Webinar Transcript: Health Systems have to fix the fundamentally broken math in order to navigate any stresses to hospital capacity as highlighted by the current COVID-19 crisis
Earlier this year, Mohan Giridharadas, CEO & Co-founder of LeanTaaS, sat down with Steve Hess, Chief Information Officer of The University of Colorado Health (UCHealth), to discuss why the EHR alone cannot solve capacity management challenges. The iQueue platform, which relies on machine learning and predictive analytics, has helped UCHealth improve capacity in their ORs, infusion centers, and now, even their inpatient beds. Missed the event? We’ve got the transcript below, plus the full webinar recording and a blog summary.
Mohan Giridharadas: Today’s presentation will be led by Steve Hess, who’s the Chief Information Officer of UC Health in Colorado, and myself. I’m Mohan Giridharadas, the founder and CEO of LeanTaaS. We’ll start with an introduction to UC Health, then talk about the impact of COVID on capacity management. Steve will cover both of those two topics, and then talk about why optimizing capacity is a difficult problem, and why EHRs are not built to solve it. He will then summarize the impact of applying this in a real world setting at UCHealth. We’ll leave plenty of time for Q&A.
Steve Hess: Thanks, Mohan. Good morning, everybody. The University of Colorado Health is an integrated healthcare system, across Colorado from the northern part down to Colorado Springs. So we’re 12 hospitals, 5 billion in revenue, about 2000 inpatient beds, and about 4 million annual ambulatory visits. That’s the footprint of UCHealth.
From an IT perspective, we are on a single enterprise Epic electronic health record across 12 hospitals, and we also host that for independent clinics and hospitals, about 7 million unique patients in our integrated platform. You can see some of the other stats — there’s a rather large footprint single enterprise electronic health record, and we’ve been on this electronic health record since 2011.
From a COVID perspective, like many of you, we’ve been challenged with the pandemic over the past 11 months or so. We’ve actually seen two surges, to date, one in March, April, 2020 where we got up to about 263 patients on a daily basis that were COVID positive, or in a rollouts census status. And then we actually flattened down, as you can see in December.
On December 1 we got back up to 468 patients. As you can see there, we had a rather rapid, rapid rise of COVID during our second surge. As many of you experienced too, there was a different disease that we’re fighting in the fall of 2020, than what we were seeing in the spring of 2020. It nevertheless challenged our capacity, challenged our operations, challenged staffing, medical equipment, IT equipment and so on. We’ve obviously dropped off precipitously since then. As of today, we’re about 110 this morning, so obviously with the vaccine rollout, where we have hopeful days ahead.
We’re not naive enough to think that there couldn’t be another surge here, and the need to keep our focus on this as important. So from a response to the pandemic perspective, like many of you, we’ve done quite a few things to respond. I’ll go through these quickly just to give you some context. From our virtual visit perspective we are rather mature; in our virtual health experience, we are early in the process.
We were seeing about 1000 virtual visits per month, pre-pandemic. And then we quickly scaled the existing solution to a little bit over 77,000 visits a month. That includes urgent care, primary care and specialty care. As you can see there, we leveled off and then got to about 40,000 in January, so this is here to stay. We’re never going back to that 1000 visits per month but it’s pretty impressive how all of us, including all of you, have responded to the pandemic.
We also have done some things around capacity management and RPM. We’ve sent patients home with Massimo devices and Bioanalysis devices and were using them to watch over them at home, trying to determine our hospital. From other COVID responses, we created our own symptom tracker on our customized mobile app. We’ve done a tremendous amount of COVID-19 testing, both PCR and antibodies. Obviously we’re like many of you, responding to the vaccine distribution, and as of this morning were almost at 150,000 doses administered.
In summary, what we saw with our COVID responses is that the IT platform, your EHR, is really critical, paramount to a successful response to this hidden enemy we’re all facing. The one thing I really want to emphasize here is that if we didn’t have that enterprise EHR, but also a capability to produce analytics and intelligence, I don’t know how we would have successfully responded to this.
We’re obviously still fighting the battle, like many of you, but the key to our response here at UCHealth was really an enterprise EHR where we can actually grab data real time and really analyze it, and predict what’s going to happen tomorrow, next week, next month. To not only look back at what happened yesterday, last week, last month, but use that data to predict how we’re going to move forward. We really used our IT platforms to look at census prediction. We actually looked at external models coming from not only Colorado, but across the nation, across the world, and layered multiple models on top of our data to really look at where we were heading into the future.
So, not only the descriptive analytics, like what happened yesterday, last week, last month, but also what we think is going to happen. Then that allows us to make those real time decisions, whether it’s around staffing, setting up surge areas, getting more medical equipment, getting additional IT equipment, and so on.
The upper right graph on this slide is actually our LeanTaaS iQueue tool we’re using to manage our daily capacity. It’s pulling in data from direct admissions, from the ED, from the OR, looking at what our demands are going to be for a limited supply of beds, but also looking at our predicted discharges, to really help our capacity managers and our nursing leaders to manage the daily capacity. Again, without this, without the EHR, without these advanced intelligence analytics tools, which Mohan is going to go into a little bit more detail, it will be extremely difficult to manage an ever changing capacity platform.
