I have nothing to disclose. I will show you some slides from our mission control command center which runs on a GE software. We will discuss how to incorporate forecasting data into our clinical operations, and then how do we track clinical teams' workload, two, I think, very important parts of our operation. To put you a little bit in context, the department I'm leading is the OHSU Mission Control Center. It has an inbound role, which I highlighted here. It's how we run our entire health system. We are about an hour and a half flight north from San Francisco, an academic center up on a hill. We are probably the only academic center in the world who has an aerial tram between our front campus and the main campus here. Our health system are three hospitals, the academic center, which includes an adult and a standalone children's hospital, and then two community partners. We manage the entire capacity. We manage the transfer center, all inbound traffic. We oversee the throughput of our patients through our health system. We do the RRT and emergency activation through our dispatch center. We run the virtual hospital, which is currently including a telemedicine ICU where we serve external and internal ICU units and a hospital-at-home unit, which we opened during COVID. And then what pertinent to this talk, we have built out over the last five years our predictive analytics center that's really a partnership with our business analytics team, which helps us to really get the operational mindset into this data mining for us. Our health system runs on both Epic and Cerna EMR system, and I mentioned the command center. So for us, this was really a journey, and I really want you to walk with me through this. In 2016, this is how we managed our capacity. We had a whiteboard. We recognized in the morning, wow, this is going to be a bad day, which was already early if you recognize that in the morning. And then we deployed. At this point, I was the director of critical care. That's how I got into this job because I was summoned to meetings every day where we kind of like, wow, this is really bad. What are we doing? We wrote it down. The nursing team had different data than I had from the attending physicians, which was really interesting from the get-go to like, why don't we ask the person who's medical legal responsible to decide if a patient needs an ICU bed or not, and not just hearsay. And then we wrote it there, and by the time it was written, it was already outdated. Like there were more transfer center calls. There were upgrades, RRT calls, patients died. So then in 2017, we made the decision we need real-time data who does that automatic and is updated in real time. So we went with GE for us to have that. So essentially, the data you see here is a much more detailed version here. It shows you by cluster what your current occupancy is, how many patients are supposed to discharge from this unit today, how many patients are coming in, the entire cluster. We see our entire health system there. It updates about every three seconds. So this allows us active problem solving. So right now, we don't have to pick up the phone and call around what's going on. We just know it. We can look at it. We can pull it up from anywhere. But what we learned during this COVID pandemic is like, wow, this like knowing how full we are doesn't help us because we're always full. And we're like, it's more the question of how many patients are waiting which we are not serving. So what we really want to develop, how do we go in the future? And like I think this is from an operational management, how do you run an ICU? How do you run your team? It's like really, for me, it's going from the toddler phase. We're living in the presence. Like I have no ICU bed right now. But how does next week look like? And the most important part there is like, how do we make next week looking better? And I will share you this a little bit. So this is what looks like sort of an unstable VF where the electrodes move around. This is what we have now. It's our nine-day hourly forecast. And I'm going to walk you through this. So the adjusted forecast census is what you see in gray. This is what like data back to 2012 seasonal variability, which for our system also includes when which services have the annual meeting and reduce their surgical schedule and weather pattern. Everything is in there. It's built by one of our economists, which used an approach much more like you try to predict the financial stock markets. Red is what we entered over the last year because with COVID and with all these waves, similar like in Columbia, every time we upgraded Epic, it seems like a wave came. We recognize that the data before COVID is almost irrelevant. So we kind of now using something which shows the acuity and the sensors over the last six weeks, because that really drives the demand on nursing staffing, depending how many ECMOs we are running, how many open heart surgeries we're doing, how many emergencies, like it's a very different nursing ratio in the ICU than not. And this is actually labeled here with the staff beds. You don't need to be a data scientist to figure out if your expected census is higher than your staff beds. It's a problem. Especially, it's a problem that like being an anesthesiologist in the OR, if the next day we just have five less anesthesiologists, we are kind of deciding which rooms are we not running and canceling these cases. Once the unit is full and you just don't have the nursing staff, you can't just not care for the patient. You need to figure out how to do that. So this is like kind of what shows you these trending and you can see that we cross up and down. Like for us, it was also a very interesting pattern to actually see that while the hour to hour, like our nursing staffing and our system goes in four hour intervals, but day and night it changes and there's no stability. It's like, it seems very random at times. And for the people who like numbers, we also put it at a number part and it's color coded is like blue means we have more staff beds or nurses than we predicting to need. If it's anything like red, orange, and like deeper, you can see where your problem points is. And then we also looked at like eight weeks out because like on the moment, like there's ways and we will talk about it, how we handle this, but this is sort of just a snapshot of an eight week forecast and shows how many day shift nurses we have and how many night shift nurses we have. And that's pretty balanced. If you look at an eight week part, a little bit up and down, and it shows you what this translate in staff beds, which also includes the acuity, like, you know, how, what is your nursing