The fact that you do interventions in the middle, is very different from the typical type of query. Because everybody wants to view their databases laboratory and laboratory keep things the same. So, everything is different, is unusual. That unfortunately is the nature of medicine. Because as you know, we don't have a unique number of people who remain in practice. Because we are a teaching hospital, we have new residents every year. Nursing turnover, probably 20 percent per year. We have new nurses coming in all the time, and every single day we have new patients here. All of the different key stake holders if we think of all of the players in healthcare, bring unique experiences and have unique needs. So, the one thing that remains constant, is our ability to collect data, measure data, and see what are our opportunities to improve on. So, let's go back. So, you saw the time. You want to a full year, you want to be able to project forward the patients. I think I interrupted you the middle you see the patients are. The patient's turnover on a daily basis. Well, but your definition of patients. Oftentimes, when I look at a patient I think of the patient encounter. Because every time that a patient comes to the hospital, patients might come to the hospital multiple times individual patients, but each individual encounter is like a clean slate. Okay. Because you are looking at process. Because we're looking at process but also outcomes. Individual processes during a clinical encounter can affect the outcomes. But one can argue, if Mr. Jones comes ten times? That person's had the same risk each time and therefore it's not fair if he gets two or three events. Because he's liable for them. Well, except for the fact that the risk might be different but the risk might be the same. Let's say for argument's sake, the risk is the same for each individual encounter. Maybe they were prescribed the right thing the first time, maybe they were prescribed the wrong thing the second time, maybe during their first encounter they received all of their preventive medication as prescribed, and maybe the second time they received half of it. Those individual encounters affect the outcome that the patient might have experienced. Most of your system interventions were not at the patient level. If you are having intervention at the patient level, then you would have to focus on the patient more but this not [inaudible] see everything. Exactly, and I think that for blood clot prevention, it definitely focuses on the patient encounter level. There are many instances such as diabetes management and control that focus on the patient, the unique patient level. I think that your focus really needs to depend on what your measure is, and it needs to be appropriate for what it is. So, I was about to say is it general that quality improvement is focused on encounter and other things on the patient, but maybe you wouldn't agree with that. I think it depends on what it is that you're measuring. I really do, and every thing is unique. There isn't a recipe book to look at all quality improvement. We should write one. We should write one, but I think that it needs to be modular and dynamic enough that we can look at different things that might influence outcomes and the quality of care that patients receive. But the benefit it is that we are collecting the data that can help inform that. So, we have the patient, we have the time horizon, what else do you want to put into your query? We need to measure the outcome, we need to have a defined outcome measure. In this particular case, we decided to use administrative data. So, ICD-10 codes that identify VT events. We've used a validated set of ICD-10 codes that are published by the Agency for Healthcare Research and Quality. Then the other thing that we would want to look at, is setting. Before we get to settings though, let's go back to the outcome. What are the choices that you have? So, we could come up with a home-grown set of ICD-10 diagnoses that haven't been validated but that our health system might choose to use. We could potentially, with a lot more effort, look at patient notes. Perhaps, developing natural language processing tools to try to identify blood clots within free text reports, radiology reports. So, do you have any sense of how many clots are missing for the claims data that are in the notes? More oftentimes than not what we see, is that events get coded in administrative data that are not true VT events. So, there's actually an over-reporting. Oops. Okay. Well, that's a problem. Isn't there the issue of sometimes you can get a clot diagnosed by imaging that's not in the clinical note and not in the [inaudible]? It's entirely possible, though those are somewhat tricky to find. There are instances where a patient is taken into the operating room, and the surgeon will compress the vein and say, "Oh! That's a clot." It will go into an operative note that may or may not get into administrative data the patient may or may not have imaging thereafter. Also not all clots cause death. Exactly. You can have a small clot in the lung that's not clinically. That's actually another area where the data has been very enlightening. We've actually seen in cases where patients who are seen at hospitals where they do more CT scans. Scans of the chest, scans of the abdomen, and also where they are screening the legs for blood clots called duplex ultrasound. Where you do more of these diagnostic studies, there's actually a higher number of blood clots diagnosed. It's not that more blood clots are happening, but that on incidental finding patients without any symptoms at all are having these tiny blood clots diagnosed and getting reported into data. So, I realize we can go on and on about this. So surprised that you've been spending years on this problem, and I'm glad you've hit me over the head about how important it is and so forth. But just to wrap up, let's just review where we've been. You talked about the problem itself, you talked about your role being both the data guy but also the domain expert on this which I don't want to undervalue. Then the question that you had about instances of VTE is a little bit unfair, because you know a lot about it. It's not a generic problem falling into your lap, but you described the understanding the problem above the line and the problem understanding were they want to go and then getting more concretely into the pico components that you actually do the query for. Have I adequately summarize where we are? You have. To your point, we could talk about blood clot prevention as one area for many days to come. But I think that it's one example that shows how many different individuals are involved in preventing something from happening for measuring something, and what the value of data really is in telling us how we're doing but also informing specific areas that we can improve our practice. I think you made that very exclusively all the way through. But all this other stuff is nice, but getting to the data and how that data drives. This is really important. Yeah. Telling us what is not usually sufficient to drive change but really why, is important as well. I think that's one of the most valuable uses of data out there. Oh! Great. Thank you very much Ben. Thank you. You're welcome.