Optimizing the Scheduling of Oncology Clinics Webinar transcript

SANDY PUETT: Hello everyone. Welcome to today’s session,. Optimizing the Schedule of Oncology Clinics. My name is Sandy Puett. I’m the marketing director for LeanTaas iQueue. Before we begin, I have a couple of logistical points. You’ll notice that you’re currently on mute. If you have any questions during the presentation, please enter them and queue the Q&A box which can be found on the lower right hand corner of your screen. We will have time to answer your questions at the end of the presentation. 

So if you have a question, please enter into the Q&A box. In addition, we’ll send out a link to the webinar recording shortly after we conclude. You can also contact us by texting your email address to 630-884-5493 or sending an email for a demo to demo@leantaas.com. With that out of the way, I’d like to introduce Mohan. Giridharadas who is the founder and CEO of LeanTaas. Let’s go ahead and get started. 

MOHAN GIRIDHARADAS:. Thanks for joining. For today’s session, what I want to do is quickly talk about the challenges that most oncology clinics face in getting patient’s schedule, talk about why this happens. There’s an underlying set of issues in the mathematics foundation for scheduling that tends to cause these problems. And three, to outline the pieces of a solution that would help address this. I’ll go through all of this fairly quickly, probably 20 or 25 minutes, and I’ll make sure we leave enough time for questions that you could type into the Q&A box, and we will field them at the end. OK? 

So to summarize the challenges, in our view there are three unique challenges in cancer centers. The first is that every patient has a unique journey through the cancer center on the day of their appointments. They come into the lab. They go see the oncologist. They go get infusion. Sometimes they don’t need to see the oncologist, they just get their labs done and go on to infusion. Sometimes they’ve got another path and so on. So each path tends to be unique. That’s kind of problem number one. Problem number two is that if you think about what a patient goes through, they wait a little bit and they get their lab work done. They wait some more and they see the oncologist. They wait some more and they get the infusion done. 

If you look at the end-to-end time that they spend in the cancer center, they spend three or four times as much time waiting than actually having someone work with them either in the lab or in the infusion center or the oncologist. So the value added time to waiting time is a one to three or one to four kind of a ratio. And the third problem that they have is this concept of a rush hour effects. Labs peak early in the morning. Infusion peaks in the middle of the day. We all know what happens with rush hour when we drive to and from work that a 20 minute drive becomes a 60 minute drive, et cetera. So whenever there’s a rush hour effect, the wait time climbs in a nonlinear manner. It becomes much, much worse not just 10% worse or 20% worse. 

And the problem with interconnected appointments is there’s a domino effect created, because the labs back up, the infusion can’t get started, or the clinicians start running behind. And when they run behind, it has a domino effect on infusion. So this ripple effect continues throughout the day. So this tends to be the set of three problems that are almost universally true in most cancer center. OK, moving forward. So let me take a minute and describe why this happens. And it’s not just a matter of how each clinic chooses to run. The entire foundation of the map underneath scheduling here is broken. And let me give you for specific reasons why it’s broken or how it’s broken. OK? 

First is the following. Any health system you go to has a concept of a resource. And the resource could be a provider. It could be a room. It could be an infusion chair. It could be an imaging machine. And they tend to get scheduled in this grid sort of a format where if someone has an eight to nine appointment, somebody colors in this eight to nine block and either writes the patient name or the MRI number or something. And as the assets get scheduled, these get filled in. This looks logical. It looks like a sensible way of doing it. It looks like it uses the resources, available resources very well. The reason it’s fundamentally broken is this is a tennis court model of reservations. 

If you call your local racquet club and tried to schedule a tennis court, and they gave you the nine to 10 slot, this is what they would do. They will color in court number 10 from nine to 10 and give it to you. The reason it works for tennis courts is the start and the end time is exactly predictable. When you get on the court, you’ll kick out somebody whose is already there. And when the top of the hour arrives, you will get kicked off the court. Unfortunately medical appointments don’t work that way. Patients never show up on time, and the appointment often runs longer than expected. So as a result what has happened is a framework that’s used to schedule predictable events has been used to schedule unpredictable events. And therefore the mathematics on it is fundamentally broken. That’s kind of issue one. 

