement in the Medicare ACO program, there naturally is ample motivation among Mercy Health’s senior executives to optimize their organization’s reimbursement from Medicare. But as the health system’s executives have been learning, along with the leaders of all the other MSSP ACOs, there are nuances and complexities around MSSP benchmarking. One of those has been around hierarchical condition categories, or HCCs, which assess the health condition of the individual patient. Originally introduced by the federal Centers for Medicare & Medicaid Services (CMS) to risk-adjust Medicare Advantage payments to participating health plans, the use of HCCs in the MSSP program is proving challenging to participating ACOs.
The challenges built into the CMS-HCC model are ones that Ladd Udy, director of population health and ACO, and his colleagues at Mercy Health System, have steadily been unwrapping as they’ve been delving into care management under the MSSP program. Udy and his colleagues have partnered with the Salt Lake City-based 3M Health Information Systems, in efforts to optimize reimbursement under the system. Udy will be speaking on the topic “Using Hierarchical Condition Categories to Manage Population Health,” on June 28 along with Donna Smith, a senior 3M consultant, at the Healthcare Financial Management Association’s annual ANI Institute, to be held at the Venetian Sands Convention Center in Las Vegas.
Udy recently shared with HCI Editor-in-Chief Mark Hagland some of the learnings that have been gleaned so far from this work at Mercy Health System, and which will be the subject of his and Smith’s HFMA ANI presentation this month. Below are excerpts from that interview.
Let’s begin by discussing the core issues around hierarchical condition categories. What are the fundamental issues for ACO leaders?
Well, to begin with, our efforts are still relatively young, and it takes a year before you even have the data back from CMS. But the focus has been around the MSSP program’s use of the hierarchical condition categories, which show the prevalence of certain types of illnesses in our population. They use that model, which encompasses both quality and cost outcomes, and the payment element of the model involves the cost outcomes tied to our benchmarks. They [Medicare officials] look at three years of historical spending on a patient, and also apply a risk-adjustment factor to determine how healthy or sick your population looks based on your claims.
And so the HCC model tells them whether you’re hitting your cost benchmarks?
Well, they use the model to determine what your benchmark is in the first place. Then they give us our target, our benchmark, saying, based on the HCC and based on your claims, this is your benchmark.
So you’re using analytics to determine your accuracy of your spending on patients?
Well, to determine how accurately we’re documenting how healthy or sick each patient is.
So inevitably, becoming successful in this area takes you back to coding, then, correct?
Yes, it does. And the other tricky part of this model is that it resets each January 1. So for example, let’s say that a patient has a diabetes and has their foot amputated. And then we see them in the clinic and assess their condition and document that, and it goes into the encounter diagnosis. So the patient will be newly risk-adjusted. But on January 1 of the next year, that resets, and goes essentially to zero. So if we don’t assess a wound or amputation, it drops off the risk adjustment for that patient. So we have to make sure that we’re correctly risk-adjusting each patient.
Tell me about the mechanics of applying the solution. Are they difficult?
Well, getting it accurate and having accurate measurement, has been a significant challenge, not just for us, but for ACOs everywhere. Everybody who hasn’t participated in a Medicare Advantage arrangement until now is struggling with this. There is a small handful of providers who’ve been getting capitated payments from Medicare—and they probably have a pretty good handle on this, because they’ve worked with this. And this could be a difference between $800 PMPM [per member per month], versus $2,500 PMPM. On a fee-for-service basis, if we simply put down “diabetic,” we’ll get paid for the claim, but overall, our risk adjustment will be off. And that’s the problem we’re facing. Our data is telling us that our Medicare population is healthier than the average Medicare population by our significant amount.
So right now, you’re digging into this, and figuring out where the problems and gaps are, correct?
Well, we’re getting closer, yes; that’s what led us into this. We saw in our data that we looked healthier than the average Medicare population. Other sources of data conflict. When you think of Wisconsin and even Illinois, you think of beer, cheese, and brats—that’s part of the culture. And anecdotally, we don’t feel we’re healthier than other populations; but we need the data. And the crux of my presentation, what we’ve done is, we’ve figured, OK, there are different ways of approaching this. We thought this was hitting us in the MSSP. But we also have contracts with the state of Wisconsin, for Medicaid enrollees. Wisconsin runs Medicaid through a managed care organization. We have 14,000 Medicaid patients, and we’re capitated, and they risk-adjust it as well.
What turned out to be the core issue or set of issues? Doctors’ ability to code more precisely and optimally?
