A very robust discussion around the challenges, opportunities, and incentives around population health management took place on Oct. 26, during the Health IT Summit in Washington, D.C., sponsored by Healthcare Informatics, and held at the Ritz-Carlton Tysons Corner in the Washington suburb of Tysons Corner, Virginia. Not only did panelists engage in lively discussion, numerous audience members gave impassioned comments and asked many questions, leading to a discussion that all those involved in agreed was a meaningful one at this moment in the evolution of the U.S. healthcare system.
Drexel DeFord, CEO of the consulting firm Drexio Digital Health, led the panel discussion, entitled “Population Health Strategies to Improve Outcomes and Coordination of Care.” He was joined by Sule Calikoglu Gerovich, Ph.D., director of the Center for Population-Based Methodologies at the Maryland Health Services Cost and Review Commission, in Baltimore; and Robert S. Rudin, Ph.D., an information scientist, at the Washington-based RAND Corporation.
“I want to start with the term population health,” DeFord said in opening the discussion. “Sometimes, it’s good just to define population health. And I think back to being a ‘recovering CIO,’ spending a lot of time talking about big data. I like the joke about population health being like teenage sex,” he said: “everyone talks about it, and everyone thinks everyone else is doing it, but a lot fewer people are doing it than people think.”
Gerovich said she is fully aware of some of the contradictions in creating population health strategies. “My background is in public health,” she noted, “and at Hopkins”—she received her doctorate from Johns Hopkins University—“the School of Medicine was on one side of the street, and the School of Public Health was on the other side, and we used to say that that was the widest street ever,” she said. “To me, population health is the seeping of public health concepts down to the provider level. Everyone is defining population health differently, but in the end, it’s thinking about prevention and outcomes.”
What’s more, Gerovich added, “In Maryland, we have all-payer rate-setting: Medicare, Medicaid, private payers, all pay the same prices. And the commission that I work with looks at the costs that different hospitals face, and regulates the charges and ChargeMasters at the hospitals. We’ve been thinking about population health, and how we can incentivize providers when they come to their door and beyond.”
(l. to r.) Drexel DeFord, Sule Calikoglu Gerovich, Ph.D., and Robert Rudin, Ph.D.
“I agree with that,” Rudin said. “The key aspect of what makes something a population health program is that it is proactive, as opposed to reactive. It’s about not waiting to treat a patient, but devising strategies to identify [at-risk] patients ahead of time. In our study, we found three goals within this population health strategy. The first basic goal is to identify high-risk patients; the second goal is to identify the subset of patients who can be helped and treated—and there’s a big drop-off between who’s doing #1 and #2. And goal three would be to identify subsets of patients and match them to specific interventions. Very few are yet doing #3. In a lot of models, the variable most predictive of hospitalization is previous hospitalization; but the whole goal is to try to prevent the first hospitalization. So the predictive capability is still very poor; really, no one’s doing it really well right now.
“I’ve always thought of it as being like a soccer field,” DeFord said: “you’re trying to intercept the patient before they cross the mid-field line—and if you can keep them far, far away from the sick end of the soccer field, even better. Meanwhile, Bob, you mentioned analytics—can you talk more about that?”
“There’s a near-consensus here: in terms of predicting who the high-risk patients are, most everyone agrees that the big bottleneck is in the data, not the analytics.,” Rudin responded. “The gap is in the data: there are major problems in using data to predict. We found three categories for data. Claims data has some advantages: it’s longitudinal, it covers multiple doctors; but it’s very limited in terms of the advanced kinds of information needed. EHR data is richer, but it’s limited by organization. It’s also very dirty. Problem-list data is pretty well known to be wildly inaccurate and incomplete. So EHR [electronic health record] data’s got a lot of problems. Also, there’s a lot of missing data. And as Dr. Williams said a little while ago, some of the most predictive data is around social rather than clinical variables. The third type of data would be around the patient him/herself. The problem is that there are so many types of data that might be relevant. For example, use of a wood-burning stove is very predictive of respiratory illness. Now, are you going to ask all of your patients whether they use wood-burning stoves? There are probably not a lot of patients in New York City that are going to answer affirmatively. There are also problems not only with input but also data output. Change, the dreaded ‘c word.’ This requires training and other elements as well. So we have a long way to go to become truly predictive.”
“I agree,” Gerovich said, “and I see that some of the data needed is at the population level. So someone needs to also be thinking about the clustering level, rather than just about the individual. My apologies to clinicians here, but they really have a hard time moving from thinking about the individual patient to the larger population. If clinicians see one false negative or data error, they tend to lose patience, too. So we have a lot of cultural change that needs to take place to help clinicians understand how this should work.”
“Yes,” DeFord said, “sometimes, clinicians are looking for a reason to not change. A group of us had dinner last night and spent a lot of time talking about the culture of medicine, and how difficult changing that is. You see the same things in the surveys and interviews you do. Can you talk about the change challenge?”
“Yes, never underestimate how difficult the tiniest change in clinical practice is,” Rudin responded. “We heard that over and over again in our interviews—that to have a good predictive model, you need good-quality inputs, and you need to do something with the outputs. And that requires understanding what your data points are, and what to do with them, and integrating them into your habits.”
“The incentives also matter,” Gerovich said. “Since I’m in the payment world, we think a lot about incentives, and what matters. So it’s not all about clinicians; it’s also about all the incentives—making sure we have the incentives for them to do the right thing. I was a nurse, and I know other clinicians, too. I think if we do the incentives right from a payment perspective, and we think a lot about care management reform, then I think there is a door opening, and in the past couple of years, things have already changed.”