We’re using some of our intelligence to manage our vaccine clinic scheduling as well, and managing by vaccine, by location, and looking at how we can maximize the efficiency of the vaccine distribution.
Mohan: Thank you, Steve. Before we jump into optimizing capacity, why it’s a difficult problem, and the EHRs. Now, in advance of this session, we received a bunch of questions. We categorized them into four broad buckets. There are some questions around COVID surge and re –entry, some questions around data and analytics, some around the OR, which you can see is an important topic, and then inpatient beds, discharges and throughput. As we go through the presentations, we will touch on all of these, but they won’t necessarily be at a high level since we’re trying to cover a lot of ground. Hopefully the more detailed questions can be addressed during the Q&A.
With that, let’s jump into the two topics I’ll cover before handing back to Steve. Optimizing capacity and why it’s a difficult problem, and the EHRs.
One quick slide on our background of LeanTaaS, we are a 200 person company, Silicon Valley and Charlotte based. We work at 300 hospitals around the country including half of the largest ten systems, in total a little over 100 systems. Here’s a partial list of our customer base.
The three core products we work on are iQueue for Operating Rooms, iQueue for Infusion Centers, and iQueue for Inpatient Beds. All of them focus on using advanced mathematics, data science, and predictive and prescriptive analytics to unlock capacity. In total we have raised more than $250 million, most of them over the last four years, to build out the suite of products. With that as a background, let me start talking about capacity optimization.
As we look in many other industries outside healthcare, whether it is package delivery or freeway utilization or airlines or airport security, the notion of advanced mathematics and analytics to manage capacity is well understood and has been deployed for decades in all of these industries.
UPS delivers 15 to 16 million packages a day, not counting Christmas week. It has to manage the number and the positioning of vans, trucks, trailers, hitches, aircraft, etc., to make all of those packages move to their destination, regardless of the fact that they have no idea on any given day, who will send a package from which point to which point. But despite that massive amount of uncertainty, they have to optimize their capacity.
Airlines have to adjust the price of every seat on every flight on every day, 300 days into the future, because price is a big determinant of whether the flight will fill up or not. So the right number of full price tickets, half price tickets, discounted tickets, credit and wireless tickets or award tickets, they have to manage that window of pricing and pull that lever. That doesn’t happen by simply laying out a seat map, and trying to fill it out. Similarly with security.
So, the whole notion of how to do this is quite well understood. And as you think about healthcare, we do have the assets that are valuable and important just like aircraft and freeway capacity. OR blocks, arguably the most expensive real estate anywhere, because a single minute of OR our time is worth hundreds of dollars. Similarly infusion beds or oncology chairs, infusion chairs in an oncology center. All of these are scarce assets in limited supply, and they are perishable assets: if you don’t use it you lose it.
If a bed is empty at 10am today, Tuesday February 9, it will always have been empty at 10am, Tuesday February 9. We don’t recover last time, just like an airline seat that’s empty when the plane takes off is forever empty. Thinking of it as a perishable asset helps a lot.
So what are the two core concepts in optimizing capacity, going each in turn? Concept number one is matching supply with demand. Concept number two is linking individual and independent services and stringing them together to actually perform an overall service. These are two important concepts, let me take each one in turn.
What does matching mean? In order to match the supply and the demand, you have to understand on the supply side, you need staff with the right skills, you need equipment, whether it’s a robot or a component imaging machine. You also need the right facilities, whether it’s an OR or procedure room and exam room and infusion chair. All of these supply elements have to be available at the same time at the same place. If any one of them is out, the healthcare service does not happen. To deliver infusion, for instance, the chair, the pump, the nurse, and the drug have to all be ready, willing, and able. If the pump is broken or the nurse calls in sick or the drug formulation is delayed, that infusion doesn’t happen.
That’s on the supply side. On the demand side, you have to have a very clear understanding of the number of patients coming in, when they will arrive, and how long their treatment will last. So you’ve got a wobbly, hard-to-predict demand signal, which changes continuously, and you’ve got a wobbly, constrained supply signal which is hard to predict as well. These two have to match in a very tight window. Let’s step back and think about how most medical appointments are made. Two people looking at a calendar online or physical and simply say, “Bob, 10 o’clock, Wednesday, that’s when we see you.” Nobody did any demand-side, stochastic math. Nobody did any supply-side, constrained optimization math, they just looked at the calendar.
So obviously it does not work out that way. When we push health systems and say, “Look at the complexity of the problem, and look at the way you’re trying to solve it, by looking at a calendar,” they rightfully push back back and say, “But it’s a hard problem, It’s difficult to predict demand or what goes when people are going to get sick or treatment they’re going to be, supplies are unpredictable, etc.”
And we agree that it is a hard problem. We show them an example of where other people have solved the hard problem.
For example, Uber — every minute of every day, on every street corner in every major city in the world. They have to match the demand for a ride, to the availability of a car and driver who is ready, willing and able to offer that. They don’t just look at a calendar and say “John will give Bob a ride from point A to point B”. What they do is active demand forecasting, based on zip code, time of day, day of week, weather, construction, sporting events, etc, and the demand model that’s very accurate and predictive.
