How predictive analytics helped reduce readmissions at UnityPoint Health
The most sophisticated analytics programs in the world won't matter for much if health systems staffs don't use them, or don't use them well. Technology is important, but integrating it into clinical workflows – while also fostering an enterprise-wide appreciation of data-driven decision making – is essential to making data tools work for better population health.
Des Moines, Iowa-based UnityPoint Health has gained some experience in recent years in the right way to putting analytics to work. With a keen focus on encouraging a culture that embraces the value of data, and keeping attuned to the unique needs of various stakeholders, the health system managed achieved a 40 percent reduction in its risk-adjusted readmission index over three years at a pilot hospital.
Those wins have encouraged the health system to continue building out a more comprehensive care coordination initiative for readmissions reduction, fueled by descriptive and predictive analytics that have been integrated into care teams' daily workflows.
First things first: "We really believe it's about being practical and choosing the right use cases," said Rhiannon Harms, UnityPoint's executive director of strategic analytics. "We don't want to build models that sit on the shelf, or later realize no one is using them. We keep those people who use the tools top-of-mind, so we can build things that will actually help them drive results."
"Industry-wide, for the past three or so years, healthcare has focused so much on getting the technology right, to the point where we can gather the data, do the modeling, deploy it," added Ben Cleveland, data scientist at UnityPoint Health. "Now it feels like we're shifting away from that and getting to a point where it can be useful."
Readmissions is an obvious place to start.
"Most folks will land on it pretty quickly due to the financial incentive involved, but it's a really rich use case for a number of reasons," said Cleveland.
"It really spans the whole care continuum," he explained. "Think of a patient going to a hospital and getting treated there, having to coordinate the follow-up care planned for their discharge once they leave the hospital, How do you facilitate that hand-off to post-acute caregivers? What information do those folks need to keep them out of the hospital? It turns into a multi-faceted problem that has decision points throughout the patient's timeline and trajectory."
At UnityPoint, the goal was to take a more holistic and forward thinking view of readmission risk, he said – moving beyond just identifying the patients most in need of more targeted interventions, but instead asking, "OK, given their high risk, now what should we do? How can we use analytics to inform the intervention after they leave? How can we assess the likelihood of success for those interventions?"
Rather than just calculating the overall risk for 30-day readmission, the analytics team at UnityPoint also tried to assess risk for every day along that 30-day post-discharge continuum.
"What we found is that some patients were at much greater risk of coming back early on after their stay, and then others tended to be more at risk later on in that 30-day timeline," said Cleveland. "Maybe their problems would compound over time, or they would miss their follow-up appointments, or they wouldn't follow directions for medication.
"So we developed a risk heat map over that 30-day timeline that visually depicts a patient's risk very quickly," he said. "You don't have to be a data scientist to interpret the output you're looking at, you can look at it quickly. For every patient we have they have their own individualized heat map that our care teams are working off of."
The program allows care teams to easily see who's at high risk for readmission. But even more importantly, forearmed with the knowledge of where on that 30-day continuum they're most at risk, it helps them know when to schedule interventions in the "heat zones" most applicable to the patient.
Another innovation was a freestanding no-show appointment model, said Cleveland.
"We incorporated it into this tool, and now what we can do is, if clinicians and care teams decide to schedule follow-ups, we'll actually compute the no-show risk," he said. "So you've identified their readmission risk over time, you've planned two follow-up appointments with their PCP – but it turns out they have a high risk of not showing up for those, so you have to augment your strategy a little more to ensure interventions actually happen."
Finally on the hospital side, "we also have a freestanding length-of-stay model that predicts how long a patient will be in the hospital," said Cleveland. "This helps folks with discharge planning, to do resource allocation. We've incorporated that into the tool."
All told, for every patient in the hospital – and for the 30 days after they're discharged – UnityPoint has "four models firing and providing individualized risk assessments in a variety of different avenues," he said.
That's hugely useful, of course, but not exactly a piece of cake to put into practice. So careful attention to care teams and clinical staff – with clear instruction about how to use the tools and clear explanation of their value – has been important to the initiative's success.
"In general, predictive analytics is pretty new to healthcare over the past few years," said Harms. "We recognize that there's training and education we need to provide as analytics professionals to our clinical and business leaders on how to use predictive analytics. It's a different way of approaching the problem than looking in the rear-view mirror with descriptive analytics. We've really tried to do some proactive work on the analytics competency and training piece.
"We've really tried to focus in on the use case and working to be a partner to our clinical and business leaders, to enable them to drive better results," she added. "It is for us about developing better solutions with the end workflow and end result in mind, and doing that in partnership with those who provide care to our patients."
Harms and Cleveland will offer a more detailed description of their predictive analytics strategies at the HIMSS Pop Health Forum, October 2-3 in Chicago.