While the 12-hospital UCHealth system is Colorado’s largest health system, the nearly 136K patients that they admit/observe annually can often come at unpredictable times—making it difficult to match demand to limited resources. Steve Hess, UCHealth’s CIO, has been leading an initiative to improve bed capacity management at the health system, whose flagship facility is the University of Colorado Hospital, located in the Denver suburb of Aurora. The UCHealth system encompasses 12 hospitals (five in northern Colorado, three in the Denver metro area, and four in Southern Colorado), 1,987 inpatient hospital beds (699 of them at UC Hospital), more than 24,000 employees, more than 5,000 affiliated or employed providers, and volumes of 136,000 inpatient admissions and observation visits per year, 85,000 surgeries per year, and more than 3.9 million outpatient, urgent care, and emergency room visits.
In order to help them master the bed capacity management issue, Hess and his colleagues have been partnering with leaders at the San Francisco-based LeanTaaS, and making use of that company’s iQueue for Inpatient Beds solution.
Since upgrading their bed capacity management system using iQueue for Inpatient Beds, Hess and his colleagues at UCHealth have achieved the following:
> A 37-percent reduction in time to complete ICU transfers
> An 8-percent decrease in opportunity days (the difference between how the Centers for Medicare and Medicaid Services (CMS) defines expected lengths of stay for specific diagnoses, versus actual lengths of stay in the UC Health system)
> A 4-percent decrease in time to admit
> A 90-percent improvement in confidence in critical capacity decisions
Recently, Healthcare Innovation Editor-in-Chief Mark Hagland spoke with Hess about the work that he and his colleagues have been engaged in around bed capacity management at UCHealth. Below are excerpts from that interview.
How did this initiative originate at UCHealth?
We have a fairly mature IT capability, but have also leveraged IT to do some things, including per our virtual health center, but also in the operational space. Our journey with LeanTaaS goes back several years. We ended up having this huge bolus of activity in terms of infusion. So about five, six years ago, we worked with LeanTaaS to optimize the infusion schedule. We were taking Epic EHR [electronic health record] data to LeanTaaS through the cloud, and they worked with us to flatten out that volume of activity. Within 90 days, we were flattening out that volume of wait time. That was an easy home run. So then we said, operationally, where else could we use that capability? So we turned our attention to the OR. We worked with LeanTaaS to optimize scheduling times in the OR, so we could optimize the schedule across the entire footprint, because that’s a very expensive resources to manage.
We implemented iQueue for ORs. We were client number two for infusion, and the co-developer of IQueue for ORs. It’s more like a marketplace now, where surgeons can give up time or request it. And the third thing was bed capacity management: that’s the Holy Grail. So you have an EHR, and you’re using it to place patients, etc., but it’s really a transactional system. So how can you give virtual command center leaders the ability to see what’s happening today, predict what will happen tomorrow, and then prescribe interventions that will impact what will happen tomorrow? So we partnered with LeanTaaS again, to develop IQueue for Beds. We had actually been pulling out spreadsheets. But that was never reliable. Now, we have visibility at the 12-hospital system level.
With regard to what opportunity days are, they involve discharging a lower-acuity patient and replacing that patient with a higher-acuity patient. So capacity leaders are now making decisions based on iQueue for Inpatient Beds.
What kind of logic did you and your colleagues develop in order to create the algorithms to run this program?
LeanTaaS has engineering, data science, and math talent, and they worked side by side with our clinicians, looked at our system, looked at the gaps we had in data, and created algorithms, and we iterated quickly on those algorithms. At the end of the day, it’s complex yet simple—it’s about getting the right patient in the right bed at the right time. The idea is that if we put a surgical patient in a medical unit where they’re not used to treating surgical patients, it will cause delays and difficulties. So getting the right patient into the right bed, we’re preventing delays. What is the typical flight path for the patient we’re about to admit? And what nursing unit should that patient go into? With what predicted length of stay? Then we can put the patients in the right days. So one particular patient should go into a unit for four days rather than nine days; we can plan better. The other dirty secret here is that large health systems—we have 12 hospitals—and prior to IQueue for Beds, we were probably doing things 12 different ways. And inside a hospital, you have capacity leaders—capacity manager is a typical title. But with different people doing that job in different ways on different days, you have inconsistency of approach. Now, with IQueue for Inpatient Beds, you create consistency across shifts. The decision-making is more hardwired using the tools.