ratio demand will need to look like. In the summer 2022, it looked probably similar, like in a lot of your systems, like staffing shortages across, like we had this predictor. And like, you know, I think we are a learning system, what we want to do, but like here we saw the wildfire coming, but we didn't react to that. We're just like, ah, it's not going to be that bad. It always works something out. It was really that bad. We just didn't have enough nurses. We actually had to really close beds. We had to close more than 10% of our ICU beds and make adjustments to that because we missed that opportunity to create, bring in more travelers and do things. This was a lesson learned over the summer for our executive leadership team. And now we are, we actually using this data to kind of level, like when you look at this from now, it's flat again, because we, we have an agreement. This is the number of beds we want to staff. And this is our nursing ratio, what do you need to do that? And this is where our hiring targets is and what we cannot fill, be filled with travel nurses. So we can make the process much more smoother and simpler and kind of proactive. The important part with all of that is also for us to like how accurate we are, like on the left side, you see what actually showed up like the cluster adult acute ICU and Pete's and how it's predicted. And you can see like over a 14 day course, how this kind of changed. And you can see that like, you know, sometimes we overestimate, sometimes we underestimate. The interesting part is at the end in the ICU nurses volunteer to take, to come in an extra shift, to take care of the patients we have while in acute care, it's less of this commitment. And part of that is also like acute care is so much more like you're like, if you call in sick, you're less than 1% of our workflows in the ICU, you're already a high, like we, it's just a different kind of margin there. And this is like, again, this, how much is scheduled. That is what in our, in our data. And we use the data from our scheduling software. And this is the expected nurse. So the in between is the same day sick call. And like, you know, it's important for us to really acknowledge that when you have a larger workforce, people call in sick because our life goes on. Like they are sick, their family members sick, something happens. And like, so we building this now into our management, how to do that. So how does it look in reality? What do we do with that? So this is just a quick one-on-one, how do you manage capacity? Like you have patients coming in, there's the scheduled and the unscheduled one. Common myth is that the scheduled one is predictable, the unscheduled one, unpredictable. It's actually most likely when you look in your system, like an hour, what's coming in through our emergency portal is stable, seven days a week and nights. We create on the schedule that's self-inflicted from us. And that is very, like, I'll show you a little bit how this looks like. The throughput is really depending, how many staff beds do you have? What is the time, the actual length of stay of the patient in that? And what is your staff, all the supportive staff? It's not just nurses, physicians, everything else, pharmacists, and also your environmental service like the cleaning crew and all of these parts. And then where do the patients go? And like we get, like, we spend hours every day talking about long lengths of stay and access to post-acute care and SNF, and we often totally ignore the fact that 80 percent of our patients, despite a quaternary care center, go home. They might need home health, but they have a bed at home, they have a home and they can go there. Twenty percent have a wide range of problems why we can't discharge them. And so this is how, what we created sort of this tipped-off pyramid, like we want to be green, open for all services, we can just schedule everything like our surgeons and procedurals want to do. We haven't been green since, like, late 2019. And then we kind of close portals, like our transfer center portal, we kind of message that the ICUs are, like, extreme high occupancy for us means we will place the patient, but there will be hours delay so that the intensivist can coach the referring. And then we kind of really go further down, like closing the transfer center for certain areas, go on ambulance divert in our emergency room, stroke, STEMI divert, and we are one of two level one trauma center. We use this very rarely, less than 1 percent of the time going on trauma divert. And as you can see, depending where we are, the risk that we need to adjust our same-day surgical schedule is high, and that kind of same-day cancellation, it has a cascade of unpleasant parts from starting the phone call to tell the surgeon that their case is canceled, but also way more important to call that patient and the family, like, hey, sorry, yes, I know we scheduled you eight weeks ago, but we can't do it. And it's really challenging around that. So the simple math right now for us being completely full or overfilled, and admission and discharges need to match, and they don't necessarily match. This is just the entire admission into our acute academic adult center by day of the week, and like, to make it easier, like, the pluses is we discharge more, the minuses, we discharge less. This is how your bell-shaped admission curve kind of gets created, Monday, Tuesday, Wednesday, most systems admit more, and then over the weekend, you try to offload it. But this is not happening anymore. We are not really offloading our health system. Like, you know, we used to have about 60 patients less on a Sunday morning than we had on a Wednesday. Now it's essentially a flat curve. And so you really need to use these data to understand what's going on. Like, so this is just, like, the number of patients waiting in our emergency room, new patients getting added to the wait list who need to be admitted. This is an uptrend since, like, this pandemic started. We are now at, like, three to seven an hour. It used to be less than one patient an hour. And that's an amazing demand. Similar in the ICU, we have, like, about 20 to 30 percent of our patients in the early afternoon are downgraded. We can pull out. But in order, like, in this, it takes us about 13 hours to find a bed for an ED boarder, and it takes us about 21 hours to find a bed for an ICU boarder, not because we just, we are not good looking for ICU patients. That's an active decision, because in the ICU, they are safe. We only pull the patients out of our ICU to backfill the next admission, so we keep the ICUs full as long as we can staff it. In the ED, we need to pull them out to keep the ED