Optimizing Infusion Scheduling – Diagnosing the current state of your infusion center webinar transcript
PRESENTER: Thanks, everybody, for joining. Let me jump directly into it. This is the fourth of a series of four webinars that we go through the various stages of how you optimize infusion scheduling. Prior webinars focused on the earlier topics, so the plan here is to quickly go through those previous topics. It will be a refresher for those of you who attended prior webinars. If you didn’t attend prior webinars this would be a quick executive summary of it, and then you can download the full webinar to get the greater detail. So today we’re going to focus on diagnosing, the bulk of the time we’ll be on “diagnosing the current state of your infusion center.”
Let’s jump into it. So who’s LeanTaaS and what do we do? We are a Silicon Valley software company, we’re up to about 100 people now, mostly focused on people with backgrounds in mathematics to do schedule optimization, software engineers to put it into a product, product managers, and process excellence, and operations experts. We’ve raised a whole bunch of money to build the algorithms, the product, the platform, and the team over the last seven or eight years. And we are focusing a narrow question? How do you unlock the capacity of scarce assets, specifically infusion chairs in this case, but there are other scarce assets we’re working on.
A lot of magic happens when you unlock capacity, access improves, shorter lead times for patients to get appointments, the wait time reduces, the operating costs go down because of the same staff and hours of operation. You can see more patients. Capital costs go down because you can get more out of your infusion centers before needing to add chairs or add satellite locations. And to the extend you’re seeing more treatments, it helps the revenue top line as well.
The two products that are in marketing infusion and operating rooms, we are working at the moment with specific health systems on building out the suite of iQueue products. So the clinic’s product is being developed in two pilots in New York and Houston. The imaging product is being built out on CT clusters at the moment, we’ll expand to MRI later. Inpatient beds and labs are also under development.
We work with many of the leading health systems across the US. They range from academic medical centers like Stanford and Penn and Emory, Duke NewYork-Pres,. UPMC, et cetera, to iconic single service providers like MD Anderson, Sloan Kettering, and [? Carano ?] stepped [? their ?] focus entirely on cancer, to standalone hospitals like. Boca Raton and standalone cancer centers like [? Starten ?] an MCC. And the infusion product has now been deployed in 116 infusion centers, nearly 3,000 chairs, continuing to grow at that pace month over month. So that’s who we are and what we do.
The one background question. I want to spend a minute on is, why can’t the EHR do this? Whether it’s Epic or. Cerner or Allscripts, it really doesn’t matter. Why is it that the. EHR can’t do it, and it can’t for three reasons, one is the entire mathematical foundation of EHRs is not enough. Appointments are treated as reservations on a specific resource, and so every EHR encourages health systems to schedule in this grid like format where you’ve got chairs across the top and times of day down the left, and an eight to nine appointment for John Smith is marked as eight to nine.
This works only for deterministic events, which is events where the start and stop time is known at the time of making the appointment, like a tennis court appointment or a spa appointment. Infusion appointments are stochastic or random, meaning the start and end time is never nailed down at the time of making an appointment. You cannot use a deterministic framework to schedule a stochastic event. The math does not work. This is why the grid looks great the previous night, and then on the day off it never works out that way. So that’s one reason, the. EHRs aren’t robust enough.
The second is everyone’s schedules on a first come, first schedule basis, and first come first schedule is wrong, because it puts the appointments in a random order on the day off. The appointments have to be optimized, meaning you have to level load by balancing arrivals and departures with people who are in the middle of their infusion treatment. Just having a first come, first serve with no understanding of how the durations fit into each other leads to an incomplete and a suboptimal schedule, which is why the staff scrambles to move patients in and out of chairs, and the waiting room starts to pile up.
And the third reason, the mathematics are not robust at all is EHR’s don’t help you optimize connecting appointments. Infusion is always the second or third step of a series of appointments of the lab and the clinic and then the infusion, and trying to link them by hand, by inspection, by saying, John Doe has got an 8 o’clock clinic appointment, and therefore should get a 9:15 infusion appointment, just does not work. The system as a whole has to be optimized. As a result of this, infusion scheduling just ends up not working out. You get the peaky profile, you get the long wait times in the middle of the day, et cetera.