You’re reducing variation, then, correct?
Yes, at the end of the day, that is our goal. The more variation we can decrease, the more predictable the pattern and the care interventions. Now, every single patient is unique; that is the beauty and the complexity of healthcare; we’re not dealing with widgets or books or sneakers, we’re dealing with people. But within that uniqueness, there are patterns. There are typical flight patterns. And so we use advanced analytics and intelligence to determine when a patient is diverting from their typical flight path. And that diversion might happen on day two of five. So using these algorithms, we can intervene with very actionable, directed steps at day one or day two, to prevent days six, seven or eight from ever happen. If you understand where the patient’s going to be discharged to, you can understand variations in individual cases, and we can guide that length of stay to stay on track, or we get derailed. And this is so complex for the human brain and a group of clinicians to manage. That’s where AI and machine learning can really help. This helps trigger early interventions.
What are some of the key learnings taking place as AI and machine learning are evolving forward in the clinical space?
We’ve layered on clinical algorithms like sepsis, respiratory distress and other issues, into our analytics. And we have all this great data, but were sending it to bedside clinicians in the form of alerts but with nothing to do to help them. They were getting alerts saying, “Hey, your patient’s deteriorating.” So iteration number two was that we sent the alerts to our virtual health center; think of it as a clinical command center. So we have physicians and nurses who, all they do is analyze the alerts and intelligence and actually do chart biopsies, to help the clinical team at the bedside to optimize care delivery. Transfer that over to the bed capacity issue, and you might have a patient who on day two it seems they will have a length of stay of five-and-a-half days rather than five. Well, what if I can send that data to an operational command center that will have physician advisers and care managers working on discharge disposition managers or can help the bedside teams themselves? So we’re figuring out how to operationalize machine learning to support the clinicians at the bedside, to create the appropriate interventions.
What is the secret of multidisciplinary success in areas like this one?
I’m a big believer of people, process, and tools. We’ve implemented the base EHR, and everyone’s on the same instance of Epic, on the same platform. So the first step is to create that base EHR to support your analytics. The second step is to build that base analytics insight layer. And the third step is to create a closed loop between the EHR, analytics insight layer, and then creating a loop back to the EHR. Bring the actionable insights back into the workflow of those doctors and nurses; that’s your tools stack. And then you need to build those expert teams that partner not only with IT but with third-party supports like LeanTaaS. We made the decision about three years ago that we needed to partner with others who are better, smarter, faster than we are. We decided to partner with people at LeanTaaS, who can supply the right experts, who can work with our operational people and clinicians, and from the IT standpoint, we will create the tools that will support the clinicians in their workflow. So we, the IT people, become the glue to make this real within the Epic EHR.
What will the landscape around this look like over the next two years or so?
What it will evolve to is that it will still be IQueue for Beds, but frankly, some of those beds will be in patients’ homes. Think about the virtual health center. The IQueue for Beds will create algorithms that cross over between the operational and clinical. So we just got a patient out of the OR, or out of the PACU post-surgery, who will be in a bed for two nights. Could the PACU patient have one night in the hospital and then we watch over them from the virtual care center, while they’re at home? So the algorithm will be the same for clinical and operational. And it will begin to erase the distinction between inpatient and outpatient.
What advice might you like to offer to those who might follow in your footsteps?
Your IT strategy has to be your operational strategy. You’ve got to start making your digital strategy your strategy, and maybe stop distinguishing between your IT strategy and your operational strategy; they should be one and the same. We’re laser-focused on bringing intelligence to people at the top of their scope. We’re losing nurses and doctors; they’re making career decisions because of the pandemic. So we need to help people to be able to work at the top of their license. So one clinician can watch over not 40 but 4,000 diabetics, using intelligence, to help us think differently about how to create that intelligence.