Then they understand the supply. They look at driving patterns, who drives when, in which zip codes, and for how long, so they have an understanding of the supply. Where it doesn’t match, they actively intervene. They change the incentives to bring drivers out of the woodwork, they ping drivers and tell them to go two miles east or two miles west because the demand is higher. If that doesn’t work they do surge pricing, to cool off the demand just a little bit and delay, so people who don’t want to pay $20 for a $10 ride, “We’ll get a coffee and wait 20 minutes until the price comes back down to 10 bucks.”
So what they’re doing is actively intervening every second of every day in every theater around the world to make the supply-demand match. That is what is needed, and it is completely absent wherever you’re looking at a health system. So that’s concept one.
Concept two is the concept of linking. When you think about a medical appointment, the entire encounter is a series of individual services. A patient goes and gets their lab work done, then they go to the clinic and see their oncologist, then they go to the infusion center and get the infusion done. A patient who comes in through the ED, turns out needs emergency surgery, so they go routed to the OR, recover in the post-anesthesia unit, and then they get put into an inpatient bed. These are four separate services that will link together.
Again let’s think back about how most health systems construct these linked appointments. They simply do it by inspection and gut feel. They say, “Mary, you need to have an oncologist appointment at nine o’clock. Why don’t you come get your labs done at eight and we’ll squeeze you in for your infusion at ten.” Nobody did connective math on it.
It turns out, linking is a whole class of mathematical problems, called topology and network optimization, where you have to find out the optimal path of a network by figuring out the sequences of nodes that you must traverse in order to get there. So think what an airline has to do to get connecting flights to work out. None of us would be happy with a 12 hour layover in Chicago, but if they gave us a 20 minute layover we would certainly miss our connecting flight.
So they run through crazy network math to figure out what should be the interval. Winter in Chicago needs a longer connection because of the de-icing trucks. Each plane has a different crew and staffing, crew have rotations, planes have rotations for servicing, etc, etc. So, they solve that problem, they take months and months to solve it. Again, inspecting and just making it up doesn’t work. That’s like ignoring reality. I could choose to ignore gravity and keep dropping my phone every day and hope that it doesn’t hit the ground, but it’ll hit the ground every day because gravity shows up every day. Similarly this math is under the surface, every day, on matching and linking. Health systems simply choosing to ignore it is not a solution. You have to confront it and you have to solve it.
Let me give you one more example. Both matching and linking happen together thousands of times a day, in a very productive manner. Think about what happens in the situation, pre-COVID when everyone is actually traveling.
When you’re in a terminal waiting for your flight, the plane pulls it docks at the gate. A bunch of passengers come out, and a little while later you board. 15 minutes after that you take off.
During that 45 minute window, here’s what happened. There’s a set of services called above wing services. A new crew, or loading a new set of food and supplies. They clean the cabin, they kick out one batch of passengers before they load the new passengers. Simultaneously, below the wing they check the flaps and main structures. They unloaded 1000 bags, reloaded 1000 new bags, they refueled the aircraft and checked the engines.
All of these services happen. They happen in a tight window, they happen thousands of times a day at a major airport like Atlanta, which has 5000 flights a day. And they happen perfectly. 99.99% was not good enough, you will still have a dozen crashes and if you don’t do it right.
What does this have to do with matching and linking? Every service has to match the resource. Given the speed of a baggage cart is only four or five miles an hour, and the airport is five or 10 miles across, you can’t reposition at the last minute. So what they do is calculate pushes for 15 minutes, which is how many push backs will happen in this corner of the airport, between nine and 9:15 and 9:15 and 9:30, based on the number of push backs and the size of the aircraft in that zone of the airport. They have the right number of doors, dogs, ramps, to do baggage, the right number of fuel trucks, the right number of catering trucks, the right number of cleaning resources. Every resource service had to match supply and demand based on push backs in that zone of the airport in a 15 minute window. That’s the level of mathematical precision around supply and demand. What does this have to do with linking?
Each of these services has to happen in a certain sequence. You cannot clean the cabin until you kick out the passengers, but you must clean the cabin before the new passengers come in. If you wait until all the passengers have left, you don’t have enough time to restock the plane. So you restock the supplies from the right side while the passengers are getting off from the left side door. The fuel truck and the baggage truck cannot be under the wing at the same time, because they collide with each other.
So there is a very elaborate dance sequence and choreography of how this thing needs to happen, which is different at each airport, but it’s a well designed choreography that needs to happen. So that’s kind of how linking comes into the picture. Why is this magic?
During the 80s, the Atlanta airport could do 500 flights a day. Now the Atlanta airport does 5000 flights a day. The airspace is not any different, and the number of runways is not fundamentally different. There used to be four runways in the 80s. Now it’s five runways. because it takes 30 years to put in a runway in a major city due to zoning and property rights and home ownership and all that. So with a 20% increase in capacity of runways, we’ve managed to do 10x more in velocity. Imagine if you could do this in health systems.
Imagine if a hospital could have five times the number of patients flowing in and out of the hospital every day, with still low wait times and high quality of care. That’s the magic of solving the matching problem and the linking problem correctly, in a detailed rigorous mathematical manner. Now, let me give you two examples.
Inpatient beds are the most challenging form of hotel room that exists. There are a million inpatient beds around the country, call it one of the most expensive hotel rooms in the country, because it’s an average of $1,000 a bed. That’s a billion dollars in bed revenues every day in the country.