Just to give you a sense of how crazy the map gets even for a modest situation with five infusion types and 70 treatments done in their day, which would be a small to medium infusion center. There are 256 possible slots, which means, assume you could seat patients at 7:00, 7:10, 7:20 all the way up to 5:30, that’s 64 start times. And if you could see 12 patients at a time, four nurses walking four patients to four chairs, that’s 256 possible slots. The number of ways you could set up the schedule is a number with 105 zeros behind it. So this is not going to happen by inspection or someone looking at an EHR scheduling screen or an Excel spreadsheet.
In fact, just from a pure probability standpoint, if you took the two biggest lotteries in the country, you’d have to win both of them seven times in the row to match getting an infusion schedule right for one single day. So this is why the mathematics is overwhelming. And on top of that, you have to take into account the fact that the volume and mix on every day of the week is different, that nurse availability drives whether a patient can get started or not. Chair availability often constrains whether a patient can be seated or not, and the variability is both expected and unexpected.
Lots of things happen that you know will happen, and lots of things happen that you could not have anticipated. So all of this needs to be factored in. And so as a result of this, the way the day unfolds is like a losing game of Tetris, where you think of each block as a duration of an appointment. Patients come in roughly in the order of that treatment. The charge nurse puts them in the right chair with the right nurse at the right time as best as he or she can. And you can see the game of Tetris unfolding in a losing manner. And this is a problem for two reasons, one is the peak only lasts three hours. The nursing shift is eight hours, therefore, by definition you’ve got a bad choice to make. Either you staff for the peak, in which case you’re overstaffed for the mornings and afternoons of the day or you staff for the average, in which case, right when you need it most, you are understaffed.
But the bigger problem is if ever a peak approaches the capacity, a system becomes mathematically unstable. What this means is every metric will go into the tank. The most easy to grasp live example is the freeways at rush hour. They’re at capacity, you cannot squeeze another car on the roads, and therefore every metric is in the tank. A 20 minute drive takes an hour. A fender bender that should delay 20 people delays 10,000 people. A fender bender that should take 10 minutes to clear, takes two hours to clear. So every metric becomes 5x, 10x, 100x, 1000x worse. Infusion is a series of fender benders waiting to happen. The clinic will run late, the lab will run late, the pharmacy will backup, a patient will call in late, nurse will call in sick.
If any of those happen early morning or late afternoon, it’s not a problem. If any of them happen in the middle of the day, it’s like a big rig crash on the middle lane of a freeway at 5:00 PM on a Monday, it’s going to be a mess for many hours. So that’s kind of what happens. Now, because of this all infusion centers virtually face these three problems. Patients wait in the middle of the day, the chairs profile starts out empty, hit’s a peak and comes down, and nurses end up missing their lunch break a lot, because this peak is right in the middle of the day. So that’s the summary of the challenge.
Now, how do you optimize this? Really quickly, so we can get down to the planning and the diagnostics. The way you optimize it, is you have to take the [? spiky ?] profile and make it a flat profile. Notice when you flatten the profile, even though there are chairs available in the middle of the day, because you’re filled up, others strings in the middle, you can still see as many or more patients. To get this right, there are five kind of hard math problems that need to be solved. First, the prediction of the volume of appointments for. Mondays, for Tuesdays, for Wednesdays, each day of the week you have to predict the census with a very high degree of accuracy.
Second the census alone is not enough. You have to predict the mix. How many one-hour treatments? How many two-hour treatments on a Monday, on a Tuesday. Third, you have to factor in the ability of each center to predict the duration accurately. So when a particular center says this is going to be a three-hour appointment, how accurate are they? Is it really tightly centered around 2 hours and 45 minutes to 3 hours and 10 minutes? Or is it spread out between one and eight days– oh, one and eight hours. So once you know the duration of accuracy, you can then incorporate it, which means you are adjusting the width of the Tetris blocks up and down based on the error around it.
