Understanding the Time Patients Spend in Your Infusion Center Webinar Transcript

CONOR O’DONOGHUE: All right. Hi, everybody, and thanks again for joining today. Just to introduce myself, my name is Conor O’Donoghue. I’m a product manager with the iQueue for infusion centers team. And I work with several leading cancer centers on infusion scheduling and predictive analytics. What we want to discuss today before we dive in is the topic of cycle times. So first, we’ll quickly talk about what that means and how we measure them. Then we’ll frame the problem and elaborate on why cycle times actually matter for your center. After that, we’ll look at the different categories of issues we see with cycle times, and then we’ll jump into a more detailed review of the different analyses we can perform to get to the root cause of those issues. Finally, then we’ll quickly look at how machine learning algorithms can be employed to anticipate future cycle times. OK, so what are cycle times and how does iQueue measure them? The cycle times, to us, is the total time that the patient spends in the infusion chair. This time period is a great reflection of when a center’s resources are being fully utilized. 

So although every infusion center is different, a typical high-level flow of what we see looks something like this. So the patient checks in. Sometime later, the patient is seated in the chair. After that, the first regimen of the infusion is administered. Then sometime later, the last regimen of the infusion is completed. And then once that’s complete, the patient checks out. Now the cycle time is the time between the patient being seated in the chair and the last [INAUDIBLE] medication being completed. So usually, we use the chair time and the last meds stop time timestamps to calculate this. Although given the variability and how each center captures their data, this can change from center to center. But during that particular period, the three main resources of the infusion center are occupied. So the chair is in use from the moment the patient sits in it to the point they complete their infusion treatment. The nurse is tending to the patient. 

Now for many centers, we appreciate that the nurse is also engaged with the patient prior to them being seated and even after the infusion is complete. But this varies from center to center. But one thing is pretty certain, that the nurse is always engaged for a period of time at the start of the treatment and then another period of time at the end of treatment. And then the final resource is pharmacy. So the pharmacy is engaged and prepping the drugs. So again, sometimes this can be started prior to the patient being seated but almost always continues after they’re in chair. Understanding this particular time period is crucial to getting a clear picture of when your resources will be occupied throughout the day. So again, cycle time is the duration of time between patient being seated in the chair and last regimen being completed for that treatment. So why do cycle times matter? Or why should we care about this particular duration? So having a strong understanding of your center’s cycle times can really help in ensuring the availability of your three main resources over the course of the day. If we had a detailed grasp of how long we can expect our various types of appointments to run, we can better plan the day to level out and distribute patients across the centers full day. But there are times when the actual cycle times can vary from what’s expected. 

This is quite common as every center has some variability in their cycle times due to money circumstances such as things like reactions, or difficult access, or long delays even in the drug prep time. But if it consistently happens, then we’ll call it issues for your center. So cycle times can vary in two ways. They can either run longer than expected, or they can run shorter than expected. When cycle times run long, there can be unanticipated bottlenecks due to chair availability. I think it’s fair to say we’ve all seen situations where one appointment, running longer than scheduled, can result in the unavailability of a chair and ultimately, a patient waiting in the waiting room longer than they should be. 

If even if this happens in just a few occasions in a day, it can cause major problems for your center, resulting in dissatisfied patients and ultimately, frustrated nurses. Now what about– what if cycle times run shorter than expected? So this is not as common a complaint as the previous case, but it still happens and can lead to operational inefficiencies. So when this occurs, the center is going to find itself underutilizing its resources, which is a major inefficiency. Let’s say we have four appointments that all run 30 minutes shorter than expected at any given day. That’s two hours of idle chair time that could have been used to treat more patients. So increasing patient access is often a major focus of infusion centers, so we really want to avoid this scenario as best as we can. 

OK, so let’s look at these two scenarios in a little more detail. So let’s first take the case of where the cycle times run longer than expected, which is what’s reflected by this chart. So just briefly to explain the chart, what we’re seeing here is a 31-chair center’s chair utilization graph for a single day. So on the x-axis here, you’ve got time of day in 10 minute increments. And then on the y-axis, you’ve got a total number of chairs in use at any of those given times. Now there are two lines on this chart. So first, the blue line shows what the center actually scheduled that day. So based on the appointments that are scheduled, can you expect the durations of those appointments? This is how many chairs were scheduled to be in use throughout the day. Then the pink line shows the actual utilization. So based on how many patients arrived, how long they actually spent in the chair– this is how many chairs were in use at every point throughout the day. Then the gray line across the top is the chair capacity. So in this case, that’s 31 chairs for this center. So on this particular day, this center had 39% of its appointments running longer than expected. We can see that in the morning hours from about 7:00 AM to about 11:00 AM, things look pretty good. And by pretty good, what. I mean is the blue planned appointments line and the pink actual appointments line are running pretty close to each other. 