operational. But we need to add other times to that. Like, this is, like, our teams on average write the order at 1139, but the next patient in this bed doesn't get, on average, admitted until 6 p.m. So, there's, like, the room turnover, which is a very familiar term for anesthesiologists. Wheels in, wheels out of an operating room. That happens in your ICU the same. Like, you know, you discharge a patient, the room needs to be clean, needs to restock. All of this takes time. And that time is idle for the system, and most importantly, idle for the patient who actually needs to go into this next bed. So, these are the things, like, when we look at our entry portals, the one we can manage is really our schedule. And thanks to this pandemic, we actually were able to level load our operating room and procedural admissions. We kind of did a very sophisticated analysis, like, the discrete event modeling. When do we reach tipping points in certain units, all of our ICU acute care? And so, we came up with, like, admission targets for certain levels. And now, every week, we decide based on our forecasting how many surgeries can we absorb. And that instead of, like, no, you can't do all your surgeries on Wednesday, but, like, you can do substantially more surgeries if you do 20 every day. And so, that's kind of how we manage that. And we also manage how many scheduled cases we allow to come into our ICUs, knowing that we have about two-thirds additional admissions out of the emergency room, out of upgrades from the operating room through the trauma system, and all these other portals. And knowing this allows us to really break the system less. And this is what we do, again, like, level loading. This is how we looked in 2015, as I mentioned. Now, we are much more like this. They're still, like, a little bit more earlier in the week, and we're actively working on shifting and incentivizing our surgeons to do surgeries more on a Friday than on a Monday. This is really, like, active leadership management around that. For the sick calls and the anticipated gaps, we have a very active incentive model. Our nurses are unionized, so they need to be part of our union contract. But we are now releasing financial incentives five, seven, eight days ahead of it so that the nursing teams can also plan and schedule that I can take an extra shift. We open and close our entry portals while EMTALA, through this entire pandemic, continues to exist. So, any time we have an open bed, we can't really say no to preserve it to something else, but, like, that's kind of where we handle it. And then we open and close overflow areas. And we used to only do this on Tuesday and Wednesday, and now we have about 36 beds open seven days a week and staff it accordingly so that we can make a little bit of a dent in the backlog of our procedural surgery and still care for our ICU patients. We did this at the beginning manually, and then we moved it in an electronic version. Like, this is a tracker which shows us the scheduled cases in the main operating room, what is new admission. This is, like, against these targets here, as you can see. And then, from our command center, these are all the patients, because, like, the acute care patient who gets a cranium that needs to go to the neuro ICU, we still need to process them and take care of them. This is now giving us a 14-day forecast. We can exactly know. It drives how they schedule cases. It has minimized to almost eliminated our need to cancel surgeries, and we can really load the surgical ORs. And it allows the anesthesia department to actually really move, shift between inpatient care to outpatient care and redistribute the workforce. And for the intensivist, we can run it so that our ICUs essentially run always around a 90 percent occupancy rate, which helps us from a staffing perspective. So with all of this, like, I was really keen that, like, I also want to have a trending ability in real time how busy our various ICUs is. So we took, essentially, like, people may remember, in 2013, the SCCM actually put out a white paper about how you should determine how many patients your ICU team can take care of. And we kind of advanced this and kind of fine-tuned it, what we think is, like, what drives actually a high-acuity ICU patient, a low-acuity patient. And we kind of put in things like these various organ failures and other things, but also, like, lines, like, in these procedure things. And what we have now is, in our command center, we can trend, like, how many patients do we have in the ICUs, how many are acute, how many low, how many we have on our ECMO team. And, like, we kind of put a number on there, like, you know, the team, like, the PA, the PNP, the attending, the fellow, the resident, the intern, together, they have so much time to care for patients. And each of these, we kind of gave a time value and kind of assess how much time is left. And then we can, like, look at this in a trending version. And currently, we are feeding our machine learning algorithm to predict, based on what we are doing right now, how will the next two shifts look like. So this is work in progress. And, like, just as an example, like, you know, three-week data, as we went into this tripledemic, the ICUs who are really staffed to go to full, and, like, we didn't overflow any ICU patients during this kind of wave, shows you that mostly our ICUs operate in a loaded, in an appropriate loaded environment. And none of the intensivists have kind of complained that they were too busy or anything. For hospitalists and the opposite, this, like, you know, they had to care for, like, 30, 40 more patients than normally. We used this to kind of really activate crisis standard of care, to change the documentation, to pull in additional resources, like opened up an extra hospitalist team. We also used this when we hit a certain kind of sensors and acuity part to offload some of the patients, like orthopedic then needs to take care of their own patients and not medicine taking care of them, and cardiology takes care of their own advanced heart failure patient and so forth. So this has been super helpful for us in kind of trending. And then also the most important part to actually understand when are we coming down, when we can scale back and can go closer to a normal operation. And again, like, this is how it looked in real time, like, where all these teams showed, like, we are overloaded, they can't really take the next admission. And then we had to really figure out how to do that. With this, I'm happy to take any questions. And thank you. Thank you.

forecasting data

clinical operations

real-time data

patient acuity

staffing needs

scheduling optimization