And finally, you are then playing super Tetris to place those Tetris blocks in a way that with this adjustment factored in, they still pack it very tightly. When you do that, magic happens for four reasons. The biggest one is, it unlocks capacity right when you need it most in the middle of the day, imagine you had several free chairs. You could accommodate patients who are running late, clinics who are running late, patients who need more time, patients would react to the medications, whatever the issue is, you’ve got chairs. It’s a bit like, if ever there were a fender bender, magically three more lanes opened up on the freeway, fender benders would not be an issue as much. The second is patients had choice. There are many slots of each duration throughout the day.
So someone who has a three-hour appointment following a clinic appointment, can find a convenient spot. Third it flattens the workload for nurses. Their workload remains consistent throughout the day. And fourth, it fits the nursing schedule. Nurses who come early, leave early, nurses who come late, leave late. In order to pull this off what needs to happen is the templates cannot just be first come first serve, or be it up by inspection, let’s do three appointments at 8:00 o’clock, five appointments at 8:30, or let’s schedule to a chair or two a nurse or to a pod. None of those methods work. What needs to happen is, a duration based schedule, where across the top you cluster the appointments into duration buckets, and you have to figure out the right width of the bucket.
In this case, we’ve set the five out of one-hour bucket, a two-hour bucket, three to five, six to eight, and nine plus. Having clustered them into buckets, you’ve figured out start intervals. The tighter the start interval, the better. So 10 minutes is really good, you don’t need to go down to five but going to 30 of one hour is simply not enough. So you get 10 minutes start intervals. You then figure out what your capacity to do simultaneous starts. That is reflected in looking across a given row. So this center between 8:00 and a 8:30, can take on three or four patients at a time. Once it becomes 9:00 o’clock, it’s too busy. They never wanted to see more than two coming in at a time. So that smoothens the flow. Fourth, each duration bucket, you have to figure out that error band that I was talking about earlier. So you get things into the right bucket.
And this is the most important part, for every start time and every duration bucket, you have to calculate how many starts you want to accommodate at that time. So the way to read this is at 7:00, at 8:00 o’clock in the morning, we want to be able to start two appointments of three to five hour durations, and two appointments of six to eight hour durations. This is the magic grid for your templates. And your scheduler should try and aim their way towards it, and if they get it even 80% or 85% right, you will magically get a flat profile. So that’s the essence of the optimization. Switching gears to planning. Planning is very important because you need to know in advance, infusion operations are very often a reactive business.
The staff knows the tsunami will hit sometime in the middle of the day, they just don’t know exactly when, exactly how bad, and for exactly how long. That has got to be solved. And so what you’ve got to be able to do with predictive algorithms is forecast how today will unfold at 10 minute, with 10 minute precision. So the way to read this shot is, the green line was what was being aimed for, a smooth ramp up, somewhat flat and a smooth ramp down. It was geared up to deal with 82 patients, and this looks like it get off to a slower start, yellow means you’re below the curve, gray means you’re on the curve. And so for a long period of time to be running right at the optimal frontier, but this will be a nice smooth day. So they can anticipate It. Well that’s for today, that’s great, but looking forward the next 30 days, you can start to anticipate where you might have problems further down the road. And this lets you think in advance and make certain moves, whether it is to move a nursing schedule as a last resort, moving a patient appointment, or just figuring out how you cope with it.
So for example, here drilling down we can already see, Thursday in this particular case, tomorrow is going to be a bit exciting towards the end of the day, but they’ve got, they’re running right on the edge of the frontier, all through the day, which means they need to make sure that the nurse compliment fully shows up. We don’t have one call out or something that happens in the morning, and that we don’t have unscheduled breaks, and et cetera. So this just gives you a way of managing the process well in advance. OK. The way you get to planning are two very different, but equally difficult problems. The first is matching demand and supply, analyzing the arrival pattern, the volume, the mix, the timing must be matched to the available supply, which is staff, equipment, chairs, et cetera. And the matching has to take place in a very, very tight window, 15 minutes, 30 minutes stops.
