What happens when the leaders of an already-busy hospital are faced with a “burning platform” set of incentives for optimizing average inpatient length of stay (LOS)? That’s the challenge that faced the leaders at the 300-bed, level 2 trauma center Parkview Medical Center, a community-based teaching hospital in Pueblo, Colorado. Several years ago, the executives at the only other inpatient hospital in Pueblo decided to downsize their hospital from 300 to 20 beds and focus almost entirely on outpatient care. As a result, Parkview Medical Center was suddenly bursting at the seams with patients, and, in a blue-collar town and with a challenging reimbursement demographic—55 percent Medicare, 30 percent Medicaid—Parkview’s leaders had to rethink how they were delivering care, in order to become hyper-efficient at managing daily patient loads.
What Sandeep Vijan, M.D., who three years ago became Parkview’s chief quality officer (and who still practices as a general surgeon, 20 percent of the time), realized quickly that he and his colleagues were lacking the ability to analyze data on some basic levels, so they turned to Ruben Amarasingham, M.D. and his colleagues at the Dallas-based Pieces Inc., which Dr. Amarasingham and his colleagues created while he was a senior executive at Dallas’ Parkland Health & Hospital System, and then commercialized to support patient care organizations nationwide. Dr. Amarasingham and his colleagues at Pieces Inc. have been supporting Dr. Vijan and his colleagues at Parkview Medical Center on this data journey, through the use of Pieces’ Reducing Length of Stay (reLOS) solution, whose algorithms leverage natural language processing [NLP] and machine learning to surface insights from unstructured data, such as free-text clinical notes, from the electronic health record (EHR). The reLOS Framework works to reduce excess LOS by helping to identify and prioritize the key discharge barriers throughout the patient’s stay and providing a reLOS checklist. At its core, the solution’s worklist identifies interventions that could occur sooner in the patient’s stay, thus expediting processes from admission through discharge.
Through leveraging the software and using it in their continuous improvement work, Dr. Vijan and his colleagues have Parkview has realized significant decreases in their average length of stay, as well as their excess length of stay, per national benchmark averages including a relative reduction in average percent of excess LOS of 88 percent, which equates to an estimated annualized cost avoidance of $5.5M because of reduction in inpatient days.
Recently, Drs. Vijan and Amarasingham spoke with Healthcare Innovation Editor-in-Chief Mark Hagland to discuss Parkview Medical Center’s journey around optimizing length of stay. Below are excerpts from that interview.
Tell me about the origins of this initiative?
Sandeep Vijan, M.D.: Several years ago, about three or four, there’s a competing hospital across town that decided to downsize its service offerings and focus more on lucrative outpatient care; it went down from 300 to 20 beds. So Parkview Medical Center was in the position of having to absorb inpatient volume for the entire county and frankly, for all of southern Colorado. So ER wait times started to increase, among other issues. And this is a blue-collar town, 55 percent Medicare, 30 percent Medicaid demographically—these aren’t people who have the luxury of leaving town for healthcare, they rely on us. So we really had to look at throughput. Unfortunately, what we lacked when we started to look at that was good data on our patients. We had good data on our processes, but not our patients. We realized we simply didn’t have the in-house bandwidth to do this.
My predecessor began the partnership with Pieces. It offered us a way to use AI to identify patients—what their clinical needs might be in terms of discharge priorities, before the doctors were able to document it in the medical record. For example, Patient X will require oxygen to go home with; Patient Y has a high readmission risk score. And by giving these pearls to case management and nursing, we were able to build workflows that allowed us to get really efficient in our care delivery model and our care coordination, and that all translates into earlier discharges. And in the implementation phase, since we rolled this out hospital-wide, we’ve able to achieve an 88-percent reduction in our excess length of stay. So I’m going to give Ruben some time here. The concept of excess length of stay deserves some articulation.
Ruben Amarasingham, M.D.: There’s the normal length of stay of the individual; excess LOS represents the amount of LOS of stay beyond what could be achieved clinically for a patient of a particular status. And we were able to benchmark that against norms. If you can reduce excessive LOS, you’re really achieving a tremendous service for the hospital and for that patient. Hospital-acquired infections and other iatrogenic events that can occur, including falls. So excess LOS is a really good element to follow. And readmissions and LOS are two sides to the same coin. If you discharge too early, you run the risk of readmission. Threading the needle around LOS is very complicated. So using AI to support the elimination of excess LOS is very useful. And that 88 percent number represents a very high level of performance. Technology can always be supportive. AI is still in its very early days here. Combined with the kind of leadership and operational quality improvement at Parkview, you can really achieve a lot.
Vijan: Ruben’s very humble, but he and his team need to take some credit. As a hospital, we don’t know what our baseline is, whether it’s good or bad, right? How do we compare with our peers? To be able to use national benchmarks that Pieces can provide, really helps. And we were well over five days LOS when we started this process; we’re down to 4.5 days. I think it was 4.85 was the national average when we started. We were well over 5, and we’ve been able to put together processes for dramatic reductions.
Pieces has had a huge hand in this; not only does their platform share lots of data; but their team is also constantly generating data. And at the core of this is the reLOS worklist that they provide; that worklist gives us a list of action items to focus on in each patient case. We’ve built our workflow around the support that the technology provides.