So what is the problem? In a hotel, they have the luxury of forcing you to check out by noon, and not allow you to check in before three or four o’clock in the afternoon. So they buy themselves a three hour window where everyone who’s leaving has left, and all those who are arriving haven’t yet arrived, so they can clean, get the rooms ready, etc.
Unfortunately, in inpatient beds the arrivals happen before the discharges. Surgery’s early morning, and people need a bed immediately after surgery, just as an example. Meanwhile checkouts or discharges don’t happen until the late afternoon. You need rounds, you need to have lab work, you need to take one more look at the CT scan or the image, you need physiotherapy to come in, you need home medical equipment, there’s some reason or the other the discharges pick up in volume late in the afternoon.
Meanwhile the arrivals have already started cranking. What this does is creates an inversion window, if you will, where the incoming rate exceeds the outgoing rate. It forces backups on the system. People are boarded in the ED, in the PACU, because they can’t find a bed, they’re boarded in the hallways. Four hours later this valence starts to subside, life looks normal, but likely it will happen again tomorrow.
This constant inversion problem creates the need to get ahead of discharges, anticipate discharges, etc. That is a classic minute-for-minute supply-demand matching problem.
Let’s switch to the OR. What is the problem in the OR? It’s a supply demand problem, and the asset being managed on a supply-demand problem is block time.
How do health systems solve it? They realize it is very hard to do, so they did a static allocation of blocks. “Dr Smith you get Mondays, every week. Dr. James you get Wednesdays and Thursdays every week or the orthopedic practice, you have Tuesday, Wednesday, Thursday for surgeries.” So the blocks, which is the asset of the OR, have been allocated in a semi-static manner. Yes, you can change the blocks, but it’s a political process and changes only a couple times a year.
Meanwhile, the demand for all OR minutes by any surgeon is massively volatile. Pretend I was an orthopedic surgeon who had blocks tomorrow, Wednesday. Whom am I going to be operating on tomorrow, Wednesday? It’s all the people I saw at my clinic over the last six or seven weeks, some percent of them who needed surgery, some percent of them who would get it done tomorrow, and I could schedule, and all happen to fit within the eight hour OR block I’ve got.
It’s impossible to make that line up. Each curve here is the actual number of minutes used by a surgeon, over a period of 52 weeks. As you can see it’s shocking. No surgeons can predict how many minutes of OR time they will need on any given week in the future, and yet the blocks are allocated as if it’s static, which is why you have this classic dichotomy or a proper contradiction, which is the following:
Walk through any hospital with 20 ORs in the middle of the day, there will be two or three ORs that are empty, guaranteed. Walk back through the same hospital later at night, there will be two or three ORs performing surgery at nine o’clock at night. So why is it that we had, during prime time, valuable real estate going idle, and yet needed to use it again late at night, when you’re paying overtime for the nursing, for the techs, for the anesthesia, etc., etc., and running the hospital on the edge while needing to perform surgeries at nine o’clock at night. It’s exactly this notion of trying to look at grids and match supplies and demand, ignoring the fact that the underlying mathematics is very complicated.
Let’s do another quick poll before we jump into the EHR. The question is, what are you using right now to address this problem. You can pick one or two of these. Are you using the existing EHR functionality, through bed boards, and reports or analytics? Do you have robust internal analytics teams that are capable of doing these kinds of analytics, are you using external vendors? Let me give you 15 seconds to answer this.
Okay, so the dominant answer is internal analytics teams, with some reliance on the EHR functionality. So let’s switch gears now and let me talk about why EHRs are not built to solve capacity management.
First, this is not a knock on EHRs. EHRs do a fantastic job getting all of the patient’s data and putting it into one place. It’s one source of truth, it allows you to integrate every aspect of prior visits, lab visits, the financial aspects, care aspects, the symptoms, etc., all in one place. So, it’s a wonderful system of record.
However, the system of record is an aircraft carrier. What you need is a speedboat. So let me describe four specific issues.
One, EHRs have made every health system think about scheduling in the vernacular language of grids. So wherever you go, you see resources across the top. This can be a provider, a room, a machine, a chair, an OR. And down the side you’ll see times of day, for each day of the week.
And so these grids are everywhere. When John Doe gets an appointment at eight to nine, that’s kind of what happens. Grid-based scheduling is fantastic for tennis courts and conference rooms. Why is that? Tennis courts, and conference rooms are what are called deterministic schedules. At the time I make the appointment, I know the exact start and end time. If Steve has the tennis court eight to nine and I have it nine to ten, he knows and I know they’ll kick him off the court at 9:01. It doesn’t matter if he’s in the middle of a set, or a game, it doesn’t matter. That was the time.
If I’m in the middle of infusion and my infusion starts late, nobody’s kicking me off at the top of the hour. So you cannot use a deterministic framework to schedule what is inherently a random or stochastic event. So the math doesn’t work. So what’s the reality and how do you realize that it doesn’t work?
These grids look great the previous night, then the harsh reality of the next day kicks in. John Doe’s running long, so you need to relocate the staff for a pink appointment, there’s no space for it so you move the yellow and put the pink there, and so the frontline is playing realtime Tetris, and that simply doesn’t work. So that’s Issue One.