You cannot just say, I’ve got 80 patients today, and matching with the ability to treat 80 patients, because it varies wildly if those 80 patients came in each per hour over a 10-hour period or came in 15 per hour for the middle four hours of the day, and the rest of– the remaining 20 were scattered through the rest of the day. So it matters a lot how you match demand and supply. And the second thing is, how you optimize connecting services? So whenever they have connected appointments, labs, clinics, procedures, the spacing has to be mathematically good enough so that it is simultaneously convenient.
You cannot expect patients to wait hours between related appointments, nor is it a fantasy where it’s wishful thinking, where we’re going to make a connection to 20 minutes knowing full well they’re going to miss it. This would be like, if they gave you a 20 minute connect time at O’Hare airport, you’re guaranteed to miss your connecting flight. So it has to be possible to execute consistently on time every day, and yet be convenient, and that’s the mathematical optimization of connected networks. OK so that gets us through all of the preamble in the plan 15 or 20 minutes.
Let’s not double down on what are the types of intelligent diagnostics? For you to answer a core question, how well am I doing? How good is my infusion center? How much headroom is there? How much better can I make it? OK? So in order to do this, you can think of four key questions you need to answer.
Question number one is what is the improvement potential. How high is a high? Am I already running as best as I possibly can or there headroom for me to get much better? So this is an important question to answer. The second one is how all my schedule is doing? All they booking appointments in an optimal manner that makes it possible for their colleagues downstream or actually delivering the infusion to stay on track and have a flat chairs profile. Question number three is, are the financial and operational results as good as can be expected from this. Right? So if you’re doing as well as you can, you might be seeing the results. And fourth is suppose something goes wrong on a specific day, how easy can I drill down into the root cause of what was challenging. Right? Because assembling the facts and having the discussion and being able to pinpoint the root cause within minutes is a powerfully important thing. OK?
So let me take each of these questions one at a time and give you our view on how we think through them. So what’s the improvement potential of my current state? With a very simple data pull, no PHI needed, there are two buckets of data. Operational parameters, how many chairs and beds do you have? What are the hours of operation,. Monday, Tuesday, Wednesday, Thursday, Friday? You can add staffing to it if you want one more level of detail, how many nurses start at 7:00 o’clock, 8:00 o’clock, 8:30, et cetera. But that’s a second order of operational parameters. And then a historical data dump, about six months of data. So you’ve got 15 or 20 episodes of Mondays, Tuesdays, Wednesdays, more than that of each day of the week.
Think of each row as one unique infusion treatment that was delivered. So what is the date? What type of infusion was it? What did you expect it to take how long? What was the scheduled appointment time? And then any other timestamps you’ve got. When did they actually check in? When did the first med start? When did the last med stop? When did they actually check out? So it’s just timestamp data for each infusion, right? That’s what you’re looking for. And what you can do with that is develop a fact based view, which is how much more volume can I absorb within the current hours of operation. That gives you a sense of how good am I, is there room for improvement. And second, how much can the wait time be reduced in the middle of the day.
So not to average the wait time across the whole day, just to take the wait time in the before condition between say, 10:00. AM and 2:00 PM, and say, can I take that down by 10%, 20%. What’s the headroom for improving that? That gives you a sense for how well am I doing compared to what I could do, with the existing infrastructure, existing chairs, existing hours, existing staff. OK? The way we do that is by looking at the six months of historical data, we can help you cluster. So we cluster the appointments into duration buckets, and we use the geometrical clustering algorithm that helps us guide into the right buckets.
So we don’t make these buckets up, we look at where the data is clustered. And so in this particular case, we said, there’s a half-hour bucket. That’s a super short infusion. That’s probably an injection clinic or something where they have patients drop and get their shots or get their blood draws, and go away. There’s a one-hour bucket, a two-hour bucket, three four to five and six plus. As we look through this. This is the typical volume on a Monday, Tuesday, Wednesday, et cetera. What we think we should plan for? This is a historical average. Now notice, we deliberately decide a planning volume that is significantly higher than the historical average.
In this case, you can see it’s 30%, 40% higher. Why is that? Because we want to make sure we can handle the most challenging Monday. We want to account for the fact that there will be add-ons, and that there will be no shows. Even the no shows, you don’t know until the last minute. So they needed a slot on the calendar, and then they would fall off. And you needed add-ons to have a room on the calendar, and they may not match exactly. So it’s usually prudent to build an optimized schedule with a significant lift compared to this. And having done that, what we can then say is let’s look at how you currently scheduled.
