Read
Michael Besedick (00:05):
Thanks everybody for having me. Yeah, so if you guys recognize Peter Sellers up here, I thought Dr. Strangelove was appropriate because it's been a game changer for me and something really fun to work on. And as I'm sure everyone that's in this room can appreciate, it's a really tough problem. There's a lot of different ways to do this, but results are mixed. I mean, if you asked anyone in this room, they could probably put a staffing plan together, but you never really know what you're going to get. And the first category is tradition really has reigned supreme. It's what's driven behavior and how you assemble this schedule for years and years and years. And some practices have this, particularly private practice groups have been well equipped because they've both looked at the staff and the schedule and looked at the P&L and have intuited what the gap between those two things are.
(00:54)
As the industry shifts, you're going to see that experience kind of erode and not necessarily seeing the same level of performance as you go from one paradigm to the other. OR schedules are, they're either flawed or they go still quickly. You can have a great guess, get the direction, you might use a block schedule, but everyone knows here any given day you're looking at a 20 to 3--0% add-on rate in most acute care settings and it's only gotten worse in the years that I've been working in this industry. So what are you left with? Well, you have some technology tools that are on the table. You have workforce optimization tools, schedule tools sound great, but who set these tools up? Was it the provider preference in setting up their schedule? Is there an alignment with what the hospital's goals are? Those are also in flux.
(01:45)
After you go through the configuration stage, you're already onto a different paradigm. So what is a better solution? Well, I personally think one is something that balances your coverage and your productivity dynamically. So something that can learn, something that can adapt to new inbound information, something that is not static, that takes you months to produce and then as soon as it's produced, you walk away and it's gone. So the other thing that you need is you need a plan where you can actually take action. So once the schedule's produced, well, what do you do with it? So I understand what my coverage needs are. What am I actually going to change? Who am I going to hire? Who am I going to fire? What open recs can I open or close as a function of this? And what we've seen in this kind of dynamic approach is usually like an eight to 10% pickup depending on the scale.
(02:36)
It depends on where you start, but this has real results and I'm going to walk you through actually a real example of how this can work for you day-to-day. So there are three kind of tools that you have in your toolbox if you want to solve this problem. You have experience, you know the facility well, or it's very small. It's something that you can just understand for yourself. You have a contract or a budget, something that you've set up with your group, or you're using OR data to actually figure this out. Really going to touch on these two because it's hard to put experience and intuition on a screen, but these are really the two predictive tools that you have in anticipating what's going to happen in your day. And let's just walk through an example of what this looks like. So this is a real OR.
(03:21)
This is an actual day, a schedule. This is something that you probably look at. Maybe it's not necessarily gantted out. Maybe you're looking at a sheet of paper, maybe you're looking at something on your phone screen where you're looking at rooms and you're looking at cases in these rooms. And I put the three o'clock, five o'clock, seven o'clock cut points. Those are the common places where you're going to really design your staffing around you need to pay attention to because those represent your eight-hour, 10-hour, 12-hour shifts. You see some staggered starts here, some strategies that we talked about employing here and you're going to convert this into a rooms running tool. We've all seen this. You can see up at the top right, there's my Gantt chart. And in the middle here, we have our concurrency, your total number of rooms running needed by hour of day.
(04:08)
And this is kind of the first step in visually understanding if I'm a director and I need to put a staff together. Okay, I need this many people working at three. I need this many people working at five. And you can see here, yeah, the schedule looks great. We're going to be done by five. How often does that happen? So the other tool that we have up here is the contract, two predictors in anticipating what your schedule's going to look like tomorrow. So you really have two paths forward here. You have your schedule, you have your contract, right? Which one are you going to choose? Are you going to default to the contract because that's what you're obligated to do. They're not even close to the same. So are you going to average them? Are you going to take the fudge factor approach? And this is a very common scenario.
(04:54)
You're overstaffed during the prime hours, but you're understaffed in the evening. So you're doing premium pay, but you see this under utilization from seven to seven. It's just a very common thing that we see. So what's an alternative? Well, everyone's showing up their prediction for rooms running analysis, but there's something unique about this approach here that I think is really kind of innovative. And if you guys have used ChatGPT or a large language model, it's not exactly like that, but the tool functionally works the same. You're integrating not just the schedule, which is the purple line to get the green line. You're also integrating historical information and you're actually integrating your last prediction and how right or wrong that prediction was and altering that. So you can see here there's clear divergence in the afternoon. So how do we modify our plan to actually cover this?
(05:49)
Here's the discrepancy here. You can see there's obvious differences in that divergence. Here's the discrepancy here between the schedule. We have to reconcile this red area and that's real money every single day. And which path that we choose, you're going to be beholden to the consequences of whatever that choice is. So what do we do? Well, after we've made our prediction, this is the other new component about this, is we can actually take that line and construct a real staffing plan that covers it in alignment with the shifts and the shift preferences that we have within our organization. So you can see here, you see this box and kind of more straightforward here, we have an alignment of, okay, we need this many eights, we need this many 10s, we need this many 12s. And then we also have what we call a relief shift.
(06:36)
And this could be a person that comes in that maybe gives lunch or break relief starting at 11:00, they work till 7, another eight hour shift or a swing shift, someone that comes in late talking through one of the strategies there for potentially having staggered starts. These are all configurable things. So if you wanted to make the person that came in at 11 be totally clinical and not reduce your room coverage at that point or reduce the amount of rooms that you needed to cover at that point, you can absolutely do this. And it's a very flexible model. You could change what the preference is on the day of week, for instance. So it's entirely up to you and you can change them in real time. So I think this is pretty innovative and essentially it adapts day to day and we have to ask ourselves the questions, well, how did we actually do?
(07:27)
Well, we can evaluate each of our models against what actually happened on the day. So what I'm showing you here is the blue line is what actually happened. I know there's a bunch of lines to keep track of, but here's our schedule, here's what actually happened on the day diverges in the same way that we would've expected. Here's what actually happened and here's our contract and here's what happened and here's what we predicted to happen. I don't know about you, but they look a lot closer. So what's the improvement? Conveniently, that 30% error rate pretty much in line with the add-on and this is just one random example that we plucked and our contract actually performed even worse. And if you can imagine a model that does this and learns every day incorporates things like vacation, weather patterns. If you had a weather event like Nashville had and seeing an extreme dip in the number of cases that you have scheduled, your model would adapt to that change.
(08:24)
So the important part of this is that dynamism and not necessarily getting stuck in doing the same thing every day and hoping for a different result. So like I said, this is a pretty important development for us in giving our clients a tool that can really adapt to the day-to-day effects of these changes. So this is really the flow. You see a 2% or a twofold improvement in your error prediction rate, this translates to real money and this is a very simple schematic of how it works. You start with your coverage estimation and then you can devolve that coverage into specific shifts and you can do this day to day looking as far in advance as really you want. Generally, we like to look about a week in advance or so and it's real dollars here.
Jeff McLaren founded Medaxion in 2008 to maximize information technology opportunities in the anesthesia market. Previously, he served as co-founder and CEO of Safer Sleep, LLC, a provider of anesthesia safety and record automation services in New Zealand and the UK. Jeff began his healthcare technology career as co-founder, President, and Chief Product Officer of HealthStream, Inc.