Issue Two is every EHR, and every EHR-based scheduling system relies on the first come first served. So if you call a health system and say, “Hey, I need an appointment for September 20, far off into the future,” they will tell you, “Great, we’re open eight to five, when would you like to come in?” Why? Because the calendar is empty, and so you can get dibs on whichever spot you want.
That sounds nice. It sounds fair. It sounds like first come first serve is a good way of doing it. It’s mathematically completely wrong.
The simplest way of understanding that, is to pretend that each day is a table upon which you must solve a jigsaw puzzle. So September 8 is a table on which a puzzle must be solved, and each puzzle piece is a patient. Once you put the piece down you can’t move it, because you don’t like to move patient appointments.
If I just handed you the pieces in random order for September 8 and made you put it down and not move it, what is the probability you solve the puzzle? Zero, absolutely zero. There is no chance you’d solve it.
Now let me tweak it a little bit. Imagine I wrote the number on the back of each puzzle piece, 1, 2, 3, 4, etc, like a paint-by-numbers, and on the September 8 table you had a cut out for where each piece should go. I hand you the puzzle pieces in the same random order, what is the chance you solve the puzzle? Nearly 100%. What changed? The mathematical and predictive intelligence to tell you where to steer each appointment.
How does this apply to healthcare? If we told you who should get a block when, for how long, which surgery should go where, which infusion appointment should start at 10am on Tuesday vs. 8:20am on Thursdays, that’s the notion of creating the cutout on each day.
The third issue with what EHRs do, is that nothing in the EHR has probability theory, simulation algorithms, to account for change. Everywhere else, where people do complex scheduling like airlines, they overbook. Sometimes they get it wrong, and when they get it wrong, they have to correct it, because once the plane is off the ground, no one’s getting on and off that plane. So we’ve all been at the airport, where they say “Hey, we’re looking for three volunteers, give up your seat, we’ll buy you dinner, a hotel room, we’ll fly you free later,” and that’s what happens.
Nothing in the EHR lets you put three appointments into two slots. Nothing lets you take a five hour treatment and put it into a three hour slot. Because of that, you are stuck in a permanent game of heads-I-lose-tails-they-win, because it doesn’t work. Now why should health systems do this? If they get it wrong, they don’t have to offer incentives like the airlines. Someone who waits seven minutes instead of two minutes, they wouldn’t even notice.
Because of the first three shortcomings, health systems have pressured EHRs, saying “Hey, we spent $100 million on you guys, what are you doing to help us?” So EHRs created a lot of reports, report writers, dashboards, alerts, predictive reports, etc. These are all nice, they’re not enough.
I could give you a perfect report on this Rubik’s cube. The top left is a white, the top right is red, the middle column is orange, red, red. I could even give you predictive analytics. If you slice the middle column forward by one notch, the vertical column will be orange, red, red. A very accurate report doesn’t help you solve the cube.
What technology needs to do is use optimization, machine learning, AI, simulation predictions, prescriptions, to tell you the five steps you need to make in order to solve the Rubik’s cube. And that’s where the magic is.
Let me switch gears and have Steve pick up on how these methods have actually been deployed.
Steve: Thanks, Mohan. The journey UCHealth has been on with LeanTaaS started in approximately 2015. Recall that we went live with our enterprise Epic EHR in 2011.
So, the natural evolution of an enterprise EHR rollout is obviously, a lot of dust has to settle after you go live. There’s a lot of optimization, there’s a lot of change management, getting people comfortable with the tools. Back in 2015, we looked at some of our oncology infusion scheduling, and what we were seeing was a bolus of activity between the hours of 10am and 2pm, to the point where we were creating significant patient delays and waits, and actually incurring quite a bit of nursing overtime in getting our patients through our infusion centers.
Again we were using Epic, all the tools, including Epic scheduling. All the templates were set up, we’re looking at two hour, four hour, eight hour infusion schedules. We were just saying, “Okay, now what do we do about this? How do we solve this problem?”
We turned to LeanTaaS as an innovation partner and said, “Help us think through this from a mathematical calculation perspective. We’re trying to figure out how to schedule different lengths of infusions in a daily schedule, to actually flatten that peak from 10am to 2pm, create a much better patient experience, and get patients through and reduce overtime.”
We actually worked with LeanTaaS and their mathematical geniuses in Northern California. What we did was create the templates in Epic, and then send that de-identified data out of Epic to the LeanTaaS cloud, and then run that through their machine learning algorithms, which then spit back out. “Here’s what you need to do to change your templates to flatten that peak.”
We did that. Within 90 days, what we saw in our first largest academic medical center infusion center was a 7% patient volume increase with no other variables that got changed. So just by applying this machine learning math on top of our existing EHR scheduling capability, we flattened that curve.
We saw 7% more visits, more patients, and we didn’t actually have to add staff, didn’t have to add chairs and so on. So then we started to roll out the infusion solution to our other infusion centers. So then what we did is we turn our attention to, “Okay, where else do we have extremely expensive real estate where we have unpredictable scheduling patterns, and we’re trying to play this game of Tetris, in a way that actually improves surgeon efficiency, OR nursing efficiency, our patient satisfaction and efficiency, and reduce our costs?” We turned to the LeanTaas team and said, “Let’s figure out this problem with the OR.”