So take a Monday, and look at the profile, and because we’ve got the timestamps, we can construct the profile of your typical Mondays, be able to see what the peak was and notice if the peak is above chairs capacity? What that means is all the charts in the infusion suite are full, and some waiting room chairs are also occupied, which is often the case in many, many infusion centers. Look at the volume, and look at, how many patient hours were done. Patient hours is an important metric, because the number of patients– so each patient sits on the chair for a certain amount of time.
When you add up the product of those two, you get the patient hours. So if you had one patient spending eight hours, and one patient spending seven hours. So there was two patients, but the total patient hours was 15. So you have to add those up, and that matters. What you want to then do is use the forecasting to say let’s take on a higher volume of patients, so its from 76 start to take on 90 something, yet get the peak lower and be able to take on more patient hours as well. So this is a way of being able to do more with the same number of chairs. What this can then tell you is, how much opportunity there is? We are happy as was reported on the flyer to do this diagnostic at no charge for the first infusion centers, first 20 infusion centers that can give us the identified data, from which we can build this assessment. So we don’t need patient names, we don’t need doctor names, we just need that time stamp sort of data. So that’s how do you diagnose your current state.
The second question, how are my schedulers booking appointments in an optimal manner? Now it turns out, this is also a very critical thing to do, because if your schedulers do first come first schedule and put patients where they see holes in the chair without regard to what the nurse load is, you end up with a day that’s not manageable. And so, there’s a concept of a compliance score. And what the compliance score does is, having determined what the optimal pattern, the way I showed you the grid answer key for how, that how appointments should be offered up, you can compare the optimal grid versus the actual gird, what the scheduler actually ended up putting on the paper.
Now you’d never expect them to be perfect, because they’ll always be the emergent case, the last minute case, the patient who had logistical problems getting complacent, et cetera, et cetera. So you expect there to be some variation, but what you can do is you can measure how closely they followed it, and penalize, create a penalty score when they overbook. And the penalty score depends on the degree to which the overbooking was harmful, meaning did they overbook many appointments in a short cluster of time.
You know, that’s going to create a problem. Or did they try and match it where the mismatch was they put a three hour appointment into a four hour slot, or a three hour appointment into a two-hour slot, which is a match for a longer or a shorter duration, which is a less harmful thing to have done, meaning they’re just off a little bit on time. So we don’t penalize under bookings, because under bookings just mean that one patients showing up, there’s nothing a scheduler can do to create demand, if you will. So once you get the compliance score, you can report on the compliance score and start to make sure that your compliance is in the 80%, 85% range, which is a safe range to be able to deliver it.
You then can start to say where they’re not matching, what can you learn? So in this particular chart, the grays are where they match the optimal recommendation. The reds are where they overbook, and the gap is the, where you’re seeing the difference. So it says plus one, when they overbook. The yellow is when they underbook. Right? Would you want them to do is to start getting strategic, to place over books next to an under book, so they sort of offset each other, to not crowd the over books too much. So you can see there were no two overbooks that were less than an hour from each other. So when they start doing this, they will end up with a higher and higher compliance score. So it’s a bit like teaching the schedulers, how to make very judicious trade-offs on how they do this. You can look at this either as a difference, or you can look at it in absolute numbers, which is how many appointments went here.
Once you do this, then it automatically creates a learning loop, where you start to say if the scheduler is always overbooking a four hour appointment at eight o’clock, but he’s getting the rest of the day right, that is not a training issue. The scheduler knows what to do, they’re just getting bulldozed into doing this overbook all the time, perhaps we should find an optimal answer that has an extra spot at that time. OK? So that’s how you measure schedule compliance. And then you can get really fine grain on this. This is a bit of an ichart, but what’s going on is, each of these is a bucket, so there’s a zero to two hour bucket, and it’s listed out 12 Mondays in a row, then there’s the two to three hour bucket, listed out 12 Mondays in a row. And this, these are on the left of the times of the day.