However, look at about from 11:00 onwards, we can clearly see that the appointments did not end when they were expected to. And the patients remained in the chair for longer than was scheduled. At 12:40 they almost ran out of chairs due to cycle times running long. Luckily, this particular center is scheduled much lower than the full capacity on this day. And we can see that by about 3:00 PM though, there are still 25 patients in the chair even though they only planned to have about 13. So this is a frustrating effect of underestimating your cycle times. And this impacts both the patients and the nurses. And this type of scenario can ultimately result in appointments running past close and nurses having to stay on way past their shift time. All right. So let’s look at the opposite problem then. The inverse of this is the cycle times running shorter than expected. So again, just to quickly explain the graph, time of day is on the x-axis, chair count on the y-axis. The blue line reflects the scheduled appointments, and the pink line reflects the actual appointments. So on this particular day, 20% of the appointments ran shorter than expected. 

So again, we can see that from the hours of about 7:00 till noon, the appointments– the actual appointments are matching up pretty well with what was scheduled. But then we see things shift, and it’s clear the appointments start to finish up much earlier than expected. As the afternoon progresses, we see all of this blue space, where they had patients scheduled to actually be in the chairs, but they’re not there. So the center here has overestimated the durations of their appointments, and now they have a bunch of unused resources that they could have fit in more appointments into. So this is problematic from a few different perspectives. One, there might well have been patients, who actually requested those times, that were turned down as the center believed them to be already booked. Two, there would have been nurses idle at these times– that those same nurses probably experienced way busier times on other days of the week. So this is quite frustrating as nurses might prefer a more evenly-balanced workload throughout the week. 

And finally, this is actually just an inefficient use of resources and that more could be done with these resources from both an operational and a financial perspective. All right. So assuming our example center here is experiencing some of these troubles with their cycle times either running long or short, how do we diagnose this problem? There can be many reasons as to why cycle times are not matching the expected durations. Here, we split this out into four main groups. So one, on the top left, we’ve got unexpected reasons they might run along. Two, on the top right, we’ve got systematic reasons they might run long. Three, on the bottom left, unexpected reasons they might run short. And then four, on the bottom right, systematic reasons they might run short. So let’s take just a quick look in more detail of each of these. 

So on the top left, unexpected reasons or appointments may run long– this would include things like reactions, difficult access, long turnaround times due to in-chair wait for medications, and even the patient needing an add-on blood or a hydration. Any of these unplanned-for events can have a significant impact on your cycle times. Now what about the systematic reasons your cycle times might run long? This could intrude the underestimation of prep time. So are you accounting for all of these things like intake, access, assessment, drug prep time? The other thing that might cause them to run long systematically is, are the treatments actually being categorized correctly? We might be greatly underestimating the lengths some of these treatments actually take. Now what about the unexpected reasons they might run short? So on the bottom left here– again, severe reactions could actually render the patient too sick to continue the remainder of a multi-regimen treatment. Also, the treatment plan could change, maybe, after a clinic appointment. And then on the bottom right, we’ve got systematic reasons they might run short. These are largely the same and sort of the opposite of the systematic reasons they might run long. So in this case, prep time would be incorrectly estimated, just it would be overestimated rather than underestimated. And then the regimens expected duration might have been categorized incorrectly again. So this time, they might be overestimating the length of their regimen. OK. So we’ve seen the different categories of reasons.

Now we need to take these and figure out a way to get to the root of the problem as any of these phenomena can greatly impact your cycle times. And ultimately, this will create operational issues for the center. OK, so let’s take a look at some deeper root cause analysis. So firstly, to diagnose this in detail and get to the cause of the issues being experienced, we need to do some detailed data analysis of your cycle times compared to various factors that might be influencing them. Typically, the first step is to review your appointments for a given time period and categorize all of your appointments into three main categories. So what we do is we categorize every appointment that was completed into one, ran long, two, ran short, or three, ran within range. Now to do this, you compare the expected duration against the actual cycle time. And we define a time window on either side of that expected duration that we consider acceptably within range. So appointments that run lower than that range are categorized as ran short. And appointments that run longer than that range are categorized as ran long. So now we have a set of appointments for a given time period that we want to analyze, categorized into ran long, ran within range, and ran short. Then when we have that set ready, we want to take a look at these appointments against a variety of different factors. So some samples that we might look into are things like treatment length, time of day, day of week, regimen provider. And these are the examples that we’ve chosen to look at today, but the scope for this analysis is much greater. And further insight can be found from looking at many different variables, specifically the patient profile variables would also be quite interesting. But we’ll start with these examples for now to illustrate how to look at patterns in your data and drill down into the root cause of your cycle time variability. All right. So first up is the appointment duration. So iQueue typically defines all your appointment types into duration groups. So based on the volume and mix, this can vary from center to center. 