ORs are infinitely more complex, because we have more variables and we have surgeons involved. It was not an easy nut to crack, but we did crack it with LeanTaaS’s help. We then rolled out that solution across all of our ORs as well.
Lastly we’re starting to roll out the inpatient beds, the acute capacity management tool, as we talked about earlier. So this has been a journey, and it’s been a wonderful partnership to leverage that investment in the EHR we’ve made and layer intelligence on top of it.
From an infusion perspective, we rolled out our infusion tools with LeanTaas across our entire footprint. You can see that it works in large academic medical center-based infusion centers. It also works in some of our smallest Community Hospital infusion centers, just as well. This is an enterprise EHR with enterprise intelligence tools from LeanTaaS, then on top of the EHR, and we rolled it out completely across the UCHealth footprint.
So what do we see? You’ve heard me mention that in the first 90 days, we saw about a 7% uptake in volumes and a decrease in wait time. That has been sustained over the years. Here are some of the graphs. COVID excluded a little bit, because that changes a lot. But in essence we’ve increased our visit volumes on an annual basis, while decreasing the infusion wait time, almost to the same degree. What we’ve seen is an increase in visit volumes, decrease in wait time, and from an investment in the LeanTaas tools, versus the revenue and patient satisfaction, we’ve seen a 13x ROI on our investment in these intelligence tools. So 13 times our revenue, versus our expense by rolling this out across the UCHealth footprint.
From the OR perspective, we rolled out iQueue tools across all of our ORs. Again, this is an academic medical center, inpatient and outpatient. Then we extended to our community hospitals, inpatient, outpatient. You can see that from the largest of the large academic medical centers to the smallest of the community centers, it does work.
Over the years, what we really moved away from is managing and monitoring block time utilization to get down to operating room utilization. Because at the end of the day, you’re going to be limited in your thinking if you’re just thinking about surgical blocks. You want to look at the most efficient use of the rooms, and you can then get more granular around staffed operating room utilization and so on.
Here’s some of the outcomes we’ve seen with the OR as well. We’ve essentially created a marketplace of OR time, where surgeons can actually give up OR time because they don’t have a patient or they are out of the office on vacation. Other surgeons or their scheduler can then pick up that OR time. Part of the iQueue tool is this essential open marketplace for OR time, which really allows for the effective and efficient use of OR time.
On some of this data, the release fill rate percentage should be higher than 17%. If a surgeon gives up that OR time, we should be able to fill that even better than that. Even with these tools, UCHealth has room for improvement. Now, what’s interesting about this is we’ve seen anywhere from a 4% room utilization to 13% room utilization, dependent upon the hospital and the inpatient versus ambulatory. So we’ve seen a 4x ROI in our OR implementation, which is significant when it comes to this expensive real estate and the amount of money we all put into our procedures and ORs.
So from an inpatient needs perspective, we’re still early in this journey. But imagine us. We’re enterprise EHR customers, but we were still essentially managing some of our daily capacity management decisions on Excel spreadsheets. With this rudimentary high school math behind the scenes, we were trying to figure out how to predict how many discharges and how many admissions we were going to get. We moved from that high school math on an Excel spreadsheet to college-level calculus with iQueue for Inpatient Beds.
Our teams are using this tool that gets real time data out of our EHR to feed the essentially hourly decisions that are getting made, based on the ED demand, the OR demand, the direct admit demand, and then the supply of our valuable resources. So we’re still early in our journey, but we fully expect to have the multiplier x ROI on these tools as well.
Mohan: Sometime last early last year, Forbes approached us, saying our angle on how to drive healthcare operations in a fundamentally different way than just doing process improvement at each step of service was unique and different. They wanted us to write a book.
So along with our president and chief operating officer, I wrote a book called Better Healthcare Through Math, which is now in your favorite format: audio, ebook or hardcopy, you can look for it on Amazon and get it. We’ve got a limited number of complimentary copies we’d be happy to send out to you, drop us an email at BHCTM@LeanTaaS.com, that is Better Healthcare Through Math, and we’d be happy to be in touch with you.
Some questions have come in on top of the original ones. Steve, this one’s for you. How does the supply for COVID injections coincide with patient payment transparency for the shot and the service, which is often not covered by insurance?
Steve: Obviously each state might be a little bit different, but I’ll speak about our experience in Colorado. First of all, the patient has no out of pocket financial responsibility for the COVID-19 vaccine. That’s the first thing everybody needs to understand. There is no patient financial responsibility, no out of pocket costs. The patients aren’t being charged for the COVID-19 vaccine.
There is an ability for healthcare systems to charge for and get reimbursed for the administration by the patient’s payer, based upon their plan. We have made a decision at UCHealth not to charge the payers for that administration fee. That’s obviously been fairly controversial and each health system will make their own decision.
What we’ve been doing is calculating the cost of our vaccine distribution rollout. Our weighted average is about $17 per vaccine dose. That includes all the IT work, all the vaccinators, all the registrars, all of the supplies — which is really what the question is about — that leads to being able to vaccinate that number of people. We’re working with the governor of the state, but also some of our key payers, to look at how we can share in some of that cost. But the goal here is to get needles in arms, and get Colorado back to as close to normal as possible. We made some decisions around that administration fee and worked with the payers in the state of Colorado, differently behind the scenes, and not let the guarantor payer registration process get in the way of getting needles in arms efficiently.