If the spec is green, it means they matched it perfectly. If it’s light yellow, they were short by one. If it’s dark yellow, they were short by two. If it’s light red, they overbooked by one, if its dark red, they overbooked by two. So as you look at this, this mosaic offers you enormous amounts of learning. So you can start to for instance see that there is a consistent under booking of a three to four hour slot at 7:00 in the morning. They don’t need this slot, they’re never ever using it. All right?
Second, you can see that the six plus hour slot, most of the weeks around 9:20 is getting under utilized. So that six hour slots should be better used, because later in the day you’re seeing they keep getting pushed into making an overbooking of a six hour appointment. So that slot should probably move. And the third thing you can see is, we’re constantly overbooking this four to five hour slot in the 12:30 window. So maybe they need another slot there. So part of building this mosaic in such a fine grained way and learning from it and doing pattern matching algorithms on it, helps you create a continuously improving optimal template for the scheduler to aim for. OK?
Question three, are the financial and operational results as good as expected? On the left is how the. Tetris game unfolds today. On the right is if you had the optimal schedule, the Tetris game would be packed. So the whitespace you see on the left are losses in productivity. It’s either a loss in chairs productivity, because a chair was open, but a nurse was not available to stop the patient. Or it’s a loss in nurse productivity. Right? Where there were no chairs, and so the nurse had nothing, could not start a new patient. When you compress the Tetris blocks, the productivity goes up as measured by patient hours divided by nursing hours. This is the single metrics that tells you if you are getting better, and in a very fine grained way, not averaged over a month, but every day, every shift, or every chunk of hours in the middle of the day looking at this ratio.
Now, if you make it better, how do you capture that improvement. You can monetize it in one of three different ways, you could either see more patients, which means get more patients done in the same day, or if you’re growing by 10%, don’t add 10% more staff, add 2% more staff and you still get it done. So those are a sign that your patient hours and nursing hours are getting better. It’s important to know how you chose to monetize it. Because in the second example, if I grow by 10%, but only add 2% staff, but I test the impact my saying has my labor budget gone down, well, no the labor budget went up by 2%. Even though there was productivity, the labor budget did go up.
So it’s important to know you chose to monetize it this way. If you choose to monetize by reducing operating costs, you could either cut the hours of operation, instead of staying open until 9:00 at night, close at 8:00 or close at 7:00. Or you could, if you’re been forced into opening satellite centers, because the main center can deal with it, suddenly now the smaller centers can be used for other things, and the volume [? of ?] [? times ?] in the bigger centers. So here, you’ve chosen to monetize this gain in productivity by essentially reducing your operating costs.
Or third, you might say, I don’t care to do more volumes, I don’t care to reduce my costs, I just want my patients to have a better experience. Then there are three specific ways, the wait time will go down, access will improve, and you could afford to have nurses spend more one on one time with patients, which is a big driver of patient satisfaction as well. Wait time is a big driver of patient dissatisfaction. So these are three different monetization strategies with seven specific actions you could take, that’s kind of how you know whether both the financial and the operating results are good.
In terms of metrics, if you chose to monetize with the first bucket, then you will see infusion revenue per nursing FTE go up. The operational metric will again be the patient hours put nursing hour. If you choose to monetize with reducing the operating cost, you aggregate nursing labor dollars would have gone down, plus the other both regular and over time. The operating metrics case was the same. If you choose to monetizing and improving the patient experience, then you will see the median wait time for patients in 11 or 12 go down significantly. The percent of patients who waited less than 15 minutes improved significantly, and in a very lagging way, because patient satisfaction surveys lag a lot, just because of the sampling and all of that. You should see your patient satisfaction improved significantly.
So the way you think about this is, the two core questions are, am I complying to the schedule? And are the financial results good? Assume you start out with a note, because I don’t have an optimal schedule, so I don’t know if. I’m complying to it. And my results aren’t as good as I want, then moving to the right is establishing an optimal schedule and getting your schedulers to stick to their optimum schedule. You will then certainly get schedule compliance high. Despite that if the results aren’t still good enough, then you might need a template refresh or you might be facing some other issues, which when we get into the diagnostic part, how to I diagnosis, you could get to. It is very unlikely that with no schedule compliance, you’re having terrific financial performance.