Issue number two is the whole concept of probability theory. Anytime there are complicated events to be scheduled, like airline seats on every flight, et cetera, you have to use the mathematics of probability theory. Every airline on every flight, every route, every day of the week, uses probability theory and predicts how many overbooks they should do because they know the pattern of no shows and cancellations and last minute add-ons and standbys, et cetera. And 99% of the time they get it right. Sometimes when they get it wrong, they’re forced to do incentives, but they use probability theory. 

Nothing in the EHR lets you use probability theory. You can’t overbook a resource. You can’t put a four-hour appointment in a two-hour slot to play the odds that something else will finish earlier. What this means is you get set up for a heads I lose, tails they win sort of a game meaning if things run longer than planned, your nurses and chairs and rooms are occupied. If things run shorter than planned, you don’t have standbys waiting to fill the slot unless you’ve got people in the waiting room. So it’s a bit of a one sided bet that works against you. The third thing that happens is every health system creates an itinerary for patients. You’ll go to the labs at 7:30, you’ll see the oncologist at 8:15, your infusion will start at 9:30, and then you’ll have some follow on work at 10:00 or 11:00. 

All of that looks great on paper. It’s the concept of a connecting flight. You’ll fly from. Atlanta to Chicago, Chicago to Portland,. Portland or Seattle or whatever the route might be. The reason that works for airlines is much as we all complain, the on time performance of the first leg of the flight is in the 80% to 90% range when you talk about plus or minus 15 minutes. That doesn’t work for the first leg of medical appointments. The on time performance is not quite the same. So imagine how your connecting journeys would work if there was a 50% to 60% chance that your first leg would run late, and another 50% to 60% chance that your second flight would run late. It’s impossible to make a connecting journey of three or four flights with that level of instability in terms of the on time performance. 

So the whole concept of creating connected appointments without using probability sets you up to have the itinerary being wrong the moment it comes off the printer. And the fourth reason, the mathematics doesn’t work is most medical appointments are made on a first come, first serve basis. If I were to call ahead and try and make an appointment for June 15, 2018, the scheduler is likely to tell me the calendar is open. Pick a spot, any spot, we are open from 8:00 to 5:00. Unfortunately that doesn’t lend itself to an optimized allocation. They should have understood the duration of my appointment, the nature of my appointment, and slotted me in correctly. 

And so because of these four reasons, the scheduling results in this paradox of rooms being empty for parts of the day while the waiting room is overflowing for other parts of the day. So this whole mismatch is what tends to happen. The reason that it gets progressively worse is that just like rush hour traffic, the wait time when you start to approach rush hour doesn’t go up by small linear amounts. It goes up exponentially. The thing that makes it particularly challenging in healthcare is that every appointment is unique. Every patient, every provider, every interaction is unique. And therefore the variability tends to be high. Unlike widgets on an assembly line where you know this step always takes one minute and it’ll be between 55 seconds and 65 seconds. That’s how long it takes. There’s almost zero variability. 

When you’ve got zero variability you can push up to the edge of the cliff. When you’ve got high variability, the stuff starts to get much, much worse earlier and earlier. So what happens is creating packed schedules for providers, for chairs, for exam rooms, et cetera, makes this problem particularly worse. And understaffing doesn’t help either. And so as a result of all of these things, it’s the combination of them that creates the scenario like a perfect storm where the wait time gets much, much worse. OK? So that kind of sets the context for it. Let me now describe how one can think about optimizing appointments with the providers. So when you think about what a clinic is trying to do, there are three distinct sets of objectives. And they often seem in conflict with each other. 

If you looked at it from a patient’s perspective, all they’re really asking for is please get me an appointment when I need it. And once you give me an appointment slot, please do see me at approximately the time you promised as opposed to giving a 10 o’clock slot and then letting me sit in the waiting room until 11:15. That’s what the patients want. What the providers want is I’d like to treat as many patients as possible, whether they’re new or returning, just want to have the most impact I can on the community. And two, I want to minimize the time I waste during the day, either waiting on patients or the rooms or staff. 

