In the previous module, we introduced the concept of lean. We drove the waste out of the system, balanced the process, and made sure that each resource was working nicely at high levels of utilization. That felt really good. But, to speak like a doctor, are there any side effects we should be aware of? The purpose of this session, is to show you that things in actual operations are a little bit more complicated compared to what we saw before. Specifically we will see that, when we introduce variability into a process, a very real element of real operations, forces can perform much worse than what is expected by a nerife analysis, ignoring that variability. Join me in a look at an endoscopy suite. This is a suite of Doctor Toyota, it's not a real example it is just illustrative. In Doctor Toyota's endoscopy suite, patients arrive for preparation and anesthesia nicely every 30 minutes. Moreover, Doctor Toyota can complete his procedures in exactly 24 minutes. He works like a robot, never a minute more and never a minute less. What does the utilization of Doctor Toyota? Remember the definition of utilization, flow rate divided by capacity. The flow rate is 1/30 and the capacity is 1/24. So our utilization is 2040 divided by 30 which is 80 percent. Let's take a look of how this would play out. I have a simulation here, run on fast, take a look. Now you're going to get the drill, low grade patients come, look how they're going to get served and most importantly, look how long they have to wait. Let's take a look at the simulation now. So what was a typical waiting time for Doctor Toyota's patient? You did not see any waiting? That's right. You did not see any waiting, as there was no waiting. You see this in this gun chart here. None of Doctor Toyota's patients ever had to wait. Doctor Toyota start the procedure, 24 minutes later he's done with the procedure. He's idle then for six minutes, and then the next patient gets wheeled in. A dream for any operations manager. You say that this is not realistic? You're right. And that's why it's just an illustrative example. But, what is realistic? If you look at your own data, chances are that the time between two patients arriving will not always be exactly 30 minutes. Some patients come late, others are early, so sometimes you have 10 minutes between two arrivals, but sometimes that number is 50 minutes. Now, say the time is on average, 30 minutes. So, we still have 12 patients arrive over the course of a six hour window. However this time, we had some variability in the arrival of patients. Similarly, Doctor Toyota is just a fictitious character. Real doctors are not robots. Not to mention that their patients were also vary in their medical conditions. So it's just unrealistic to believe that you can do an endoscopy in exactly 24 minutes for each patient. You might have an average procedure time of 24 minutes, but sometimes you're a little faster and sometimes you're a little slower. So we have not changed our average demand rate, we have still 12 patients arrive over a six hour window, and we have not changed our average capacity, it still takes us on average 24 minutes per patient. Consequently, our utilization is still unchanged at 80 percent, all we did is we introduced a little bit of variability into the process, or to put it differently, we made the example more realistic. Now look what will happen to the patient flow. Again, I want you to have your eyes on patient waiting times. Previously, none of our 12 patients had to wait. Now we see some patients actually have to wait a fair bit. Again looking at the gun chart is helpful. Look at patient 5 for example, she spends almost three times as much waiting than being in the procedure. That is not just unpleasant for the patient but depending on her anaesthesia, could be really problematic from a medical perspective. We also notice a significant space problem. On the bottom of the slide, you see the number of patients in or right before the endoscopy suite. Halfway through the day, we we're running short in space. We just don't seem to be processing the patients fast enough, but we should not jump to the conclusions here too quickly. This is not a story of insufficient capacity, after all, our utilization is only 80 percent. Yes. Temporarily, we have periods of too little capacity, but we also had periods of idle time. We just cannot seem to match supply with demand. And there is a reason for that, and that reason is variability. A utilization of 80 percent is fine in a deterministic world, but in a world of variability it is problematic. In an endoscopy suit, we pay the price for the variability in the form of inventory build up, or maybe more accurabely, our patients are paying the price for the variability, patients get buffered. Since they already have an I.V. inserted and maybe even have received the first medication, they will not walk away from the procedure. Our flow rate is exactly the same as our demand rate, every patient that arrives to the endoscopy suite will leave it as a treated patient. This is not true for all health care services. Let's look at an example of the emergency room. The process flow diagram is at the very highest level, exactly the same as in the endoscopy suite. Patients arrive, they wait, and eventually, they will be admitted to the emergency room. What happens when the emergency department gets busy? Just like in the endoscopy suite, the waiting room fills up. Again patients get buffered, we are holding patients in inventory. But, two additional things happen in the emergency department. First, patients get tired of waiting, and they simply walk out of the emergency department. In the emergency medicine literature, this is known as a left without being seen or LWBS for short. In operations management, we say that the customer abandons the queue. We did some studies on this and found that the left without being seen rate ranges between one point five percent and nine percent depending on how sick patients are. The sickest are the least likely to quit, which makes sense as they need the care the most and will also experience the shortest waiting times. This is bad for the hospital, and it can mean financially a foregone revenue. Second, when things get really busy, the emergency department has to go on diversion. Practically, this means that the inflow of ambulance arrivals has slowed down. Ambulance diversion has been demonstrated to be a major problem in the US, and this is associated with significant risk to the diversion patients. In some regions, it's also remarkably common. In a 2011 study published in JAMA, the research team measured the diversion hours of 4 California hospitals. Take a look. Diversion in these hospitals is not some rare event. It is a common occurrence happening several hours each day. As this study showed, the patients in the ambulance, the study looked at cardiac problems, had a significant higher mortality than the patients that were not exposed to diversion. In some, we see that the hospital is affected by variability either in the form of a buffer inventory, or by suffering a loss in patient volume. I like to call this phenomena buffer or suffer. You will always be worse off in the presence of variability. You will either have to create some buffers, or you will suffer a loss in the flow rate. To better understand the buffer or suffer effect, consider the following like super stylized example. A practice is facing a demand of either one or two patients in a given period of time, chances are 50/ 50. Moreover, the capacity of the doctor is also somewhat uncertain and can either be one or two patients in the same time period, again, 50/50 odds. Statistics tells us that the expected demand is, a one point five patients per unit of time, and the expected capacity is also one point five patients. Thus, we have a beautifully designed process, was 100 percent capacity utilization. It looks like, on average, we have enough capacity. But working with averages is a dangerous thing. Let's look at this example in more detail. There really are four scenarios, demand can be one or two, and capacity can be one or two. Now recall the definition of flow rate. Flow rate is a minimum of demand and capacity. So in the first case, a minimum of 1 and 1 is 1. In the second case, the minimum of 1 and 2 is 1. In the third case, a minimum of two and one is one. Only in the last case do we actually serve two patients. The average flow rate is not one point five, but one point two five. We almost lost 20 percent of our capacity to the gods of variability. Why is that? Short of having a buffer, in order to serve a patient you need demand, i.e. the patient, and capacity, i.e. the provider, at the same moment of time. If you have excess demand, it gets lost. If you have excess capacity, it sits there idle. You lose either way. If we could buffer the amount, we could simply take a patient from that buffer if demand is less than capacity, and we could put a patient into that buffer when demand exceeds capacity. This is what I mean with buffer or suffer. Buffer or suffer means that in a world of variability, I will either have to buffer my demand and inventories, while I will have to accept idle time at my resources. This is why healthcare systems, like big waiting rooms and full appointment books. In this first, less not variability. We saw that having enough capacity on average does not do the job. Unless we are able and willing to buffer demand, we will suffer a loss in patient volume. So be very careful when you see reports that use aggregate high level patient flow data. A little bit of idle time is actually not a bad thing. So it protects you from the evil spells of variability. So we really observe that lean and variability they just don't like each other. That means that you either have to get rid of variability, or you have to say goodbye to some of the idea of the hospital being lean.