What that usually means is the capacity is sufficient. That means if you’ve got twice as many chairs as you really need, then you can kind of schedule however you want and it just works out, because there are plenty of chairs and plenty of nurses. So usually that’s the case, but this becomes the pathway for you to figure out where you are and where you need to go. OK? So let’s move to the fourth and last question, which is, can I easily get to the root cause of why a specific day was challenging?
Having worked now with many, many infusion centers, pulling the data do a fact based diagnostic in most infusion centers we’ve seen, is actually very, very difficult. As a result, anecdote drives diagnostics. There’ll be a prevailing view that Dr. X always runs late, and because Dr. X is running late, this is what is happening. Or we just started this new clinical trials, and five patients showed up, and they had to stay in their chair very long, and that threw my schedule off. So you’ll have 50 of these reasons floating around. And it’s very hard to distill what actually made a difference, and what actually did not make a difference. And then anecdote, and recency, what happened yesterday, last week, tends to dominate what should have happened.
So what we want to propose is a different way, a true root cause diagnostic method. So let me walk you through two examples, where you start out with an innocent question, hey! The wait time was high on. August 21st, what happened? So starting with that, let’s try and click through how in five minutes you can get to a better answer. So you say, all right was this unusual for a Tuesday, because August 21st was a Tuesday. Pull up very quickly all the Tuesdays of the month, and say that’s slightly worse than most, but not dramatically worse than most Tuesdays.
Next question, were the volumes high? Quickly pull up all the Tuesdays, no, in fact, the volumes were low. So volumes was not the culprit. Then you say, if you add-on a lot, so quickly pull on the purpose of the add-ons, the greens are the cancellations, and so you can start to say, yup, the add-ons were somewhat high, a bet they contribute a little bit, but it was not the worst add-on day we’ve ever seen by any stretch. Then you stay, were the patient’s late, did that cause everything to get messed up, because patients started showing up late because the clinics were running late. And it turns out, green is on time arrivals. The darkest color is late, and go, nah! That was not really a big problem.
Then the next question becomes, that the schedulers comply? Well the compliance on that day was 80%. So it was not great, it was horrible, but it wasn’t, it wasn’t great. So that probably contributed. So now let’s try and understand in what way did they not comply. So when you look at this, they did a huge bunch of overbookings in the one to two hour range, and interestingly, simultaneously there are a whole bunch of under bookings in the zero to one hour range. So that would say, that the schedulers classification of what should be in the zero to one bucket, and what should be in the one to do bucket was potentially a little bit suspect.
Maybe it so happened that all the appointments that showed up were one- to two-hour appointments, or it happened that maybe we need to go through the classification of which regimens fit in which bucket, and rethink what happens here. OK? And then you say, all right, does that explain how the day worked out? It turns out that when I go back to the previous page, the bulk of the overbooking was in the nine to 11 window, which you can see is exactly where it started to go south. OK? So this is a way where literally clicking through five or six targeted questions and having the detailed facts in your fingertips, lets you get a much more nuanced answer than if you walked around asking people, hey! Why did Tuesday go bad. You have got eight other reasons why it went bad.
Let me give you one more example to show you similar starting point, but different diagnostic path. So here, turns out August 3rd was significantly worse than all the other days around it, right for a month on either side pretty much or a month before it, and so you say, let’s start with that, what happened. Well, it turns out, let’s look at what the wait time did by hour of day. So what was the wait time at 8:00 o’clock, 9:00 o’clock, 10:00 o’clock. When you look at it, it’s a flat wait time. This is a sign of a healthy profile. The wait times are not climbing up in the middle of the day. But on August 3rd, suddenly it jumped at 10:00 o’clock, and then things continued to snowball after that. So whatever went wrong, went very badly wrong between 9 and 10.