From a staff perspective, they want to use the treatment rooms and the examination rooms as efficiently as possible. And they don’t like the chaos that happens in the middle of the session where rooms are full, patients are in the waiting room, providers back up, et cetera. This is kind of the three sets of objectives that are going on. And it feels like it’s difficult to satisfy all of them. And so providers often take a bit of a rack and stack approach where they’ll fill up the exam rooms, and they’ll get to the patient when they get to the patient. That works great for the provider. It doesn’t work so well for the patients. So this is the nature of the conflict that is automatically set up. So let me try and parse that down a bit more. 

Many health systems try and take a cookie cutter approach. New appointments will be 90 minutes, returning appointments will be 60 minutes, and jam that template down every single provider. Providers tend to not like this, because each one practices differently. Their patient mix is different. The kind of disease they treat are different. And so having a cookie cutter template while it looks like it works great for administration doesn’t work so well for the providers. So let me just now for a moment assume that you’ve got five types of, you’ve got new patients and returning patients. Let’s assume that there are two types of new patients. A particularly complex case where the provider would expect to spend 40 minutes with the patient one on one in an exam room. And the overall duration would be about 100 minutes from the time the patient walked into the lab. Imagine a more routine new patient where the time spent by the provider is slightly less and so correspondingly 30 and 90. For return, assume we’ve got three types. 

A complex return, where we are still actively treating and it’s a 30 to 70, or more medium one 20 to 60, and a routine follow up, which says the results are fine, labs are fine, everything looks good, see you again in three months, so something like that. So you can imagine the provider only needing to spend 10 minutes. OK? So with these five types of treatments, let’s try and plan what a provider session should look like. So let’s assume the provider has a half-hour day session four hours, and assume you need to give them about minutes free per hour just so they can catch up on nodes, make a phone call, et cetera. So over the four-hour session, they’ve got 200 minutes of active provider time. Now, these are five types of appointments I laid out. And at the bottom is the amount of time the provider needs to spend with the patient. 

So pretend for a moment, we stack their template with only new complex patients. How many patients could they see? Obviously they could only see five if that’s all they were seeing. If they were seeing, if you stack their template with only this type of patient, they could see six. And so on all the way to where if you stack their deck with only routine simple cases, they could see as many as 20. Obviously the actual mix would be some combination. So whichever way you cut it, you would imagine that a provider in this half a day session is going to see somewhere between 5 and 15 or 20 people whatever that number turns out to be. Let me just now highlight how complicated this gets and how quickly it gets complicated. 

Even with this simple combination, there are 277 possible arrangements where in an arrangement I mean five of these are nothing else, it’s one option or one 40 minute, six 30 minutes, and two 20 minutes, all adds up to 200 minutes. That’s another way to add up to 200. If you now said, OK, how many ways can I arrange this? It turns out there are 1 million distinct ways of arranging the possible templates, of arranging the possible sequence of appointments to satisfy this requirement. So obviously it’s a daunting number. There is no chance that a scheduler scheduling them on a first come, first serve basis with some sort of a good arrangement or a nurse or a clinic manager trying to shuffle who the provider sees first is going to explore all million possibilities to come up with the best one way of doing this. 

Now imagine the complexity if you’re simultaneously trying to optimize for eight or 10 providers sharing 15 or 20 common rooms. So this is why the problem very rapidly becomes mathematically intractable. OK? So let’s talk about what the solution needs to do. First, the appointment models have to get sorted out. Meaning every provider has got a different rhythm for how they treat patients. Second, we’ve got to set up templates which are optimized for each provider based on the specific practice they have. The number of patients, the mix, et cetera. Third is how you allocate rooms. The rule of thumb let’s give provider X three rooms, and provide a Y two rooms is actually incorrect and I’ll show you why. And then being able to report and calibrate on a continuous basis daily, weekly, monthly, et cetera, based on operating metrics like wait time, average length, the mix, and rebalancing as needed. 

What we’re trying to achieve out of all of this is the thing that you are prioritizing more patients seen with less wait time, a smoother flow and in general a higher level of staff satisfaction. OK? So how do you do this? Let’s talk about how you should think about the choreography of patients going through a clinic. If you broke an appointment down, there are three distinct phases. There’s a set up phase, but what I mean by setup is rooming the patient, doing the vitals, talking about their meds, their current details, et cetera. Then there’s the active physician touchdown. One on one provider and patient in the room talking through it. And then a close up, maybe next steps, may be medication instructions, maybe follow up appointments, any update instructions. OK? 

