[MUSIC] In our last section, we looked at a kind of a formal way to talk about whether you should put your intervention or your decision support into place. Now we're going to look at it a little bit in more qualitative pieces, if you will. [COUGH] If you're going to go ahead, what is it that you need to be thinking about beyond those variables that I mentioned in that decision tree? Well here's the list of, these are the minimum things you need to be thinking about before you do a decision support intervention. I separate them into kind of above the line and below the line concerns. So first, you have to make sure that you have somebody who owns the problem. Your IT people should never be in charge of decision support. It'll just alienate the clinicians and they'll be upset. You need a clinician or public health or population health, somebody who's in the domain area who they will take ownership of the problem. Not necessarily the technology, but the problem. And they're the ones who the IT people are going to go to to ask about whether they're getting the domain issues correct. The champion is a little bit different, the champion is a person who kind of convinces the other folks to go along with decisions support. So the owner could be the department chair, but the champion could be a venerated official, doesn't have to be the same person. And buried in this thinking about the owner, the champion, and all that, is the notion of governance of the process of putting decision support into place. Now you also have to think about who your user, remember we spent a lot of time on who's the user and what decision do they have, so you have to have a user, then you have to be thinking about what decisions do they have? So what population are we talking about? What's the objective for that population? So I might want to reduce the number of allergic medication bad outcomes. So the outcome though is the orders, right, so that, because the allergy orders lead to the bad outcome. So if I want to reduce bad allergy outcomes, the objective, I need to reduce the number of drug allergy drug orders. So that leads to the behavior that I want to support, that leads to that outcome. So to reduce the number of drug allergy orders, I need to get to positions to change their behavior, their ordering behavior. And finally, again, above the line I need a way to evaluate whether whatever it is I'm putting into place, do they address these issues that I just laid out, the objective, the outcome, and the behavior. Below the line, now, we have a work flow, [COUGH] if we talk about drug allergies where, which, who's work flow are we talking about? Is it the pharmacist, is it the ordering clinicians, is it the nurse who's giving the medication? So we need to see which exact step in the work flow are we going to be targeting. You have to find the right place and right time in that work flow to target. And then there's this notion of information system context. If the computer is going to be interacting with the user, which computer is it, either will it be. And then we need to think about the actual final intervention. Why does the user actually looking like, what is the user actually looking at? What is that interface that's often called the widget? So if you look at the Bili Light decision, the owner is whoever's in charge of the care of the newborns, it could be the neonatal service, it could be the general pediatric service, it could be the department chair. Who's the champion, it could be any one of those people, again there has to be governance, which is the Bilirubin problem going to be the one that we're going to tackle or is it going to be a drug allergy order or is it going to be a wait-based ordering. Which problem are we going to tackle? So you need the champion, and then the identified user would be again, the intern or whatever else is going to be making a decision about newborn babies who are yellow. Continuing the target population's clearly, healthy newborns, remembers that we said the graph does not apply to pre-term babies, or babies with other risk factors for kernicterus. The objective is to minimize brain damage, the outcome is proportion of babies getting the appropriate treatment, preferably the lights rather than the exchange transfusion, and the behavior is ordering the correct treatment. And finally, evaluation method. You could use a dashboard to see what's going on in real time, or you could use some sort of ongoing report. But wait, there's more. So what's the work flow? The work flow is, well, interpreting the lab results, right? The bili comes back from the lab, the doctor's gotta figure out what to do. Do you want to wait a few hours until they've consult the laboratory system, or should they be notified right off the bat? Clearly a little triage decision right there, don't you think. So the information system context is probably the EHR, because that's the mechanism through which most clinicians see lab values. And the targeted behavior is, again, following the machine's recommendation. There's still more, all right? So the intervention is using that nomogram. The interface is the graph. And the widget is both the superposition of the babies, but Bilirubin levels on the graph, plus a button to take action. So there are many different types of widgets. I'm not going to go through the whole list, you can read them yourself. On the right hand side, I'm showing you where some of the examples we've been playing with live. At the very top, are basically displays of information, general static information that either from some textbook or an article. The machine might find the article for you but the article is the same no matter who's reading it. In terms of keeping professional skills up to date, there are videos. I've divided the other decision support interventions and widgets, which are all kind of patient responsive. The owner say patient specific but certainly response to the patient's data. They divide them to implicit knowledge and explicit knowledge. So explicit knowledge is kind of straight forward. In the case of the drug alert, the machine puffs up and says you're about to kill the kid, that's a bad thing to do, don't do it, here's some options. The case of the public health, it's a little bit more subtle. You may remember in the whole graphic, there was a section of the poster, or the report, that had the vertical bars and then there were these threshold levels. Those thresholds are express knowledge, right? Above the level is bad, below that level is good. In the middle you have sources of implicit knowledge, so if you look at the cardiogram display, nowhere does it say that if the cardiac display is one thing, you should do this, or should do that. But the fact that it's saying that, if you want to make a decision about how to treat chest pain, these are the data to look at. That's a lot of knowledge about chest pain. It's just that it's not explicit knowledge. If that knowledge were to change about which information we needed, so let's say echos were no longer required. Somebody would have to go in and go into the program and boldly remove that code that talked about the echo. In the case of the drug alert, there might be a line of knowledge that needs to be changed, whether a drug causes or doesn't causes type of allergic reaction. So the way you modify the widget depending on new knowledge is also divided between implicit and explicit knowledge. These we'll review them later on, but I just want to get you hearing this language now. No surprise, there's a framework to put this all together. In fact, the verbal pieces I've been telling you about come from this. So this is a book put out by HIMSS in the 2000s. I went through that book and kind of boiled down about 50 pages of text into one picture. You're very welcome. And I hope by showing you the picture, and showing you the relationships, it becomes a little bit easier to think about. Now why have I put this diagram into a talk called you should part two? The reason is that if you can't articulate these pieces of your problem, you don't understand your problem. In the flip side, if somebody comes to you and does articulate all these pieces you have a pretty good chance that they'll understand what they're talking about and you have considered what they're doing. They still don't talk on the consequences, but by and large this will be great heuristic for thinking about which of the pieces of decisions support, or to be honest any intervention you want to consider. So we're going to look, if you look at the public health dashboard, which is not a typical, a type of thing to call decision support. The owner or chief stakeholder would be the public health officer but the champion might be the epidemiologist. The user epidemiologist or public health officer, maybe somebody else might support the office. The information system, the environment. This case, the CIS application, there really aren't any for public health. It's a sad story, right? The life of a public health practitioner when it comes to informatics and IT is a whole bunch of different modules that they just have on their desktop and they have to figure out how to use, so I apologize for that. But within the notion of surveillance for health problems on an annual basis, this would be one of those modules, that this is the diabetes module. The interface, the widgets are maps and graphs and in point in fact there are similar maps and graphs for other problems. So this is a nice example with the same interface is used for multiple problems of the same sort. Which means that the designed action and outcomes are similar to the desired actions that say, well, I'm going to focus on this. The outcome is actually focusing on, let's say, that area of the Bronx, and the objective is lowering diabetes morbidity in New York City. How you evaluate all this. You can measure each one of those items I just talked about. What work flow are we talking about? Well this is an annual problem, so probably this will sort of budget discussion going on and the work flow step here is the consideration of greater diabetes dollars. As opposed to let's say the [INAUDIBLE] dollars with something, some other program. If there's a contention across programs for funding, I would want to be advocating for the problem that the public health officer cares about. In many ways, that should not have been a painful process, and you should be able to take any project either that you are working on, or that you've been part of, or that people are saying that you should use. And you should go through this little exercise to see whether or not you can fill in all the little pieces.