Mohan: Fantastic, thank you, Steve. There’s a question here, which came in before the session. How do you integrate capacity across multiple health systems in a region, given that you guys run so many hospitals across all Colorado?
Steve: That’s a great question. We are an integrated healthcare system. Since all our hospitals and ambulatory clinics are on the same EHR, we do actually have an integrated centralized Transfer Center. The idea there is that the capacity management tools from LeanTaas that we layered on top of the enterprise EHR are being managed by centralized teams. Those teams are able to see all of the capacity across the different hospitals across our different regions of Colorado.
What I like to emphasize is that the intelligence tools from LeanTaaS will work at a local level, will work at a regional level, or work at a system level depending on where you are in your healthcare system integration platform. We’ve actually made the decision to centralize a lot of the capacity management, so we can make decisions from the direct admit perspective — do we want to send them to all our academic medical centers, or to another hospital? We also have a virtual health surveillance capability that allows our doctors and nurses to see patients in other hospitals, to take care of them in a standardized way, so whether it’s a local process, or a regional process, or a system, enterprise process you have, these tools will work. The ability to see data in a transparent way, and actually have the intelligence to make those go-forward decisions, can be set up based upon your individual workflows.
Mohan: Terrific, I’ll take the next question, around improving early discharge.
This is right at the heart of our inpatient beds product, and we spent literally two years on just that narrow math problem. Here’s why the conventional approaches to it don’t work. Most efforts to improve early discharges focused on the flow, “Can we get the process right? Can we predict based on the patient that this patient is showing these characteristics, and therefore will need more time to discharge, and so we should double down on them?”
All of those turned out to be very noisy data, because patient for patient, the variation is simply too high. So we’ve come up with a mathematical approach that works.
We’ve decided that each unit in the hospital, the 24-bed general medicine unit though the 16-bed ortho recovery unit, whatever that might be. Each unit has its own fingerprint, just like all 7 billion people on the planet have their own fingerprint.
So what’s in the fingerprint? The patterns of arrivals; how many come in; what type of patients come in; how long they stay; what is the typical length of stay and what is the discharge process; how accurately are they predicting the original estimated discharge vs the actual discharge; do they tend to run late, what happens? By looking at these patterns, we have created a fingerprint model for every single unit.
For every single day of the week, we run 24 mathematical models continuously for each hour of the day. We’ve got a unique fingerprint on how each unit behaves on each part of the day, throughout the day. These models are continuously updated in real time, and are constantly retrained and calibrated again. What this does is give us a stunningly accurate prediction of the discharges that are likely to happen, unit by unit.
That changes the entire event management process on its head. What tends to happen today is, there’s a 7am meeting, where everyone shows up with a spreadsheet that says, “I’ve got a 24 bed unit, 20 are occupied, I’m expecting six in and four out.” So everyone is doing abacus-like arithmetic, a plus b minus c, and trying to run the hospital backwards. What our models are doing is playing out the chess game for the remaining 23 hours of the day, all the way up to this time tomorrow, and predicting when openings are likely to happen. Now your patient placement people, instead of playing reactive six-year-old’s chess where they see one move and literally one move, know how the chess game is going to unfold over the next several moves, and can make intelligent counterintuitive decisions. For example, “This patient should be placed, there isn’t the place right now, rather than putting them in an off service unit, let me wait 30 minutes, because I’m confident a bed in the right unit is going to open up and I can put them in there.”
Also, all health systems reorder reactive transfers, and boarding at the ED as a crisis after it has happened. By predicting it, we can stand up surge capacity well in advance which works out much much better. The key in the discharge is understanding it’s a unit by unit fingerprint game, and it requires sophisticated modeling around that.
Steve, I think the next question for you is on rescaling. How do you rescale all clinical people, both licensed and non licensed workforce on the fly, given the demands are ever changing?
Steve: First of all, going into COVID, we actually looked at all of our clinicians from that competition skill license perspective, and got them into essentially labor pools. Those allow us to pull from the ICU, or inpatient ICU, or ambulatory to inpatient and everything in between. We tried to meet that demand-supply matching through our central labor pool. What’s important to understand is we actually leveraged the EHR, and the intelligence, to get to the point where we were predicting what the surges look like, what the volumes looked like, and then try to stay one step ahead of it from a staffing perspective and a medical equipment perspective.
Obviously, when you start talking about the ORs, you kind of have a common labor pool. Talking about the med surg ICU, you actually have a little bit of overlap there, but obviously we got some ICU versus the med surg, and then you obviously have ambulatory. It’s coming down to some of the slides Mohan showed around that supply and demand matching, making sure you understand what the physical care areas are, what your medical equipment is and then what your staffing is and then matching it as best you can. I would say that you want to try to stay within those carriers as much as possible, but COVID’s challenge us all to think creatively about matching multiple staffing pools across a common footprint of supply.
Mohan: Steve, the next question is for you as well. When you think about data and analytics, how do you think about the role of your internal analytics teams versus external tools like ours?
Steve: It’s a great question, and the poll answers indicated that many are solving this problem through internal analytics teams. The reality is I have an analytics team as well. I have a very large EHR team and a very large analytics team. If you’re anything like UCHealth, your IT teams never, ever have enough resources for what the organization demands.
