On Tuesday at the Health IT Summit in Nashville, sponsored by Healthcare Informatics and taking place at the Sheraton Downtown Nashville, attendees were offered a very vigorous discussion of the opportunities and challenges around the leveraging of data analytics in patient care organizations. The panel discussion, entitled “Forward-Thinking Examples of Analytics in Modern Healthcare Settings,” was moderated by Michael Hamilton, vice president, analytics, at Albany (New York) Medical Center.
Hamilton was joined on the panel by J.D. Whitlock, vice president, enterprise intelligence, at the Cincinnati-based Mercy Health; Bob Cawley, CIO at the Glens Falls, N.Y.-based Adirondack Health Institute, an independent, non-profit organization that partners with providers and patient groups to transform health in northern New York state; and William Paiva, executive director at the Stillwater, Oklahoma-based Center for Health Systems Innovation at Oklahoma State University.
“As we had agreed to do,” Hamilton said, “we wanted to discuss forward-thinking examples of analytics. A good place to start might be to discuss some of the challenges we all face in our current system of data.”
“There’s no shortage of challenges, in that regard,” Adirondack’s Cawley said. “The important thing to know is that we’re a network of providers, not a healthcare system. And that has a lot of implications for how we implement technology. We were formed in 1987, but underwent a significant transformation in 2007, when we started the Adirondack Home Health Network. It was then that we got everyone on interoperable EHRs [electronic health records], connected to a RHIO [regional health information organization], and established a regional clinical quality dashboard, and a regional all-claims database. In addition,” he said, “New York state stated a DSRIP program [Delivery System Reform Incentive Payment Program] in 2014. Adirondack joined as a participating provider system. Among the key areas of focus in the DSRIP program has been reducing avoidable high-cost utilization including ED utilization and inpatient admissions.”
What’s more, Cawley noted, “We’re also very rural, with only 700,000 people across 11,000 square miles. We have five or six hospitals; the largest is 300-350 beds, while most are small. Within the hospitals and primary care, we have pretty good adoption of primary care and adoption of the RHIO, but even within primary care, there are 15 different EHRs we’re dealing with. And nursing home, long-term care, behavioral health, if they are on an EHR, they’re not connected to the RHIO. And given that 25 percent of our providers are financially challenged,” that fact in itself is very challenging, he added. “We don’t have the resources to connect all those entities, so we’re relying on the RHIO to connect all of us. The other main source of data is the state government; since DISRIP is a Medicaid program, the primary payer for us is the state of New York, so we’re attempting to get data that way. So, trying to get all this information together is the first challenge,” he said. “If you’ve ever tried to integrate data from EHRs, you’re familiar with the amount of remediation work needed, and what’s needed to achieve the fantasy of marrying clinical and claims data and then use it.”
“I’ll answer that question a little bit differently,” Paiva said. “We’re blessed that by the time the data comes to us, it’s cleaned up as much as possible. So we don’t deal with all the informatics issues. So once we’ve got the data, what do we actually do with it? We’ve spent the last three years figuring out what to do with the data and how to provide tools meaningful to the healthcare system. A few things we tend to focus on, as it relates to getting value out of your data: whatever tool or product you develop needs to be clinically relevant and timely. You have to provide tools that allows people to do something they couldn’t do otherwise. So it needs to be clinically relevant and to help them to do something they couldn’t do before.”
Further, Paiva said, “We need to stop thinking about it as an either/or—either the physician or the tool. It can be both. We were able to reduce error rates for diagnosing breast cancer by 85 percent, when an algorithm we developed was used together with a pathologist diagnosis process. These tools are really designed to help clinicians. And the last thing I would say is that you need some sort of benefit to physicians beyond the clinical—either financial, to help manage a population, or to manage some kind of MACRA or CPC+ credit,” he said, referring to the Medicare Access and CHIP Reauthorization Act of 2015, and the Comprehensive Primary Care Plus program. “Here’s one story of a tool we developed that has been successful,” he continued. “We had surveyed rural physicians on their challenges. And one of the ones they mentioned was managing diabetic patients in rural areas, where they don’t have many specialists. So we focused on diabetic retinopathy, because only 10 percent of diabetics in rural areas get eye tests. We asked, could we develop a tool from the data collected in the primary care setting, to predict whether the patient has diabetic retinopathy? So we built an algorithm based on primary care visits, comorbidities, and other data. It solves a rural-physicians problem and addresses requirements under MACRA and MIPS [the Merit-based Incentive Payment System]. That’s an example of how to build tools that work and are useful.”
Albany Medical Center’s Hamilton said that, “As we continue to grow, we have affiliated hospitals operating with different EHRs, and that’s a challenge. And then, physician engagement and adoption across the enterprise, are critical. Otherwise, you’ll continue to see articles saying that data science is dead. And that’s because you didn’t plant the seeds, you just tried to reap the harvest, and tried to solve problems that didn’t exist.”