Yes. We were only putting a single diagnosis on a claim, just enough to get paid, and weren’t putting down secondary diagnoses on a regular basis, because we didn’t have a reason to do it until now. And we’ve found we weren’t alone there, either, a lot of ACOs are in the same boat.
Can you provide a couple of examples of typical situations involving sub-optimal coding on the part of physicians?
A typical one is when a physician codes for diabetes without complications, as opposed to diabetes with retinopathy or nephropathy—those are typical secondary diagnoses that are very common and yet which often physicians neglect to code for.
And what is the financial difference between the coding without complications and the coding with one or more complications?
Diabetes without complications doesn’t get a risk adjustment factor at all. With complications, it could be 0.37. So, consider that the average risk per patient is 1; so that could be as much as 1.37, when complications are added in. That could make that much of a difference. And Medicare rebalances every year, so that the average risk for the average Medicare patient is 1.0. And Diabetes with complications adds 0.37. The starting point for each patient is 0, and the average ends up being 1. Certain things like acute cancers have a 2.5 risk adjustment, for example.
How far do you think you’ve been off on average, altogether?
Well, that’s tough to determine, because it doesn’t affect our fee-for-service payments, only our benchmark.
How would you describe your “off-ness,” qualitatively speaking?
I would say overall that the bulk of our Medicare population looks to be 20 percent healthier than average.
Does that mean you could be losing 20 percent of your ACO payments?
No, because the risk adjustment is only one component of the overall spend.
But it’s significant?
It is, yes.
When did you and your colleagues begin the work to analyze your coding in this area?
We first started seeing the discrepancy about 18 months ago now. But for a long time, we didn’t really know what to do about it. So that led us to discuss what to do, and how we could impact that score and make sure it’s accurate. So for the last 8-9 months, we honed in on the strategy. And what we’ve done is that we’ve figured out how we can identify which patients, in the medical record, have a condition that hasn’t yet been assessed in the calendar year. And we’ve created a pop-up alert for this for the provider, so that when the patient is in the clinic setting, the alert will tell the provider, hey, this is something to assess. We made it active for all patients, they’re all a part of the registry. It looks at the patient’s active problem list in Epic. And if they have an active problem that hasn’t been listed as an active diagnosis within the calendar year, this alert pops up. And the same alert comes back next year.
How many patients has the alert fired for so far?
I’m looking at a report from back in April, right now. And as of that time, we had been live for six weeks, and it had already populated for about 1,000 patients within six weeks. We had it fire for all patients, not just Medicare. So we knew that risk adjustment was hitting us in other contracts, too, Medicaid, Blue Cross—so we made it go live for all patients. But if that rate held, we’re looking at 8,000 patients a year who have one of these conditions that needs to be assessed.
What percentage of your patient population is that?
It’s going to be 8,000 out of 150,000 patients—probably about 5 percent of patients who have one of these conditions on their active problem list.
So are you working with doctors on improving their coding?
Yes. First, we brought 3M in to educate providers in the mid-fall; we had a mandatory meeting, which meant that about 40 percent showed up. They educated them on what the model was and why it was important, and we subsequently went live with best practices, via a one-page sheet. The providers are the ones who have to list the encounter diagnosis; a staff member can’t do that.
What have the biggest learnings been in all this so far?
A couple of things. One is that there’s not a single way best way to approach this. I’ve had conference calls with people in ACOs at other organizations. Everyone seems to have a different approach. There’s not a single solution that works for everyone. And it does take some patients, because the true score that matters comes from CMS. We can measure on our own, but if it doesn’t matter their score, it doesn’t count. And we only get a report from CMS once a year. So it’s challenging, and there are no slam-dunk solutions at the moment. But not doing anything will have negative implications for most providers, whether they know it or not.
Given all this, what would your advice be for fellow leaders of ACOs?
To do something. Do something that’s reasonable for your organization and fits with your culture. You have to determine the urgency. If you’re already in capitated payment, this is very urgent. If you haven’t taken any downside risk yet, it’s important, because you eventually will have downside risk, so you’ll need to do it, but it may not be super-urgent yet. But it’s important, and has required a lot of physician education. We’ve had to do it over with a number of doctors. So yes, it requires several rounds of education. And our approach is young, and I’m sure we’ll tweak it more with time. And as we come closer to accepting downside risk, it will become more important.
Can CIOs, CMIOs, and other informaticists play an active role in this?
Yes, sure. We have an internal IT person who works on the Epic pop health suite of tools, Healthy Planet. That’s the person who put together and built the best practice alert, and determined when it the alert would fire. And without her, we wouldn’t have known what it was possible to do and when.