One I think is going to be really important for people to understand is that what we’re talking about here is not report writing. I have an excellent analytics team. They’re certified in all the different Epic modules, they’re certified in the Epic databases. We actually have a data warehouse that’s been fed using Epic, and we’re using Power BI from Microsoft for visualization, and the team is awesome. But what they’re really good at is actually descriptive analytics, as in producing reports, and Power BI visualizations that showed us what happened yesterday, last week, last month, last quarter, last year. It’s incredibly helpful, and it’s really important and it’s necessary.
But what we’re not great at is the math. It’s the advanced analytics, the intelligence on what’s going to happen tomorrow, and even more importantly, not only what’s going to happen tomorrow, but what do you do today to impact tomorrow? So that’s how we move from descriptive analytics to predictive analytics to prescriptive analytics, and what Mohan and the LeanTaaS team deliver is mathematical data science. They’re not delivering report writing, they’re not delivering analytics, they’re delivering predictive intelligence prescriptive intelligence that allows us to make current decisions.
A couple of things. First of all, one of the natural change management concerns people will have is, “I just invested 200 million up to 2 billion in an enterprise EHR. Why can’t that do this for me?” And then you’ll also have a natural organic reaction to, “I have a 30, 50, 70 person analytics team. Why can’t they do this as well?”
First of all, this is not report writing, this is not analytics, this is data science and math. Second of all, this is not a project where you do it, and then you implement it, and then you just let it ride. This is not a set it and forget it. The machine learning aspect of what we’re doing with Epic, LeanTaas and the UCHealth team is around constantly learning, constantly tweaking the algorithms. The machine learning is truly that. It is a machine learning capability that’s constantly looking at our increased volumes, and our days and our templates and so on, and informing us, predicting what’s going to happen but then also prescribing to us, what should we do to change our templates today, our schedule today, our staffing today, our equipment today, to impact what’s going to happen tomorrow.
The folks on Mohan’s team are not Epic report writers, they’re not Meditech report writers, they’re not Cerner report writers. They’re mathematicians, they’re data scientists. Colorado is a beautiful place, a gorgeous place to live, and there’s nothing better from my perspective, but data scientists aren’t coming to Colorado. For us to attract and retain data scientists, mathematical geniuses like the folks on Mohan’s team, isn’t feasible. Even if we turn to the University of Colorado, there’s some brilliant people here, but to attract and retain mathematicians, data scientists is a very different game than trying to attract Epic report writers and analytics folks. You should carefully analyze what you need within your own organization and truly look at your ability to attract and retain people beyond your report writers and Power BI analysts.
Mohan: One last question. Let me get this. This is specifically interested in OR capacity management.
Let me describe just like I did for the beds how we think about OR capacity. First of all, every health system, before they work with us, relies entirely on surgical block utilization. Block utilization is a helpful metric, but it is mathematically completely incorrect for doing operating room capacity management.
Why is that? Because an average is an average, so there could be two surgeons with identical 75% block utilization metrics. One of them achieves that 75%, by running their OR very efficiently three days of the week, and on the fourth day that they have block time that they abandon. That is 75% utilization. Someone else could have highly variable length surgeries, like a neuro spine surgeon who doesn’t know what he or she will experience until they get into the case. So sometimes the case is two hours and sometimes it’s five. They can be working really hard every single day and still have 75% utilization.
From the first surgeon, you can take away one of their four blocks and not hurt their practice. From the second surgeon you cannot. Looking at block utilization is actually completely incorrect.
What we’ve done is come up with a mathematical concept, which we have filed patents on, called Collectable Time. We can analyze the patterns of surgeons using time and figure out how much of it could be collected back without damaging that practice. Notice the five minutes here and there are first case delays and turnover times. Those all sound great, and process improvement efforts go after it. But if you save five minutes here and there, you won’t be able to squeeze in an extra case. Collectable Time is large chunks of continuous time that will be left idle: a four hour delay, not using the whole afternoon, an entirely abandoned block, a block that was released too frequently. We mined the data for Collectable Time, and based on that suggested reallocations.
Think of that as portfolio management of a financial investor, you should have such a substantial percentage in stocks and bonds and so on. So having done the portfolio allocation, we built a marketplace on top of it. Think of that as daytrading.
What our marketplace does is mine the data of all the 100 surgeons and says, “You got a block three days, three weeks from now. I don’t think you’re going to use it, because your typical patterns would have started to fill that block out six weeks in advance.” So a text message is proactively sent out to that surgeon, who can release the block with one click. When they release it, it becomes available to someone else to do it and all the constraints — is it a robot room, not a robot room, what kind of surgery, what kind of equipment — is all mathematically captured behind the scenes.
By creating this liquid marketplace, people are more willing to be good actors and give up their time, because they know when they do need it, they’ll get it. The liquid marketplace is like day trading on top of a collect framework, which is like portfolio analysis. It completely changes the way even highly, highly well-running OR organizations with 70 or 80% block utilization before are finding five and six points of improvement. We can almost guarantee hundreds of thousands of dollars per OR year in impact, which leads to the 13x and 14x ROIs Steve was talking about earlier.
Thank you very much for attending. Thank you, Steve, for your time as well.
Please feel free to drop us an email if you need more information: firstname.lastname@example.org