Mercy Health’s Whitlock said, “To Bob’s point about interoperability challenges, integrating ambulatory EHRs, is extraordinarily difficult and takes a lot of time. The good news is, we’re down to the last few practices on this. The bad news is, we’re struggling with those final integration steps. So the ambulatory EHR integration struggle is real. And then to William’s point about making sure you’re developing tools that can actually be used, I agree-don’t go spending a year doing something that’s not going to be used.”
Instead, Whitlock said, “Make sure you have your clinical business leadership ready for what you’re actually building. Meanwhile, I think that one thing we’re doing a nice job with is that we have an enterprise performance dashboard with 34 indicators, tied to clinical, operational and strategic factors or elements, and we publish it monthly. And our leadership from the top is into it, and in fact, it came from them. The downside is that my enterprise data warehouse team spends half of its time producing this, and it’s rather manual. The good news is that it’s really well-accepted throughout our organization, because it’s sponsored and supported by our regional leadership.”
Moving Towards Predictive Analytics
Inevitably, forward-thinking forms of analytics tend to be built out of a drive towards predictive analytics, the panelists agreed. “When you think about forward-thinking analytics, a couple of buzzwords or buzz terms come up—machine learning, and predictive analytics in particular,” Hamilton said. “William, what are your thoughts on that?”
“We’ve moved quite a bit into artificial intelligence and machine learning,” Paiva responded, “because I was getting frustrated with descriptive analytics projects. What was happening,” he said, “was that all the projects involving descriptive analytics ended up with the same basic conclusion—that there’s variance in healthcare—whether in drugs, readmissions, or whatever. In our case, we decided to try to develop predictive tools to put into the hands of physicians and other clinicians and administrators, in order to help them better manage the health of populations. So far,” he said, we’ve focused on two populations—patients with CHF, or congestive heart failure—and those with chronic kidney disease, or CKD. We’re trying to predict which CHF patients will decompensate, and which CKD patients will end up in unplanned dialysis. And there’s actually zero information in the literature on this; so we started looking into artificial intelligence and machine learning.”
Paiva went on to speak of the broader picture, noting that “Investments in artificial intelligence have increased from $100 million per quarter, three years ago, to about $1 billion per quarter invested now. In the last three years, the healthcare industry has been the largest recipient of those monies. Within healthcare,” he added, “there are two categories that have received the most funding—clinical risk analytics and diagnostic imaging analytics.” Even so, he said, “There’s been a fundamental lack of education in terms of healthcare people understanding analytics and analytics people understanding healthcare; that lack of understanding runs both ways. So we launched an 18-month program within our medical school, in which the students learn about topics such as design thinking and minimal viable products, and other subjects, and they learn about those topics before they begin their clinical rotations. We’ve also launched a healthcare informatics program within medical school.” Both of those innovations, he said, relate to the fact that “There remains a gap between the [IT] people developing products and innovations, and healthcare people not even knowing what they are. Look at the org charts of all our hospitals,” he said. “How many chief digital officers are sitting on the management team? How many chief analytics officers are even reporting to a senior executive? Healthcare organizations have to catch up to where we’re going.”
Responding to Paiva, Whitlock said, “In terms of predicting chronic kidney disease, of all the things to predict, that’s a pretty good one. And one interesting thing—what was most interesting about this to me was that, for the patients with all the right labs, they did a nice job of predicting the progression of kidney disease, but that was only 10 percent of the potential people affected by it.”
“That’s why you need to move into artificial intelligence and machine learning, to support better analytics,” Paiva responded. “The reality is that if you’re doing predictive modeling, it’s a challenge. And the reason we went after CHF and CKD is that that’s where the money is. So I would say that, in addition to all my other criteria, it’s always good to follow the money.”
“And we just talking about the fact that predictive analytics aren’t always the best option,” Hamilton said. Sometimes, bundled analytics are good.”
“That’s right,” Whitlock responded. “If you want to do some advanced analytics around hospital length of stay and cost, and want to take post-inpatient-stay and look at where you’re losing money or making money on your bundled payments, and line it all up, since the data is coming from different places, that can be done.” Further, he said, “A key factor turns out to be around discharges to SNFs”—skilled nursing facilities. “You’d be astounded at the variation among our facilities based on culture—some are at 10 percent in terms of discharges to SNFs, while some are 15 percent, some are at 30 percent.”
“We took a somewhat different approach to that,” Cawley said. “When you’re looking in particular at ED [emergency department] use, so much of what’s driving avoidable ED use isn’t even medical, it’s socioeconomic, it’s behavioral health. So we’re trying to put together work groups that are multidisciplinary, representative work groups, to try to figure out what’s going on. And one of the hospitals did an ED study, found their high utilizers, and found that they were doing all clinical assessments, but the utilization of the high utilizers had nothing to do with the clinical or medical factors.”
“I agree,” Hamilton responded. “A lot of what we spend a lot of our time on is getting the right data, and then getting it into the right format to do predictive analytics on it, and we don’t have unlimited budget or time, and we have to focus on the biggest return on investment, and focus on the easy wins.”