But for the example we see here on the right hand side, the center has 6 expected duration groups. So we’ve got a zero to one hour, a one to two hour, a two to three hour, three to four hour, four to five hour, and then greater than five hours. Now if we check each of these groups for any high percentages of ran long appointments, we might uncover a systematic problem with how we’re categorizing those appointments. So these graphs show for each appointment group the actual duration on the x-axis and the number of appointments that lasted that long on the y-axis for a given time period. In this case, I think we’re looking at about three weeks of data for this center. So the light blue color reflects appointments running short. The medium blue bars reflect appointments running within range, and then that dark blue bars represent appointments that ran long. 

So we quickly scan across the groups. Notice the two-hour group in the top-middle graph. We can see that 54% of the scheduled two-hour appointments in this time period ran long. So that means that the majority of the appointments the center is scheduling for two hours are actually running longer than that. Yeah, so with an insight like this, the center can re-examine all of their appointment types that they schedule for two hours, and see, maybe, if there’s a subset of two-hour appointments that consistently run long, and perhaps consider scheduling some of them for three hours instead. Otherwise, they will just continue to underestimate the actual cycle times for these appointments. And that will ultimately lead to bottlenecks in resource availability, which is going to frustrate both your patients and your nurses. OK. The next analysis we would want to look at then is cycle times by time of day. 

So here is a graph on the right-hand side– has time of day on the x-axis and the number of appointments on the y-axis. So for each time of day across the centers operating hours, we’re basically showing the total count of appointments scheduled at that time. I think this is for over about a four week period. Again, the different blue colors denote whether they ran short. So short is light blue, within range is that medium blue, and then ran long is that dark blue. So notice the first 7:00 AM time slot, the very first bar on that graph. There is high variability here in how these actual durations performed against their expected durations. So quite a lot ran short. And also, a significant amount ran long. And then what about this 9:00 AM time slot here? So about 35% of appointments that are scheduled at this time slot are running long. So in general, the morning schedule appointments tend to perform worse for this center in terms of actual cycle times than the afternoon or evenings. 

And this is quite an important insight. So does the center have specific guidelines around scheduling particular appointments or types of patients in those morning hours? And if so, are they scheduling that group of appointments that are [INAUDIBLE] expect the duration? So a simple analysis like this can basically help us– point us in the right direction of where that problem truly lies. Next analysis that we want to look at is cycle times by day of week. So this chart here shows all of the appointments that ran long, ran within range, and ran short, broken down by day of week. Again, this represents about a four-week period. So the red represents ran long. The green is within range, and the blue is ran short. So this analysis can help us pinpoint any specific problem there is when it comes to cycle times. So we can clearly see that on Tuesdays and Wednesdays, over 50% of appointments ran long, while on Fridays, actually, over 58% of appointments ran short. So variability by day of week like this could be due to something like a particular regimen being consistently scheduled for that weekday or even perhaps the center might only do clinical trials on specific weekdays, which can be associated with greater complexity or greater variability in the cycle times. But the patterns by weekday can help us further drill down into which appointments are consistently running long on. Tuesdays and Wednesdays. Is there a particular provider who always works on those days? Are a lot of patients who have add-ons or need fluids because of treatments that are given on those days? 

Overall, having this detailed understanding of your cycle times by their week and which days are the problem days, both for running long and for running short– this can help us make decisions when we re-examine the best scheduling strategy across all the days of the week, which leads us on nicely to our next analysis, which is cycle times by regimen. So this particular analysis is one that often uncovers systemic issues with how appointments are assigned their expected duration. So what we’re seeing here on the right-hand side is the nine most common regimens for a particular center. So each tile represents one regimen. So across the top, we’ve got abraxane, rituximab, carboplatin. In the middle, we’ve got cisplatin, FOLFIRI, orencia. And then on the bottom, we’ve got paclitaxel, [INAUDIBLE],, and [INAUDIBLE]. So the area of the tiles represents the total number of appointments over a four-week period, and then that’s broken down by red for ran long, green for ran within range, and blue for ran short. So you can see the top three look pretty good, as in the tile is made up mostly of green space, meaning that most of these appointments ran within range of their expected duration. However, take a look at those middle two, so cisplatins and FOLFIRIs. So in both of these cases, over 50% of these appointments ran longer than expected. And then look at the bottom left, so what about that paclitaxel? So over 45% of those appointments actually ran shorter than expected. Most of the time, we see infusion centers of some form of a regimen scheduling cheat sheet, essentially like a guide for schedulers to understand how long to schedule a particular appointment for based on the regimen. This type of analysis can quickly identify where that cheat sheet might need updating or adjusting. So it’s clear here that they should be booking more time for the cisplatins and FOLFIRIs than they are. And they can also probably afford to book slightly less time for their paclitaxels. 