A similar model for return. What’s varying here is the touch time of the provider. Remember if it was– it could be 30 or 40 minutes. It could be 10, 20, or 30 minutes in the return case. Now let me describe how one thinks about optimization. The only thing here that’s not substitutable is the provider, because the provider can only be in one place at one time. The setup, two nurses could simultaneously do setups on two patients. Nurse one could escort patient one to room one, at the same time as nurse two is escorting patient to to room two. Similarly with the close up. So the bottleneck and nonsubsituitable resource is the physician. So the simplest way to think about this is to think of the provider as the soloist in an orchestra. So what does a soloist do? They don’t hang around waiting for the conductor to organize the music or the musicians to get their instruments tuned or the patrons to sit-in their seats. They wait until all of that is done. They come on to stage. They do their magic in 10 minutes on the stage, and then they leave the stage. 

Imagine when they left the stage, if they went to the auditorium next door which has been busy getting ready for them, and the moment they get on the next stage that auditorium is ready as well. And they can do their magic again for 10 minutes and move on to the next stage. That’s kind of what we are trying to create for the oncologist, which is show up, do what you need to do, and by the time you move to the adjacent treatment room that patient is prepped and ready for you. And try and run that throughout the day. So what you’re trying to do is mathematically balance the patient flow such that the provider can serve wave top to wave top, always encountering a fully prepped patient in a room that’s ready for them. Obviously this will never work out exactly as planned things will run ahead and behind schedule and so on, but setting the framework up as close as you can do to perfection like this gives you a much better shot. 

So what does this look like in real life. For a given provider, what you have to do is analyze the historical pattern and figure out how many news, how many returns, how many short news, how many long news, et cetera. And then sort out the right sequencing for it. Why does the sequencing matter. Remember it’s one out of the million possible sequences. The reason it matters is you’re trying to provide adequate support to do the set up portion of it– the first leg of this three part thing. The second leg of it is the oncologist. Therefore the mathematical constraint is they cannot be in more than one place at one time. So no dark green can overlap with another dark green. Now a light blue can overlap with another light blue with a dark green cannot, because the provider is only in one place at one time. 

You are also trying to set it up, so that there are no gaps in the provider schedule, because that would be not an efficient use of the most scarce resource you’ve got. And the third leg is the follow up that can also overlap. What happens when you sequence this right is you get a sense for how many simultaneous rooms you need. So notice this provider busy as they are, at points in time, they need three treatment rooms to be active. But at points in time, they only need one treatment room to be active. And so it’s this ebb in flow of rooms that accidentally creates a lot of bottlenecks. Because looking at this, one could automatically say assume I have three providers practicing in a small clinic, and this is their pattern throughout the day. 

One could easily imagine a room allocation model that says, let’s give X three rooms, let’s give Y three room, and let’s give Z two rooms. And that’s the way we’ll use our eight treatment rooms. But if you look at it mathematically, a better model might be give. X two rooms, give Y two rooms, give Z only one room, and keep three rooms at float rooms which can be used by any of the providers. And they are occupied, but they can be used by any of them. So depending on how the layout is, these could be the three central rooms that are equally convenient for all three providers, et cetera. Why does this work? It works because it unlocks hidden capacity. If you think about Atlanta. Hartsfield airport, it has five runways. 

One could easily have a runway allocation model that says let’s give delta two runways because they’re the biggest and busiest airline in Atlanta, let’s give American one, let’s give United one, and let’s all the other airlines share the last runway. This looks logical and if you spent the day in the control tower looking at the runways, they’d all look fairly busy. At every point in time, there’d be a delta plane getting on the runway, getting off the runway, et cetera. So it would feel like a good allocation. If you then change the model and set all five runways are in a pool, and as each plane comes in we’ll assign a runway to it. The capacity of the airport will go up by 25% to 30%. So reserving slices of capacity like rooms or runways kills the overall volume and throughput in a way that’s quite invisible and hard to grasp, but that’s how the mathematics unfolds. OK? 