In a comment from an audience member who had just given a presentation at the Summit prior to the population health panel, Marc S. Williams, M.D., director of the Genomic Medicine Institute at Geisinger Health System in Danville, Pa., said, “The points you made about the data are worth exploring a bit more. First, of all, when you have reliable data and present that to the clinicians, we do go through a kind of Kubler-Ross process of denial and anger and acceptance—and many times, if you have good, reliable data and show that data to clinicians—a lot of times, there doesn’t have to be a lot of cultural impetus to change, you’ll change because you’re not as good as your peers,” and those kinds of revelations immediately produce behavioral change, Dr. Williams said. “And the other thing that’s interesting—getting reliable data that’s clinically meaningful, can change behaviors,” he added. “Second, we have an Einstein problem: Einstein famously said, not everything that can be counted is meaningful, and not everything that’s meaningful can be counted. So when you go to clinicians and say, you’re going to be measured on this—the clinicians may say, that’s not clinically meaningful. So at Geisinger, we’re trying to worth through that. The thing is that outcomes measures are built around the lowest common denominator.”
“That’s in some ways the meaningful use challenge, right?” DeFord responded. “Because some of the measures are seen as not even mattering.”
“I agree, the measurement has to change,” Gerovich said. “We have process measures: do you do this for a heart attack patient? And with some measures, everyone has reached 95 percent, so the measure becomes meaningless. And we have to decide what the mean population health and quality measures are, and let providers figure out the specifics.”
And the Summit’s co-chair, Dave Levin, M.D., said this: “As we discussed over dinner, I believe everything comes down to people, process, and technology. And I’m really excited about the concepts and promise of population health; but I’m really fearful that we’re screwing it up. There’s so much focus around the technology and the analytics, around population health. And in many organizations, it’s one neuron connected to t, and when the population health neuron fires, it leads to the ‘I need analytics’ neuron. But the real need is for actionable data,” he said. “Health systems are rushing out to buy these tools, but they’re not looking at their care processes or care coordination, and basically know nothing about patient engagement. So we’re going to end up repeating a lot of the mistakes of the past. And you can always tell an organization is in trouble when they say, ‘We need coordinators for our care coordinators’! So we need to step back a bit. So the question is, do you see what I see, which is people focusing too much on technology and not enough on people and process?”
“I completely agree with that,” Rudin responded. “In terms of people, processes, and technology, this isn’t just a healthcare thing; there’s a long history of experience with the idea that if you just bolt technology onto bad processes, or as they say, pave the cow path, you won’t really change anything. So you have to co-develop the processes and the technology in parallel; they have to be designed together. And that’s not just true of healthcare. I would say also, I think there’s a lot of confusion about what these terms mean. And we’ve heard a lot from providers that in terms of coordination of care, there’s a lot of smoke and mirrors out there. They would look at products out there, and see that it wasn’t what they had in mind. Or vendors were doing vaporware tests to test the market. So we really need to define what these terms mean; and if you simply identify patients at high risk, that’s not actionable.”
Additionally, Gerovich offered, “Even if you give the data to people, they need a focus. In Maryland, since I’m a state official, we think a lot about the care coordination. There are good things that we need to leverage, and we need to think about what we can do collaboratively. Health information exchange is great in Maryland; all of our hospitals are connected; and when the patient comes to the ED, you can see the patient’s history, and send alerts. So we’re trying to connect to providers, to see what they need at the point of care. And you can’t do everything all at once. And we need work in terms of long-term version and how we get there.”
“We’re singing from the same sheet of music all the time about people, process, and technology,” DeFord said. “And as a recovering CIO, I’ve implemented lots of systems—EHRs, revenue cycle, etc. And what you realize when you’ve implemented them, is that you’ve taken a train wreck, and you’ve implemented a fast, efficient train wreck. So you see the problems that were always there, and you then get to fix those problems. I’m a big believer in the Toyota Production System and Lean—and I think this idea of really looking at processes, and figuring out what pieces you can reengineer, sometimes at very low cost, is a really great first step. I don’t know that we necessarily do that very well—but then do the technology insertions once you’ve gotten that figured out. It’s a real challenge.”
Rudin added that, “One thing we’ve heard is that when provider organizations are taking on risk in bundled payments, the providers have to decide how to divvy up all those payments, and that’s not always easy. And one provider system, they weren’t able to execute contracts, because they couldn’t figure that out. So the idea would be that if you could make the process more efficient, you could make some changes. So maybe if you could get someone to a community health center, that might be more efficient; but that’s the theory.”
“We’re really in this transition from a fee-for-service-based system to value-based payment, and it’s like a cow with two hooves over the fence and two hooves still on the ground on the one side, and we’re all wrestling with that transition.”
Meanwhile, an audience member who identified herself as a physician said this: “As I’ve dealt with the challenges of adopting one EMR after another, and after 30 years, I think, we need to focus on getting a good patient history. And what I’m finding—let’s find the path together. I could create a big rap sheet on things IT has done that has made patient care difficult, I could go on and on, but this has been a struggle for us in adopting this new technology, and we’re all guilty. What I hate,” she added, “is the fragmentation of care that I think is worse than ever. I can’t pick up a chart now and get the story anymore. We’re getting all these different codes and three different ways of saying hyperlipidemia. And it’s hard to read notes from specialists and know what’s going on. So let’s focus on the getting the history of the patient.”