Another way of analyzing your cycle time is by provider. So this one will help us to quickly identify certain providers and if they’re not taking into account things like prep time when they note the times in their orders. So this particular graph shows us the total appointments for a four-week period, broken down by provider, which is indicated across the top. And then underneath, we can see that the appointments are grouped into ran long in red, again, ran within range in green, and ran short in blue. The y-axis then shows the percentage for each group of the total appointments for that provider. So here we can easily see straight away Dr. Murray and Dr. Yan’s appointments are consistently running longer than expected, over 50% in both cases. And then what about Dr.. Harris here on the end? So those appointments are consistently running shorter than expected. Now might it be worth examining how these providers are indicating duration on their orders? Now, we appreciate that not all centers have a process by which the ordering provider is indicating the expected time, but some certainly do. And for those that do, each provider could have a very different method of arriving at that time indication. And the infusion center really needs to establish a way to standardize this or else it’s going to lead to these kind of problems. 

So we’ve seen a multitude of ways just now of how we can look at cycle times in order to get an in-depth picture of what some of the root causes of our cycle time variability can be. But the variety of factors shown here involves just tells us that looking at this through visual inspection alone, factor by factor, might not actually be enough to uncover the true hidden trends in the data that are driving the variability in your cycle time. So here’s where machine learning can come in. So basically, we can leverage machine learning to better predict your center’s cycle times by taking into account all of the factors we’ve just discussed and more. So we start by getting the data and running it through a machine learning algorithm. So we can input provider, time of day, regimen, day of week, and even your own structured notes. So often there is valuable information within that unstructured notes field, which is that a free-text field that we see a lot of centers have in their EHR. So we have even prototyped a script here that can look at that free-text field and pull out the important information about the employment while ignoring any of the noise that might be in there. 

So these five items here, again, are just a sample of what can be used to anticipate cycle times. But basically, as you can see, the data goes in, the machine learning does its magic, and out pops the expected duration. So we believe leveraging machine learning in this way could potentially unlock a much more accurate strategy for defining your expected durations, which will then, in turn, better predict your future utilization. And ultimately, it’ll help the center plan their days more effectively. So that brings us up to the end of today’s short webinar. So I just wanted to summarize what we discussed and the overall message of today’s talk. So number one, cycle times are important. So firstly, it’s important to understand them, and it’s also important to calculate them correctly. Once we have a deep grasp on our center’s cycle times, only then can we understand what it takes to fully utilize our resources to avoid reserving more space than needed or using more resources than anticipated. Number two, there are many variables. So today, we went through just a small few factors that can have an impact on your cycle times.

 But there are many, many more. And lots of them are hidden somewhere in your data, just waiting to be analyzed. And then finally, machine learning can help. So new innovations in machine learning approaches can really be used on this particular problem, especially given the multitude of variables involved. So that takes us to the end of our presentation. Thanks again, everybody, for listening. And I hope you got something out of it. And yeah, I’d like to open it up now to any questions that there might be coming in. 

MODERATOR: So if you have any questions you’d like to ask Conor, you can click on the Q&A button at the bottom of your console, and we’re more than happy to answer them as they come in. I see that we already have a couple. Oh, let’s see here. How are wait times related to cycle times? My center has identified long in-chair wait times as a pain point for patients. How will this impact cycle times? 

CONOR O’DONOGHUE: Right. Right, great question. So yeah, this will ultimately increase cycle times, and it could be an indicator that maybe your pharmacy is overloaded by the current patient flow. One possible way to address this would be to optimize the appointments starts on the template. If they are spread out more level-loaded across the day, your pharmacy will be able to turn around the drug faster, reducing the total time the patient is in the chair waiting, essentially. But yeah, great question. 

MODERATOR: iQueue is working with so many different cancer centers. Is it possible to know how well my center is doing compared to other centers? What amount of variability is considered normal? 

CONOR O’DONOGHUE: OK. So the answer to that one really depends on a lot of factors for each individual center. So the centers are so different– things like workflow, timestamps, patient population, types of treatments that you’re administering. So we would say there are two good rules of thumb here. So first, compare the utilization of what’s run long versus what’s run short and then determine how uneven that summation is. Is that causing bottlenecks? Or are you perpetually underutilizing your resources because of this? And then secondly, again, while each center is different and has their unique workflows and ways of even capturing the cycle time, we do find that on average, the percentage of patients running long is no more than 25%. So that’s like 1 in 4 patients. If you feel this is low for your center, it could be worth digging into a root cause analysis on this. But yeah, great question. Thanks.

MODERATOR: Yeah, it looks like those are all the questions that have come in. So thank you, Conor, for presenting today’s webinar. You did a great job. Thanks again for joining us.