What have been the biggest challenges in this initiative, and how have you overcome them?
Vijan: That’s like the hammer-and-nail question or those of us in the quality area. One of the biggest initial areas was creating the awareness that length of stay is important. For clinicians, the concept of efficiency is an alien one. They don’t understand that we have to do things in a timely manner, and that every extra day of length of stay is the opportunity to be able to care for another patient; and also for unfortunate events to occur. And we had to create multidisciplinary rounds. At first, the clinicians didn’t understand the value of it. We’re going to sit in a circle and sing kumbaya? Because the case manager can read my notes and figure it out. So that was an eye-opening revelation for the hospitalists; they themselves didn’t realize the value of multidisciplinary rounds. So you have a patient admitted with pneumonia who may or may not require oxygen at discharge. At one point in their hospital stay do we make that decision? Pre-Pieces, it was potentially on discharge day; and that takes another 24 hours to set up after you jump through the hoops of the vendors and insurance companies. So if my case managers can start working on that early, say, day 2, then that’s already set up.
So it’s a clinical determination that triggers a set of processes, correct?
Yes. Across the country, we have the same antibiotics and the same general treatment protocols; the challenges are mostly process-oriented. So in order to get that patient certified for oxygen at home, they have to do a walking test to test their oxygen levels. And that’s the responsibility of respiratory therapy, sometimes, nursing. And so getting processes have to be triggered earlier. And this is where Pieces comes in. You can think organically, I’m going to try to get this going on my own; but it’s harder than you think, without a tool. And the RELOS work tool is a line listing by unit. It’s a column with action items to think about, based on what we know about a patient from the physician notes. So it’s a bridge between what the physician is thinking about the care of the patients, and the actionable items that the care team has to put in place, for an efficient discharge. So it gives us T-minus-1 and a T-minus-2 reports, one and two days before anticipated discharge, and receiving those reports helps our teams to prioritize what to do first.
Tell me about the artificial intelligence element in this?
Amarasingham: One of the core components related to what Dr. Vijan is talking about, is the use of NLP to read the doctors’ notes and other information in the record, to understand and proactively anticipate what’s going to be needed. So much for the healthcare delivery system is based on communication, and both doctors and nurses have very little time, and when you think about how little time there is, AI can help to create a bridge to facilitate information-sharing, to make sure that care is efficiently delivered.
And one of the things we’re starting to see is that with the EHR being adopted, we in the healthcare delivery system started building a digital footprint of care, but it was really transactional and required doctors and nurses to put in a lot of information. Now with AI, we’re moving towards more fully guided care and more intelligent assistance, and that will bring back some of the joy of medicine. That’s part of what we’re trying to do with Pieces is to try to provide more and more intelligent assistance. Now, in addition to the NLP, there’s a lot of predictive modeling we can use to predict when a patient might most appropriately be discharged. Using structured data—lab data, medications, vital signs, all of those forms of data have discrete elements. You can get a certain amount of information from that data; when you combine it with laboratory data, medications, and changes in patient status documented in the EHR—that combination of all-access data to help predict, is going to be more and more helpful.
In all of this work, you’re trying to get the full picture of a patient’s situation. A doctor can come into the room and make an assessment based on the record and talking to other physicians involved in the case. And when they bring all the elements together, they develop a hypothesis. But think about all the data points there, including a lot of data that’s hard to put into context. AI is taking more and more of what I call unobserved, dark-matter data and making it visible. And increasingly, it will begin taking data from sources outside of the EMR. And to the extent that these machine systems can get exposure to all that data, their capability to provide assistance will become greater. But yes, NLP data combined with structured data, allow you get more insights.
What have the biggest learnings been so far, and what will be your next steps at Parkview Medical Center?
Vijan: The greatest opportunity there was showing us where our processes were breaking down. Where our opportunity lies for the future is in care pathways. So ,based on a Pieces prediction on hospital day 1, is the trigger for a care pathway, where certain things—such as a transition from intravenous to oral medications, or the setup for oxygen, or a heart failure clinic appointment in our heart failure clinic—triggering those early on and efficiently, rather than waiting several days into the hospitalization, God forbid.
So often, Mrs. Smith is still occupying a hospital bed simply because pre-discharge processes are inefficient, correct?
That’s right; unfortunately, that situation is all too common. Often, Mrs. Smith is lying in a bed for a day or even a day and a half, simply because of process-based delays. Now, that said, we cannot control the insurance approvals, but if we know that the approvals occur from 9 to 5 on Monday to Friday, we need to make sure we seek the approvals during those hours. At least you can start having those conversations.
What will the next couple of years look like for you in this work?
I think part two will be embracing every evidence-based element in care delivery. What we’ve tackled has been the glaring, low-lying-fruit opportunities. Further reductions from here are much harder, and they require resources from the hospital. Can we provide things on Saturdays and Sundays? That becomes a staffing challenge for case management, for example. Or resources available to the OR? There are much harder challenges ahead that we have to address on an organizational level on how we deliver care, and we have to have a discussion, because ultimately, that becomes a people discussion: how many FTEs do we have to deploy to make that happen? That’s where this is headed.