So now we need to understand why? Now you say, OK, what was the actual, and what was the plan. The blue was how they should have scheduled. The purple was how they actually execute it. So you can start to see they start to fall behind in a little bit, little bit, little bit, and then at 10:00 o’clock, suddenly they fall behind a lot. And because they fell behind a lot, it took a long time to recover. They didn’t recover until about 2:30 in the afternoon. So you say, OK, why did that happen? Now you go in, and start looking at add-ons. So you think it was add-ons, and they go, Yup! They do have more add-ons. Do you think that caused the problem? Well, let’s see when the add-ons came. No, the first add-on was at 11 o’clock. It had already hit the fan before them.
So the first add-on cannot be blamed for a problem that happened between 9 and 10, because the first add-on only happened at 11:00. So now you say, OK, what does the chair utilization looks like? Utilization overall, despite this, ended up being OK. So the wait times were probably just bad in small, in a few pockets, but they were bad enough that it caused the whole day to look bad. So it wasn’t a chair allocation or utilization problem. Go back to the appointment scheduling, and here again you can see they overbooked in a huge bunch right here, and essentially did not give the operation a chance to get going, because the 9:30 to 11:00 put themselves so far behind the [? eight-ball ?] they could not reasonably recover before the end of the day.
So that’s two examples of how you could start with an innocent question like why did the wait time or why did a particular day go bad? And have all of this in your fingertips, so within five or six clicks, you have a very thoughtful answer and potentially a measure you could take to counteract it going forward, but it involves taking aside a few schedulers and telling them what to do or taking aside you know, this nurse manager, about preceding or not preceding patients, et cetera. So you can come up with an immediate remedial actions very, very quickly.
So we’re right at the 40 minute mark, let me stop and leave it open for four questions. You can type questions into the Q&A box if you want. OK? Here’s a question. How do you get the data. This sounds like a heavy lift for our already slammed IT department? Good question. There is nothing manual about this. The data can get automatically pushed. EHRs have all of this data inside there, because every appointment has timestamps. So even if you don’t know many of the intermediate steps, it’s there in the EHR. It’s easy to write an automated script that pulls that data out every day and pushes it over. So there is no double entry, there is no manual entry, there is no interface needed with an EHR, which means this works with any. EHR that you have [? on. ?] OK.
What is the ROI for the solution? We’ve implemented solutions that end up returning enough to barely pay for the solution. So the ROI varies a lot, and it varies depending on many, many things like 340B pricing or not, staffing levels, volume, growth, reimbursement levels for other things, et cetera. We have consistently found that the return tends to be four to five times the size of the– the cost of the solution. And so from a financial return on investment standpoint, we have not had any difficulty with justifying it after the fact. The uncertainty is obviously at the start of launching, and so what we have chosen to do is to guarantee the product by saying if for whatever reason you felt it was not going to give you the returns you needed, even 30, 60, 90 days after going live with it, you just turn it off, and we will refund everything you’ve ever spent on that product. So that’s our way of eliminating the uncertainty at the front end. At the back end, we’re very confident of a four to five return.
Keep the questions coming, if you have any. We are, we are interested in diagnostic, what do you need to get started? I covered that, but let me quickly go back to that page again. Yeah, so this is the page I was looking at, push to, yeah. So starting from a simple data pull, this is the steps we need to get started. Operational parameters as bucket A, which is number of chairs and beds, and then the hours of operation. And if you can give us a nursing roster that would be good. And then B is the historical data dump, roughly six months of data, one row per infusion. Essentially, no patient information, no position information, no drugs information, just the date, type of infusion, expected duration, and then as many timestamps as you can muster, all this can be put in a excel spreadsheet like format and shipped to us if you wanted to take advantage of the diagnostics. OK.
MODERATOR: Yeah, I don’t, I don’t see any more questions coming in. So thank you, Mohan for presenting today’s webinar, and thanks to all of you for joining us today. Couple of quick reminders, a recording of this webinar should appear in your inbox in about 24 hours. Also if you’re viewing this webinar to receive CE credit, you need to complete a CE survey, that will be emailed to you. And looks like that’s it, so again, thank you, Mohan for hosting today. And thanks to all of you for joining us.
PRESENTER: Thank you.
MODERATOR: Thank you.