So now what do you do? What you need to do is then on a daily basis actively manage it. Where you can look at the cancer program, look at the provider, look at for any set of providers, how the rooms are going to unfold. So you can anticipate periods of time when you’re going to be really busy and all the rooms are going to be full and then start to say here’s what a provider may have some gaps here’s where there’s a spike in the number of rooms needed. And therefore be on top of it, and more importantly, looking forward, not just today, but tomorrow, the next week, et cetera, so you can manage it. Right? Once that happens and you’ve created this template for each provider, the trick becomes to monitor the compliance against it. 

So looking at the provider historically you can say what I’m trying to do for this provider is set up a typical session where they see 11 patients, two of whom are like this, three of whom are like this and so on. And over a period of days, weeks, months monitoring how close you are to it where red indicates you overbooked that particular type of appointment, green implies you were right on target, and yellow implies you’re under booked, you are under utilizing that slot. And as time unfolds, you can start to see where you’ve got chronic patterns of overbooking and under booking certain types of appointments which then helps you tweak the template for each provider better. And after getting each provider’s template correct, you’ve got to be able to figure out the set of providers who practice simultaneously so that you get the rooms correct. OK? 

That’s a walk through of how we do the compliance on this. So what does it take to pull all of this off together in one coordinated manner? First is prediction algorithms have to get much, much better. You can’t just look at volume and mix simply on an Excel spreadsheet and say, yup, this provider sees eight patients, two of whom are new. The volume mix timing has got to be much more precise. Second this optimization has to happen where you are optimizing the sequence of appointments for any given provider and then taking the set of providers, optimizing the relative sequence across them so that you can start to steer new providers rather than just saying which days would you like to practice, being able to say ideally I’d like you to practice, the new provider, like you practice on. Wednesday afternoons and Thursday afternoons, those are the two best days because your template will complement my existing providers on those days. 

Third is to simulate. What happens is no matter how well you run the math, life never works that way. Each patient will run late. The provider will run late. The parking lot will get full. Stuff will happen. And so what simulation does is it says, there may be hundreds reasons why things go wrong, but finally it manifests in only one of two ways. Patient punctuality, meaning the patient was either early, on time, or late regardless of whether it was their fault or not. And duration, it either ran shorter than expected, equal to expected, or longer than expected. That’s it. Those are the only two variables that manifest. And so what simulation algorithms do is they say let me test a million different permutations and combinations of templates to find the one that’s the most resilient to these shocks of punctuality and duration. And when you find the template that is most resilient to shocks, it is better for you. 

Because then when stuff goes wrong as it inevitably works, your template is a little more resilient and bounces back. The way to think about this is rather than doing process improvement to fix all the potholes on the road, you’re designing a better shock absorber for the car. So you can deal with the potholes because there will always be potholes. Having done that, it has to coexist with the EHR so that your schedulers are not getting into a different screen. They’re continuing to navigate and schedule patients exactly as they are. And the final aspect is it has to learn no matter what you do templates will need to be tweaked and learn, and you have to learn from them on a regular basis because the volumes will change, the mix will change, the provider practices will change, the set of providers practicing on a given day will change, the number of examination rooms will change, the support staff will change. 

All of that continuously changes. And this is the learning loop that you’ve got to do if you’re going to execute a mathematics and lean based optimization of our oncology clinic runs. OK? So what do you have to do? On the provider templates, going through those five steps I just talked about. Start out by estimating the volume and mix accurately. Figure out the right sequence of news and returns, simple and complex, work out the various permutations and combinations, put them into the EHR, and continuously monitor expected versus actual. On room allocation, start out with the right number to each provider which may typically be smaller than you’re currently allocating. Figure out the right mix of fixed and float rooms. If every room is floating, it’s too chaotic. If every room is fixed, it’s wasteful. And so the right combination of fixed and floating is key. 

Figure out the various permutations of this that shows you the points in time when you run out of rooms. And have root cause problem solving as to why we run out of rooms on certain, during certain hours and certain days of the week. And then have the room allocation guidelines and managing it correctly. And then continuously measure room utilization in and out of the rooms. Often your EHR captures room-in and room-out. Some people have RTLS like systems to capture it. But even if you didn’t have precise location data and room data, you can make progress on this. And finally on the staff scheduling. Once you do the three part template, you can figure out how many MAs and nurses you need to keep each provider on track. 

Figure out the start times and shifts or shift lengths for the support staff. And then again if you have to keep doing the various permutations and combinations to get there. And then each health system has got its own cadence for how much in advance shifts need to be published, when you can call for overtime. Do you have a floor pool? Don’t have a floor pool– all of those things matter. And then finally compare actual versus planned to get there. So that’s the way all of this works out. Just so you know we are launching two alpha deployments of this at two of the leading cancer centers in the country, at Memorial Sloan Kettering and at MD Anderson. We all confirmed and ready to go. All the prework is taking place right now. 

We’ll start in January, and in each center, we have picked a practice of 20 oncologists to work with. So we’ll work with them through all of the puts and takes around their templates, the room, the staff, and so on. And within a period of about four months, we put in five, just to have a safe up side. We think it’s closer to four. We’ll have a very good point of view on exactly how this thing should drive. OK? Let me pause for questions as they’re coming in. 

SANDY PUETT: OK. We have one question so far, which is asking about how we’re going to share the slides and the webinar recording. These will be emailed out to you as a link, and you can access them at your leisure. You’re welcome to forward them to anyone in your organization who you think will be interested. 

MOHAN GIRIDHARADAS:. Or email us to what you can see on the screen right now or text us with your email address for two things. One, we’ll send you the link to the webinar. And two, if you want to participate in a bit of a deeper discovery session, specifically your clinic, where the wait time is your issue or provide a template to your issue, we’re happy to engage and think aloud with you on that topic. 

SANDY PUETT: We do have one other question, which is around how do clinics get physician and scheduler buy-in on this process? 

MOHAN GIRIDHARADAS: This is a very interesting one. A schedule of buy-in is pretty easy, because all you’ve got to do is help them use the existing questions to steer a new versus return, simple versus complex. And they use their existing EHR and their existing scheduling methodology. So it’s not a big shift for scheduler. It’s just emphasizing the importance of putting a square peg in a square hole and a around peg in a around hole. So scheduler buying is pretty easy. Provider buying is a little more challenging. 

What we find helps us is that paradoxically providers resist the cookie cutter model. And the next best alternative that most of them are encountering is a cookie cutter cramdown that says, this is how you will schedule. And so they actually welcome the flexibility of saying I treat patients slightly differently. I like to spend more time. I like to, I like gaps between my patients, so I can catch up on notes or I don’t like gaps, because I write all my notes at the end of the day. So all of those can be created as mathematical constraints. 

And so providers actually like that flexibility. Where you run into some resistance is where providers know that they’re bucking the system but choose to do so. We’ve had one case where the provider would deliberately know they could only see 12 or 13 patients in a day, but they would create 20 or 25 appointments because they just felt that they needed to see as many patients as possible. And the patients would wait three and four hours, but because they were brilliant and absolutely one of the leading experts in that particular form of cancer in the country or possibly the world, patients did not complain as much. So sometimes providers will deliberately go outside the guardrails, and you have to just kind of manage your way through those. 

SANDY PUETT: One other question that’s come up which is, can you explain why we can schedule overlap on MAs? I understand the physician, but was not clear on double booking the. MAs, the assistants. 

MOHAN GIRIDHARADAS: Yeah. The way you can double book on the MAs is the following. We typically use the set up and leave a larger window for it. So the actual MA task for a set up maybe three minutes, let’s say, or five minutes. And we leave a 15 minute window to get it done. So when you overlap, all you’re saying is sometime between 9:00 and 9:15, please get this patient set up, and sometime between 9:10 and 9:25, get this other patient set up. So it looks like there’s an overlap between 9:10 and 9:15, but because this setup envelope is bigger than the actual task time for the MA, it allows for the overlapping. And what that does is it creates a right staging and readiness for the provider so that just like the soloists, they can come onto stage and perform their magic. 

SANDY PUETT: All right. I think that’s all of our questions. All right, everyone. Our time is up for today. Huge thank you, Mohan, for leading us through. Of course, thank you all for participating. Keep an eye on your inbox for a link to the recording of the session, the slides, as well as for announcements about future webinars. Please complete the survey that will be presented to you as you leave the session. It should only take you about 15 seconds. You can text your email address to us at 630-884-5493 or you can send us an email for a demo request at demo@leantaas.com. Thanks again.